Datasets:

suolyer

/

pile_pubmed-central

like 1

Introduction {#Sec1} ============ Tumor formation is a multistep process that involves the acquisition of oncogenic traits and is opposed by diverse tumor suppressor mechanisms. It is well established that cellular senescence is one of such tumor suppressor mechanisms. Senescence is an antiproliferative response that controls cell balance in a variety of physiological and pathological settings, halting proliferation and triggering clearance of damaged cells^[@CR1]--[@CR3]^. In the context of cancer, senescence acts as an effective tumor suppressor barrier, blocking the expansion of potentially oncogenic cells in premalignant lesions^[@CR4]^. We have recently shown that SIX1, a member of the SIX family of homeobox transcriptional regulators, is a negative regulator of senescence, which controls the expression of key senescence regulators such as the cell cycle inhibitor p16INK4A^[@CR5]^. Work in Drosophila and vertebrate animal models has established that SIX proteins, and their cofactors of the EYA family, play a critical role during organogenesis, most notably in muscle, kidney and diverse neurosensorial structures^[@CR6]^. In humans, alterations in SIX or EYA proteins are linked to the Branchio-Oto-Renal (BOR) syndrome, a developmental disease characterized by renal and otic defects^[@CR7]^. In addition to its physiological role in organogenesis, it has also been shown that SIX1, and other SIX proteins, act as oncogenes in a variety of tumor types, including lung, breast, brain and colorectal tumors. SIX1 is frequently overexpressed in these tumors and it has been associated to several traits critical for tumor formation and progression, such as proliferation, angiogenesis, invasion and cancer stem cell function^[@CR8],[@CR9]^. Of note, studies on SIX1 in cancer so far have focused mostly on carcinomas, and thus the knowledge about the role of Six1 in tumors of non-epithelial origin is much more limited^[@CR10],[@CR11]^. Considering the role of senescence as a tumor protective barrier and the link of SIX1 to senescence in fibroblasts, we set here to investigate the role of SIX1 in fibroblast transformation and tumorigenesis, in connection with cellular senescence. To this end, we used a cellular model of oncogenic transformation and tumorigenesis based on mouse primary fibroblasts. The analysis of tumors with SIX1 overexpression indicate that the oncogenic effect of SIX1 is associated with the repression of a senescent gene signature and the induction of a dedifferentiated tumor phenotype mediated, at least in part, by the stemness regulator Sox2. Further studies with human glioma cells have confirmed these observations and clearly support the link of the pro-tumorigenic effect of SIX1 with senescence escape and SOX2-mediated self-renewal. Results {#Sec2} ======= SIX1 promotes fibroblast tumorigenesis {#Sec3} -------------------------------------- To investigate the impact of gain of function of SIX1 in immortalization and oncogenic transformation in a genetically defined model, we have used primary Mouse Embryo Fibroblasts (MEF). These cells represent a well-established cellular model for these studies, as they can be immortalized and transformed with a small number of well-defined genetic alterations^[@CR12]^. SIX1 was ectopically expressed in early passage wild-type MEF with or without expression of an shRNA against p53, using retroviral transduction. As expected, p53 knockdown was sufficient to immortalize early passage MEF. Increased SIX1 levels did not alter significantly the colony formation ability of shp53 MEF, and neither was it sufficient to allow efficient immortalization of wild-type MEF in the absence of shp53 (Data not shown). Next, immortalized fibroblasts with or without ectopic SIX1 were retrovirally infected with the activated form of the Ha-Ras oncogene, RasV12. (For simplicity, shp53/RasV12 cells are hereafter designated V/RAS, while shp53/SIX1/RasV12 cells are named SIX1/RAS, Supplementary Fig. [S1](#MOESM1){ref-type="media"}). The impact of SIX1 gain of function on transformation in this model was first investigated in anchorage-independent growth assays using soft agar, which showed that SIX1/RAS cells were able to form significantly higher number of colonies than controls without SIX1 overexpression (Fig. [1a](#Fig1){ref-type="fig"}). Of note, SIX1 ectopic expression alone was not sufficient to confer anchorage independent growth in these assays (Data not shown). To evaluate the effects of SIX1 overexpression in tumorigenicity *in vivo*, V/RAS and SIX1/RAS fibroblasts, together with controls lacking RasV12 expression, were injected subcutaneously in immunodeficient mice. Both types of RasV12-expressing fibroblasts formed tumors in all cases, as expected, while immortal fibroblasts with SIX1 overexpression in the absence of RasV12 failed to form tumors, consistent with the *in vitro* data. Of note, tumors with SIX1 overexpression showed a faster growth rate and reached larger size and weight at the end of the experiment (Fig. [1b--d](#Fig1){ref-type="fig"}), indicating that SIX1 promotes tumor growth in this context. All the tumors were diagnosed as encapsulated subcutaneous fibrosarcomas by histopathological analysis. They contained cellular atypias such as hypercromatic or giant nuclei or multinucleated cells, among others, irrespective of their genotype (Supplementary Fig. [S2](#MOESM1){ref-type="media"}).Figure 1SIX1 promotes tumorigenesis in fibroblasts. (**a**) Colony formation efficiency in anchorage-independent growth assays after seeding 2 × 10^5^ cells of the indicated types (n = 3 independent assays). (**b**) Representative image of tumors formed in xenograft assays in immunodeficient mice. (**c**) Kinetics of tumor growth in xenograft assays (n = 20 tumors in 10 mice from one representative experiment of a total of two independent experiments). (**d**) Weight of tumors excised at the end of the experiment (n = 3 for each type). All the graphs show the average and standard deviation of the data. Regulation of senescence-associated genes during SIX1 tumorigenesis {#Sec4} ------------------------------------------------------------------- In order to understand the molecular basis of the increased tumorigenic phenotype of SIX1-overexpressing fibroblasts, we performed a differential expression analysis in tumors with or without SIX1 overexpression, using RNASeq. We have recently shown that SIX1 is a negative regulator of senescence^[@CR5]^, and this response is a barrier against tumorigenesis^[@CR4]^. Thus, our first objective was to determine if suppression of senescence could play a role in the tumor-promoting action of SIX1. To this end, we interrogated the differential expression results from the RNASeq with respect to genes associated to SIX1 in senescence^[@CR5]^. Our previous results in primary human fibroblasts have shown that p16INK4A, a key effector of senescence, is one of the major downstream effectors of SIX1 in this response. SIX1 directly represses p16INK4A expression to suppress cellular senescence, and SIX1 down-regulation contributes to p16INK4A induction during senescence^[@CR5]^. To assess *p16Ink4a* expression in tumors with SIX1 overexpression, we performed an exon-specific analysis of the RNASeq results to discriminate the *p16Ink4a* transcript from the *p19Arf* transcript, also encoded by the *Ink4a/Arf* locus. Of note, we found a dramatic downregulation of the *p16Ink4a* transcript, but not of the *p19Arf* transcript, in the SIX1/RAS tumors, in a reverse situation to senescence. The results from RNASeq were validated by QPCR in a larger series of tumors of each genotype (Fig. [2a,b](#Fig2){ref-type="fig"}). Consistent with the RNA results, immunohistochemistry and Western Blot analyses showed undetectable levels of p16Ink4a protein in the SIX1-overexpressing tumors, in contrast to readily detectable levels in the control tumors (Fig. [2c,d](#Fig2){ref-type="fig"} and Supplementary Fig. [S2](#MOESM1){ref-type="media"}). In these assays, we also confirmed the overexpression of SIX1 in SIX1/RAS tumors by immunohistochemistry, Western Blot and QPCR. No significant changes were observed for the rest of Six proteins expressed in fibroblasts. Notably, *Eya2*, the major cofactor for SIX1, was dramatically upregulated in control tumors relative to SIX1/RAS tumors, probably reflecting a selective pressure to activate the SIX/EYA pathway in fibroblastic tumors (Supplementary Fig. [S3](#MOESM1){ref-type="media"}). To obtain further insight of the relevance of SIX1-associated senescence in this context, we extended this analysis to a selection of the most significantly up or down-regulated genes in senescence triggered by SIX1 silencing (SIS)^[@CR5]^. Interestingly, we observed that many of the genes most upregulated in SIS were significantly downregulated in the SIX1-overexpressing tumors from this study, and the reverse was true for SIS downregulated genes, suggesting a general inverse correlation of the SIS gene signature in SIX1-dependent senescence or tumorigenesis (Fig. [2e](#Fig2){ref-type="fig"}). Using QPCR and immunohistochemistry, we validated these results for *Pax3*, a gene upregulated in SIS and significantly downregulated in SIX1-expressing tumors (Fig. [2f](#Fig2){ref-type="fig"}). To investigate changes in senescence-related genes along the different steps of the tumorigenesis process, we also analyzed their expression in immortalized and transformed fibroblasts. In addition, we derived cell lines from tumors of both genotypes, to evaluate potential selection of features during tumor formation. QPCR and Western Blot in this set of cells revealed different patterns of expression for the selected senescence-related genes (Supplementary Fig. [S4](#MOESM1){ref-type="media"}). *p16Ink4a* expression was already dramatically reduced in immortalized and Ras-transformed SIX1-expressing cells before tumor formation and this pattern was retained in tumor-derived cell lines. In contrast, expression of *Pax3* became readily detectable only after RAS expression in control cells, but not in SIX1-overexpressing cells. Collectively, these results indicate that SIX1-senescence genes display a reverse pattern of expression during tumorigenesis and support the notion that blockade of senescence contributes to the oncogenic effect of SIX1 in this model.Figure 2Altered expression of senescence-associated genes in SIX1-tumors. (**a**) RNASeq results for the two transcripts of the murine *Ink4a/Arf* locus obtained from exon-specific analysis in tumors with or without SIX1 overexpression. Exon 1 alpha is specific of the *p16Ink4a* transcript and exon 1 beta is specific of the *p19Arf* transcript. (**b**) QPCR analysis of the expression of *p16Ink4a* and *p19Arf* in tumors with or without SIX1 overexpression. (**c**) Immunohistochemical detection of p16Ink4a and SIX1 in the indicated tumors. The left panel shows representative stainings and the right panel shows a quantification of positive cells (V/RAS n = 3, SIX1/RAS n = 5). Scale bar, 50 µm; inset, 20 µm. (**d**) Western blot analysis of the indicated proteins in tumors with (S) or without (V) SIX1 overexpression. (**e**) Differential expression of SIX1-senescence-associated genes in the murine xenografts from this study and in senescent human fibroblasts. The graph shows the expression fold change of the indicated genes in SIX1-expressing tumors relative to control tumors based on the RNASeq results (Tumor) and sh-SIX1 senescent fibroblasts relative to control fibroblasts (Senescence, data from^[@CR5]^). (**f**) QPCR analysis and representative immunohistochemistry of *Pax3* in the indicated tumors. Regulation of stemness and differentiation genes {#Sec5} ------------------------------------------------ To gain further insights into the genetic basis of the tumor phenotype caused by SIX1, we analyzed the global results obtained with RNASeq. The analysis of differentially expressed genes using a volcano plot identified *Sox2* as the gene most significantly upregulated in SIX1/RAS tumors (approximately two thousand fold increase, Fig. [3a](#Fig3){ref-type="fig"}, Supplementary Table [S1](#MOESM2){ref-type="media"}). Sox2 is a master regulator of stemness and cell fate during early embryogenesis and in some adult tissues. It also has important roles in cancer, where it has been linked to promoting proliferation, invasion and maintenance of cancer stem cells^[@CR13],[@CR14]^. Interestingly, SIX1 is also linked to stem and progenitor cell specification both in development and cancer (see Introduction). With this background, we considered interesting to explore in more detail the link of SIX1 to SOX2 in tumorigenesis. First, we validated the differential expression of Sox2 in tumors with or without SIX1 overexpression. QPCR analysis confirmed high levels of *Sox2* in SIX1/RAS tumors compared to undetectable levels in V/RAS tumors (Fig. [3b](#Fig3){ref-type="fig"}). Further analyses by immunohistochemistry immunofluorescence and Western Blot confirmed these results (Fig. [3c--e](#Fig3){ref-type="fig"} and data not shown) and, collectively, they clearly showed a dramatic increase in *Sox2* transcript and protein in SIX/RAS tumors. We also investigated Sox2 levels in cell lines representing different steps of our tumorigenesis experiment, as shown above for senescence-related genes (Supplementary Fig. [S5](#MOESM1){ref-type="media"}). We found that *Sox2* levels were modestly elevated in immortal fibroblasts with SIX1 overexpression (near the limit of detection by QPCR), but a much more dramatic increase occurred after expression of RasV12, which now could be easily detected by both QPCR and Western Blot. The differential expression of Sox2 was further retained in tumor-derived cell lines. To test if the increased levels of Sox2 associated to SIX1 overexpression had a functional impact, we introduced in V/RAS and SIX1/RAS cells the reporter construct SORE6-GFP, which contains a SOX2/OCT4 response element linked to GFP^[@CR15]^. Consistent with the previous results, FACS analysis showed a significant increase in the activity of the SORE6-GFP reporter in SIX1/RAS cells, confirming that SIX1 overexpression in transformed fibroblasts is accompanied by a significant increase in Sox2 levels and activity (Fig. [3e](#Fig3){ref-type="fig"} and Supplementary Fig. [S6](#MOESM1){ref-type="media"}).Figure 3Upregulation of Sox2 in SIX1-overexpressing tumors. (**a**) Volcano plot of RNASeq results. FC, expression fold change in SIX1-tumors relative to controls; -log padj, minus logarithm of the adjusted p value. The horizontal dotted line indicates padj = 0.05, the vertical dashed lines indicate FC 2 and −2. (**b**) QPCR analysis of *Sox2* expression in tumors with (SIX1) or without (V) SIX1 overexpression, (V/RAS n = 4, SIX1/RAS n = 6). (**c**) Immunohistochemical detection of Sox2 in the indicated tumors. The left panel shows a representative staining and the right panel shows a quantification of positive cells (n = 6 for each tumor type). Main scale bar, 50 µm; inset scale bar, 20 µm. (**d**) Immunofluorescence of SIX1, Sox2 and p16Ink4a in the indicated tumors. Scale bar, 100 µm. (**e**) Flow cytometry detection of the activity of the Sox2-responsive SORE6 cassette in the indicated fibroblasts transduced with SORE6 or empty vector (CMV). Given the role of Sox2 as key regulator of stemness, next we interrogated our differential expression data for further indications of changes in the differentiation state in SIX1 tumors. Interestingly, Gene Set Enrichment Analysis (GSEA) using the Hallmark collection identified significant enrichments in several categories related to differentiation, such as Myogenesis and Adipogenesis (positive correlation) or Epithelial Mesenchymal Transition (negative correlation) (Fig. [4a,d](#Fig4){ref-type="fig"} and Supplementary Table [S2](#MOESM3){ref-type="media"}). A similar analysis using the Gene Ontology collection also identified several genesets related to muscle among the most significantly positively correlated terms (Supplementary Table [S2](#MOESM3){ref-type="media"}). Notably, this expression profile associated to SIX1 in tumors recapitulates to some extent the physiological role of SIX1 during differentiation, since SIX1 has been implicated in both myogenesis^[@CR16]^ and adipogenesis^[@CR17]^. To validate this differentiation-related gene signature, we analysed by QPCR a selection of genes included in the leading edge of the categories Myogenesis (*Cdh13* and *Fst*) and Epithelial Mesenchymal Transition (*Myl9* and *Fbln2*), confirming the downregulation of mesenchymal markers and induction of muscular lineage markers in SIX1/RAS tumors (Fig. [4b,c,e,f](#Fig4){ref-type="fig"}). In support of these conclusions, RNASeq results showed that additional mesenchymal markers, such as *Thy1*, *Snail*, *Meox1* or *Fn1* were clearly down-regulated in SIX1/RAS tumors, while the cancer stem cell marker *Prom1* (also known as *CD133*) and markers of epithelial or other cell types such as *Ocln, Cldn1, Pkp1*, were clearly induced in these tumors (Supplementary Fig. [S7](#MOESM1){ref-type="media"}). Interestingly, the histological analysis of SIX1/RAS tumors also indicated features consistent with de-differentiation and loss of mesenchymal phenotype. First, Sirius Red staining (Fig. [4g](#Fig4){ref-type="fig"}) indicated a reduced presence of collagen-rich stroma in SIX1 tumors, consistent with a less fibrogenic phenotype. Also, while control tumors contained predominantly cells with fibroblastic elongated morphology that formed bundles, SIX1/RAS tumors contained mostly cells with rounded, de-differentiated morphology, which were distributed more randomly (Fig. [4g](#Fig4){ref-type="fig"}). Collectively, these results indicate enhanced cellular plasticity in SIX1-expressing fibrosarcomas, as shown by the activation of Sox2 and additional markers of stemness or alternative cell linages, and the concomitant loss of mesenchymal markers.Figure 4De-differentiation in SIX1-overexpressing tumors. (**a**) GSEA enrichment plot for the geneset "Epithelial Mesenchymal Transition" from the Hallmark collection. NES, normalized enrichment score; FDR, false discovery rate. (**b**,**c**) RNASeq results (Reads per million) (**b**) and QPCR analysis (**c**) of *Fbln2* and *Myl9*, two genes present in the leading edge in the analysis shown in a. (**d**) GSEA enrichment plot for the geneset "Myogenesis" from the Hallmark collection. (**e**,**f**) RNASeq results (Reads per million) (**e**) and QPCR analysis (**f**) of *Fst* and *Cdh13*, two genes present in the leading edge in the GSEA analysis shown in d. (**g**) Representative Sirius Red staining (top, quantification on right panel) and P-Erk immunohistochemistry, used here as a cytoplasmic marker (bottom) of the indicated tumors. To evaluate if the differentiation phenotype linked to SIX1 in our experimental tumors could also be observed in human mesenchymal tumors, we interrogated gene expression data from soft tissue sarcomas included in the TCGA database. Using the whole set of soft tissue sarcomas, we found a significant inverse correlation of *SIX1* with *MYL9* and *FBLN2*, two of the mesenchymal genes most significantly downregulated in the mouse SIX1-expressing tumors, and a tendency to positive correlation with *SOX2*. Interestingly, the subset of myxoid liposarcomas, characterized by highly frequent SIX1 overexpression (95% of tumors), also showed strong inverse correlation between *SIX1* and the mesenchymal markers *MYL9* and *FBLN2*, although no clear correlation with *SOX2* was observed in this case (Supplementary Fig. [S8](#MOESM1){ref-type="media"}). Sox2 contributes to the oncogenic activity of SIX1 *in vitro* {#Sec6} ------------------------------------------------------------- To characterize further the functional link between SIX1 and SOX2 in fibroblastic tumors, we used lentiviral shRNA to silence SIX1 in V/RAS and SIX1/RAS cells. Efficient knock-down of SIX1 resulted in down regulation of Sox2, as shown by Western Blot and QPCR, suggesting a direct link between SIX1 and Sox2 (Fig. [5a,b](#Fig5){ref-type="fig"}). Given the oncogenic role of Sox2 in different tumor types^[@CR14]^, we hypothesized that Sox2 could contribute to the increased tumorigenesis of SIX1-expressing transformed fibroblasts. To test this notion, we determined the impact of the manipulation of SOX2 in anchorage-independent growth assays. Silencing of Sox2 in SIX1/RAS cells, using two independent shRNAs, caused a significant reduction in the number of soft agar colonies. Conversely, SOX2 overexpression in V/RAS cells resulted in increased colony number, which nevertheless did not equal that of SIX1/RAS cells (Fig. [5c](#Fig5){ref-type="fig"}). These results indicate that Sox2 upregulation contributes significantly to the oncogenic effect of SIX1 in our model, even though additional factors must also play a role in this phenotype.Figure 5Sox2 is regulated by SIX1 and contributes to its oncogenic activity *in vitro*. (**a**) Western blot analysis of SIX1 and Sox2 in the indicated fibroblasts with or without expression of shSIX1. (**b**) QPCR analysis of *SIX1 and Sox2* expression in shSIX1 fibroblasts, relative to vector-infected cells, n = 2. (**c**) Colony formation efficiency in anchorage-independent growth assays of the indicated cell types (n = 3 except for SIX1/RAS and shSox2-A, n = 2). Western blots on top of the graphs show SOX2 levels in each cell type. (**d**) Chromatin immunoprecipitation (ChIP) of ectopic SIX1 and the histone mark H3K4me3 in the SRR2 enhancer of the Sox2 locus. The data shows enrichment of binding of the indicated antibodies relative to non-specific IgG (n = 2). SIX1 binds to a regulatory element in the Sox2 locus {#Sec7} ---------------------------------------------------- Next, we tried to determine the mechanism responsible for SOX2 overexpression in tumors with elevated SIX1. Based on the role of SIX1 as a transcription factor, we asked if SIX1 was present in regulatory regions of the Sox2 locus, using chromatin immunoprecipitation. This assay revealed specific binding of SIX1 to the Sox2 SRR2 downstream enhancer but not to the upstream SRR1 enhancer (Fig. [5d](#Fig5){ref-type="fig"} and Supplementary Fig. [S9](#MOESM1){ref-type="media"}). SIX1 binding to SRR2 was higher in SIX1-overexpressing cells, coinciding with increased levels of the active chromatin mark H3K4me3 in this region. The SRR2 enhancer is active in pluripotent and neural stem cells^[@CR18]^, as well as in specific subpopulations of tumor cells^[@CR19]^. Interestingly, it contains the sequence TCACG that matches the SIX1 consensus binding motif^[@CR20]^. These results showing binding of SIX1 to the SRR2 regulatory element suggest that SIX1 could participate in transcriptional regulation of Sox2. SIX1 controls senescence and SOX2-mediated self-renewal in glioma cells {#Sec8} ----------------------------------------------------------------------- To establish if our results could reflect a general link between SIX1 and SOX2 in cancer, we decided to investigate this connection in a different tumor type. To this end, we focused in glioma, because SIX1 and SOX2 are frequently overexpressed in these tumors and they are both specifically enriched in glioma stem cells^[@CR21]--[@CR25]^. To study the link between SIX1 and SOX2 in glioma, we used the cell lines U251 and GNS166^[@CR26]^, representative of differentiated and stem-like phenotypes respectively. Both cell lines express high levels of SIX1 and SOX2. Notably, silencing of SIX1, using two independent shRNAs, led to a marked reduction in SOX2 transcript and protein in both glioma cell lines, in line with the results in mouse transformed fibroblasts (Fig. [6a,b](#Fig6){ref-type="fig"}). In agreement with the key role of SOX2 in glioma cancer stem cell renewal, and the reduction in SOX2 expression caused by shSIX1, knockdown of SIX1 had a clear functional impact in glioma cells, leading to a dramatic reduction of their self-renewal capacity, as measured by their ability to form oncospheres (Fig. [6c,f](#Fig6){ref-type="fig"}). Interestingly, in keeping with its role as a senescence regulator in other cell types^[@CR5]^, SIX1 silencing also caused a clear reduction in proliferation in both cell lines, which was accompanied by the induction of markers of cellular senescence, such as Senescence-Associated Beta Galactosidase activity and the acquisition of an enlarged morphology (Fig. [6d--f](#Fig6){ref-type="fig"} and data not shown). These results indicate that SIX1 controls SOX2-mediated self-renewal, proliferation and senescence in glioma cells, and they support the existence of a general link between SIX1 and SOX2 in tumors.Figure 6SIX1 controls SOX2 expression, proliferation and self-renewal in human glioma cells. (**a**) Western blot analysis of SOX2 in U251 cells infected with two shSIX1 vectors or control vector. (**b**) QPCR analysis of *SIX1 and SOX2* expression in U251 and GNS166 cells infected with two shSIX1 vectors, relative to vector-infected cells, n = 3. (**c**) Quantification of oncosphere forming capacity in shSIX1 U251 cells after 7 days in culture, relative to control pLKO U251 cells (n = 3). 1ry NS, primary neurospheres. (**d**) Proliferation of shSIX1 GNS166 cells. The increase in cell number (cells at day 5 relative to initial number of cells) was calculated for each cell type and the value for control pLKO cells was designated as 100% (n = 3). (**e**) Quantification of Senescence-Associated Beta Galactosidase activity (SABGal)-positive cells in GNS166 cells expressing the indicated vectors (n = 3). (**f**) Representative images of oncospheres (top) and SABGal staining (bottom) as described in panels c and e. Discussion {#Sec9} ========== We have recently shown that the homeoprotein SIX1 is a negative regulator of senescence^[@CR5]^, an intrinsic tumor-suppressive response^[@CR2],[@CR4]^. Moreover, evidence from animal models and human tumors suggests that SIX1 can act as an oncogene in different types of cancer^[@CR8],[@CR9]^. With this background, we have analysed here the contribution of SIX1 to fibroblastic tumors and its link to senescence, using a cellular model for immortalization and oncogenic transformation based on primary mouse fibroblasts. The tumor-suppressive role of senescence is mediated by a proliferation block in potentially oncogenic cells within premalignant lesions, which halts their progression to full-blown malignant tumors. This antiproliferative effect can be reinforced by the immune--mediated clearance of senescent cells^[@CR2]^. Consistent with this tumor-suppressive effect, ablation of senescence is considered a prerequisite for tumor formation. Our transformation and tumorigenesis assays with mouse fibroblasts show a clear protumorigenic effect of SIX1, in line with the evidence in other cancer types^[@CR8],[@CR9]^. Interestingly, the transcriptomic analysis of these tumors has revealed the reverse regulation of a gene signature associated to SIX1-mediated senescence, which includes p16Ink4a, a key mediator of senescence caused by SIX1 loss or other triggers^[@CR2],[@CR5]^. These findings underscore the relevance of senescence as a tumor protective mechanism. Moreover, they are consistent with the notion that, in addition to other well-established functions of SIX1 in tumors such as promoting proliferation, angiogenesis and invasion, blunting of senescence may be an important component of the pro-tumorigenic action of SIX1. In addition, our study shows that SIX1 promotes stem-like or undifferentiated phenotypes in fibroblastic and brain tumor cells, linked to the regulation of the stemness factor Sox2. This protein plays a key role in the control of pluripotency, self-renewal, proliferation and cell fate in early development and in some adult stem cells. Notably, SOX2 also plays a similar role in the context of tumors, where it has been linked to maintenance and self-renewal of cancer stem cells^[@CR13],[@CR14]^. Likewise, it has been reported that SIX1 is enriched in cancer stem cells in tumors such as breast carcinoma, glioma and Wilms tumor, and it can promote the acquisition of cancer stem cell features, associated to activation of the TGF-ß or Wnt pathways^[@CR25],[@CR27],[@CR28]^. The direct link to SOX2 shown here identifies a novel mechanism by which SIX1 could contribute to self-renewal in cancer stem cells. SOX2 has been associated to tumor initiating cells in a variety of human tumors including mesenchymal^[@CR29]--[@CR31]^ and brain^[@CR21]--[@CR23]^ tumors studied here, where it influences proliferation, tumorigenicity and self-renewal. In line with our observations, inspection of datasets from the TCGA collection reveals a significant correlation between SIX1 and SOX2 expression in different tumor types, including glioma, soft-tissue sarcoma, prostate and esophageal cancer (Supplementary Fig. [S8](#MOESM1){ref-type="media"}). Collectively, these results suggest that SIX1 could play a widespread role in the regulation of SOX2-mediated stem phenotypes in cancer. Interestingly, previous studies have also pointed to a possible link between Six1 and Sox2 in development, during formation of neurosensory structures in the inner ear and the olfactory epithelium. Overlapping expression patterns have been reported for both proteins, but a direct link has not been unambiguously demonstrated in this context^[@CR32]--[@CR34]^. Taking together these observations and our current results, it is feasible that the link SIX1-SOX2 we have identified in tumors might recapitulate physiological regulatory circuits in action during organogenesis. Our results indicate that SIX1 binds to the Sox2 SRR2 enhancer, suggesting that SIX1 might regulate Sox2 transcription in mouse cells. The significant enrichment in SIX1 tumors of pathways that have been implicated in SOX2 regulation, such as mTOR^[@CR23]^ or Hedgehog^[@CR35]^ prompted us to consider additional indirect effects. However, pharmacological inhibition of either pathway failed to have any significant impact on Sox2 expression in this model (Supplementary Fig. [S10](#MOESM1){ref-type="media"} and data not shown), further supporting the relevance of the direct transcriptional regulation of Sox2 by SIX1. Recent results have shown a link between senescence and cellular plasticity or regeneration in different settings^[@CR36]--[@CR40]^. These reports suggest that senescent cells can promote cellular plasticity in neighboring cells through paracrine mechanisms mediated by the SASP. However, the cell-intrinsic impact of senescence on cellular plasticity is less clear^[@CR40]--[@CR42]^. Our results open the possibility that SIX1 might play a role in coordinating both processes in tumor cells, and future work should address this interesting question. Methods {#Sec10} ======= Cell culture {#Sec11} ------------ Cells were cultured with Dulbecco's Modified Eagle Medium (DMEM) (GIBCO, Grand Island, NY, USA), containing 10% foetal bovine serum (FBS) and 0, 5% antibiotics (penicillin and streptomycin) at 37 °C in 5% CO~2~. For soft agar colony formation assays, 2 × 10^5^ cells were resuspended in a warmed solution of complete medium containing 0,3% agarose and plated in 60 mm dishes with a solidified bottom layer of 0, 5% agarose in complete medium, in triplicate. After two weeks, colonies larger than 100 µm were counted. To derive cell lines from tumors, portions from freshly excised tumors were washed with PBS, minced with a scalpel, digested with TrypLE Express (Gibco, Waltham, Massachusetts, USA), and seeded on 100 mm dishes with DMEM/10%FBS. The day after, non-attached cells and tissue debris were removed and the attached cells were expanded. Oncosphere formation assays with glioma cells were performed as described^[@CR23]^. Senescence-associated Beta Galactosidase staining was performed as described^[@CR43]^. Cell proliferation assay {#Sec12} ------------------------ Cells were seeded in 12-well plates (2,5 × 10^3^ cells per well), recovered by trypsinization and counted with a hemocytometer at days 1, 3 and 5 after plating. Retroviral and lentiviral transduction {#Sec13} -------------------------------------- Early passage wild-type MEFs were used for viral transductions essentially as described^[@CR5]^. The vectors used were pRetroSuper mouse p53 (a gift from Rene Bernards), pWZLHygro-SIX1 (generated in this study), pWZLBlast-SOX2 (a gift from Matthew Meyerson, Addgene plasmid \# 26351), pLXSN-RasV12, pLKO-shSIX1^[@CR5]^ and SORE6-GFP (a gift from Lalage Wakefield^[@CR15]^). Western blot {#Sec14} ------------ Western Blot analysis and preparation of total protein lysates from cells in culture were performed as described^[@CR44]^. 30 µg of total lysate (for cell lines) or 100 µg (for tumors) was loaded on each lane. Protein lysates from tumors, were obtained through homogenization with a pestle in RIPA lysis buffer (10 mM NaPO~4~ pH 7.2, 300 mM NaCl, 2 mM EDTA, 0.1% SDS, 1% Deoxycholate, 1% NP40). The primary antibodies used are described in Supplementary Table [S3](#MOESM1){ref-type="media"}. Quantitative PCR {#Sec15} ---------------- Total RNA was isolated with TriReagent (Invitrogen, Carlsbad, California, USA). Quantitative real-time PCR was performed as described^[@CR45]^, using SybrGreen. Primer sequences are described in Supplementary Table [S3](#MOESM1){ref-type="media"}. Human *SOX2* was analysed using the TaqMan probe human *SOX2* (Hs01053049_s1, Roche). Tumorigenesis assays {#Sec16} -------------------- 2 × 10^5^ cells in 100 µl of PBS were injected subcutaneously in both rear flanks of 6-week-old female athymic nude (*Foxn1*^*nu*^*/Foxn1*^*nu*^) mice. Tumors were measured every 2 days with a caliper using the formula V = A × B^2^/2, where A is the longest diameter and B is the perpendicular diameter. Fourteen days after the injection, (or when the tumors reached the maximum allowed size), the mice were euthanized and the tumors were excised, weighed and processed for immunohistochemistry assays, protein and RNA extraction, and cell line derivation. All animal experiments were performed in accordance with relevant guidelines and regulations, approved by the ethics committees of the Instituto de Investigaciones Biomédicas and the Spanish Research Council (CSIC), and authorized by the Madrid Regional Government (Reference: PROEX 362/15). Immunofluorescence {#Sec17} ------------------ Immunofluorescence with cells in culture was performed essentially as described in^[@CR45]^. For tumors, antigen retrieval was performed with Envision Flex Target Retrieval Solution High pH (DAKO, Les Ulis, France), blocking and incubation with primary antibodies (Supplementary Table [S3](#MOESM1){ref-type="media"}) was done in DAKO diluent solution. Secondary antibodies (Donkey anti rabbit AF-488, donkey anti-rabbit AF-555, donkey anti-goat AF-546, 1:500, ThermoFisher, Waltham, MA) were used in DAKO diluent solution with DAPI diluted 1:1 with PBS. Finally, the slides were mounted with ProLong (Invitrogen, Carlsbad, California, USA) and analyzed with a confocal spectral LSM710 (Zeiss) microscope. Immunohistochemistry {#Sec18} -------------------- Immunohistochemical analysis of tumors was performed essentially as described^[@CR46]^. Further details are described in Supplementary Methods. Flow Cytometry {#Sec19} -------------- V/RAS and SIX1/RAS cells were transduced with the vectors SORE6-GFP and pMX-Cherry (as an infection efficiency control). Two days post infection, cells were trypsinized, washed and resuspended in PBS (4 × 10^5^ cells/200 µl) and analysed in a Cytomics FC 500 MPL cytometer for detection of GFP and Cherry fluorescence. The MXP software was used for the cytometric analysis. RNA Seq {#Sec20} ------- Total RNA from three tumors of each genotype was used for RNASeq analysis. A detailed description of the methods used for sequencing and data analysis is included in Supplementary Methods. Chromatin immunoprecipitation {#Sec21} ----------------------------- Chromatin immunoprecipitation was performed essentially as described in^[@CR45]^. The antibodies used were: SIX1 (HPA001893, Sigma), and H3K4me3 (Ab8580, Abcam). The primers used for PCR were: GGTGGTCGTCAAACTCTGCTAATT, AGAGTCTCGGAGAATGCCCT (SRR1) and ATTTATTCAGTTCCCAGTCCAAGC, CCCTCTCCCCCCACGC (SRR2). Statistical analysis {#Sec22} -------------------- Statistical significance was calculated using unpaired two-tailed Student's t-test (\*\*\**P* \< 0.001, \*\**P* \< 0.01, \**P* \< 0.05). Supplementary information ========================= {#Sec23} Supplementary Information Supplementary Table S1 Supplementary Table S2 **Publisher's note:** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Electronic supplementary material ================================= **Supplementary information** accompanies this paper at 10.1038/s41598-018-38176-0. We thank René Bernards, Lalage Wakefield and Steve Pollard for sharing reagents and cells; Orlando Domínguez, Genomics Unit, CNIO, Madrid for help with RNASeq analysis; Juana María Flores, UCM, Madrid for help with histopathological analysis, and the Histopathology Unit, CNIO and the Histology Facility, CNB-CSIC, Madrid for histological analysis. This work was supported by grants SAF2015-65960P (MINECO-FEDER) to IP, and CP16/00039, PI16/01580 (Instituto Salud Carlos III-FEDER) to AM. JA-I is the recipient of a predoctoral fellowship (PRE20160375) from the Department of Education, University and Research of the Basque Government. C.d.L., S.M.-A., C.E., I.M., J.L.-A. and I.L.-A. performed experiments with mouse fibroblasts and tumors, J.A.-I. and A.M. performed and supervised experiments with glioma cells. I.P. supervised the project and secured funding. C.d.L. and I.P. wrote the manuscript with input from all the authors. The RNASeq data generated in this study has been deposited at Gene Expression Omnibus (GEO) with the accession number GSE113385. Competing Interests {#FPar1} =================== The authors declare no competing interests.

1. Introduction {#sec1-nanomaterials-09-01749} =============== A diverse group of infectious diseases, commonly known as neglected tropical diseases (NTDs) are predominant in the underprivileged parts of the world, especially in developing countries. These diseases are widespread in tropical and subtropical areas due to poor hygiene and insufficient health infrastructures \[[@B1-nanomaterials-09-01749]\]. Currently, more than 20 different types of NTDs are prevalent in 149 countries, affecting approximately 1.4 billion people worldwide \[[@B2-nanomaterials-09-01749]\]. Leishmaniasis, one of the most neglected tropical diseases, is currently affecting around 12 million people worldwide and 350 million people are under the risk of infection in 98 developing countries \[[@B3-nanomaterials-09-01749]\]. Leishmaniasis has recently earned more public attention due to its high infection and morbidity rate. The London declaration on NTDs was made to eliminate Leishmaniasis as a public health problem by 2020 \[[@B4-nanomaterials-09-01749]\]. Leishmaniasis occurs due to obligate protozoan parasite of the *Leishmania* species \[[@B5-nanomaterials-09-01749]\]. There are almost 51 species of parasites, out of which 21 are pathogenic and cause Leishmaniasis \[[@B6-nanomaterials-09-01749]\]. Some of the species that cause Leishmaniasis includes *L. donovani*, *L. amazonensis*, and *L. aethiopica* etc. Leishmanial parasites exist in two major forms: round and elongated. The round parasite is small and non-motile, while the elongated parasite can move with the help of flagella \[[@B7-nanomaterials-09-01749]\]. Leishmanial transmission occurs when a sand fly sucks blood from an infected individual (human or animal) ([Figure 1](#nanomaterials-09-01749-f001){ref-type="fig"}) \[[@B8-nanomaterials-09-01749]\]. The parasite transformation occurs as it changes from the amastigote stage to the promastigote stage, taking about 4--25 days \[[@B9-nanomaterials-09-01749]\]. The disease results in the development of ulcers and also affects other bodily organs \[[@B10-nanomaterials-09-01749]\]. Leishmaniasis exists in three major forms, namely as mucosal leishmaniasis (ML), cutaneous leishmaniasis (CL), and visceral leishmaniasis (VL). In ML, the symptoms take more time to appear, approximately 1--5 years. The symptoms include runny nose, ulcers formation, breathing problems and nose bleeding \[[@B11-nanomaterials-09-01749]\]. In CL, the symptoms appear few weeks after the person is bitten by sand fly \[[@B12-nanomaterials-09-01749]\] and in the most common type, VL symptoms appear in about 2--6 months, and include weakness, weight loss, fever, enlarged spleen, liver enlargement, lesions, and swollen lymph nodes \[[@B13-nanomaterials-09-01749]\]. Among endemic regions of the world, 0.2--0.4 and 0.7--1.2 million cases of VL and CL have been reported, respectively. Approximately 75% of the global estimated prevalence of CL has been reported among certain countries, for example, in Algeria, Afghanistan, Colombia, Syria, Brazil, Iran, Ethiopia, Costa Rica, North Sudan, and Peru, while more than 90% of VL cases have been reported in Bangladesh, India, South Sudan, Ethiopia, and Brazil \[[@B3-nanomaterials-09-01749],[@B14-nanomaterials-09-01749],[@B15-nanomaterials-09-01749]\]. The leishmanial parasite has ability to take control of the immune system of the affected individuals, which enables the disease condition to persist for a long time and develop into a chronic infection \[[@B16-nanomaterials-09-01749]\]. Basically, the parasite imbalances the host immunity due to its uncontrolled growth inside the macrophages, leading to the eradication of innate, as well adaptive, immunity of the host. There are two ways by which leishmanial parasite manipulates the immune system; by one way the parasites hide in long-lived macrophage cells surviving hostile conditions \[[@B17-nanomaterials-09-01749]\]. The other way is that the parasite mediates a cell signaling pathway in macrophages which inhibits T-helper cells' (Th2) cytokine responses, specifically interleukins, IL-5, IL-4, and IL-13, leading to downregulation of the protective immune response \[[@B18-nanomaterials-09-01749]\]. Hence, the parasite has the ability to switch between a pro-inflammatory Th1-type healing response to an anti-inflammatory Th2-type non-healing response, which prioritizes their survival and growth inside the macrophages \[[@B19-nanomaterials-09-01749]\]. Additionally, the parasite has also the ability to inhibit the intracellular leishmanicidal activity by decreasing the production of reactive oxygen species, nitric oxide, and pro-inflammatory cytokines leading for their better growth and survival by reduced proliferation of CD4+ and CD8+ T cells, which eventually leads to an enhanced Th2 response \[[@B20-nanomaterials-09-01749],[@B21-nanomaterials-09-01749]\]. Furthermore, several co-inhibitory molecules, such as CTLA-4, PD-L1, CD200, and Tim-3, have shifted the balance of the immune system towards the non-healing Th2 response \[[@B19-nanomaterials-09-01749]\]. The lack of knowledge regarding the Th1 to Th2 cell shift in the host immunological response is due to the unidentified host or parasitic factors that contribute to the severe pathology of Leishmaniasis. Due to the lack of demarcated entities for protective immunity of the host, the generation of vaccines for the parasite has been a difficult task for researchers. Several *leishmania* vaccine candidates have been developed and evaluated in native and recombinant form, like gp63, gp46, TSA, PSA2, LACK, LmsT1, Leish111f, and m2, to kill parasites. However, none of them have shown any outcomes towards prophylaxis \[[@B22-nanomaterials-09-01749],[@B23-nanomaterials-09-01749]\]. Hence, the lack of prophylactic measures has been a concern in the elimination of this NTD. Although the control measures for the elimination of Leishmaniasis are limited, yet two strategies have been applied, such as classical therapeutics interventions and vector management through insecticides for the control of the *leishmania* parasite in disease-endemic regions \[[@B24-nanomaterials-09-01749]\]. The currently available therapeutic interventions are not effective antileishmanial drugs, besides their enhanced number of cases with relapse and repercussions, have made the current situation critical for the elimination of leishmaniasis \[[@B25-nanomaterials-09-01749],[@B26-nanomaterials-09-01749]\]. Thus, the search for safer, more efficient, innovative, cost-effective therapies is urgently needed for treating Leishmaniasis. During the last decade, nanotechnology-based drug delivery systems have been used to enhance the performance of drugs in treating several diseases. Combined use of a nanocarrier system with the antileishmanial drugs is a new and promising approach as these nanocarriers can penetrate the macrophages' cells and reach the infectious parasite, enabling targeted and efficient delivery \[[@B27-nanomaterials-09-01749]\]. A diverse group of nanocarriers also serve as a method of enhancing efficacy, regulating pharmacokinetics, and reduced drug toxicity with sustained release of the drug. Nanotechnology has enabled advancements in pharmacology by providing treatment to various forms of leishmaniasis by targeted delivery of drugs \[[@B28-nanomaterials-09-01749],[@B29-nanomaterials-09-01749]\]. Various nanotherapeutics have been approved by the Food and Drug Administration (FDA) and are currently available for clinical use \[[@B30-nanomaterials-09-01749]\]. Therefore, in this review we attempt to comprehensively compile the various recent reports on nanotechnology-based approaches for the treatment of Leishmaniasis with emphasis on the utilization of various nanosized techniques and nano-drug conjugation systems. 2. Conventional Treatment Strategies against Leishmaniasis and Their Limitations {#sec2-nanomaterials-09-01749} ================================================================================ With research advancements, several diagnostic and therapeutic lines have been developed against leishmaniasis. Among all forms of leishmaniasis, VL could be fatal as it affects the organs if not treated properly within two years \[[@B29-nanomaterials-09-01749],[@B31-nanomaterials-09-01749]\]. The diagnosis of leishmaniasis is usually done by examination of the tissue of lesions under microscope. Several molecular-based diagnostic techniques, such as polymerase chain reaction (PCR) and real-time PCR, have also been developed with high sensitivity and specificity \[[@B32-nanomaterials-09-01749],[@B33-nanomaterials-09-01749]\]. However, the main issue associated with these molecular tools is the lack of their availability in health centers of underdeveloped countries. The other method includes the culturing of parasites in different media, such as Novy-MCNeal-Nicolle \[[@B34-nanomaterials-09-01749]\]. The treatment of leishmanial disease has always been a challenge for researchers. In the early 1950s, sodium stibogluconate and meglumine antimoniate were a few of the first anti-leishmanial drugs, but they had many side effects associated with their intake \[[@B35-nanomaterials-09-01749]\]. With advancement in technology and studies, now, there are many therapeutic drugs available for leishmaniasis treatment, including pentamidine, miltefosine, paromomycin, amphotericin B, and its lipid formulations \[[@B36-nanomaterials-09-01749],[@B37-nanomaterials-09-01749],[@B38-nanomaterials-09-01749]\]. The typical chemical compounds employed for anti-leishmanial drug development includes antimony sulfide, doxorubicin, quillaja, saponin, and phosphatidylserine. Although patients with compromised immune systems, cardiac diseases, and organ transplantation cannot be given drugs like pentavalent antimonial \[[@B39-nanomaterials-09-01749]\], it has been recently found that amphotericin B (AmB) is the most effective drug for anti-leishmanial activity. AmB was initially used as an antifungal compound consisting of deoxycholate salt \[[@B40-nanomaterials-09-01749]\]. AmB with encapsulation of liposome has been found to be more effective than AmB alone. Liposmal AmB is less toxic than AmB alone, although it becomes costly \[[@B41-nanomaterials-09-01749]\]. Miltefosine (Impavido) is the only orally administered anti-visceral leishmanial drug. However, the issue associated with miltefosine is that it cannot be used in pregnant and feeding women because it can harm the developing fetus in the womb \[[@B42-nanomaterials-09-01749]\]. A new drug, Humatin has been developed recently with similar efficiency as AmB but with a limited number of side effects \[[@B43-nanomaterials-09-01749]\]. Vaccination is another approach that is employed for leishmanial treatment. The convention vaccine for leishmanial immunization made use of the killed parasite as an antigenic component but its efficacy was low \[[@B44-nanomaterials-09-01749]\]. Later, another approach for fighting the leishmania parasite was introduced, known as a peptide vaccine \[[@B45-nanomaterials-09-01749]\]. The approach is based on the utilization of a minimal pathogenic component to generate long-lasting immunity against the deadly parasite. The choice of epitope is very crucial in peptide vaccine development and, therefore, different in vitro and in silco analysis are conducted to determine the immunogenicity of the peptides \[[@B46-nanomaterials-09-01749]\]. The potential peptides are compared and the best epitope candidates are used for vaccine development by combining multiple epitopes. Peptide vaccine is a promising approach for leishmanial treatment but the challenge is that it is degraded very easily in the body by the immune system \[[@B47-nanomaterials-09-01749]\]. In conclusion, all the available antileishmanial treatments have some limitations or side effects associated with them. Chemotherapeutic drugs are expensive and the parasite has developed resistance against them. Clinical mishandling of medicines in a majority of underdeveloped countries has played a key role in the development of widespread resistance against leishmanial disease \[[@B48-nanomaterials-09-01749],[@B49-nanomaterials-09-01749]\]. In addition, to date there is no effective vaccine available on the market to prevent leishmaniasis \[[@B50-nanomaterials-09-01749]\]. Thus, it is very important to develop alternative drugs via adopting novel strategies that can effectively control this fatal disease. 3. Nanotechnology: A New Horizon for Treatment of Leishmaniasis {#sec3-nanomaterials-09-01749} =============================================================== Innovations in interdisciplinary sciences have been moving the translational sciences to the next level for better control of infectious diseases. Nanomedicine (the use of medical applications of nanotechnology for human welfare) is one of the promising fields in this area that has been continuously growing, keeping up hope for highly sensitive diagnostic tools and better drug delivery for various infectious diseases in the near future \[[@B51-nanomaterials-09-01749]\]. As the traditional antileishmanial drugs have low tolerability, long treatment duration, and are difficult to administer, a tremendous upsurge has been observed in the development of novel nano-biopharmaceuticals that can cure leishmaniasis. The field of nanotechnology has played a vital role in revolutionizing the process of delivering drug in the field of medicine. The nanotechnology employs the use of various drug-loaded nanocarrier systems, such as metallic nanoparticles, liposomes, nanoemulsions, nanosphere, solid-lipid nanoparticles, nanocapsules, polymeric nanoparticles, and nanostructured lipid carriers and nanostructured layered films for efficient drug delivery to the target sites for the treatment of leishmaniasis ([Figure 2](#nanomaterials-09-01749-f002){ref-type="fig"}) \[[@B52-nanomaterials-09-01749],[@B53-nanomaterials-09-01749],[@B54-nanomaterials-09-01749],[@B55-nanomaterials-09-01749]\]. These nanocarrier systems enable targeted delivery, increased bioavailability, and reduced toxicity of drugs \[[@B56-nanomaterials-09-01749]\]. Nanocarriers enclose the drugs that provide targeted delivery and also protect the drug from being metabolized \[[@B57-nanomaterials-09-01749]\]. The absorption and distribution profile of nanocarriers greatly depends on the physicochemical properties, i.e., the size, hydrophobicity, targeting molecule, and their charges. Many processes, like uptake and entry of nanocarriers into cell and their further interaction with immune system, are dependent on the size and charge of the nanocarriers. Anther property is the hydrophobicity, which controls the absorption and distribution of nanocarriers by effecting the immune cells' interaction, protein interaction, particle clearance, and protein charge \[[@B58-nanomaterials-09-01749],[@B59-nanomaterials-09-01749]\]. The charge is used in binding plasma protein, protein interactions, membrane damage, and in immune cell stimulation \[[@B60-nanomaterials-09-01749]\]. These drug-loaded nanocarriers enter the cell by phagocytosis, a form of endocytosis in which the cell engulfs particles larger than 0.75 µm in diameter. Macrophages, neutrophils, and monocytes are capable of phagocytosis and, therefore, sometimes are referred as "professional phagocytes" \[[@B61-nanomaterials-09-01749]\]. Leishmaniasis is a particularly interesting disease to be treated with drug-loaded nanocarriers since the parasites exclusively infect the highly phagocytic cells known as macrophages. In this way, the macrophages take up the drug-loaded nanocarrier by phagocytosis, where they will directly act on the parasites ([Figure 3](#nanomaterials-09-01749-f003){ref-type="fig"}). This allows the drugs to reach an effective intracellular concentration, along with a reduction in toxicity and dosage of drugs \[[@B62-nanomaterials-09-01749],[@B63-nanomaterials-09-01749]\]. Furthermore, there are two types of vectoring procedure, active and passive vectoring, which could affect the distribution and uptake. In active vectoring, a specific compound is added or attached to the surface of the nanoparticle, whereas passive vectoring is the inherent capacity of cells when they recognize the foreign particles to organisms \[[@B61-nanomaterials-09-01749]\]. 4. Drug-Loaded Nanocarrier Systems for Treatment of Leishmaniasis {#sec4-nanomaterials-09-01749} ================================================================= Various nanocarrier systems have been synthesized and used in the controlled drug delivery in treatment of leishmaniasis. Each of these nanocarrier systems have their own advantages and disadvantages, as discussed in [Table 1](#nanomaterials-09-01749-t001){ref-type="table"}. Among the traditional nanoparticles, the most preferred are liposomes and polymeric nanoparticles as they are easily and rapidly internalized by macrophages in the liver and spleen \[[@B64-nanomaterials-09-01749]\]. The most commonly employed nanocarriers in curing leishmaniasis are liposomes due to its unique properties. They are able to load and deliver both hydrophobic and hydrophilic drugs by surface functionalization, which is used to improve drug targeting. Additionally, the fate of liposomes and the leishmaniasis parasite is the same. The positively-charged liposomes are readily taken in by the macrophages. Since the macrophages can recognize sugar molecules, liposomes are surface functionalized with sugar to improve macrophage targeting. However, liposomes face some limitations, as well. They are not stable. They could result in toxicity because the drug can leak from the liposomes into the blood supply \[[@B65-nanomaterials-09-01749]\]. Nanoemulsions are one of the best drug delivery systems due to their simple preparation, ability to solubilize hydrophobic drugs, physicochemical stability, and easy scale-up \[[@B66-nanomaterials-09-01749]\]. Polymeric nanoparticles are also a widely used nanoparticle system for the treatment of leishmaniasis. They have the properties to overcome some of the drawbacks of liposomes \[[@B67-nanomaterials-09-01749]\]. The have small size, low toxicity, and are cost effective as they can be used to deliver more than one drug. They have the ability to design biodegradable systems and can be surface functionalized. Among the polymeric nanoparticles, poly lactide-co-glycolide (PLGA) is the most commonly employed as it is biodegradable and biocompatible. One important difference between liposomes and polymeric nanoparticles is their stability \[[@B53-nanomaterials-09-01749]\]. Unlike the unstable nature of liposome, polymeric nanoparticles do not face the limitation of drug leakage into the blood supply \[[@B68-nanomaterials-09-01749],[@B69-nanomaterials-09-01749]\]. There are some advanced nanocarrier systems, such as metallic nanoparticles, dendrimers, and carbon-based nanomaterials. They need to be studied well in order to know their advantages and drawbacks. One important advantage of dendrimers is their ability to load more than one drug due to their branched structure, enhancing drug bioavailability \[[@B70-nanomaterials-09-01749]\]. Along with the advantages of nanocarriers as efficient antileishmanial drugs there are also some challenges to overcome. One of the prominent hurdles is the high cost of these nanoformulations. Hence, their commercialization and high scale production is not economically feasible. On account of their economic feasibility, solid-lipid nanoparticles (SLNs) are better because they are made of triglyceride lipids whose production scale up is less expensive then phospholipids \[[@B71-nanomaterials-09-01749]\]. 4.1. Liposome Nanoparticles {#sec4dot1-nanomaterials-09-01749} --------------------------- Liposome nanoparticles are nano-sized spherical vesicles that are made up of bilayer phospholipids which provide an aqueous support which serves as a carrier for the adherence of both hydrophilic and lipophilic drugs \[[@B79-nanomaterials-09-01749]\]. Liposome usage has many advantages over conventional drugs: they have increased retention in the body and stay in circulation for a longer period of time \[[@B80-nanomaterials-09-01749]\]. Liposomes have the ability of sustaining the drug release, controlling the release, and reducing the drug dosage and its frequency of dosage \[[@B81-nanomaterials-09-01749]\]. Due to these properties, liposomes are largely employed for the study of leishmaniasis treatment and, hence, are the most anticipated clinical drug. In case of anti-leishmanial treatment, the drugs are encapsulated in a liposomal layer which enables their efficient intracellular delivery to the leishmanial amastigotes. These liposomes have the ability to penetrate the macrophages by the process of phagocytosis and directly deliver the drug to the site of the parasite \[[@B82-nanomaterials-09-01749]\]. The nanosized liposomes are an emerging approach used in many disease treatments, particularly for the delivery the chemotherapy drugs. The liposomes provide enhanced pharmacokinetic properties along with target precision which provides a major advantage \[[@B83-nanomaterials-09-01749]\]. The best example of liposome is AmB lipsome, a formulation of AmB liposome, which reduces the side effects of amphotericin B \[[@B84-nanomaterials-09-01749]\]. In the AmB formulation there are different types, such as AmB colloidal solution, AmB liposome, and AmB lipid network, amongst which the liposomal AmB has been proven to be most effective \[[@B85-nanomaterials-09-01749]\]. According to one study, the formulation of the liposome can overcome the disadvantages of conventional drugs. The different formulations, including miltefosine, paromomycin, and meglumine antimoniate, have been developed against leishmania by the process of freeze drying. The liposomal drug was administered by subcutaneous injection in mice and the efficiency was studied, which turned out to be 90% \[[@B86-nanomaterials-09-01749]\]. In another study it has been reported that the macrophages have receptors on their surface that play a role in the control of cellular functions, such as recognition, activation, secretion, and endocytosis \[[@B87-nanomaterials-09-01749]\]. The liposome, incorporated with a ligand, interacts with the receptors of macrophages enabling the uptake of the liposomal content \[[@B57-nanomaterials-09-01749]\]. The liposomes conjugated with mannose and 4-sulphated acetyl galactosamine have been found to be effective against leishmanial activity \[[@B88-nanomaterials-09-01749]\]. 4.2. Lipid Nano-Capsules {#sec4dot2-nanomaterials-09-01749} ------------------------ Lipid nano-capsules (LNs) are nanocarriers that range in size from 20 to 100 nm and they mimic the lipoproteins. Lipid nanocapsules consist of a core of lipids and a surfactant membrane surrounding it. It is a hybrid structure made from the use of liposomes and polymeric nanocapsules \[[@B89-nanomaterials-09-01749]\]. LNs are made by use of a solvent-free method which provides the stability and increased bioavailability. The major advantage of LNs is that they deliver the drug on site, minimizing the dosage by many folds and reducing the side effects of toxicity \[[@B90-nanomaterials-09-01749]\]. In one study for the development of LNs, the core was made of hydrophobic olive oil and the outer shell was made of a hydrophilic component, chitosan \[[@B91-nanomaterials-09-01749]\]. The miltefosines are alkylphospholipids that are used in the treatment of leishmaniasis destroy the Ca^2+^ homeostasis of parasite. The formulation of NPs loaded with miltefosine provided enhanced effectiveness against leishmaniasis, which was demonstrated by the injury to the promastigotes. The stability and sustained drug release were also ensured by the LNs. The LN oral drugs have been developed with advancements in technology \[[@B92-nanomaterials-09-01749]\]. 4.3. Metallic Nanoparticles {#sec4dot3-nanomaterials-09-01749} --------------------------- Metals have been used in medications since early history. There is a wide range of metallic nanoparticles that are being used for antileishmanial activity providing minimal toxicity and high efficacy \[[@B73-nanomaterials-09-01749],[@B93-nanomaterials-09-01749]\]. The metallic nanoparticles first came into existence in the 1850s. A study was conducted for the treatment of visceral leishmaniasis with iron oxide (Fe~3~O~4~) nanoparticles coated with glycine (peptide), which encapsulated the AmB drug. A 10--15 nm nanoparticle size was used, which enabled the controlled release of AmB, reducing the parasitic content in the spleen of treated subjects \[[@B94-nanomaterials-09-01749]\]. Glycine-coated nanoparticles could be employed further in leishmanial treatments. Zinc oxide nanoparticles (ZnONPs) are massively produced and used. A study was conducted in which ZnONPs were employed in varying concentrations (0.18, 0.37, 0.75, and 1.5 µg/mL) against the amastigote form of leishmania, *L. donovani*, in vitro culture. The results were analyzed by colorimetric assay which suggested that ZnONPs exerted a cytotoxic effect on the amastigote cells, causing hinderance in their proliferation and suppression of activity of *L. donovani*. The study suggests that ZnONPs could be a cost-effective means against anti-leishmanial drug development \[[@B95-nanomaterials-09-01749]\]. Sumaira, et al. \[[@B96-nanomaterials-09-01749]\] prepared ZnONPs from *Verbena officinalis* and *Verbena tenuisecta* plant leaf extracts. The results suggest that *V. officinalis* had more phenolic content. Both plant ZnONPs were tested for anti-leishmanial activity, where the *V. officinalis* ZnONPs had better activity due to the greater phenolic content and smaller size as compared to *V. tenuisecta*-mediated ZnONPs. Silver has always been very useful in medications since early times. Silver colloidal solutions were initially used for treating infections; approximately 650 different diseases and illnesses were treated \[[@B97-nanomaterials-09-01749],[@B98-nanomaterials-09-01749]\]. Later, with advancements, nanotechnology helped to develop nanosilver or silver nanoparticles (AgNPs). Various studies have been conducted on the biogenic synthesis of silver AgNPs and their mode of actions in different biomedical applications \[[@B99-nanomaterials-09-01749],[@B100-nanomaterials-09-01749]\]. Antileishmanial activity of AgNPs was checked, obtained from a fungus source, *Fusarium oxysporium*, and evaluated by a group of researchers \[[@B91-nanomaterials-09-01749]\]. The results of the study were promising as AgNPs led to the death of promastigotes enabling its apoptosis. In further studies it was found that the AgNPs release reactive oxygen species (ROS) that cause damage to the membranes of promastigotes. In the case of amastigotes, the AgNPs led to a reduction of infected macrophages. AgNPs had a direct damaging ability against amastigotes \[[@B91-nanomaterials-09-01749]\]. Another AgNP study suggested that the anti-bacterial activities of silver nanoparticles help fight leishmaniasis. In this study, the effects of AgNPs were checked against leishmanial parasite morphology, infectivity, metabolic activity, survival abilities, and proliferation rates. AgNPs led to impairment of morphological characteristics and infectivity rates of the parasite. Additionally, the metabolic activity and proliferation was reduced by 1.5-fold \[[@B90-nanomaterials-09-01749]\]. Overall, the AgNPs could be a new therapeutic source for the treatment of leishmaniasis \[[@B90-nanomaterials-09-01749],[@B101-nanomaterials-09-01749]\]. The use of nanoparticles under ultra-violet (UV) and infrared (IR) light have high toxicity by generating ROS, causing the death of the parasites. In one study, the antileishmanial effects of some nanoparticles, such as AgNPs, gold nanoparticles (AuNPs), titanium dioxide nanoparticles (TiO~2~NPs), ZnONPs, and magnesium oxide nanoparticles (MgONPs) were evaluated on leishmania major parasites \[[@B28-nanomaterials-09-01749],[@B102-nanomaterials-09-01749]\]. Increased antileishmanial activity was observed for AgNPs, followed by AuNPs, TiO~2~NPs, ZnONPs, and MgONPs under UV and IR light conditions as compared to dark. Thus, both of these light-improved antileishmanial properties of these nanoparticles must be considered in future studies \[[@B28-nanomaterials-09-01749]\]. Similarly, in another study, chitosan-derived TiO~2~NPs were used as an effective antileishmanial agent against amastigote and promastigote forms of the parasite. Chitosan-derived TiO~2~NPs were loaded with meglumine antimoniate to enhance the activity of TiO~2~NPs. The activity of the nanoparticles was checked via a UV spectrophotometer. The formulation was found to be effective against amastigote as well as promastigote forms of the parasite \[[@B92-nanomaterials-09-01749]\]. Overall, metallic and metal oxide nanoparticles provide a promising approach for the reduction and treatment of all types of leishmanial activity \[[@B92-nanomaterials-09-01749],[@B103-nanomaterials-09-01749]\]. 4.4. Polymeric Nanoparticles {#sec4dot4-nanomaterials-09-01749} ---------------------------- These nanoparticles are made from various types of biocompatible and biodegradable colloidal particles. Their size ranges from 10 to 1000 nm \[[@B104-nanomaterials-09-01749]\]. They carry drugs by different approaches, like adsorption, encapsulation, dissolution, entrapment, or by chemically binding the drug on the surface of polymeric nanoparticles (PNPs) \[[@B105-nanomaterials-09-01749]\]. Advanced physicochemical properties of PNPs lead to improved bioavailability, enhanced cellular dynamics, biodegradability, and controlled drug delivery \[[@B74-nanomaterials-09-01749]\]. Polymers are the most widely studied and researched form of carriers used in nanomedicine. It was first used in 1979 for cancer therapy when polyalkylcyanoacrylate nanoparticles were used to adsorb anti-cancer drugs \[[@B106-nanomaterials-09-01749]\]. PNPs include synthetic polymers, such as poly (lactic acid) (PLA), poly (glycolic acid) (PGA), poly (lactide-co-glycolide) (PLGA), poly (caprolactone) (PCL), poly (cyanoacrylate) (PCA), and natural polymers, such as gelatin, albumin, chitosan, and alginate \[[@B107-nanomaterials-09-01749],[@B108-nanomaterials-09-01749]\]. Among these polymers, PLGA has been mostly used in drug delivery and in tissue engineering. PNPs are present in two different forms: nanospheres and nanocapsules, polymeric or reservoir systems, respectively. In nanocapsules, the drug is encapsulated in a cavity surrounded by a polymer membrane, whereas the drug is not confined in a cavity, but it is dispersed uniformly in the case of nanosphere \[[@B109-nanomaterials-09-01749],[@B110-nanomaterials-09-01749]\]. PNPs appear to be a great choice for delivery of drugs and proteins to target cells because of their easy permeation due to their small size, and these polymers can be designed in various molecular designs with many applications \[[@B75-nanomaterials-09-01749]\]. PNPs deliver drug to the targeted site by the following three mechanisms:Through an enzymatic reaction which lead to polymer degradation at targeted site resulting in the release of the drug.Through swelling of the PNP followed by hydration and drug release by diffusion.Through detachment of the drug from the polymer \[[@B111-nanomaterials-09-01749]\]. PNPs are being studied for their use as a drug carrier for the treatment of leishmania. Different types of PNPs are being studied on mouse models, in in vitro studies, to investigate the treatment for leishmania \[[@B112-nanomaterials-09-01749],[@B113-nanomaterials-09-01749]\]. A studied conducted by Roy et al. \[[@B76-nanomaterials-09-01749]\] studied the effect of nano-encapsulated diterpenoid lactone and andrographolide on albino mice. Poly(DL-lactide-co-glycolic acid) nanoparticles and polyvinyl alcohol (PVA) was loaded in a 50:50 ratio. The results showed significant antileishmanial activity on mouse models using 4% PVA on using 1/4 of the pure compound dosage. The authors suggested that this could provide a cost-effective chemotherapy of leishmaniasis. A recent study was conducted to investigate the efficacy of carbohydrate-functionalized PLGA (poly lactide-co-glycolide) nanospheres for the treatment VL in mice \[[@B114-nanomaterials-09-01749]\]. PLGA nanospheres were prepared by nanoprecipitation and surface functionalized with three different types of carbohydrates, i.e., mannose, mannan, and mannosamine. Co-culturing of these PLGA nanospheres with macrophages led to the activation of immune-modulatory and pro-inflammatory responses in the host, which triggered the killing of parasites. The authors reported that the mannan-functionalized PLGA nanospheres were more effective against VL parasites as compared to the mannose- and mannosamine-functionalized PLGA nanospheres \[[@B114-nanomaterials-09-01749]\]. 4.5. Solid Lipid Nanoparticles (SLNs) and Nanostructured Lipid Carriers (NLCs) {#sec4dot5-nanomaterials-09-01749} ------------------------------------------------------------------------------ Solid lipid-based nanoparticles are a relatively new class of nanocarrier. Solid lipid nanoparticles (SLNs) and nanostructured lipid carriers (NLCs) belong to this class and differ from each other based on their matrix \[[@B111-nanomaterials-09-01749]\]. SLNs are nanospheres made from lipids which remain solid at body temperature and are stabilized by emulsifiers \[[@B115-nanomaterials-09-01749]\]. Their size is less than 1000 nm \[[@B116-nanomaterials-09-01749]\]. The have various advantages as they protect the drug against harsh environmental conditions, their large-scale production is easy, using high-pressure homogenization technique, they are biocompatible and biodegradable \[[@B117-nanomaterials-09-01749]\]. However, they have some limitations as well, i.e., SLNs have low drug loading efficacy due to its crystalline structure and there is a chance of drug expulsion during the storage of the crystalline structure and initial burst release can occur \[[@B118-nanomaterials-09-01749]\]. A study revealed that chitosan-coated SLNs carrying AmB were synthesized for chemotherapy of *Leishmania* infections. Their antileishmanial activity showed that the SLNs have a strong effect than formulations of AmBisome and Fungizone, available on the market. Additionally, this study showed SLNs are safer than market products, by evaluating acute toxicity study in mice \[[@B77-nanomaterials-09-01749]\]. In another study, in vivo efficacy of SLN-loaded paromomycin sulfate was investigated against *L. tropica* in a mouse model. It was found that parasite propagation and switching towards the Th1 response was more effectively inhibited by using PM loaded with SLNs as compared to when paromomycin sulfate was used alone \[[@B119-nanomaterials-09-01749]\]. Nanostructured lipid carriers (NLCs) are referred to as the second generation of solid lipid-based nanoparticles. They are the combination of both solids and lipids, unlike SLNs. It does not have a definite crystalline structure, but rather has different sized moieties. NLCs have better loading capacity since they do not have a crystal structure. Thus, drug expulsion and burst release is not faced in the case of NLCs \[[@B120-nanomaterials-09-01749]\]. A recent study was carried out to make a formulation of a veterinary drug named buparvaquone by using NLCs \[[@B117-nanomaterials-09-01749],[@B121-nanomaterials-09-01749]\]. Another study prepares Amphotericin B lipid nanostructured carriers in order to increase the therapeutic efficacy and reduced toxicity of Amphotericin B, which is the only main treatment against leishmaniasis. Different formulations were synthesized and the selection criteria were particle size and particle size distribution. The in vitro release profile of the AMB-loaded NLCs showed 65% drug release within 24 h. The results of the study showed that delivery of AMB through NLCs is preferable over using Amphotericin B alone \[[@B122-nanomaterials-09-01749]\]. 4.6. Nanotubes {#sec4dot6-nanomaterials-09-01749} -------------- Nanotubes are cylindrical hollow molecules that are synthesized from inorganic and metallic materials. A number of studies have been conducted which prove nanotubes are excellent nanocarriers. Anti-leishmanial efficacy of AmB associated with carbon nanotubes was examined in a study. The authors found this formulation to have better targeted killing of *L. donovani* compared with free AmB \[[@B77-nanomaterials-09-01749],[@B123-nanomaterials-09-01749]\]. Another study developed betulin associated with CNTs as an anti-leishmanial formulation. The study reported better cytotoxicity of the new antilieshmanial formulation compared to the control group \[[@B124-nanomaterials-09-01749]\]. A study used a formulation of linked AmB, an antileishmanial drug, with functionalized carbon nanotubes (f-CNTs) to lessen the drug-induced toxicity. This formulation was able to inhibit parasite growth more effectively than AmB. This drug carrier improves the drug efficacy. Additionally, there was toxicity observed in the kidneys or livers of mice \[[@B125-nanomaterials-09-01749]\]. The use of carbon nanotubes in drug delivery have not been designed for humans yet, but are in the preclinical stage. 5. Nanovaccines: An Emerging Approach of Nanotechnology for Combating Leishmaniasis {#sec5-nanomaterials-09-01749} =================================================================================== To date, chemotherapeutic drugs act as the main treatment of leishmania, including amphotericin B, paromomycin, fluconazole, antimony-containing compounds, and pentamidine. However, these therapeutic drugs have their own limitations, such as toxicity, longer regimens, low efficacy, and drug resistance \[[@B24-nanomaterials-09-01749]\]. Other therapies, which could be effective for eradicating leishmania, are vaccine based. There are two types of leishmanial vaccines being prepared: first-generation and second-generation. First-generation leishmanial vaccines are comprised of live vaccine while second-generation vaccines are made using recombinant technology \[[@B126-nanomaterials-09-01749],[@B127-nanomaterials-09-01749]\]. To date, there is no licensed vaccine for leishmaniasis. Three types of leishmanial vaccines, Leish-F1, F2, and F3, designed at the Infectious Disease Research Institute (IDRI) are in clinical trials \[[@B128-nanomaterials-09-01749]\]. These are formulated on the basis of the selective antigen epitope properties of leishmania. Recombinant leishmania vaccines are also being investigated at the Sabin Vaccine Institute \[[@B50-nanomaterials-09-01749]\]. With the advent of nanotechnology, there is now a new approach of synthesizing vaccines using nanoparticles as carriers of antigen preparation. Nanoparticles could provide a safe, efficacious, and efficient delivery system for vaccines. According to one study, solid lipid nanoparticles can serve as an efficient tool to synthesize leishmanial vaccine \[[@B129-nanomaterials-09-01749]\]. Delivering antigens and adjuvants using nanoparticles have different purposes:To help increase uptake by of the antigen, loaded in nanoparticles, by the antigen-presenting cells \[[@B130-nanomaterials-09-01749]\].To activate a stronger immune effect as a simultaneous delivery of the antigen by different NPs to the same APC, activating the immune response strongly as compared to the free antigen and adjuvant \[[@B131-nanomaterials-09-01749]\].To activate Th1-type immune response \[[@B132-nanomaterials-09-01749]\]. In 2005, a group of researchers prepared a nanovaccine by loading recombinant *Leishmania* superoxide dismutase in a chitosan nanoparticle using the ionotropic gelation method in mice. The study assesses the loading efficacy and size of nanoparticles loaded with SODB1. The results showed higher cell-mediated immune response and higher IgG2a levels, on using stable chitosan nanoparticles, which could be used as a nanovaccine for leishmaniasis \[[@B133-nanomaterials-09-01749]\]. Another study uses nanoliposomes used as the nanocarrier for soluble Leishmania antigens (SLA). Results showed that parasites decreased in the footpad and spleen of a mouse model injected with this formulation when compared with the control group \[[@B134-nanomaterials-09-01749]\]. Although there is no nanovaccine commercially available for leishmaniasis, studies have led to higher efficacy by using nanoparticles as vaccine carriers and as adjuvants to form nanovaccines, which could be potentially decrease the number of leishmaniasis cases. 6. Conclusions and Future Outlooks {#sec6-nanomaterials-09-01749} ================================== Despite several treatment options, there is not even a single efficient option that would effectively control the incidence of leishmaniasis. The drugs available for treating leishmaniasis face many side effects, such as high cost, toxicity, and resistance to parasites. Various studies are being conducted to enable the use of nanotechnology for devising nanomedicines and nanovaccines for treating leishmaniasis \[[@B44-nanomaterials-09-01749],[@B51-nanomaterials-09-01749],[@B108-nanomaterials-09-01749],[@B135-nanomaterials-09-01749]\]. Various nanomaterials are being studied for the development of safe and cost-effective drugs for treating leishmania. Many studies have revealed that the potential efficient agents for leishmanial treatment could be liposomes, PLGA nanoparticles, carbon nanotubes, and SLNs that enhance the parasite-targeted drug delivery. Nanovaccines are a relatively new concept in treating leishmania; although no vaccine is yet available, but studies are on-going to find efficient nanovaccines. Despite many studies that have been conducted to find nanotechnology-based efficient drugs for leishmaniasis, they are all still in the preclinical stage, except for one liposomal drug (AmB) which is commercially available \[[@B136-nanomaterials-09-01749]\]. The commercial aspects of nanomedicines are a major concern for researchers. Any drug delivery system's most desired characteristic is its commercial feasibility. The cost of drugs can affect the resources and scale up of drug development. AmB is currently the most cost-effective drug in leishmaniasis treatment. The other nanoparticle-based drug delivery strategies are under the process of development and trials only, and their production cost has not yet been analyzed \[[@B137-nanomaterials-09-01749]\]. There is, however, need for advanced studies and research to develop effective drugs with low cost against leishmania disease. Conceptualization: K.S. and S.A.; methodology: Z.K.; software: S.A.; validation: K.S., Z.K., and C.H.; formal analysis: S.A; investigation: I.A.; resources: S.A.; data curation: C.H.; writing---original draft preparation: Z.K. and K.S.; writing---review and editing: C.H., S.A.; visualization: I.A.; supervision: S.A.; project administration: S.A. This research received no external funding. The authors declare no conflict of interest. {#nanomaterials-09-01749-f001} {#nanomaterials-09-01749-f002} {#nanomaterials-09-01749-f003} nanomaterials-09-01749-t001_Table 1 ###### Advantages and limitations of nanocarrier systems. Nanocarrier Advantages Limitations References ---------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ----------------------------------------------------------------- Liposomes Ability to carry either, hydrophilic or hydrophobic drugs, biocompatible, biodegradable, stable, possibility of surface functionalization Toxic because the drug can be leaked or displaced into the blood stream; High production cost \[[@B72-nanomaterials-09-01749]\] Polymeric nanoparticles Biocompatible, low toxicity, biodegradable, cost-effectives, possible surface functionalization, avoids leakage of the drug Difficult to scale up \[[@B73-nanomaterials-09-01749],[@B74-nanomaterials-09-01749]\] Solid lipid nanoparticles (SLNs) Protect drug against harsh environmental conditions, easy scale up, biocompatible Low drug-loading efficacy due to its crystalline structure, there is a chance of drug expulsion during the storage of the crystalline structure and initial burst release can occur \[[@B75-nanomaterials-09-01749],[@B76-nanomaterials-09-01749]\] Nanoemulsions Stable, Carry both hydrophobic and lipophilic drugs Toxicity of surfactants \[[@B77-nanomaterials-09-01749]\] Metallic nanoparticles Antibacterial, Antifungal properties, Stable, Uniform structure Toxicity \[[@B78-nanomaterials-09-01749]\] [^1]: These authors contributed equally to this work.

[^1]: Dr. Maisel\'s present address is the Department of Pathology, University of Colorado Medical Center, Denver

Background {#Sec1} ========== The Ebola epidemic in 2014 and 2015 in the West African countries of Guinea, Liberia and Sierra Leone was unprecedented in its size, over 28,000 people were infected and around 11,000 died \[[@CR1]\]. In Sierra Leone, as of April 2016, 14,124 cases were registered, 3,956 of whom died \[[@CR2]\]. As the epidemic took hold in Sierra Leone, Guinea and Liberia, reports were rife of instances of community resistance to medical intervention, communities' mistrust and avoidance of healthcare centres, and stigmatisation of health workers and survivors \[[@CR3]--[@CR6]\]. Wilkinson and Leach argue that such resistance and mistrust must be understood through consideration of a long history of structural violence \[[@CR7]\]. First as a central port in the Atlantic slave trade and then as a mining colony for the British Empire, a substantial proportion of Sierra Leone's encounters with outsiders were primarily extractive \[[@CR8], [@CR9]\]. The country's postcolonial past was similarly marred by years of oppressive and corrupt rule, which deprived the majority of the population of access to education, employment and basic healthcare services \[[@CR10], [@CR11]\] This was followed by a brutal ten year civil war (1991--2002), in which rebelling factions fought against authorities as symbols of a "rotten system", but in the process all sides of the conflict systematically perpetrated violence against the civilian population \[[@CR12]--[@CR15]\]. Following the war and up until the Ebola epidemic erupted in 2014, the country was undergoing a long and difficult process of post-war reconstruction, during which citizens, government and foreign partners were attempting to build trustworthy governance structures at national and local level \[[@CR10], [@CR16]\]. The continuing poverty (Sierra Leone is ranked 181^st^ in the Human Development Index) and inequalities experienced by citizens and this deep rooted historical legacy of distrust in government and outside agencies can help explain challenges faced by those attempting to contain the spread of the Ebola virus \[[@CR7]\]. The nature and rapidity of the spread of the disease also made clear the importance of understanding the socio-cultural dimensions of disease and the historical context which shapes them \[[@CR7]\]. The initial response to the outbreak was undertaken by national government, national and international non-governmental organisations (NGOs), followed by the global UN led responses. Whilst prioritizing treatment centres, international organisations also focused their responses on surveillance, case management, safe burials, contact-tracing and community sensitization \[[@CR17]\]. At first, interactions with communities focused on "sensitisation", emphasising that local populations lacked knowledge on Ebola and that "traditional practices" spread disease \[[@CR18]\]. As Chandler et al note, these strategies paid "little attention to the historical, political, economic, and social contexts in which they are delivered...\[and\] they reinforce external perceptions that local beliefs and practices are barriers to be overcome through persuasion or counterbalanced with incentives" (\[[@CR18]\], p.1275). As the epidemic developed and lessons were learned, examples emerged of changes in the engagement of local populations, as some aspects of the response adapted to deal more effectively with socio-cultural dimensions, taking seriously the importance of understanding communities' needs and constraints and building trust. For example, safe burials that prevented people from washing dead bodies were made more acceptable by seeking the approval of local leaders, discussing the practices with the family of the deceased, burying the body in the presence of the community and including components of burial rituals such as Muslim shrouds on coffins \[[@CR19]\]. The Ebola response's learning curve was thus characterized by a gradual move away from attempts to correct misinformation towards the engagement of communities. In a recent analysis of the epidemic, Laverack and Manoncourt found that in a context of fear, mistrust and resistance, due to both the structural and political context as well as experiences of poor quality health care, utilising bottom up approaches to communication, which included a respect for local knowledge, was more effective \[[@CR17]\]. As anthropologists working in these contexts have argued throughout the epidemic, rather than viewing customs and culture as barriers of resistance, it is more useful to consider how culture can adapt, something that, as Richards et al suggest, ought to be done in continuous consultation with local communities \[[@CR20]\]. Reflecting on the West African Ebola epidemic, Abramowitz et al highlight how "under conditions of extreme stress, culture can be flexible and supple in response to extreme circumstances and the arrival of new information (such as public health messages), and make allowances in extraordinary conditions" (\[[@CR21]\], p.4). The Ebola epidemic, in other words, rendered evident the benefits of engaging deeply with affected countries' social, cultural and political context in order to understand communities' response to the disease and to work with them to find ways to deal with crises. The speed with which the epidemic grew and its extent demanded not only a rapid and coordinated response from international agencies and local authorities, but also called for the fast-tracked development of vaccines and other potential preventative and treatment technologies. This required the establishment of prophylactic vaccine trials in an epidemic context, which presents particular challenges for both the communities and the trialists. The setting up of such trials requires learning lessons from the historical and political context and from the experiences of the response to the outbreak, as described above, especially in terms of rooting communication strategies on deep understanding of socio-cultural contexts as adaptable and dynamic and on effective and active engagement of community members. In addition, the setting up of trials can benefit from learning from anthropological research on clinical trials related to other stigmatised diseases in resource poor contexts. There is a growing body of work that is exploring complex political economy and social justice questions that emerge in communities where clinical trials are taking place \[[@CR22]--[@CR24]\]. These address broader ethical questions beyond the focus on informed consent. Molyneux and Geissler, for example, argue that this often does not take into account broader issues of inequities in wealth, health status and health care resources between the researchers and the researched (\[[@CR24]\]; see also \[[@CR22], [@CR23], [@CR25]\]). Dugas and Graham similarly suggest that standardised consent processes may be inadequate in contexts where translation is required, where socio-economic inequalities are stark and where collective consent may be important alongside individual consent \[[@CR26]\]. Beyond these issues, there has been a growing body of research on the local effects of medical research on participants and their communities \[[@CR24], [@CR27]--[@CR29]\]. Such research has revealed the salience of social relations developed between researchers and participants as well as the prevalence of rumours and concerns \[[@CR28]--[@CR36]\]. This body of research provides a foundation for developing strategies for engaging local populations and participants in a meaningful way grounded in the specificities of ethics and social relations in resource-poor contexts. Given the nature of the Ebola epidemic, and the specific fears and concerns that the disease invokes, as well as the political history of Sierra Leone, the setting up of a vaccine trial during the Ebola epidemic required having an in-depth understanding of the epidemic and its effects as well as building trust within communities. With the prospect of such new encounters with biomedicine for people in communities with very little (and often no) experience of medical research, there is also an urgent need to determine what rumours, concerns, fears or mistrust emerge from dialogues with community members and individuals involved in vaccine trials and the ways in which such actors can be engaged in addressing these and building trust. This paper discusses the role of social science (anthropology) and community liaison in supporting the establishment of the EBOVAC-Salone prophylactic Ebola vaccine trial in Kambia District, Northern Sierra Leone, during July to August 2015, towards the end of the Ebola epidemic. Through an exploration of notions of power, fairness and trust emerging from analysis of the Sierra Leonean context and through ethnographic research, the paper discusses three situations in which the EBOVAC-Salone social science and community liaison teams worked together to ensure effective communication, to develop appropriate recruitment strategies and to address rumours about the trial. The EBOVAC-Salone trial {#Sec2} ======================= In December 2014, the Innovative Medicines Initiative (IMI) awarded funding from the Ebola + programme to a consortia of research institutions to support the development, manufacturing and deployment of a prime-boost prophylactic Ebola vaccine regimen, in partnership with private industry \[[@CR37]\]. A portion of the IMI2 funding awarded to the EBOVAC1 consortium is being used to implement the EBOVAC-Salone study in Sierra Leone. In addition to EBOVAC-Salone, IMI2 has also funded the related EBODAC (Ebola Vaccine Deployment, Acceptance and Compliance) consortium whose objective is to conduct community engagement and communication to support the prime-boost Ebola vaccine trials and roll-out of vaccines, if proven effective. The EBOVAC-Salone study aims to assess the safety and immunogenicity of the Ad26.ZEBOV/MVA-BN-Filo prime-boost regimen in a population affected by Ebola. The study is being carried out in Kambia District in Northern Sierra Leone. Stage 1, an open label, safety and immunogenicity study of the prime boost vaccine in 43 healthy adults aged 18 years or older has completed the vaccination phase. The first participants to volunteer for the study were screened for eligibility on 20 September 2015 and vaccination began on 8 October 2015. In stage 2, a randomised, controlled, safety and immunogenicity study, 688 individuals will be screened and will be randomised to the prime boost vaccine or a control vaccine.[1](#Fn1){ref-type="fn"} Adults, adolescents and children aged one year or older will be included in this trial. Plans for additional stages are being finalized in consultation with the Sierra Leonean authorities and international health agencies. EBOVAC-Salone is coordinated by the London School of Hygiene &Tropical Medicine (LSHTM) working in partnership with Sierra Leone's Ministry of Health and Sanitation and the College of Medicine and Allied Health Sciences. The EBOVAC-Salone study is one of several vaccine trials set up in West Africa during and in the immediate aftermath of the epidemic \[[@CR38], [@CR39]\]. Methods: the community liaison and social science system {#Sec3} ======================================================== Engagement with the community in Kambia in the run-up to stage 1 of the EBOVAC-Salone study was conducted by a community liaison team and a social science team. The two teams were recruited locally. The community liaison team, comprising nine locally-recruited staff, supervised by two LSHTM staff members, was responsible for implementing EBOVAC-Salone's community engagement strategy. The team was complemented by an external communications manager who monitors rumours and concerns about the trial beyond the Kambia District and also provides information about the trial at national and international levels. EBOVAC Salone was the first vaccine trial of any kind to have taken place in the study area, and the country as a whole has had limited experience of medical research. The community liaison team members thus received background training on clinical trials, with a particular emphasis on the difference between communication to support clinical trials and communication and social mobilisation for routine or proven interventions. The community engagement strategy for Stage 1 involved engaging with all levels of the community -- from elected and traditional leaders to individual households - through a variety of channels. These included undertaking one-to-one engagement with key stakeholders, holding public meetings in partnership with influential civil society leaders, organising community meetings supported by local and traditional leaders, conducting house-to-house sensitisation visits, and hosting radio chat shows and phone-ins on local radio within Kambia District. A Participant Advisory Group (PAG) was also set up, serving as a means through which study participants were encouraged to openly discuss their experiences of participating in the EBOVAC-Salone study with each other. The purpose of the latter group was to address concerns, share views, expose perceived barriers to enrolment and to share information where necessary. This group is entirely participant-led and at the first meeting the participants elected a Chairman and representative committee. For Stage 1, trial participants were invited to join the group by field workers who visited their homes in the initial seven days following vaccination. The frequency, structure and agendas of these meetings were decided by the group members themselves. The PAG is continuing to support stage 2, where 99 participants have enrolled to date. The social science team was made up of four locally recruited research assistants, a data analyst and a transcriber, led by an LSHTM social scientist based in Kambia. The team was responsible for conducting anthropological research to examine community and participant perceptions and experiences of the EBOVAC Salone vaccine study, including any rumours and concerns around the trial and the vaccine. In addition, the team explored the socio-cultural context and perceptions of illness and disease in the community where Stage 1 was taking place. The methods used for this research were: ethnographic observation, focus group discussions, in depth interviews, and exit interviews with participants as they completed their visit to the vaccination clinic. Ethical approval for this study was granted by the ethics committees of LSHTM and Sierra Leone. Ethnographic observations were carried out in key areas in Kambia and trial clinics to explore community dynamics; social interactions around the clinic; community narratives about EVD and the clinical trials; and to trace any ongoing rumours and concerns emerging at community level. Focus group discussions were carried out with both community members and with trial participants to discuss perceptions and experiences of the trial. In depth interviews were carried out with 6 Stage 1 Participants and 16 Stage 2 participants, interviewed after prime and boost vaccination, and with approximately 20 key stakeholders in Kambia. At the time of writing the Stage 2 research was ongoing. The data for Stage 1 was analysed through thematic and framework analysis. Framework analysis was used to answer the specific research objectives. A coding tree was developed according to the key research questions. Interview transcripts were coded according to this coding tree. Thematic analysis was used to develop emerging themes not identified by the research objectives. The insights for this paper are based primarily on the ethnographic research carried out in Kambia in the three months preceding the opening of the trial. As the epidemic continued to ravage the country, and as Kambians lived under military curfew and constant news of Ebola cases, the trial team was renovating clinics, building laboratories and holding high-level conversations with key stakeholders. During the preparation phase for Stage 1, ethnographic encounters in local markets, *attaya* [2](#Fn2){ref-type="fn"} bases, *poyo* [3](#Fn3){ref-type="fn"} bars, shops and *okada* [4](#Fn4){ref-type="fn"} parking grounds enabled the research team to develop a rich and complex picture of Kambia's community dynamics, the everyday struggles and concerns of its inhabitants during the outbreak, their beliefs and fears about medical intervention and their views on the study that was about to open in their town. Results: communications and social science in action {#Sec4} ==================================================== During the preparation phase, the social science and community liaison teams developed a feedback system to inform rapidly research-driven communication strategies. Weekly meetings had three main purposes. Firstly, the social science team provided feedback on community engagement plans based on their research on the socio-cultural context, local community dynamics and perceptions of the vaccine trial. Secondly, the meetings brought up any issues encountered by the trial team or the community liaison staff that required further research by the social science team, such as the design of the recruitment strategies outlined below. Thirdly, the social science team reported on any rumours or concerns identified in the community. These rumours and concerns were communicated to the community liaison team anonymously in order not to breach confidentiality and to maintain the independence of the social science research. Following feedback, the community liaison staff brainstormed strategies to respond to concerns or rumours. Such strategies depended on the specific issue raised, but usually involved considering different and creative avenues for discussion with the community on the issue, reviewing messaging to actively engage with the issue at hand, and determining who the best person in the team was to respond and through which channel. When an urgent or potentially harmful rumour had been identified, this was reported immediately to the trial manager in Kambia and to the principal investigators. This section of the paper discusses three examples of challenges faced by the trial team during the set-up phase of the study. These case studies reflect how lessons gained from knowledge of the country's context, the experiences of the Ebola response and knowledge gained during previous trials helped inform the community engagement strategies and responses to issues as they came up. The examples are arranged through three key themes: power, fairness and trust. Each case study begins with a reflection on the context facing the trial team, then offers insights from the ethnographic research and finally considers the implications for community engagement strategies. These case studies elucidate the importance of research-driven mobilisation strategies, as well as the inevitable limitations on trialists' ability to engage with all aspects of complex community dynamics. Power {#Sec5} ----- The notion of "community" can often hide its heterogeneous nature and constant, underlying negotiations and contestations of power. The commonly used notion of 'community' used for example in public health campaigns and mobilization strategies can in fact hide complexities engendered by struggles over status, authority and economic resources. Questioning whose voice is heard, and the ways in which community 'leadership' is contested and in many ways produced by external interventions are at the heart of an understanding of communities as heterogeneous in and in flux. In Sierra Leone, Mariane Ferme has identified an accepted division between public and secret spheres, that is, a widely shared assumption that politics happens in different places, some public and some secret, and that covert strategies play a fundamental role in the working of politics \[[@CR40]\]. This means that political intentions are permanently ambiguous, something that Ferme sees as "one of the defining features of postcolonial subjectivity in Sierra Leone" (\[[@CR40]\], p.161). In what she defines as the "dialogics of publicity and secrecy", overt acts of obedience are often paralleled by covert defiance. This first case study shows how these complicated, contextual and often hidden configurations of power and struggles over economic resources can impact on a vaccine trial's mobilization strategy. The challenges posed by community power dynamics became clear from the start of the engagement process. Community engagement for the EBOVAC Salone study began two months before the start of the trial with a series of community meetings, often chaired by various local authorities. The meetings were attended by the community liaison team, who presented information about the trial and who were ready to answer questions from the audience, and by the social science team, whose role was to observe, take notes and provide support if required. The meetings brought together groups of people identified as "key stakeholders", as representatives of various important societal groups, such as area chiefs, market traders, teachers and civil society activists. At the meetings, stakeholders were provided with information about the trial and were given time to ask questions, raise concerns and offer suggestions. A few days after one of the community meetings about the EBOVAC Salone Trial, the social science team was approached during daily ethnographic observations by a small group of 'stakeholders' who had attended one of the meetings. In exchange for the promise of absolute confidentiality, they asked that their concerns be heard. They reported that, while they had been asked in the EBOVAC Salone public meetings to discuss the vaccine trial with their communities and constituencies through smaller local meetings they felt unable to do so. They asked to meet at a separate time to explain their concerns. The next day the team met the group at one of their homes, and they began telling their story. They recounted how they felt that their position as respected leaders in their community had been undermined and that, as a consequence, their ability to mobilise their communities to participate in the trial was curtailed. They attributed the erosion of their power to the 'selfishness' of those they identified as 'big' leaders in Kambia, whom they blamed for failing to share the financial resources coming into the district. They voiced an assumption that powerful leaders must have 'eaten' significant amounts of Ebola funds before it could reach their communities.[5](#Fn5){ref-type="fn"} Emphasising the correlation between financial power and ability to command respect in their communities, these 'small' leaders argued that due to their economic struggles, they found themselves having to resort to borrowing money from their community members, which diminished their ability to command respect. Regardless of the veracity or legitimacy of these allegations, these assumptions about the role of powerful authorities were widely reported during ethnographic research and they created significant amounts of discontent and quiet mistrust in authority figures in Kambia, as evidenced by the 'small leaders' private confessions. Despite dancing and clapping at public meetings, these 'small' leaders argued that they were unable and unwilling to act as spokespeople for local matters such as the vaccine trial, because they felt that their communities no longer respected them. Although it was unclear whether this was in fact the reason for their decision to refrain from supporting the vaccine trial through the organisation of area meetings, the community leaders utilised these conversations to voice their dissatisfaction with the present structure of authority. Their arguments point to the importance of taking into account what are essential, but often hidden and unspoken, contestations of power within communities. The "dialogics between publicity and secrecy" exemplified by the reluctance to hold public meetings despite their public assertions of support, offers a salient example of the broader importance of taking into account the heterogeneity of community. Whilst community engagement meetings appeared to be received positively in Kambia, such manifest support could not be assumed to translate into community-wide acceptance. Insights such as those from the "small" leaders reveal that established power structures are not indisputably representative, nor are appointed leaders necessarily trusted by all sections of the population. Internal struggles over status and economic resources, as well as mistrust and disputes over rightful loci of authority must be taken into account. This cognizance of the contested nature of power undoubtedly creates its own challenges. Undoubtedly, having key and recognised stakeholders at the EBOVAC Salone community meetings was extremely valuable and its significance for a large number of Kambians cannot be underestimated. However, recognising that power is not always straightforward and that communities are fragmented is an important foundation for building more nuanced, sensitive and genuine engagement. Ethnographic observation's illustration of the deep-rooted contestations of power put forward by the "small" leaders and the political economy underpinning them, cannot be directly addressed by a community engagement team in a clinical trial. However, the social science and community liaison teams' ability to identify such dynamics made it possible to initiate internal discussions about how to reconcile the importance of involving established authorities while diversifying engagement strategies to reach different sections of the community. This meant for example holding targeted sensitisation sessions in areas such as markets and popular meeting places, and holding radio talk shows in addition to key community leader-led meetings. Fairness {#Sec6} -------- Notions of fairness are inevitably shaped by local understandings of morality, historical legacies and individual experiences \[[@CR41]\]. In Sierra Leone, conversations around fairness often revolve around the widespread assumption that access to resources relies on having a strong *sababu*, that is, a connection with people in positions of power. Lacking a *sababu* is frequently cited as a key reason why large numbers of Sierra Leoneans live in poverty, with limited access to jobs and basic healthcare \[[@CR42]\]. Strong critiques of "connectocracy", furthermore, are paralleled by assumptions that nepotism is the only way to access resources. Beliefs about fairness were also central to the unfolding of the ten year civil war, for example as discontent surrounding chiefly abuse of powers in terms of forced community labour demanded of young men, featured prominently in combatants' accounts of their motivations for joining rebelling forces \[[@CR43]\]. In clinical research the notion of fairness emerges at several stages of the research, from concerns about equal treatment of control and treatment groups, to issues of compensation and the addressing of potential medical complications. In the initial stages of trial design, which this paper focuses on, engaging with local ideas of fairness was an especially important in trialists' discussions surrounding the design of a volunteer recruitment strategy. The senior investigators and trial team expected the issue of participant recruitment in EBOVAC Salone to be challenging, especially in an epidemic context, in a country with little experience of clinical trials. Consequently, the design of a recruitment strategy was discussed at length, combining senior trialists' expertise of vaccine trials in other contexts with conversations with local staff and anthropological insights. As Ezekiel et al. have pointed out: fair selection of the study population is a key benchmark of ethical clinical research in developing countries \[[@CR44]\]. As such, the trial team had to consider not only their own understandings of what would be an effective trial design that would ensure that the results it provided would be scientifically valid and representative of the community as a whole, but also how the Kambia community would perceive the selection of volunteers. Initial discussions centred on a number of design options, including individuals coming directly to the clinic to volunteer or having a key authority figure canvass for volunteers. Given the context discussed above, these options raised concerns that the trial would be liable to complaints of unfairness, as well as the fact that the enrolled participants would be likely to be unrepresentative of the population as a whole. Firstly, given assumptions that access to resources is assumed to be based on "connectocracy", there was the potential that, given the limited number of participants required, people could have assumed that the "big ones" were picking themselves and those they knew. On the other hand, using key authority figures to compile a list of volunteers, or asking key stakeholders to be the first volunteers could have led to accusations of forced participation. Given Sierra Leone's history of conflict, the potential for participant recruitment to be likened to forced conscription also added sensitivities for the trial recruitment process. In consideration of these concerns, and after extensive consultation with the principal investigators, trial sponsor and the trial manager, the community liaison team opted for a public lottery of household numbers followed by individual visits through which people would be offered the opportunity to volunteer in the vaccine trial. This involved inviting community members to observe and take part in the selection of 100 households (to begin with) from a bag containing all household numbers in Kambia town.[6](#Fn6){ref-type="fn"} These households were then visited individually by community liaison officers who explained in detail the purpose and procedures of the trial and voluntary nature of the process and provided these randomly selected households with the opportunity to volunteer. Notions of fairness, like power, are contested and there are limitations in the extent to which any design can avoid criticism. For example, as initial fears surrounding the new vaccine trial subsided, some community members expressed worries that the lottery system would not be able to taking into account those who were eager to take the vaccine.[7](#Fn7){ref-type="fn"} However, as one civil society activist noted, using a lottery system meant that "*bad name done komot dae*" (the trial has avoided a bad reputation).[8](#Fn8){ref-type="fn"} Had it not been done through a public lottery, she argued, allegations of unfairness would have been widespread.[9](#Fn9){ref-type="fn"} Trust {#Sec7} ----- A major challenge in preparing for this vaccine trial in Sierra Leone during the Ebola epidemic was the high levels of fear and mistrust which, as discussed, had develop through a history of oppressive rule and conflict. The epidemic was accompanied by a plethora of stories and rumours that starkly exposed the lack of confidence in government authorities, medical practitioners, and external agencies \[[@CR6], [@CR45], [@CR46]\]. This was made clear by the widespread belief that Ebola was a man-made, population control strategy in view of the next Presidential elections; that people were killed inside ambulances by being asphyxiated by chlorine; or that new cases were fabricated in order for Ebola response workers to profit from the protracted crisis \[[@CR6], [@CR47]\] These rumours have too often been treated as a sign that populations are "misinformed" at best, "ignorant" at worst, and are thus simply in need of better information in order to change their behaviour \[[@CR18]\]. However, these rumours reflect broader anxieties rooted in a much deeper socio-political context, and explain resistance to the Ebola response at the height of the epidemic. Anxieties surrounding the government's and international partners' plans reveal fractures in citizens' trust in the healthcare sector, which is seen as corrupt and inefficient, and reflects the fragility of the social contract rebuilt in the aftermath of Sierra Leone's civil war. Understanding public perceptions, rumours and concerns in this fragile context and creating multiple forums for dialogue and trust-building, were an essential foundation for more trial-specific community engagement. As communities learned about the planned vaccine trial in Kambia, the ethnography revealed that concerns surrounding the Ebola outbreak and the response to it were initially transposed to ideas about what the trial would involve. One evening, when the social science team discussed perceptions of the vaccine trial with a group of young men, one of them said he had a concern he had previously been ashamed to share. He said that he had heard that there was a "world blood bank" that was lacking type O blood. He argued that survival at the Ebola treatment centres had been determined by one's blood type, as those with type O blood were killed for their blood. Survivors were those whose blood type was not needed by the world blood bank and who were thus allowed to return to their communities alive. He reported that he had been afraid during the Ebola outbreak because he knew his blood type was O. Having heard in several community meetings that blood would be taken during the process of the vaccine trial, he asked whether this was actually a continuation of the blood stealing that he believed had been going on since the outbreak of Ebola. In other ethnographic encounters in Kambia's key congregation areas, the social science team frequently recorded links being drawn between the epidemic and the vaccine study by referring to the impending trial as "Ebola Phase Two". As with the young man's concern about blood stealing, this title referred to the concern that the trial might be a plan to "finish off" those who had managed to escape death during the outbreak. As one trial participant put it when discussing the rumours he had heard that "\[the vaccine\] is a slow poison the white man has made to kill us".[10](#Fn10){ref-type="fn"} While maintaining a commitment to the confidential nature of research conversations, the social science team alerted the community liaison team when these fears and articulations of mistrust emerged. The community liaison team responded by visiting areas such as the town's market where the idea of the vaccine being "Ebola Phase Two" had taken root. Being aware of the particular nature of anxieties surrounding blood-donation in the context of the Ebola outbreak meant that the community liaison team encouraged debates and conversations through community meetings, one-on-one conversations and radio shows to discuss the role of blood taking in the vaccine trial. They provided explicit information about the destination and use of blood samples and the fact that samples could be destroyed if the participant so wished once the study was over. They also invited questions, challenges and suggestions. Representatives of different societal groups who had previously attended community engagement events hosted their own meetings, which the community liaison team attended as guests, encouraged open and often heated debates, and also enabled the creation of spaces for expressing and confronting anxieties rather than simply rejecting them as misinformation. In another example, community meetings and ethnographic encounters brought up the issue of the insurance that would be provided for participants in case of any long-term side-effects from the vaccine. This had sparked worried conversations about the likelihood that if the trial provided insurance it meant the trialists expected people to die in the process. Concerns identified by the social science team were fed back to the community liaison team. The team developed a strategy to help allay fears around this rumour, including a new message about the provision of insurance that also emphasised the safety of the vaccine. An influential individual in that particular community, who had worked with the community liaison team, volunteered to assist with disseminating the message and dispelling the rumour through discussions in the area where the rumour had emerged. Such rumours represent more generalised concerns about medical interventions; they are not simple misunderstandings but are rather rooted in histories of exploitation and mistrust. Therefore, whilst immediate messaging was seen to be important, the main role of the teams was to develop communication to understand community concerns, and the drivers behind these concerns, and to work to build trust through ongoing active and inclusive dialogue. Discussion {#Sec8} ========== Carrying out vaccine trials in the context of an infectious disease epidemic with high mortality in a developing county recovering from years of internal conflict brings significant challenges. The EBOVAC Salone trial was set up during the Ebola outbreak as affected populations were trying to make sense of the disease and its devastating impact on families and communities. In addition, the epidemic has laid bare, and sometimes exacerbated, mistrust in the healthcare system, elected officials and external health interventions. In order to understand some of the concerns surrounding the trial, it was crucial that the teams took into account how the Ebola epidemic has affected social relations and power as well as local ideas about the Ebola response, including interactions with Western medicine and experiences of community engagement. Even more broadly, it was important for the two teams to work together to engage effectively with the community dynamics and power constellations highlighted by anthropological research. Such understandings of the community in which the trial was being set up have helped the teams to ensure that communities and participants are given an opportunity to voice concerns and to work with them to address mistrust. Knowledge of contested power, for example, allowed the team to listen to the voice of "small leaders" and to take into account their potential to disengage from the trial. This paper has offered illustrations of how social science and community liaison team worked in concert to shape engagement strategies and volunteer recruitment mechanisms. The examples discussed here show the value of research-driven communications and offer important lessons for future trials. Firstly, they demonstrate the importance of real-time social science research in the setting up of a vaccine trial. Social science researchers can act as independent "eyes and ears of the trial", listening to fears, concerns and suggestions. Secondly, an in depth understanding of community power dynamics highlights the importance of diversifying communication methods and avenues for engagement. As shown by the concerns of the "small" community leaders, contestations of power can be muted and hidden but can nonetheless affect the relationship between the vaccine trial and the community. As such, using a variety of communication channels, rather than relying solely on established leadership, can ensure messages reach different sectors of the community. Thirdly, an understanding of community dynamics and local social norms through ongoing dialogue can help inform the set-up of a trial from the start as evidenced by the establishment of a public lottery for volunteer recruitment. Finally, a focus on listening to and understanding rumours revealed deeper concerns about health interventions stemming from histories of mistrust, rather than simply being "misunderstandings". Conclusion {#Sec9} ========== This paper reveals that lessons learned from ethnographies of other trials about political and social inequities, social relations and rumours and concerns have been heeded. However, they also highlight very specific and local concerns in the context of the Ebola epidemic, which can only be revealed by on-going and "close to the ground" social science research and effective and dynamic community engagement. Notes {#Sec10} ===== More information on the EBOVAC Trial can be found at: [http://www.ebovac.org](http://www.ebovac.org/) and <https://clinicaltrials.gov/ct2/show/NCT02509494?term=EBOVAC&rank=1>. COMAHS : College of Medicine and Allied Health Sciences LSHTM : London School of Hygiene and Tropical Medicine The vaccine being used as the active control is the WHO-prequalified Meningitis ACWY vaccine. This is given as a prime dose at Day 1 and a placebo as the boost dose, except in children under the age of 2 years who are given 2 doses of the active control vaccine. Chinese gunpowder tea Palm wine Commercial motorbikes These interpretations, and the expectations that they raised, must also be understood in the context of the traditional practice of paying 'kola', a 'greeting gift', to authorities and gatekeepers, such as chiefs in order to begin community consultations, as well as the widespread payment of traditional authorities and land custodians in order to facilitate business, such as in the mining industry \[[@CR48], [@CR49]\]. Before carrying out a lottery, the social science team mapped Kambia Town using Global Positioning System (GPS) technology and produced a list of all households in the town, as this was not available at the time. Community Focus Group Discussion, Kambia, 23 November 2015 Key Informant Interview, Kambia, 19 October 2015 The fact that there may have been accusations of unfairness seems to suggest a particular eagerness (or the expectation of eagerness) to participate amongst Kambia residents in the initial stages, and during the epidemic. This begs the question of what people's motivation for joining may have been, what perceived incentives may play a role and whether certain selection biases may have been at play. Determining what factors influence participants' decisions to enroll in the EBOVAC trial is a key component of the social science research taking place throughout the running of the trial. This reflects the contribution that qualitative research can make by highlighting potential ethical issues as they arise during a clinical research project. Exit Interview, Kambia, 20.10.2015 The authors would like to acknowledge invaluable research assistance by Mahmood H. Bangura, Kadiatu Bangura, Rosetta I. Kabia and Mohamed L. Kamara. In addition the authors would like to thank the EBOVAC Community Liaison team: Abdul T. Deen, Isata D. Kamara, Mohamed Kamara, Mohamed T. Sesay, Hassan Jalloh, Suad Koroma, Adama Conteh, Kadie Allieu. Funding {#FPar1} ======= This project has received funding from the Innovative Medicines Initiative 2 Joint Undertaking under grant agreement No 115854. This Joint Undertaking receives support from the European Union's Horizon 2020 research and innovation programme and the European Federation of Pharmaceutical Industries and Association. Availability of data and material {#FPar2} ================================= In order to protect the privacy of research participants, the raw qualitative data collected for this project will not be publicly available. Authors' contributions {#FPar3} ====================== LE, SL, ES, AT, and TM were involved in the conception, design and drafting of the paper. LE and AT were involved in the data collection and data analysis. SL, BL, BG, DWJ, HL were involved in critical revision of the paper and in giving final approval for submission. All authors read and approved the final manuscript. Competing interests {#FPar4} =================== The authors declare that they have no competing interests. Consent for publication {#FPar5} ======================= Not Applicable. Ethics approval and consent to participate {#FPar6} ========================================== This research was approved by the London School of Hygiene and Tropical Medicine Ethics Committee and the Sierra Leone Ethics Board. Written consent was sought from participants interviewed. Where participants were illiterate, consent was carried out with an impartial witness' signature complementing the participant's thumb-print. All interviews have been anonymised.

Introduction {#S1} ============ In intentionally unvaccinated individuals, the incidence of vaccine preventable diseases (VPDs) is higher ([@B1]--[@B4]). The vaccination coverage in young adults for the 2016--2017 season is 33.6%, well below the national target of 70% vaccination coverage ([@B5]). University campus clinics are dealing with an increased incidence of VPDs ([@B6]--[@B9]). Despite over two centuries of vaccine availability ([@B10], [@B11]) and annually approved vaccination schedules for children and adults ([@B12]) the resurgence in incidence of VPDs has grown substantially ([@B3]), placing at direct risk individuals both vaccinated and non-vaccinated. Recent campus outbreaks ([@B13]), coupled with efforts ([@B14]) to highlight vaccine successes, brings to the forefront the longstanding college community concern about vaccination requirements on university campuses. Individuals can choose to not be vaccinated by filing a vaccination exemption request. Typically, the exemption requests are submitted for young children by their parents. While there is no law on individual vaccination status, certain institutions such as public schools, and professions such as the military require an updated immunization status. Every state has well developed criteria for allowing a vaccination waiver. The criteria are broadly grounded in religious, philosophical, or medical reasons ([@B15]). In some states, such as Michigan, there has been an increase in the number of vaccination waivers among K-12 students ([@B16]). This makes the population at direct risk to contract VPDs those students who were vaccine exempt due to the signing of a vaccination waiver on their behalf as children ([@B17]--[@B19]). As those who were under or unvaccinated as children now enter college, especially those college programs that do not require an updated vaccination status ([@B20]), they pose an increased risk of disease susceptibility ([@B21]) by living and interacting in close quarters, leading to the emergence and reemergence of serious infectious disease outbreaks ([@B22]). The empirical literature regarding the perception toward vaccination among those students who may or may not have been exempt from vaccinations as children is limited, and not much is known about their perception toward vaccinations. The objectives of this study were to characterize the sociodemographic characteristics of current university students with and without vaccination waivers as children and identify their personal perceived benefits, barriers and influencers of vaccination. Findings from this study may help university campus clinic programs to create targeted perception-based vaccine messages. Additionally, the findings could also guide the development of program practices geared toward catching-up those students who were vaccine exempt or not up to date on their vaccinations. Materials and Methods {#S2} ===================== Study Sample {#S2-1} ------------ In this cross-sectional study that took place from fall 2015 to spring 2016, we recruited students at all campus locations of a public university in the rural Midwest of the United States. The university does not require vaccinations for students except for in certain medical programs. The survey was sent to all the university students. A survey reminder email was sent to non-respondents two weeks after the original email. The invitation emails included background information on the study and the importance of student participation, the names of the researchers, and a link to complete the survey. The email also stated that by clicking the link to complete the survey, they were providing informed consent. The survey was administered electronically by the University Administration using the QuestionPro platform and de-identified data was provided to the authors for study analysis. The study was approved by the Ferris State University Institutional Research Board. Instrument {#S2-2} ---------- The survey consisted of 15 questions. Drawing on the Health Belief Model, the instrument was designed to identify students' perceived benefits, barriers, and influencers of vaccination ([@B23]), and measures were drawn from relevant empirical literature. Vaccination benefits were those identified by the World Health Organization and include effective control against disease, prevention of the onset of disease, protection of the unvaccinated community, savings of time and money otherwise lost to disease, and safety for those giving and receiving the vaccinations ([@B24]). The barriers included in the study were personal adverse experiences or that of a parent or guardian, the associated out of pocket costs, fear/pain of needles, lack of transportation, risk of adverse event greater than perceived benefit, and for moral or religious reasons ([@B25], [@B26]). To identify the cues to action ([@B27]), a construct of the Health Belief Model which identifies ways to activate willingness to engage in healthy preventive behaviors, we identified influencers such as educational material, low cost, ease of access, safe to use and administer vaccines, follow-up on compliance, and vaccination status of parent/guardian ([@B28]). Other variables on the instrument included current vaccination status, which was the primary outcome of interest and sociodemographic characteristics such as gender, age, academic college in which enrolled, financial assistance status, current level of education, education level of parent/guardian, individual insurance status, regular access to primary care provider, and vaccine counseling and concerns related to vaccination leading to autism. We selected these sociodemographic characteristics based on available literature regarding vaccination trends ([@B26], [@B28], [@B29]). Vaccination waiver status was also measured. This question refers to having had a waiver signed for any vaccination as required for school, college, or employment. Statistical Analysis {#S2-3} -------------------- Percentages for each level of the sociodemographic characteristics statistics were computed for the respondents with and without the vaccination waivers. Chi-square tests were run to test for associations between sociodemographic characteristics and the vaccination waiver status. Three questions required respondents to rate their perceived benefits, barriers, and influencers on a Likert scale of least important (rank = 1) to most important (rank = 5). Since the barriers, influencers, and benefit questions on the survey instrument were measured on a Likert scale the median values, which serves as an appropriate measure of central tendency in order to fairly characterize the responses from the instrument, are reported ([@B30], [@B31]). In addition, the results of the Mann--Whitney *U*-tests, which are appropriate for testing differences in median values between two groups ([@B32]), are also reported. Cronbach α on all scales that measured the benefits barriers and influencers of vaccination from the instrument are reported. All statistical tests were performed using SAS 9.4. Results {#S3} ======= Participant Characteristics {#S3-1} --------------------------- A total of 1,089 students from the university student population of 14900 students responded to the survey. Of these 1,089 respondents, we dropped 121 respondents as they had an incomplete response to the vaccination waiver status question, which was the primary outcome of interest in our study; and another 4 respondents were dropped because they were minors, resulting in a final sample of 964 respondents. Approximately 9% (*n* = 79) of the 964 participants responded "Yes" to having had a vaccination waiver. Of the respondents who had a vaccination waiver, 72% (*n* = 57) were females, of which 43% (*n* = 34) were in the age group of 18--21 years, and about 31% (*n* = 24) of them were from the College of Health Professions. Over 30% (*n* = 24) of them had a parent/guardian whose highest level of education was a Bachelor's degree and over 72% (*n* = 57) had not experienced any change in their insurance status that could have affected their access to vaccines in the last three years. Statistically significant chi-square tests were detected for whether the respondent was up to date on vaccinations \[χ^2^(2) = 75.62; *p* \< 0.001\], thinks that resurgence in VPDs is related to decline in vaccination rates \[χ^2^(3) = 59.05; *p* \< 0.001\] and thinks that vaccinations can cause autism \[χ^2^(3) = 60.53; *p* \< 0.001\]. These findings are represented in the Table [1](#T1){ref-type="table"}. ###### Participant characteristics by vaccination exemption status. Category Waiver status = yes (*n* = 79; 8.20%) Waiver status = No/unsure (*n* = 885; 91.80%) ------------------------------------------------------------------------------------------------- --------------------------------------- ----------------------------------------------- **Gender** Male 22 (27.85%) 266 (30.06%) Female 57 (72.15%) 614 (69.38%) Transgender 0 (0.00%) 4 (0.45%) Missing 0 (0.00%) 1 (0.11%) **Ages** 18--21 34 (43.04%) 340 (38.42%) 22--25 15 (18.99%) 243 (27.46%) 26--29 6 (7.59%) 93 (10.51%) 30--33 6 (7.59%) 57 (6.44%) 34--37 4 (5.06%) 40 (4.52%) 38 or older 14 (17.72%) 112 (12.66%) **Academic college** Arts and sciences 9 (11.39%) 153 (17.29%) Business 13 (16.46%) 125 (14.12%) Education and human services 14 (17.72%) 80 (9.04%) Engineering technology 3 (3.80%) 84 (9.49%) Health Professions 24 (30.38%) 210 (23.73%) Kendall college of art and design 6 (7.59%) 52 (5.88%) Michigan college of optometry 1 (1.27%) 13 (1.47%) Pharmacy 8 (10.13%) 143 (16.16%) University college 1 (1.27%) 20 (2.26%) Missing 0 (0.00%) 5 (0.56%) **Which of the following forms of financial assistance do you currently receive?** Governmental scholarships (TIP, etc.) 21 (26.58%) 202 (22.82%) Ferris Scholarships 34 (43.04%) 423 (47.80%) Other scholarships (community, church, rotary, etc.) 12 (15.19%) 132 (14.92%) Reserve Officers' Training Corps (ROTC) 0 (0.00%) 2 (0.23%) Student Loans 46 (58.23%) 592 (66.89%) Family Support 22 (27.85%) 286 (32.32%) Not Sure 1 (1.27%) 8 (0.90%) None of the above 12 (15.19%) 115 (12.99%) **Current level of education** Undergraduate 58 (73.42%) 626 (70.73%) Graduate/First Professional 21 (26.58%) 256 (28.93%) Missing 0 (0.00%) 3 (0.34%) **Highest level of education completed by a parent/guardian** High school 16 (20.25%) 243 (27.46%) Associate's degree 14 (17.72%) 204 (23.05%) Bachelor's degree 24 (30.38%) 256 (28.93%) Graduate level degree (Master's) 16 (20.25%) 126 (14.24%) Terminal degree (Ph.D., MD., etc.) 5 (6.33%) 30 (3.39%) Unsure/prefer not to answer 4 (5.06%) 25 (2.82%) Missing 0 (0.00%) 1 (0.11%) **Experience regarding access to physician and vaccine counseling** Had REGULAR access and was counseled by primary care provider 54 (68.35%) 597 (67.46%) Had REGULAR access and was NOT counseled by primary care provider 7 (8.86%) 113 (12.77%) Had LIMITED access and was counseled by primary care provider 5 (6.33%) 27 (3.05%) Had LIMITED access and was NOT counseled by primary care provider 2 (2.53%) 48 (5.42%) Had NO access to primary care provider but received vaccine counseling 0 (0.00%) 4 (0.45%) Had access to primary care provider but never received vaccine counseling 2 (2.53%) 37 (4.18%) Not sure about access to vaccine counseling 9 (11.39%) 58 (6.55%) Missing 0 (0.00%) 1 (0.11%) **Currently up to date on vaccinations** **\*\*\*** Yes, currently up to date 51 (64.56%) 752 (84.97%) No or unsure, but plan to be updated 5 (6.33%) 94 (10.62%) No, and do not plan to be updated 23 (29.11%) 38 (4.29%) Missing 0 (0.00%) 1 (0.11%) **Changes in insurance coverage over the last 3 years affecting access to vaccines** Yes, I now have insurance 6 (7.59%) 87 (9.83%) Yes, I have improved insurance 7 (8.86%) 59 (6.67%) Yes, I no longer have insurance 1 (1.27%) 46 (5.20%) No, there have been no changes to my insurance 57 (72.15%) 554 (62.60%) Not sure 8 (10.13%) 134 (15.14%) Missing 0 (0.00%) 5 (0.56%) **Think resurgence in vaccine preventable diseases is related to decline in vaccination rates** \*\*\* Yes, for sure 25 (31.65%) 536 (60.56%) Yes, probably/not sure, could be related 38 (48.10%) 313 (35.37%) Not at all related 16 (20.25%) 29 (3.28%) Missing 0 (0.00%) 7 (0.79%) **Think vaccinations can cause autism** **\*\*\*** Yes, for sure 13 (16.46%) 21 (2.37%) Yes, probably/not sure, could be related 36 (45.57%) 243 (27.46%) Not at all related 30 (37.97%) 615 (69.49%) Missing 0 (0.00%) 6 (0.68%) *Chi-Square test for Ho: there is no difference between samples by vaccination exemption status; all numbers rounded to two decimal places. \*p \< 0.05; \*\*p \< 0.01; \*\*\*p \< 0.001*. Distribution of Responses to the Benefits, Barriers, and Influencers of Vaccination {#S3-2} ----------------------------------------------------------------------------------- The distribution of responses are reported in Table [2](#T2){ref-type="table"}. The biggest identifiable benefit was effective control against disease. The out of pocket cost associated with vaccination was the most reported barrier and vaccines that are safe to use and easy to administer was the most important influencer of vaccination. ###### Distribution of responses to the benefits, barriers and influencers of vaccination. Benefits, barriers and influencers 1 = least important 2 = somewhat important 3 = neutral 4 = somewhat important 5 = most important Missing --------------------------------------------------- --------------------- ------------------------ -------------- ------------------------ -------------------- ------------ **Benefits** Effective control against disease 41 (4.24%) 25 (2.5%) 77 (7.99%) 172 (17.84%) 633 (65.66%) 16 (1.66%) Prevents onset of disease 46 (4.77%) 41 (4.25%) 100 (10.37%) 185 (19.19%) 573 (59.44%) 19 (1.97%) Protects the unvaccinated community 144 (14.94%) 84 (8.71%) 150 (15.56%) 118 (12.24%) 446 (46.27%) 22 (2.28%) Saves time and money otherwise lost to disease 137 (14.21%) 101 (10.48%) 131 (13.59%) 117 (12.14%) 456 (47.30%) 22 (2.28%) Safe to use and receive 61 (6.33%) 74 (7.68%) 98 (10.17%) 155 (16.08%) 557 (57.78%) 19 (1.97%) **Barriers** Personal/parent/guardian adverse experiences 560 (58.09%) 82 (8.51%) 85 (8.82%) 71 (7.37%) 118 (12.24%) 48 (4.98%) Out of pocket cost 442 (45.85%) 138 (14.32%) 128 (13.28%) 83 (8.61%) 131 (13.59%) 42 (4.36%) Pain or fear of needles 492 (51.04%) 127 (13.17%) 122 (12.66%) 65 (6.74%) 119 (12.34%) 39 (4.05%) Transportation and location 584 (60.58%) 111 (11.51%) 97 (10.06%) 65 (6.74%) 58 (6.02%) 49 (5.08%) Moral or religious reasons 764 (79.25%) 45 (4.67%) 32 (3.32%) 27 (2.80%) 39 (4.05%) 57 (5.91%) Risk of adverse event greater than benefit 517 (53.63%) 100 (10.37%) 95 (9.85%) 69 (7.16%) 130 (13.49%) 53 (5.50%) **Influencers** Educational materials 105 (10.89%) 61 (6.33%) 174 (18.05%) 177 (18.36%) 420 (43.57%) 27 (2.80%) Low cost vaccines 106 (11.00%) 81 (8.40%) 163 (16.91%) 176 (18.26%) 407 (42.22%) 31 (3.22%) Parent/guardian vaccination status 289 (29.98%) 132 (13.69%) 170 (17.63%) 115 (11.93%) 219 (29.72%) 39 (4.05%) Ease of access to vaccine programs and facilities 78 (8.09%) 74 (7.68%) 177 (18.36%) 211 (21.89%) 395 (40.98%) 29 (3.01%) Safe to use and easy to administer 61 (6.33%) 53 (5.50%) 114 (11.83%) 190 (19.71%) 516 (53.53%) 30 (3.11%) Follow-up on vaccine compliance 134 (13.90%) 136 (14.11) 222 (23.03%) 149 (15.46%) 286 (29.67%) 37 (3.84%) Benefits, Barriers, and Influencers by Vaccination Waiver Status {#S3-3} ---------------------------------------------------------------- The Cronbach α on all scales that measured the barriers, influencers and benefits ranged between 0.80 and 0.85, which is over the acceptable 0.70 as suggested by Nunnally and Bernstien ([@B33]). The five benefit and influencer questions were statistically different between respondents with and without waivers. The two barrier questions that produced significant differences between respondents with and without waivers were about personal or parent or guardian adverse experience and risk of adverse event greater than benefit. The median rating from each of the sections on benefits, barriers, and influencers are represented in Table [3](#T3){ref-type="table"}. ###### Median responses to the benefits, barriers and influencers for respondents with and without vaccination waivers. Benefits, barriers, and influencers of Waiver status = Yes (*n* = 79) Waiver status = No/unsure (*n* = 885) --------------------------------------------------- -------------------------------- --------------------------------------- **Benefits** Effective control against disease 4.00 5.00\*\*\* Prevents onset of disease 4.00 5.00\*\*\* Protects the unvaccinated community 3.00 4.00\*\*\* Saves time and money otherwise lost to disease 3.00 5.00\*\*\* Safe to use and receive 3.00 5.00\*\*\* **Barriers** Personal/parent/guardian adverse experiences 3.00 1.00\*\*\* Out of pocket cost 2.00 2.00 Pain or fear of needles 1.00 1.00 Transportation and location 1.00 1.00 Moral or religious reasons 1.00 1.00 Risk of adverse event greater than benefit 4.00 1.00\*\*\* **Influencers** Educational materials 3.00 4.00\*\*\* Low cost vaccines 3.00 4.00\*\*\* Parent/guardian vaccination status 2.00 3.00\* Ease of access to vaccine programs and facilities 3.00 4.00\*\*\* Safe to use and easy to administer 4.00 5.00\*\* Follow-up on vaccine compliance 2.00 3.00\*\*\* *Mann--Whitney U-test for Ho: there is no difference between samples by vaccination exemption status; all numbers rounded to two decimal places. \*p \< 0.05; \*\*p \< 0.01; \*\*\*p \< 0.001; the median response ranks on the Likert scale correspond to; 1 = least important, 2 = slightly important, 3 = neutral, 4 = moderately important, 5 = most important*. Discussion {#S4} ========== The results suggest statistically distinct underlying differences between vaccination perceptions of students with and without vaccination waivers. Of particular attention is the result that among the students who had received a vaccination waiver, close to one-third of them were not up to date on their vaccinations and did not plan to be updated. This could be explained as a reflection of the identified barriers that include out of pocket cost associated with the catch-up schedule ([@B34]) or the fear associated with the risk of adverse experience with vaccination schedules ([@B20]). Likewise, the proportion of those that are not sure if they think the resurgence in VPDs is related to decline in vaccination rates may be related to the transfer of commonly held negative perceptions associated with vaccine effectiveness ([@B35]). Despite the retraction of the study which incorrectly suggested that vaccinations lead to autism, it continues to remain a popular idea ([@B36]). In the San Diego, CA, USA measles outbreak started by an intentionally unvaccinated young adult boy, over 75% of the children were intentionally not vaccinated and one of their top reasons was the belief that vaccinations cause autism ([@B37]). Our study finding for whether the respondents think that vaccinations can cause autism aligns with other studies ([@B37]) and affirm that it continues to be a negative health belief, as observed in our study participants, and more so amongst those students who have had a vaccination waiver. Interestingly the comparison of groups with limited access and with or without being counseled by a primary care provider suggests an opportunity for campus clinics to educate students on vaccination benefits and subsequently address the concern of our study that adults who were under or unvaccinated as children are now entering college in a communal environment. The finding that 65% of students that had a vaccination waiver reported as being up to date on their current vaccination further reinforces the power of health education but may also be a reflection of the high number of respondents from the College of Health Professions that run programs that require them to be up to date on vaccinations. As demonstrated by their scores on the benefits of vaccination questions, respondents, with and without waivers, ranked vaccinations as an effective way to control and prevent the onset of disease as the highest ranked benefits. This finding aligns with another study among university students that found the number one priority of vaccination to be protection against diseases that are life threatening or that greatly impact the quality of life ([@B38]). This finding is important because it provides the university healthcare community with insight into the viewpoint of the reasons why a young adult population views vaccinations as beneficial. Hence, university health centers could devise vaccination campaigns that specifically focus on the benefits rated as most important in order to decrease vaccination waivers among students. The five benefit questions were statistically different from each other when comparing respondents with and without waivers. The respondents with waivers rated vaccination benefits lower than the respondents without waivers. This finding would indicate that those respondents with the waivers perceive fewer benefits of vaccination as those with no waiver. The barrier questions that produced significant differences between respondents with and without waivers both dealt with adverse events of vaccination. An informative comparison was the first barrier question that measured parent/guardian adverse experiences. This finding suggests that from a barrier perspective, the opinions of students tend to reflect their parental perceptions. Other studies also support this finding. For example, a random-effects model analyzed over 45,000 parent--child relationships and found a high correlation between parent and child attitudes and perceptions ([@B39]). However, intergenerational perceptions regarding vaccinations may be affected by race. A study regarding vaccination practices found that family traditions and the intergenerational transfer of health practices was an important source of vaccine confidence for whites, whereas for many African Americans, traditional family perceptions weakened vaccine confidence ([@B40]). Another study, particularly aimed at African American students at Howard University, found that intrapersonal factors were important when deciding whether the student would participate in a vaccination program ([@B31]). The study suggested that the optimal time to try to persuade a university student to obtain a vaccination is at the point of contemplation, during which many students understand the efficacy of the vaccine but find it difficult to obtain ([@B31]). The challenge for campus vaccination programs for students is to reign in negative intergenerational perceptions to vaccination and develop strategic, targeted messaging that inform, educate, and encourage students to confront the perception of adverse effects associated with vaccinations. Vaccine safety was identified as the most important perceived cue to action or influencer amongst the students in both groups. This finding is consistent with other studies ([@B41]) among university students and may suggest persistent fear, reflecting distrust in healthcare professionals. The questioning of the safety of vaccinations has given the anti-vaccination community a forum, and with the advent of websites and blogs this adverse view of vaccinations can spread quickly in a viral manner ([@B11]). This survey finding related to safety as the number one influencer of vaccination among those with and without waivers is another justification why not only vaccine receivers but even public health-care workers need to be properly educated on vaccination safety. A recent cross-sectional study found that there is still a major need for health-care workers to be properly educated on vaccination safety ([@B42]). This study suggests that improving public health worker education regarding safety of vaccinations may have the potential to decrease the number of vaccination waivers and subsequently increase vaccination coverage. An informative comparison among the influencers of vaccination found a significant difference among low cost vaccines between the two groups. This finding is of particular relevance to rural university communities, like that of our study population. For example, in our rural community, approximately 21% of the residents are categorized as poor with a median income is \$41,500. Being that the university is in a rural part of the country where the nearest metropolitan area is over 50 miles away, many students may rely on the student health center for all their preventative medical needs. Given this demographic, a community outreach program to assist students in understanding the out-of-pocket costs of vaccination and the coverage their insurance company provides for vaccinations could be useful. In addition, educating students about cost-free vaccination services that local health departments may offer and teaming up with local health departments to offer little to no cost vaccinations might be an effective way to lower the cost barriers for students. Limitations {#S4-1} ----------- We were able to meet the objectives of the study through the statistical procedures we used, and the large sample of current university students. The high Cronbach scores support the reliability of the study instrument ([@B33]) that assessed an array of benefits, barriers and influencers of vaccinations that were of interest to our study. Administering the survey online likely yielded more openness and full responses ([@B43]). The study was limited, however, in that, it was cross-sectional and only surveyed one university, thus results may not represent students in communities across the United States. In addition, the survey did not distinguish between under vaccinated and unvaccinated populations. This is a limitation because some individuals may choose to follow alternate vaccination schedules ([@B17]), leaving them temporarily under vaccinated. Up-to-date status on required vaccinations was not defined. This may have resulted in mixed results, for example based on whether or not an individual considered the HPV vaccination a requirement for up-to-date vaccination status. Another limitation with this question is that students may not know their vaccination status. Additionally, students were asked about changes in health insurance affecting access to vaccines. Many students are covered under their parents' policies and are unaware of their insurance status, resulting in misclassification bias. One last potential limitation of this study is non-response bias. The response rate of approximately 7% may not accurately represent the study population and limits generalizability. However, the results serve to build the body of literature on perceptions of college students toward vaccination, especially in those with a vaccination waiver. In the future, alternate recruitment methods would be considered to obtain a better representation of the population under study. The authors are extending the current study to identify factors that could affect vaccination waiver status among students. Conclusion and Implications {#S4-2} --------------------------- In the midst of the deepening confusion over vaccination safety and the resurgence of VPDs ([@B44]), the results of our study evidence distinct variation in perceptions toward vaccination amongst students who have had a vaccination waiver and those who did not have a vaccination waiver. Furthermore, the identified perception of benefits, barriers, and cues for action or influencers could inform education and information needs of vaccination efforts offered by campus clinics. The results of this study would enable a targeted, appropriate, and relevant approach to facilitate the students' ability to make an informed choice about vaccination safety, and subsequently, the uptake of vaccination services on university campuses and vaccination services offered by local health departments. Our study suggests that making vaccinations convenient, affordable, and accessible is vital for a university community, particularly, in a rural setting. For students who are uncertain about whether to get vaccinated, proper educational materials that dispel myths and misconceptions and further the understanding of the risks and benefits of vaccination should be distributed to target these individuals. Potentially, greater availability of such information among a university community would diminish barriers and assist students in reaching an affirmative decision to participate in a future vaccination campaign. Ethics Statement {#S5} ================ This study was approved by the "Ferris State University Institutional Review Board." The study was exempt from documentation of written informed consent. Author Contributions {#S6} ==================== EJ is the project principal investigator. He conceived the study plan and drafted and revised the draft manuscript. DW performed primary analysis and drafted the initial manuscript along with revising the draft manuscript. BA analyzed the data, wrote the statistical analysis, and revised the draft manuscript. All authors reviewed and approved the final manuscript as submitted. Conflict of Interest Statement {#S7} ============================== BA serves as a consultant/trainer for SAS, Inc. The remaining authors have no conflicts of interest to declare. **Funding.** This study was funded by a Ferris State University Office of Research and Sponsored Programs grant. [^1]: Edited by: Allen C. Meadors, Independent Researcher, United States [^2]: Reviewed by: Jo Ann Shoup, Kaiser Permanente, United States; Junfeng Wang, University of Illinois at Springfield, United States [^3]: Specialty section: This article was submitted to Public Health Education and Promotion, a section of the journal Frontiers in Public Health

Introduction ============ Sufficient vitamin D intake as well as adequate vitamin D synthesis in the skin is required to control calcium homeostasis a bone turnover. The effects of vitamin D on human health are diverse \[[@B1]\] but not yet fully investigated. However, vitamin D has been implicated in the risk of overall mortality \[[@B2]\], cancer \[[@B3]-[@B12]\], diabetes \[[@B13]-[@B15]\], musculoskeletal disorders \[[@B16]\], mental \[[@B17]\] and physical performance \[[@B18]\], hypertension \[[@B19]\], cardiovascular diseases \[[@B20]\], and autoimmune diseases \[[@B19],[@B21]\]. Although, many benefits of vitamin D are ubiquitously known, recommended intake (RI) and more importantly upper limits (UL) have to be considered to prevent adverse effects such as vitamin D intoxication. Intoxication may occur at 25(OH)D concentrations above 500 nmol/L \[[@B22]\], while 75 nmol/L are considered as adequate \[[@B23],[@B24]\]. In Germany, vitamin D intake from natural food sources \[[@B25],[@B26]\] as well as vitamin D synthesis in the skin is low \[[@B23]\], which subsequently leads to low 25(OH)D serum concentrations. In Germany this was once reported in a population study of Hintzpeter and co-workers \[[@B26]\] and is now detailed with a novel mathematical bottom-up model of 25(OH)D concentrations \[[@B27]\]. Building on this knowledge, the aim of our study was to develop a novel vitamin D fortification model, taking into account all vitamin D sources, different carrier products suitable for fortification and various fortification scenarios to fulfill requirements of risk considerations as well as both intake recommendations and 25-hydroxyvitamin D concentrations. Earlier fortification models have been published by Flynn and co-workers \[[@B28]\], Rasmussen and co-workers \[[@B29]\] or by Hirvonen and co-workers \[[@B30]\]. Models from Flynn and Rasmussen consider the safe upper limit for vitamin D fortification per energy unit. While in the Flynn model only vitamin D intake from natural food sources is considered as the basis for estimating the fortification levels, the Rasmussen model also takes into account the vitamin D intake from supplements. Whereas these two models give fixed values based on equations, the Hirvonen model developed the association between the risk of exceeding the UL and the fortification level. This is important for risk managers in order to decide on the acceptable risk. Our model, however aimed to combine advantages of previous models and add another, yet unconsidered, but significant aspect to vitamin D fortification modeling. It includes not only food intake from natural food sources and supplemental habits, but also vitamin D synthesis in the skin. In order to allow risk considerations, we defined various fortification scenarios for different dietary intake of natural food sources (5^th^ percentile, mean and 95^th^ percentile). Still, we have to mention that our study does not include estimates on an individual level, as it only considers average data. Methods ======= Model is based on three core dimensions --------------------------------------- A bottom-up model of 25(OH)D serum concentrations \[[@B27]\] as a function of sun exposure, food and supplements predicts and considers both vitamin D sources and vitamin D status of the average population in Germany. Output values are 25(OH)D serum concentrations of an average German individual for each month of the year and for each German federal state. For detailed description of the model and its results please see "New perspectives on vitamin D sources in Germany based on a novel mathematical bottom-up model of 25(OH)D serum concentrations" \[[@B27]\]. Having a detailed understanding of the contributing factors and the resulting 25(OH)D concentrations, one can develop new perspectives on food fortification scenarios for Germany. The fortification model depicted in Figure [1](#F1){ref-type="fig"} describes scenarios for a vitamin D fortification of bread, milk and juice. Zero point of each axis describes a minimum fortification scenario, while the fortification level rises with increasing distance from zero point. The maximum fortification scenario is indicated by the dotted cube. Depending on the chosen value of each axis, the size of the cube represents fortification intensity. This model does only stand for information for the average German population, but is not capable of considering fortification scenarios for individuals. {#F1} The x-axis -- vitamin D intake -- represents daily vitamin D intake through natural food sources as well as supplements \[[@B27]\]. While individuals of the 95^th^ (zero point) percentile have high vitamin D intake, individuals of the 5^th^ percentile (furthest from zero point) have low vitamin D intake. The more vitamin D the population consumes on average the less the food has to be fortified. The y-axis -- carrier foodstuff consumption -- plots consumption habits of the general population for foodstuffs, which this model considers to be fortified. Zero point of y-axis belongs to the 95^th^ (high carrier intake scenario) percentile of food consumers who consume large quantities of considered fortified carriers and the 5^th^ (low carrier intake scenario) percentile, furthest from zero point, belongs to those who consume very little of respective foodstuff. The average consumption of considered foodstuff in the population is represented by the "mean carrier intake" scenario. As there are differences between men's and women's nutritional habits, in all "high carrier intake" and in all "mean carrier intake" case scenarios, higher consumption volumes and thus lower fortification levels were used for reasons of conservative considerations. In all regarded carriers, men are those who consume more than women. Only in the "low carrier intake" scenario, we used consumption quantities of those, who consume less. For all carriers considered this means women. An exemption to this is milk as in the 5^th^ percentile of milk consumers, men consume less than women. The "low carrier intake" (5^th^ percentile) scenario means that 95% of the individuals of the population would have an additional vitamin D intake through fortified food, which lifts their intake and thereby their 25(OH)D serum concentrations to a targeted level (z-axis). The z-axis -- intake recommendation or recommended 25(OH)D level -- describes the recommended vitamin D intake or a 25(OH)D serum concentration to achieve. Zero point of z-axis refers to individuals, who tend to meet RI values of the Institute of Medicine (15 μg per day, IOM) and thereby reach 25(OH)D serum concentrations of 50 nmol/L which is defined as the lower value for adequate circulating 25(OH)D level by the IOM \[[@B31]\]. For the UL we considered people older than 8 years with an upper intake limit of 100 μg \[[@B31]\] or with a serum threshold of 75 nmol/L \[[@B23],[@B24]\]. Seasonally varying fortification introduced as alternative approach to constant fortification --------------------------------------------------------------------------------------------- This model is being developed in two approaches. The first approach aims to define a constant fortification (z-axis as intake recommendation) of different carrier foodstuffs throughout the year, as is common practice. For the constant fortification model, the vitamin D serum concentration model \[[@B27]\] was only used in parts. That is because 25(OH)D serum levels are not used as calculation base, but only vitamin D intake from natural food sources plus an average vitamin D intake from vitamin D supplements. To calculate constant fortification (f~c~) levels of carrier foodstuff, the difference (δ~i~) between recommended vitamin D intake (I~r~) and actual vitamin D intake through natural food sources and supplements (I~a~) is divided by intake of considered food to be fortified (F~i~), which are bread and milk as well as juice. This model can be easily adapted to all fortifiable food sources, but we only considered the three mentioned products. The underlying rationale of choosing these foodstuffs was on the one hand the goal to choose a carrier that is consumed by most people in Germany (here bread) and to choose carriers, for which there are many fortification experiences in other countries (here milk as well as juice). The more the population consumes considered carrier foodstuff (F~i~), the less will it be fortified. The higher the recommended intake levels (I~r~), e.g. the 20 μg recommendation by the German Nutrition Society (DGE) \[[@B32]\], the more will the carrier be fortified. f~c~ is calculated for all scenarios, which are determined by the different characteristics of the three axes depicted in Figure [1](#F1){ref-type="fig"}. The dimension of f~c~ is μg per 100 g of considered fortified foodstuff, which is why we include the normalizing factor of 100 in the function f~c~. For bread, only bread and rolls were considered. Pastries such as baked goods, cakes, cream pies or pizza as well as cereal and grain products were not considered here. Among the category milk we subsumed milk, milk mix drinks as well as milk products such as yoghurt or buttermilk. Juices contain all fruit juices and fruit nectars, but not vegetable juices and juice drinks such as apple juice spritzer. For detailed input parameter of the fortification model, see Table [1](#T1){ref-type="table"}\[[@B23],[@B24],[@B27],[@B31]-[@B34]\]. $$f_{c} = \delta_{i} \cdot \frac{100}{F_{i}} = \left( {I_{r} - I_{a}} \right) \cdot \frac{100}{F_{i}}$$ ###### Input parameters of the model **\#** **Parameter** **Value or comment** **Source** ---------------------------- ------------------------------------------------------------- ------------------------------------------------------------------------------------ ------------------------------------------- 1 **Vitamin D serum concentration model**     1.1  25(OH)D concentration Varies per month and per federal state of Germany \[nmol/L\]; average: 45 nmol/L Brown et al \[[@B27]\]. 1.2  Vitamin D intake through food Varies per gender; mean average men: 3.4 μg and mean average women: 2.8 μg per day Brown et al \[[@B27]\]. 2 **Carrier foodstuff consumption**     2.1 Bread Men/Women National Nutritional Survey II \[[@B34]\]   5^th^ percentile 46 g/43 g   Mean intake 180 g/134 g   95^th^ percentile 377 g/270 g 2.2 Milk Men/Women National Nutritional Survey II \[[@B34]\]   5^th^ percentile 16 g/22 g   Mean intake 222 g/203 g   95^th^ percentile 712 g/555 g 2.3 Juice Men/Women National Nutritional Survey II \[[@B34]\]   5^th^ percentile 0 g/0 g   Mean intake 270 g/232 g   95^th^ percentile 1,200 g/1,000 g 3 **Intake recommendation or 25(OH)D recommendation**     3.1 Intake recommendation       IOM 15 μg IOM \[[@B31]\]   DGE 20 μg DGE \[[@B32]\]   Upper Limit (UL) 100 μg IOM \[[@B31]\] 3.2. Recommended 25(OH)D conc.       IOM 50 nmol/L IOM \[[@B31]\]   Bischoff-Ferrari et al., 75 nmol/L Bischoff-Ferrari et al \[[@B24]\],   Domarus et al.   Domarus et al \[[@B23]\]. 4 **Others**     4.1  Conversion factor fortified food to 25(OH)D serum increase 2.32 nmol/L per 1 μg O'Donnell \[[@B33]\] The second approach aims to level 25(OH)D serum concentrations by varying the fortification amount per fortified foodstuff throughout the year (z-axis as 25(OH)D target level). For the varying fortification model, the vitamin D serum concentration model \[[@B27]\] was used as a whole. That is because 25(OH)D serum levels were used as calculation base. To calculate varying fortification levels f~v~ of the different carrier foodstuffs, the difference (δ~c~) between recommended 25(OH)D levels (L~r~) and actual 25(OH)D levels (L~a~) is divided by the conversion factor (c~f~) as well as by intake of considered food to be fortified (F~i~), which are bread and milk and juice. The conversion factor (cf) was derived from O'Donnell's review on efficacy of food fortification on serum 25-hydroxyvitamin D concentrations \[[@B33]\]. O'Donnell included not only milk-based fortified food, but also different dairy based products, bread and orange juice, which reflects the vitamin D carrier portfolio chosen in our study. O'Donnell reported conversion factors indirectly as she mentions base and final 25(OH)D concentrations after a certain additional vitamin D intake. Out of 9 trials, 7 provided sufficient data for calculation of the absolute mean change from baseline in 25(OH)D. We used these 7 trials for conversion factor calculation for fortified food. The normalizing factor 100 is used likewise. Actual levels (L~a~) are specific for each month of the year and are derived from the vitamin D serum concentration model \[[@B27]\]. $$f_{v} = \delta_{c} \cdot \frac{100}{c_{f} \cdot F_{i}} = \left( {L_{r} - L_{a}} \right) \cdot \frac{100}{c_{f} \cdot F_{i}}$$ To get an impression of the new ("n") 25(OH)D serum concentrations in case of vitamin D food fortification (L~nx~) for each month "x" of the year, we calculated resulting vitamin D concentrations as a function of actual ("a") (25)OHD concentration per month "x" (L~ax~) plus its rise due to consumption of fortified food (F~i~). As a monthly change of the fortification intensity is not practical we defined cluster of seasonally changing fortification levels (summer time and winter time). $$L_{\mathit{nx}} = L_{\mathit{ax}} + \frac{c_{f} \cdot F_{i} \cdot f_{v}}{100}$$ ### Software tools All models were calculated using Excel version 2007. Macros were programmed in Visual Basic version 6.5. Pictures were created using PowerPoint version 2007 and Think-Cell version 5.2. ### Ethics approval This study was approved by the local ethics board of the University Medical Center Hamburg. There was no need for further ethics approval as the study is only based on publicly available data (see Table [1](#T1){ref-type="table"}: Input parameters of the model). Hence there were no direct participants in our study which is why no written informed consent for participation in the study needed to be obtained. Results ======= Ideal fortification levels vary by underlying conditions -------------------------------------------------------- The Figure [2](#F2){ref-type="fig"}A/B portrays the vitamin D fortification level of 100 g of bread in the two approaches described in Figure [1](#F1){ref-type="fig"}. E.g. Figure [2](#F2){ref-type="fig"}A "mean carrier intake-IOM-mean scenario" (indicated as the dotted box in Figure [1](#F1){ref-type="fig"} and Figure [2](#F2){ref-type="fig"}A) means that 100 g bread has to be constantly fortified with 6.5 μg in order to provide someone who is an average bread ("mean carrier intake") and an average vitamin D consumer ("mean") with 15 μg a day ("IOM"). The average bread intake is approximately 180 g, the average vitamin D intake is 3.1 μg (for both men and women). 6.5 μg multiplied with 1.8 (180 g divided by normalizing factor 100) plus 3.1 μg and 0.3 μg of supplements (men and women averaged) is approximately 15 μg (differences due to rounding). Figure [2](#F2){ref-type="fig"}B follows the same calculation logic, but is not based on a constant fortification level, but on seasonal variations of 25(OH)D concentrations. Scenarios of Figure [2](#F2){ref-type="fig"}B thus aim to keep vitamin serum concentrations on a constant level throughout the year. E.g. Figure [2](#F2){ref-type="fig"}B "mean carrier intake-75 nmol/L-mean scenario" means that in January, 100 g bread has to be fortified with approximately 11.3 μg. Someone who consumes 180 g bread ("mean carrier intake") and who has an average vitamin D intake ("mean") thus has a daily vitamin D intake need of approximately 23.7 μg (20.3 μg plus 3.1 μg plus 0.3 μg) to yield a 25(OH)D concentration of 75 nmol/L. We calculated fortification levels not only for bread (Figure [2](#F2){ref-type="fig"}A/B), but also for milk and juice, Figure [3](#F3){ref-type="fig"}A/B. In Figures [2](#F2){ref-type="fig"}B and [3](#F3){ref-type="fig"}B, we indicated months, during which no fortification is required in order to guarantee targeted 25(OH)D concentrations. In case fortification levels drop below the zero level, the fortification level was set to level 0. These months were the base for defining cluster of seasonally changing fortification levels. {ref-type="fig"} depicts fortification levels for bread that are needed to either increase an individual's vitamin D intake to a recommended level (IOM, DGE or to the UL, *figure****A***) or to increase an individual's 25(OH)D concentration to a preferred level (50 nmol/L or 75 nmol/L, *figure****B***). The three columns represent the three scenarios of individuals with different carrier intakes (here: high, mean and low bread intake). **A.** Fortification levels for bread to meet intake values of nutritional guidelines of the IOM, of the DGE or to reach the UL. For definition of the "high carrier intake", "mean carrier intake" and "low carrier intake" scenarios see Figure [1](#F1){ref-type="fig"}. Please note the break in the "low carrier intake" scenario for the UL. The x-axis depicts recommended intake levels, while y-axis shows fortification levels in μg per 100 g bread. White (5^th^ percentile), grey (mean intake) and black shaded (95^th^ percentile) bars reflect the current vitamin D intake levels (natural food sources and supplements). **B.** Fortification scenarios to reach concentrations of either 50 nmol/L or 75 nmol/L. The x-axis depicts varying fortification levels throughout the year, while the y-axis shows fortification levels in μg per 100 g bread. The dotted line belongs to individuals who tend to reach a 25(OH)D concentration of 50 nmol/L and the solid line belongs to the 75 nmol/L goal. Below each graph, months are indicated, during which no fortification is required. White (5^th^ percentile), grey (mean intake) and black (95^th^ percentile) labeled graphs stand for the current vitamin D intake levels (natural food sources and supplements).](1475-2891-12-151-2){#F2} {ref-type="fig"} is analogous to Figure [2](#F2){ref-type="fig"} and depicts fortification levels of milk and juice that are needed to either increase an individual's vitamin D intake to a recommended level (IOM, DGE or to the UL, *figure****A***) or to increase an individual's 25(OH)D concentration to a preferred level (50 nmol/L or 75 nmol/L, *figure****B***). The x-axis depicts recommended intake levels, while y-axis shows fortification levels in μg per 100 g of the respective carrier. White (5th percentile), grey (mean intake) and black shaded (95th percentile) bars reflect the current vitamin D intake levels (natural food sources and supplements). Shown here is only the scenario for individuals with a mean carrier (milk and juice) intake.](1475-2891-12-151-3){#F3} Effects on 25(OH)D concentration serve as basis for risk assessment ------------------------------------------------------------------- The modeled effect of vitamin D food fortification on 25(OH)D serum is shown in Figure [4](#F4){ref-type="fig"}A and Figure [4](#F4){ref-type="fig"}B. Figure [4](#F4){ref-type="fig"}A depicts the scenario of people who have an average vitamin D intake from food and supplements as well as an average consumption of the carrier foodstuff. The two graphs in Figure [4](#F4){ref-type="fig"}A are equal for all fortified foodstuffs, as our goal was to set 25(OH)D serum concentrations to the same level for each carrier product. In some months, the right graph shows no difference between old and new serum concentrations, as there is no fortification from May to September (see Figure [2](#F2){ref-type="fig"}B and [3](#F3){ref-type="fig"}B). For risk assessments, Figure [4](#F4){ref-type="fig"}B aims to assess the effect of a fortification level based on average carrier intake on people with extreme dietary intake of vitamin D as well as extreme dietary intake of the carrier product. Driving factors for these extreme estimates are high vitamin D intake (95^th^ percentile) and high fortified foodstuff consumption (95^th^ percentile) for the upper limit as well as low vitamin D intake (5^th^ percentile) and low fortified foodstuff consumption (5^th^ percentile) for the lower limit. It becomes obvious that bread has the most stable effect on 25(OH)D serum concentrations among the three foodstuffs when estimating to the upper or to the lower limit. Considering the upper limit, the effect of fortified food overrides all other vitamin D sources. This especially holds true for milk and juice. On the other side the lower limit considerations (5^th^ percentile) show almost no change, as this group of the population consumes almost none of respective foodstuffs. The effect of different consumption habits in the general German population for these three carriers is shown in Figure [4](#F4){ref-type="fig"}C. It shows vitamin D intake multipliers due to differences between 5^th^ percentile intake, mean intake and 95^th^ percentile intake. Very prominent is vitamin D intake multiplier of juice when comparing the "low carrier intake" with the "high carrier intake" scenario, which is due to the fact that the 5^th^ percentile of juice consumers is near 0 g per day. Most resistant against a change of the scenario is bread, which is due to the fact that the spread between low bread and high bread consumption is smaller than with any other considered carrier. {#F4} Discussion ========== We compared the conventional approach of constant food fortification with a new one that takes into account seasonal variations of 25(OH)D concentrations. To our knowledge this is the first model that is able to find a fortification level, which is needed to either lift an average individual of the German population to a certain predefined 25(OH)D serum concentration or to raise the intake of an individual to a recommended intake. For means of risk assessments, this model considers several scenarios to estimate upper, mean and lower fortification levels for individuals with different intake. The novelty of the model in hand is based on the fact that it considers 25(OH)D serum concentrations rather than only intake recommendations and thus aims to level the 25(OH)D level throughout the year. Our model is programmed in a way that it can be easily adapted to all countries and all vitamin D carriers as long as input parameters are available for respective nations. Although the fortification model is based on simple mathematics, some aspects of the method of the model (4.1), the assumptions and input parameter (4.2) as well as the results (4.3) remain up for discussion and have to be further validated. The method of the model ----------------------- Our approach is different to previously published models from Rasmussen et al. \[[@B29]\], Flynn et al. \[[@B28]\] or Hirvonen et al. \[[@B30]\] in two ways. First, these models add a specific level of vitamin D per 100 kcal. Our model adds a specific level of vitamin D per 100 g of food, but also considers intake of the German population. Due to the combination of intake in gram and fortification levels in vitamin D per 100 g, the calculation results in a similar logical approach, as our model is likewise able to define fortification levels for people with low, mean or high vitamin D intake. Second, our model not only considers fortification levels to meet certain intake levels, but also takes into account seasonal variations of 25(OH)D levels due to sun exposure. When making risk assessments, the second reason might be considered a shortcoming of our approach, as vitamin D synthesis in the skin may override the effects of nutritional and supplemental intake \[[@B35]-[@B37]\]. It could be thus raised to question, whether upper limit considerations are meaningful in the second approach. Opinions are divided concerning the contribution of sunlight as influencing factor for 25(OH)D serum concentrations. Diffey et al. \[[@B35]\] state that the sun may make up to 56% percent during summer times, while Shariari et al. \[[@B36]\] report that sun may contribute to vitamin D concentrations by more than 90%. As only average sun exposure habits are available as input parameter for Germany, risk assessment statements in the seasonal variations approach may be put up for debate. The assumptions and input parameter ----------------------------------- Our model is based on a set of input parameter, see Table [1](#T1){ref-type="table"}. While some parameters such as recommended intake levels are non-country specific, some parameters are, e.g. foodstuff consumption. When adapting our model to other countries the availability of these country specific parameters are a key prerequisite. Nevertheless, data availability might be a challenge in some countries. We made every effort to refrain from input assumptions wherever possible. Nonetheless, an element of uncertainty remains the conversion factors of fortified food. In our model we used the systematic review of O'Donnell et al. \[[@B33]\] that determines the effects of vitamin D--fortified foods on serum 25-hydroxyvitamin D 25(OH)D concentrations, because the carrier products assessed in O'Donnell's study match the vitamin D carrier portfolio chosen in our study. Yet these conversion factors have been subject of various discussions. Other researchers such as Vieth \[[@B38]\] claim that the conversion factor is lower at around 0.5-1.5 nmol/L per 1 μg vitamin D. However, our model is designed in a way that recalculation based on adapted input parameters, e.g. a lower conversion factor, can easily be performed. The results ----------- The results of our fortification model (see example calculation in chapter "results"), stating that an average individual needs approximately 23.7 μg a day to reach a concentration of 75 nmol/L, are in line with other observations \[[@B7],[@B11],[@B24],[@B39]-[@B44]\]. Other publications are of similar statements, proposing 25 μg to obtain an adequate serum 25(OH)D in the absence of UVB irradiance \[[@B45]\] or to raise the level by up to 25 nmol/L \[[@B46]\], which is comparable to our results. However the intake needed per day to reach a concentration of 75 nmol/L remains controversial since other publications suggest required intake levels of 40 μg per day and higher \[[@B47],[@B48]\]. Furthermore it is important to note that determining fortification levels based on an average individual's behavior implies that only a certain proportion of the entire population reaches the desired serum concentrations. If the goal is to lift the serum concentration of almost the entire population to a desired level, fortification levels have to be set much higher. However, in this case risk implications gain in importance. Concerning food products to be fortified, one can argue, whether bread is a suitable carrier for vitamin D as there are only few, but promising experiences \[[@B49],[@B50]\]. However from a nutritional point of view in Germany, bread makes sense in a couple of dimensions. Bread is a basic and a perishable foodstuff in Germany and it is the only food that does not show a consumption decline in the elderly population \[[@B34]\], which is of special importance to prevent osteoporosis. All age categories and all social classes consume bread and the difference between mean intake and the 95^th^ percentile is low compared to other potential vitamin D carriers \[[@B34]\]. This makes the amount of vitamin D intake through fortified foodstuff controllable. Additionally, bread is not a peak product such as juice (frequently consumed during some seasons, like summer time) that could potentially boost vitamin D concentrations to a maximum due to increased intake \[[@B34]\]. Hirvonen et al. \[[@B30]\] also show that bread is an efficient vitamin D carrier when looking for a solution to reduce the proportion of people with low vitamin D intake and which is safe in avoiding the risk of exceeding the UL. Still it remains open, whether vitamin D fortified bread alone can be the solution to alleviate vitamin D deficiency in Germany, as some studies show that food fortification with vitamin D is more efficient when a wide variety of foods are fortified with a low concentration \[[@B30],[@B51]\]. The risk of overdose is higher for those, who consume larger quantities of certain foods, when only some foodstuff is fortified with high vitamin D concentrations \[[@B30]\]. The more food is fortified with lower concentration, the less likely is overdosing, as nobody can consume high quantities of all foodstuff that is fortified \[[@B30]\]. This is also in line with Välimäki and co-workers \[[@B51]\]. Considering the two other proposed foodstuffs to be fortified, milk as well as juices are common carriers for vitamin D \[[@B52]-[@B56]\], which however does not necessarily make these products suitable for fortification in Germany. A reason against milk and juice as carriers is the fact that the quantity spread of consumption for these foodstuffs is rather high \[[@B34]\]. In Finland, for example, this holds true for young women, who are not reached by the current milk fortification policy \[[@B57]\]. Bottom estimates (5^th^ percentile) in Figure [4](#F4){ref-type="fig"}B shows almost no difference for milk and juice, as the 5^th^ percentile almost consumes nothing of those carrier products. This does not hold true for bread as even the 5^th^ percentile consumes at least some bread. Top estimate (95^th^ percentile), reflects the quantity spread in consumption habits, especially for milk and juice. This is subsequently reflected in massive 25(OH)D concentration increase. One has to mention that these extreme estimates reflect very unlikely scenarios. However, these estimates are useful for risk considerations as they represent the maximum 25(OH)D concentration increase. Regardless of the fortification strategy and its potential beneficial impact on the health of the general population, one has to keep in mind two things. First, food fortification per se is not allowed in Germany. There are only few exemptions allowed for general fortification. Among them are margarine, blended fat products as well as dietary food products. Second, vitamin D food fortification poses the risk of a vitamin D intoxication, though it appears to have been caused by excessive vitamin D fortification of dairy milk \[[@B58]-[@B60]\]. Furthermore, intoxication is not the only risk, which might go in hand with vitamin D food fortification. Although the therapeutic window for a safe supplementation of vitamin D is extremely wide, some groups could be at risk. The body regulates the biologic activation of cholecalciferol through control of 1α-hydroxylase activity \[[@B22]\]. This, however, does not apply for the safe supplementation of the active hormone (calcitriol) for example for people with chronic kidney disease, as the therapeutic window is relatively small here \[[@B61]\]. Conclusions =========== We compared the conventional approach of constant food fortification with a new strategy that takes into account seasonal variations of 25(OH)D concentrations. We managed to show that bread as carrier product may be a suitable base. In terms of risk management, however, bread alone is probably not sufficient, as the risk of overdose with a single fortified product is higher than the risk with several fortified carriers \[[@B30],[@B51]\]. Our model is programmed in a way that it can be easily adapted to all countries and all vitamin D carriers as long as input parameters are available for respective nations. To our knowledge, our model with its approach is unique and may help many countries, where the population is prone to vitamin D deficiency and which are searching for a strategy to improve the vitamin D status of their population to realize associated benefits \[[@B62]\]. General vitamin D intake and respectively 25(OH)D concentrations of the German population is low. A possible reason might be that food fortification is still prohibited in Germany. With this novel model in hand, it is possible to conceive vitamin D fortification strategies for different foodstuffs and model its impact on 25(OH)D concentrations. We propose to critically discuss the strategy of constant food fortification and show considerations for a seasonal variation of food fortification to balance 25(OH)D concentrations on an certain level. Abbreviations ============= DGE: German nutrition society (Deutsche Gesellschaft für Ernährung); IOM: Institute of Medicine; RI: Recommended intake; UL: Upper limits. Competing interests =================== The authors declare that they have no competing interests. Acknowledgements ================ The authors thank Dr. Scheidt-Nave, Dr. Mensink as well as Dr. Hintzpeter for their helpful comments and critical discussions.

INTRODUCTION {#s1} ============ Human hepatocellular carcinoma (HCC) is one of the most common and morbidity cancers worldwide. In China, patients with liver disease such as chronic hepatitis are at the highest susceptibility for developing HCC, which accounts for 55% of new HCC cases worldwide \[[@R1]\]. Hepatitis B and C virus are the most risk factors for HCC, especially for the tight relationship between hepatitis C virus and high HCC incidence \[[@R2]\]. The traditional therapeutic methods for HCC mainly include surgery, chemotherapy and radiotherapy et al, which are feasible for patients who are in the early phase, but not for most patients with advanced stage of liver cancer \[[@R3]\]. Thus, the traditional therapeutic methods are difficult to acquire desired effect for advanced or metastasis cancer. It is urgent to explore the novel treatment strategies for improving HCC patients\' survival and living quality. Recently, oncolytic adenovirus represents a great promise drugs to treat human cancers. So far many targeted strategies based on oncolytic adenovirus were designed and showed potent antitumor activity in various preclinical studies \[[@R4]--[@R6]\]. Our previous designed cancer targeting gene-virotherapy strategy (CTGVT), which combined the superiority of gene therapy and virotherapy, will bring the new hope for tumor therapy \[[@R5], [@R7]\]. Based on CTGVT, the novel oncolytic adenovirus system, ZD55 (ZD55-gene), was engineered through deleting adenoviral E1B55-kD viral protein \[[@R8]\]. It was previously reported that the adenovirus mutant *dl1520* (also named ONYX-015) with E1B55-kD deletion could preferentially target and lyse p53-dysfunctinal tumor cells but not in the adjacent normal cells \[[@R9]\], however, further studies denied this view point and proved that the adenovirus mutant can enhance the viral mRNA late nuclear transport and oncolysis for tumor selectivity \[[@R10]\]. ZD55 system was similar with ONYX-015. It not only can selectively replicate in cancer cells and kill them, but carry exogenous antitumor gene \[[@R8]\]. Preclinical data showed that ZD55-gene exhibited significant antitumor effect in multiple types of cancer models whether in tumor cell lines or in mice models through the oncolytic action of virus itself and increased expression level of the carried antitumor gene \[[@R4], [@R11], [@R12]\]. However, ZD55 lacks the targeting ability for specific tumor type such as liver cancer. Thus, to improve the specific killing effect of oncolytic adenovirus on one type of cancer, one common strategy to design oncolytic adenoviruses is to use cancer or tissue-specific promoter to control the expression of viral essential gene for replication, which is the transcriptional targeted strategy \[[@R13], [@R14]\]. It causes the viral gene selectively expression in tumor cells, then the virus could only replicate in and kill tumor cells \[[@R7], [@R15]\]. Besides advanced therapeutic strategy for HCC, more important factor for improving the cure rate of HCC patients is early diagnosis. Fortunately, the current early diagnostic technologies were greatly improved by the diversified serum marker, image modalities, and histologic detection, which led to the outstanding prognosis \[[@R16]\]. GOLPH2, a Golgi membrane glycoprotein GP73, is one of glycoprotein discovered in recent years. Many results demonstrated that GP73 is an excellent marker for HCC diagnosis, and its sensitivity and specificity are better compared with the common liver cancer marker α fetoprotein (AFP), which reach 75% and 97% separately, while 58% and 85% for AFP \[[@R17]--[@R19]\]. In previous study, the tumor-targeting gene-viral therapy was performed by oncolytic adenovirus-mediated the transgene gene expression regulating by AFP promoter and proved certain efficacies in HCC model \[[@R20], [@R21]\]. Due to the outstanding character of GOLPH2, we attempt to identify the liver cancer targeting and therapeutic efficiency of GOLPH2-regulating oncolytic adenovirus for cancer gene-viral therapy. The novel GOLPH2-regulated oncolytic adenovirus GD55 was first designed, in which endogenous E1A promoter was replaced by GOLPH2 promoter to regulate E1B- 55kD- deleted ZD55. It is unreported in the present studies. Meanwhile, we also constructed the adenovirus GD55-EGFP carried green fluorescent protein (EGFP). The experimental results *in vitro* showed that the GD55 has the better specificity of antitumor proliferation ability than that of ZD55, and exhibits the targeting antitumor effect in HCC cells with the lesser side-effect to liver normal cells. Further animal experiments *in vivo* showed that GD55 has good suppression effect on liver cancer growth in xenografted HCC mice. In conclusion, the study has successfully screened the specific GOLPH2 promoter core region for HCC, and first constructed oncolytic adenovirus vector GD55 for targeting HCC. The preliminary results indicated that GD55 has excellent liver cancer specific and acts as the candidate of the individual targeting cancer gene-viral therapy for HCC patients, which lay on the foundation for future clinical liver cancer individual therapy. RESULTS {#s2} ======= Identification of GOLPH2 promoter and its high activity in liver cancer cells {#s2_1} ----------------------------------------------------------------------------- The 2.6 kb fragment upstream of GOLPH2 gene was first cloned into pGL3-basic named by p-2618/-19 by Dr. Peng, which indicated higher fluorescent intensity compared with control sequence in the EGFP reporter construct, and exhibited potent promoter activity in transient transfection assays \[[@R22]\]. We first detected the activity of long GOLPH2 promoter p-2618/-19 in liver normal epithelial cell QSG-7701 and hepatocarcinoma cell lines Huh7, Bel-7404, Hep3B with luciferase reporter assay. It was verified that all the hepatocarcinoma cell lines showed significantly higher GOLPH2 activity despite some variation compared with the normal cell line (Figure [1B](#F1){ref-type="fig"}). Among them, Huh7 cells displayed the highest GOLPH2 promoter activity. To map the minimal core promoter region and avoid potential regulatory inhibitory regions in 2.6 kb GOLPH2 promoter, we generated a series of promoter deletion truncations by PCR (Figure [1A](#F1){ref-type="fig"}) and subcloned into pGL3-basic, then measured their transcriptional activity in Huh7 cells. As shown in Figure [1C](#F1){ref-type="fig"}, the 659 bp truncation of the minimal core promoter p-677/-19, spanned from -677 to -19 bp, exerted the highest promoter activity. Subsequently, we measured the activity of the 659 bp truncation in various HCC lines, and found that the promoter activity is obviously higher in HCC lines than that of normal cells (Figure [1D](#F1){ref-type="fig"}). The HCC cell-specific expression of GOLPH2 protein was detected by western blotting analysis (Figure [1E](#F1){ref-type="fig"}). These results proved that the GOLPH2 promoter could be used to regulate the oncolytic adenovirus and drive the expression of exogenous genes specifically in hepatocarcinoma cells. {#F1} Detection of GOLPH2 expression in HCC specimens {#s2_2} ----------------------------------------------- The HCC specimens and the paracancerous liver (PCL) tissues were collected from 12 patients after undergoing clinical operation, and GOLPH2 expression was detected by immunohistochemistry analysis. The expression of GOLPH2 was positive in all HCC specimens, including strongly positive, moderately, and weakly positive (Figure [2A](#F2){ref-type="fig"}). Nevertheless, GOLPH2 expression in the PCL tissues was significantly weakened compared with HCC tissues, further indicating the liver cancer specificity of GOLPH2 expression. {#F2} Construction of GOLPH2-regulated oncolytic adenovirus GD55 and its replication in hepatocarcinoma cells {#s2_3} ------------------------------------------------------------------------------------------------------- Human GOLPH2 promoter is active in liver cancer cells, but low in normal liver cells (Figure [1](#F1){ref-type="fig"}), which makes GOLPH2 promoter a promising candidate for designing HCC-specific oncolytic adenoviruses. Here, we constructed a cancer-targeted dual-regulated oncolytic adenovirus GD55, which characterizes with the replacement of the normal E1A transcription regulatory elements using 659 bp truncation of GOLPH2 core promoter, and the deletion of E1B 55kDa gene. The genome of the recombinant adenoviruses was depicted in figure [3A](#F3){ref-type="fig"}. In GD55, E1B 55kDa gene was removed and similar with oncolytic adenovirus ONYX-015, which was reported to enhance the viral mRNA late nuclear transport and oncolysis \[[@R10]\]. {#F3} The essential of the oncolytic virus is able to selectively replicate in tumor cells and kill them. Thus, efficient viral replication and high progeny production can greatly contribute to the antitumor ability of oncolytic adenoviruses. Previously, we constructed the oncolytic adenovirus ZD55 with E1B 55kDa gene deletion, and proved its potent replication ability and antitumor effect *in vitro* and *in vivo* \[[@R8]\]. To compare the replication ability between dual-regulated virus GD55 and ZD55, liver cancer cells (Huh7, BEL7404, and Hep3B) and normal cells QSG-7701 were infected by indicated GD55 and ZD55. The results showed that higher expression of E1A in all GD55-infected liver cancer cells than that of normal liver cells (Figure [3B](#F3){ref-type="fig"}). Further, production of infective viral progeny was quantified by the TCID~50~ method in ZD55- or GD55-infected liver cancer cells. As shown in Figure [4A](#F4){ref-type="fig"}, liver cancer cells infected with GD55 produced higher titers of viral progenies than that of normal liver cells, and the titer of viral progenies in GD55-infected liver cancer cells is 10-time high than in the ZD55-infected cells. Interestingly, GOLPH2-regulated GD55 has higher viral replication efficiency than ZD55 (Figure [4B](#F4){ref-type="fig"}). Thus, the novel dual-targeted virus GD55 replicated more efficiently and produced greater amounts of progeny viruses than ZD55 in liver cancer cells. {#F4} GD55-mediated EGFP expression in liver cancer cells {#s2_4} --------------------------------------------------- In consideration of the specificity of E1A expression mediated by GD55, we also evaluate its ability as a vehicle for exogenous gene transfer. We constructed the recombinant virus GD55-EGFP and observed the tumor targeting ability of dual-regulated oncolytic adenovirus through specific EGFP expression in HCC cell line infected with SD55-EGFP, which was dose-dependent ways in 48 hr postinfection (Figure [5](#F5){ref-type="fig"}). Meanwhile, few QSG-7701 normal cells express green fluorescence in same infection. Therefore, the corresponding bright field showed the obvious appearance of cytopathic effect such cell rounding and detachment, signifying that GD55-EGFP caused more cell death in liver cancer cells, but had no significant cytotoxicity to normal cells. These findings indicated that the novel dual-regulated oncolytic adenovirus GD55 system effectively mediated the tumor-specific transgene expression and exerted the tumor cell cytotoxicity, but less in normal cells. {#F5} Tumor-killing effect of GOLPH2-regulated GD55 *in vitro* {#s2_5} -------------------------------------------------------- Further tumor-suppressing activity of the GOLPH2-regulated GD55 was investigated *in vitro* by MTT analysis. Three liver cancer cell lines and normal cell line QSG-7701 were infected with GD55 and ZD55 at various MOIs. The data of cell viability showed that GD55 had a higher antitumor capacity for hepatocarcinoma cells than that of ZD55 at a dose-dependent manner (Figure [6A](#F6){ref-type="fig"}). The cell growth suppressive rate reached about 75.31% in Huh7 cells infected by GD55. However, the novel oncolytic adenovirus GD55 caused slight cytotoxicity in normal liver cells. To further monitor the cytotoxicity of GD55, liver cancer Huh7 and normal QSG-7701 cells were infected with GD55 or ZD55 at various MOIs. After 2 days, the cytopathic effect was detected through staining adherent cells with crystal violet. The results showed that GD55 induced the stronger cytotoxicity in Huh7 cells compared with ZD55, while GD55 led to about 20-fold attenuation in killing QSG-7701 than in Huh7 cells (Figure [6B](#F6){ref-type="fig"}). Similar results were obtained from other liver cancer cell lines (data not shown). {#F6} The kinetics of cytotoxicity induced by GD55 also was evaluated in liver cancer cells and normal cells. As shown in Figure [7](#F7){ref-type="fig"}, the cytotoxicity effect of GD55 on the liver cancer cell lines was much more obvious than that of ZD55 with a time-dependent manner. There was little effect on normal liver cells infected by either GD55 or ZD55. Taken together, our data demonstrated that GOLPH2-regulated oncolytic adenovirus GD55 can eliminate liver cancer cells more effectively than common oncolytic adenovirus ZD55 *in vitro*. {#F7} Antitumoral efficacy of GOLPH2-regulated GD55 for liver cancer in an animal model {#s2_6} --------------------------------------------------------------------------------- Our *in vitro* experiments showed that GD55 could kill liver cancer cells. Further, we investigated the tumor-suppressing capacity of GD55 for liver cancer cells in nude mice. Figure [8](#F8){ref-type="fig"} showed that tumor volume of different treatments of PBS, ZD55, and GD55. The data indicated that GD55 significantly and completely inhibited the growth of Huh7 cancer xenograft, and the antitumor effect is comparable with ZD55, even ZD55 also exerted the excellent suppression effect of tumor growth. Moreover, the tumor volume of mice was 752 mm^3^ and 194 mm^3^ at 7 weeks after treatment with PBS and ZD55 mice, respectively, while tumor size in GD55-treated mice was greatly reduced. It was noteworthy that the tumor xenograft in three of eight mice completely disappeared in GD55-treated group in the seventh week. {#F8} The histo-pathological test by Hematoxylin and eosin staining showed that GD55 caused more profound cell death and symptoms of necrosis in tumor mass than ZD55, and that GD55 didn\'t lead to any obvious damage to liver tissue, which is essentially same to PBS-treated group (Figure [9](#F9){ref-type="fig"}). TUNEL assay indicated that GD55 treatment induced more extensive apoptosis in tumor tissue than in ZD55 or PBS treatment. IHC staining for CD31 also showed that vessel growth was significantly suppressed by GD55 treatment compared with ZD55 or PBS treatment, suggesting the antiangiogenesis effect of GD55. Further, we detected tumor cell apoptosis induced by GD55 with flow cytometric analysis. The percentage of apoptotic cells was determined by annexin V staining. As shown in Figure [10A](#F10){ref-type="fig"}, the Huh-7 cells infected with GD55 showed much higher percentage of cell apoptosis (25.9%), whereas, the lower percentage of cell apoptosis was observed in the liver cancer cells treated with ZD55 (8.05%) or PBS (1.0%) respectively. Morphologic observation of tumor tissue was also detected by TEM analysis (Figure [10B](#F10){ref-type="fig"}). The significant apoptosis was observed in tumor tissue treated with GD55, followed by ZD55 than PBS, as the characteristic of apoptosis, including nuclear collapse, the appearance of nucleus deformation and condensed chromatin in lumps at the inner side of nuclear envelope. The *in vivo* results suggested that an increased tumor-suppressing efficacy of GOLPH2-regulated GD55 for liver cancer than traditional oncolytic adenovirus ZD55, and the inhibitory effect on tumor growth was due to cell apoptosis and anti-angiogenesis. {#F9} {#F10} DISCUSSION {#s3} ========== Recent many clinic trials using oncolytic virus highlights the value and hope for cancer-killing viruses. Oncolytic virus has emerged as one of the most prospective anticancer drugs because the engineered viruses can successfully tackle cancer cells without damaging normal tissue \[[@R15]\]. Several excellent genetically modified oncolytic virus including vaccine virus JX-594, herpes simplex virus OncoVEX-GM-CSF, adenovirus H101, and reovirus Reolysin have been successfully used to treat various cancers in clinic and obtained specific cancer-killing effect \[[@R14]\]. In addition, oncolytic virus can induce the immune response against the tumor through releasing antigens when virus attacked cancer cells \[[@R23]\]. Decade ago, we first put forward to the cancer targeting gene-viral therapy strategy based on the combination of oncolytic virus therapy and gene therapy \[[@R11]\], and constructed the genetically modified adenvirus ZD55, showing the characteristic of cancer-killing virus \[[@R8]\]. But it is lack of specificity for some kind of cancer. In this study, we described a novel oncolytic adenovirus named GD55 based on ZD55, in which GOLPH2 promoter was used to restrict the E1A expression in liver cancer cells. Given that numerous papers published showed the GOLPH2 is a valuable diagnostic marker for hepatocellular carcinoma, GD55 could be a powerful anticancer agent to liver cancer from epithelium. Previous studies indicated high expression of GOLPH2 specifically in human hepatocellular carcinoma cells \[[@R24]\]. In consistent with them, our experiments proved that liver cancer cells tested, including Huh7, Bel-7404, and Hep3B, have higher GOLPH2 promoter activity than normal cells (Figure [1](#F1){ref-type="fig"}). GOLPH2-regulated GD55 induced more efficient adenovirus replication both in E1A expression and proliferation of virus progeny in liver cancer cells than that of E1B-55kDa deleted oncolytic adenovirus ZD55. These results may account for the efficient infection of GOLPH2-regulated GD55 for hepatocellular carcinoma in this study. In view of the high expression of GOLPH2 in liver cancer cells, we next observed the infective efficiency of GOLPH2-regulated adenoviral vector to liver cancer cells. There a significantly higher expression of EGFP gene in GD55-EGFP-infected liver cancer cells than in normal cells, indicating that GOLPH2-regulated adenoviral vector has a potent specifically infectivity for liver cancer cells. On top of this, just like ZD55′s \[[@R8]\]and BioVex\'s \[[@R25]\] products, the liver cancer-targeted adenoviral vector GD55 can also be armed with a panel of anti-cancer genes which would have the potential for treating HCC, such as GM-CSF that enhance the body\'s immune response against tumors, TRAIL or IL-24 for tumor inhibition and apoptosis, endostatin for anti-angiogensis, and antitumoral microRNA, and prodrug activating enzyme et al. We expected that GD55 carrying anti-cancer genes will obtain the synergy anticancer capacity for further liver cancer therapy. Simultaneously, our results showed that GOLPH2-regulated GD55 is more sensitive to liver cancer cells and exerted much more efficiently antitumor effect than that of simple oncolytic virus ZD55. Accordingly, animal experiments also confirmed that GOLPH2-regulated GD55 had an increased inhibiting ability of tumor growth for liver cancer cells in nude mice xenografts. Preliminary data showed that treatment of GOLPH2-regulated GD55 induced more apoptosis of tumor cells and on tumor tissues, which is consistent with studies on some other types of cancer-targeting oncolytic adenovirus \[[@R26]\], although the underlying detailed mechanism is still unknown. Besides, there are still some challenges that need to handle during the clinical trials of cancer targeting gene-viral therapy. Most of all, the immune system can be a double-edged sword, prone to attacking the cancer tissues as well as the oncolytic viruses \[[@R27]\]. Thus, how to avoid immune response to viruses is an important issue, such as transcriptional targeting strategy using cancer-specific promoter to restrict viral replication in cancer cells. This made the modified viruses to infect efficiently the tumor microenvironment and lead to prolong anticancer effect. Another hurdle is the generation of the drug-resistance to oncolytic viruses \[[@R28]\]. It is inspiring that researchers achieved the significant therapeutic efficacy through oncolytic viruses joining chemotherapy and radiotherapy in the arsenal of cancer treatments \[[@R15], [@R29]\]. Although mostly GOLPH2 be reported as a novel marker for hepatocellular carcinoma and chronic hepatitis B or C, there are still some attentions. Figure [2](#F2){ref-type="fig"} showed the differential expression levels of GOLPH2 in HCC patients, which implied that potential anticancer effect of GD55 on various types of HCC patients will vary with different activity of GOLPH2. A subset of patients with high GOLPH2 expression will respond to GD55. Thus it is essential to detect the GOLPH2 expression in HCC patients before using the GD55. Moreover, there are still some reports that GOLPH2 as a novel promising tissue biomarker for prostate cancer \[[@R30]\], gastric cancer \[[@R31]\], pancreatic cancer, AIDS progression \[[@R32]\]. This would enlarge the applied range of GOLPH2-regulated GD55 and implies for efficient gene-viral therapeutic effect to other cancers. In summary, we have successfully engineered oncolytic adenovirus GD55 regulated by GOLPH2 promoter, and have proved that use of GOLPH2 promoter can restrict the GD55 replication in the liver cancer cells. The GOLPH2-regulated GD55 efficiently inhibit the growth of liver cancer cells both *in vitro* and *in vivo*. With the progress of GOLPH2 as a promising marker for hepatocellular carcinoma, the novel dual-regulated oncolytic virus GD55 can become an effective anticancer agent candidate for future therapy of human hepatocellular carcinoma. MATERIALS AND METHODS {#s4} ===================== Cell lines and culture {#s4_1} ---------------------- The HEK293 cell line was obtained from were purchased from ATCC (American Tissue Culture Collection, Rockville, MD). Human normal liver cell line QSG-7701, Human hepatocellular carcinoma (HCC) cell lines Huh7, BEL-7404, and Hep3B were purchased from Shanghai Cell Collection (Shanghai, China). The HEK293 and QSG-7701 cells were cultured in phenol red-free Dulbecco\'s modified Eagle\'s medium (DMEM; Gibco BRL, Grand Island, USA) supplemented with 10% fetal bovine serum (FBS, Gibco BRL). The Huh7, BEL-7404, and Hep3B cells were grown in DMEM (GIBCO BRL) supplemented with 10% FBS. All cells were incubated at 37°C in a humidified air atmosphere with 5% CO~2~. Construction of human GOLPH2 promoter truncations and luciferase reporter assay {#s4_2} ------------------------------------------------------------------------------- A 2597 bp genomic fragment (−19 to −2618 bp) extracted from genomic DNA of Hela cells was inserted into pGL3-basic Firefly luciferase reporter vector (Promega), which was kindly presented from Dr. Tao Peng from GIBH, CAS \[[@R22]\]. Promoter deletions designated as p-2618/-19, p-729/-19, p-677/-19, p-613/-19, p-413/-19 were generated by PCR and subcloned into pGL3-basic plasmids. The luciferase reporter assay for various GOLPH2 promoter truncations was done according to the Dual-glo luciferase assay kit (Promega). Briefly, cells were seeded in 96-well culture plate overnight before transfection. GOLPH2 promoter plasmids (100 ng/well) were transfected into normal liver cell line QSG-7701, HCC cell lines Huh7, BEL-7404, or Hep3B with Lipo2000™ (Life Technology) together with pRL-TK (10 ng/well) as an internal control to normalize transfection efficiencies. After 36 h transfection, cells were harvested. The luciferase reporter assay was done according to the Dual-glo luciferase assay kit (Promega). All constructs were performed in triplicate and repeated three times. Results from one representative experiment were shown. Identification of GOLPH2 expression in HCC specimens {#s4_3} ---------------------------------------------------- HCC specimens and paracancerous liver (PCL) tissues were obtained from the First Affiliated Hospital of Wannan Medical University by surgical resection in March 2014. The patients, average age 51 years old (39--72), did not received chemotherapy or radiotherapy prior to surgery. The resected specimens were diagnosed pathologically as HCC. The specimens of HCC and PCL tissues were fixed in 10% formalin prior to prepare the paraffin-embedded sections, and the expression of GOLPH2 was detected by immunohistochemistry with the mouse anti-human GP73 antibody at a dilute concentration of 1:50 and the avidin-biotin-peroxidase complex reagent and diaminobenzidine tetrahydrochloride (DAB) solution (Vector Laboratories, Burlingame, CA). The sections were then counterstained with hematoxylin. The result of expression scores was determined by scoring the stained cell ratio and the stained intensity for each specimen \[[@R27]\]. Briefly, the five medium-power fields (20 × objective lens) were selected and the number of positive cells was counted under microscope. The stained intensity scores were decided as follows: staining identical to the negative control was defined as 0, while yellow, brown, and tan staining was defined as 1, 2, and 3, respectively. The stained scores of positive cell ratio were decided as follows: the score was 0 when all cells were negatively stained, the scores were 1, 2, and 3, respectively, if the proportion of the positively stained cells was 1/3 or less, 1/3 to 2/3, and 2/3 or more. Recombinant adenovirus construction and production {#s4_4} -------------------------------------------------- The adenovirus shuttle vector pSD55, pCA13-EGFP, pXC2, and adenoviral packaging vector pAdeasy are conserved in our lab. The oncolytic adenovirus plasmid pZD55 with E1B55-kDa gene deletion and oncolytic adenovirus ZD55 were constructed by our group previously \[[@R8]\]. The human GOLPH2 promoter (GP73) from pGL3 was further subcloned into Xho I and SnaB I site of plasmid pXC2 by replacing endogenous E1A promoter to generate the pXC2-GOLPH2. Then, the fragment GOLPH2-E1A obtained from pXC2-GOLPH2 with Xho I and Xba I enzyme and was inserted into pSD55 after the same site to form the adenovirus shuttle plasmid pSD55-GOLPH2, the double-regulated virus plasmid. The reporter gene EGFP expression cassette cut from pCA13-EGFP by Bgl II was inserted into pSD55-GOLPH2 using the same site, and the pSD55-GOLPH2-EGFP was constructed. All constructs were confirmed by restriction enzyme digestion and DNA sequencing. Generation, identification, purification and titration of adenovirus {#s4_5} -------------------------------------------------------------------- The oncolytic adenovirus GD55 and GD55-EGFP were generated by homologous recombination Escherichia coli strain, BJ-5183. Briefly, the shuttle plasmid pSD55-GOLPH2 or pSD55-GOLPH2-EGFP and adenovirus packaging backbone plasmid pAdeasy-1 were co-transfected into BJ-5183, and the recombinant adenovirus genome pGD55 or pGD55-EGFP was obtained. After the digestion with Pac I, pGD55 or pGD55-EGFP was transfected into HEK293 cells using Effectene transfection reagent (Qiagen, CA). After homologous recombination in HEK293 cells, the GOLPH2-regulated E1B 55kD gene deletion oncolytic adenovirus GD55 and GD55-EGFP containing enhanced GFP-expressing cassette were generated. Each recombinant adenovirus was isolated with plaque purification and amplified in HEK293 cells. The adenoviruses were harvested and purified by ultracentrifugation in a cesium chloride gradient. The virus titers were determined using the tissue culture 50% infectious dose (TCID50) method in HEK293 cells and calculated as plaque-forming units (pfu)/ml. Assay of reporter gene virus infection and viral progeny {#s4_6} -------------------------------------------------------- Normal cells and liver cancer cells were seeded in 24-well plates at a density of 5 × 10^4^ cells. Then, cells were infected with GD55-EGFP at different MOI (multiplicity of infection) of 1, 5, 10. 48 hr later, EGFP expression was observed under an Olympus fluorescence microscope with Olympus camera DP70. To determine the viral progeny, cells were seeded in 6-well plates and infected with GD55 or GD55-EGFP at a MOI of 5, After 5 hr of infection, cell medium was removed. Cells were washed with phosphate-buffered saline (PBS), and then 2 ml of fresh medium was added. 48 hr later, cells and medium were collected, and virus supernatant was collected by three freeze-thawing cycles and centrifugation. Virus yield was determined by TCID50 assay in HEK293 cells according standard protocol. Cell viability assay {#s4_7} -------------------- Various cells were seeded in 96-well plates at a density of 1 × 10^4^ cells/well and infected with viruses at the indicated MOIs and times. The cell viability rate was measured by the 3-(4, 5-dimethylthiazol-2-yl)-2, 5-diphenyltetrazolium bromide (MTT; Sigma Chemical Co., St. Louis, MO, USA) assay according the formula: cell survival=(absorbance value of infected cells-blank)/(absorbance value of uninfected control cells). Briefly, 20 μl MTT (5 mg/ml in PBS) was added into each well. 4 hr after cells incubated at 37°C, the mediums and MTT were removed, and 150 μl of Dimetyl Sulfoxide was added. The absorbance value at 595 nm was read with a DNA microplate Reader. Four replicate wells were done at each MOI, and every experiment was repeated three times. Cytopathic assay {#s4_8} ---------------- Normal cells QSG-7701 and human HCC cells Huh-7 were seeded in a 24-well plate. When cells were grown to a subconfluent level, and infected with different adenoviruses at the indicated MOIs. After 96 hr incubation at 37°C, the medium was removed and cells were stained with 2% crystal violet in 20% methanol for 30 min. Then, the plates were lightly washed with tap water, naturally dried and documented as photographs. Flow cytometric analysis {#s4_9} ------------------------ The Huh-7 cells were trypsinized after different treatments, and washed once with complete medium. Aliquots of cells (5 × 10^5^) were resuspended in binding buffer and stained with fluorescein isothiocyanate (FITC)-labeled annexin V and propidium iodide (PI) (Beyotime Biotech., China) according to the manufacturer\'s instructions. A fluorescence-activated cell sorting (FACS; Becton Dickinson) assay was immediately performed after staining. Western blot analysis {#s4_10} --------------------- Cells with different treatments were washed three times with ice-cold PBS, and lysed in buffer (50 mM Tris-HCl, pH8.0, 150 mM NaCl, 100 μg/ml PMSF, 1% TritonX-100) for 30 min at ice. Cells were collected and cell debris was removed by centrifugation. The protein concentration of cell lysates was measured by bicinchoninic acid (BCA) protein assay kit (Beyotime, China). Protein samples were diluted with sodium dodecyl sulfate/polyacrylamide gel electrophoresis (SDS-PAGE) loading buffer, and then separated by 10% SDS-PAGE and transferred onto 0.45 μm pore size nitrocellulose membranes (Millipore Corp., MA, USA). Then, nonspecific reactivity was blocked with 5% fat-free dry milk in Tween 20-containing Tris-HCL buffered saline overnight at 4°C. The membrane was incubated with primary antibody for E1A and GOLPH2 protein (Santa Cruz Biotech., USA). After incubation in the dark with IR Dye 800 conjugated IgG secondary antibodies (Rockland Inc., UK), immunodetection was performed by using an Odyssey infrared imaging system (LI-COR Biosciences Inc., Lincoln, NE, USA). The same membrane was then used for detecting the expression of GAPDH using primary antibody against human GAPDH (Santa Cruz Biotech., USA). Animal experiment {#s4_11} ----------------- All procedures for animal experiments were followed according to the guide for regulations and standards of Experimental Animal of the U.S. Department of Agriculture and the National Institutes of Health. Female BALB/c nude mice of 4--5 weeks old were purchased from the Animal Research Committee of the Institute of Biochemistry and Cell Biology (Shanghai, China). Huh7 tumor xenografts were established by subcutaneously inoculating 5 × 10^6^ Huh7 cells (suspended in 100 μl PBS) into right flank of each nude mouse. When the tumors reached 80--120 mm^3^, mice were randomly divided into three groups (*n* = 8). The tumor xenografts were intratumorally administrated with 100 μl of PBS with or without 2 × 10^8^ pfu of ZD55 or GD55. The injections were repeated three times every other day, and a total dosage 6 × 10^8^ pfu of adenoviruses per mouse. Tumor growth was monitored by periodic measurement using calipers. Tumor volume (V) was calculated by the formula: V (mm^3^) = length (mm) × width (mm)^2^/2. Tumors were harvested on day 4 posttreatment for histopathological and transmission electron microcopy (TEM) analysis. Animals were sacrificed when the diameter of tumors reached 2 cm. In this experiment, no mice with tumor xenografts were observed to die from tumor loading. Immunohistochemistry and TUNEL assay {#s4_12} ------------------------------------ Tumor tissues were fixed in 4% formaldehyde, dehydrated with gradient ethanol, and embedded in paraffin wax. Tissue sections (5 μm) were then dewaxed and rehydrated, were stained with hematoxylin and eosin. For immunohistochemical (IHC) analysis, deparaffinized tumor sections were treated with rabbit monoclonal anti-CD31 antibodies at 1:500 dilutions. The slides were then washed with PBS and incubated with goat anti-rab IgG antibodies-HRP polymers with 1:100 dilution. The slides were washed with PBS and incubated with the avidin-biotin-peroxidase complex reagent (Vector Laboratories, Burlingame, CA) and detected with DAB solution. The sections were then counterstained with hematoxylin. For terminal deoxynucleotidyl transferase-mediated dUTP-bioth nick end labeling (TUNEL) assay, an in situ apoptosis detection kit (Sino-American Biotechnology Co., Luoyang, China) was used in tumor tissue sections. Briefly, after incubation with proteinase K, tumor sections were dewaxed and rehydrated. Then, endogenous peroxidase was blocked with 3% H~2~O~2~, and sections were incubated with equilibration buffer and terminal deoxynucleotidyl transferase (TdT) enzyme. The sections were then incubated with antidigoxigenin-peroxidase conjugate. Peroxidase activity was detected by the staining of DAB solution. TEM analysis {#s4_13} ------------ Tumor samples (1 mm^3^) were fixed in a phosphate-buffered mixture with 2.5% glutraldehyde for 2 hr, and rinsed by PBS 15 min for three times, followed by 1 hr of fixation with 1% osmium tetroxide. After rinsed in 0.1 M phosphate buffer, the tumor samples were subjected to a graded series of dewatering treatment with ethanol and propylene oxide, and then embedded in pure acetone with embedding buffer at a ratio of 1:2 at room temperature for 3 to 4 hours, followed by embedded in pure acetone with embedding buffer at different gradient of temperature and time. The sections were suffered to double-stain with uranyl acetate and lead citrate, examined and photographed with a JEOL 100CX transmission electron microscope. Statistical analysis {#s4_14} -------------------- Absorbance value in the MTT assay and mean tumor volume are presented as mean ± SD. Statistical analysis was performed using a one-way analysis of variance (ANOVA) and compared at a given condition by two-tailed t test. The difference between data were considered to be statistically significant when *P* \< 0.05 (\*), to be very significant when *p* \< 0.01 (\*\*) and to be very much significant when *p* \< 0.001. We thank Dr. Tao Peng of Institutes of Biomedicine and Health of CAS for kindly providing long GOLPH2 promoter plasmid pGL3-L1. This work was supported by the National Nature Science Foundation of China (No. 81272687), Zhejiang province public welfare technology applied research projects (No. 2014C33275), Zhejiang Provincial Natural Science Foundation of China (No. LZ13H160004), Hi-Tech Research Development Program of China (863 Program, No. 2012AA020806), the Grant for 521 talent project of ZSTU, Zhejiang province health department (No. 2014RCA022), and Technology Bureau of Hangzhou (20130733Q36).

Background ========== Phosphatidylserine (PS) is a naturally occurring phospholipid present in the inner leaflet of mammalian plasma membranes. In humans, PS is most concentrated in the brain where it comprises 15% of the total phospholipid pool. PS has been shown to play a key role in the functioning of neuron membranes, such as signal transduction, secretory vesicle release and cell-to-cell communication \[[@B1]\]. The administration of PS extracted from bovine cortex (BC-PS) has positive effects on brain function. BC-PS was shown to improve learning and memory in age-associated memory impaired subjects \[[@B2]\], to enhance behavioral and cognitive parameters in geriatric patients \[[@B3]\], and to improve cognitive performance of Alzheimer\'s disease (AD) patients \[[@B4],[@B5]\]. Although the primary objective of clinical studies involving BC-PS was to test efficacy, no significant adverse events were reported following oral administration of BC-PS at doses of 300-400 mg/day for up to 3 months \[[@B2]-[@B8]\]. In the largest double-blind, placebo-controlled trial \[[@B3]\], comprising 494 participants, only one subject dropped out because of an adverse event, as compared to seven drop-outs from the placebo group. Due to safety concerns about potential contamination by bovine spongiform encephalopathy (BSE) prions in recent years, alternatives to BC-PS, such as soy derived PS (SB-PS), have been developed. SB-PS however, differs considerably in its fatty acid composition as compared to mammalian brain PS, and while SB-PS was shown to attenuate both physical \[[@B9]\] and mental stress \[[@B10],[@B11]\], further research is required to determine its ability to promote cognitive functioning \[[@B12],[@B13]\]. The safety of SB-PS was tested in a double-blind placebo controlled study on 120 elderly \[[@B14]\]. Hematological and biochemical parameters along with vital signs and adverse events were evaluated after 6 and 12 weeks of treatment. The study conclusion was that SB-PS is safe for administration to older persons if taken up to a dosage of 600 mg/day \[[@B14]\]. To gain the benefits of mammalian PS without the attendant risks, a safe, marine-sourced PS with omega-3 long chain polyunsaturated fatty acid (LC-PUFA) attached to its backbone was developed. This compound was recently found to improve the symptoms of children with impaired visual sustained attention \[[@B15]\] and to protect middle-aged rats from scopolamine-induced deleterious effects \[[@B16]\]. However, the biochemical and hematological tolerability of PS with omega-3 attached to its backbone have not been presented until now. In the present report we describe the safety of a novel formulation of PS with omega-3 LC-PUFA, mainly docosahexaenoic acid (DHA), attached to its glycerol backbone (PS-DHA), in non-demented elderly. The relevant safety parameters were derived from biochemical and hematological variables, examination of vital signs and adverse events monitoring during the 15 weeks double-blind placebo controlled study and from examination of vital signs and adverse events monitoring during the 15 additional weeks of the open-label extension phase. The efficacy of this novel formulation has also been tested in this population and the results of the double-blind phase are reported elsewhere \[[@B17]\]. Briefly, the study results indicated that PS-DHA may improve cognitive performance in non-demented elderly with memory complaints as PS-DHA administration resulted in an improvement of verbal immediate memory and in higher responders rates in comparison with the placebo group \[[@B17]\]. Methods ======= Subjects -------- Participants were recruited through advertisements in senior citizens homes, hospitals, and newspapers. Approximately 700 elderly were screened for enrollment to the study. Out of them 157 non-demented participants with memory complaints met the previously described inclusion criteria \[[@B17]\]. Briefly, eligible participants were non-demented men or women between the ages of 50 and 90 years, with complaints of memory loss \[[@B18]\] and no evidence of a condition that could produce cognitive deterioration including AD, Parkinson\'s disease, stroke, normal pressure hydrocephalus, and other brain lesions including tumors, renal, respiratory, cardiac, and hepatic disease, diabetes mellitus, endocrine, metabolic or hematological disturbances unless well controlled, and malignancy not in remission for more than two years. Concomitant use of drugs or supplements affecting cognitive function was prohibited. The study was conducted according to the principles of the Declaration of Helsinki and good clinical practice. The protocol was approved by the Ethics Committee of the Sourasky Medical Center, Tel-Aviv, Israel, and all volunteers gave written informed consent prior to participation. Study Design ------------ The study was designed as a single-center, randomized, double-blind, placebo-controlled, 15 weeks study, followed by an open-label extension of additional 15 weeks. At the first, double\--blind, phase, participants were randomized according to a computerized process based on 6 and 8 blocks, in a 1:1 ratio stratified by gender, to receive three capsules per day of PS-DHA or a matched identically looking placebo (cellulose). The daily PS-DHA dosage provided 300 mg PS and 79 mg DHA+EPA (DHA:EPA ratio of 3:1). During the second, open-label, phase, participants consumed one capsule a day of PS-DHA. The daily dosage provided 100 mg PS and 26 mg DHA+EPA. PS-DHA (Vayacog™) was supplied by Enzymotec, Migdal HaEmeq, Israel. Safety was evaluated by clinical laboratory assessments including biochemical and hematological parameters at baseline and endpoint of the double-blind phase and by adverse events recording, physical examination and measurement of vital signs and weight at baseline, week 7 and endpoint (week 15) of the double-blind phase and at the end of the open-label extension (week 30). Blood samples were analyzed by the American Medical Laboratories (AML), Herzliya, Israel. Adverse events were monitored and recorded at each visit and by telephone contact every other week. Laboratory Parameters --------------------- Biochemical parameters consisted of potassium, sodium, calcium, phosphorus, chloride, glucose, creatinine, blood urea nitrogen (BUN), bilirubin, total protein and lipid profile (total cholesterol, triglycerides, HDL, and LDL), alanine-aminotransferase (ALT), aspartateaminotransferase (AST), and alkaline phosphatase. Hematology consisted of red blood cell count, hematocrit, hemoglobin, white blood cell count and differential, platelets, MCV, MCH, and MCHC. Physical parameters ------------------- The parameters assessed were weight, resting systolic and diastolic blood pressure, and pulse rate. Statistical Analysis -------------------- Two-sided Student\'s t-test for dependent samples was used to analyze changes between different points in time in the tested parameters, in the whole group and in each gender separately, for both arms. Two-sided t-test for independent samples was used to analyze differences between arms in the change between baseline and week 15 in blood parameters, vital signs and weight, in the whole group and in each gender separately and to detect any difference between groups in the frequency of adverse events. Pearson\'s chi-square test for categorical variables was used to analyze the differences between groups in the number of participants who reported adverse events. In the analysis of differences between and within groups, P values were adjusted for the number of parameters analyzed using Bonferroni correction. SAS statistical package (version 9.1) was used for all analyses. Results ======= Study Population ---------------- A total of 157 participants underwent randomization (79 were assigned to PS-DHA treatment and 78 to placebo treatment). One hundred and thirty one participants completed the double-blind study (66 in the treatment group and 65 in the placebo group). Drop-outs were distributed equally over the two arms and reasons for discontinuation were generally similar across the treatment groups (Table [1](#T1){ref-type="table"}). Average age of participants who completed the double-blind study (± SD) was 72.42 ± 8.02 in the PS-DHA group and 72.73 ± 8.25 in the placebo group. There were no significant differences between treatment groups in the incidence of existing disorders including cardiovascular disease and endocrine or metabolic disorders. ###### Reasons for study discontinuation Reason PS-DHA Placebo --------------------------- -------- --------- Protocol violation 2 1 Withdrawn consent 5 6 Adverse events\* 5 5 Severe adverse events\*\* 1^\#^ 1^\#\#^ Sum 13 13 \* Adverse events are specified in Table [6](#T6){ref-type="table"} \*\* Classified by the study physician as not related to the study treatment ^\#^Hospitalization due to hyponatremia ^\#\#^Hospitalization due to atrial fibrillation and epigastric pain Sixty one participants from the PS-DHA group (PS-DHA continuers) and 61 from the placebo group (PS-naive) continued into the open-label 15 weeks extension period. One participant from the PS-DHA naive group dropped out from the study due to protocol violation. Average age of participants who completed the open-label extension (± SD) was 72.36 ± 7.93 in the PS-DHA continuers group and 72.73 ± 8.31 in the PS-DHA naive group. Safety parameters ----------------- Physical parameters values during the double-blind phase are presented in Table [2](#T2){ref-type="table"}. No significant changes from baseline in resting systolic and diastolic pressure, resting pulse rate and weight were found between treatment groups. A statistically significant increase in weight (0.45 ± 0.04 kg) was detected in the PS-DHA group during the course of the double-blind phase. However, this elevation did not survive Bonferroni correction. Physical parameter values of PS-DHA continuers, who completed the open-label phase, are presented in Table [3](#T3){ref-type="table"}. A statistically significant decrease in resting diastolic BP (3.1 ± 0.3 mmHg) and a slight increase in weight (0.61 ± 0.05 kg) were observed following 30 weeks of PS-DHA administration. ###### Physical parameters values at baseline and following 15 weeks of double-blind treatment PS-DHA Placebo ---------------------------------- ---------------------- ---------------------- -------------------------- ---------------------- ---------------------- -------------------------- ---------------------------- **Variable** Baseline (mean ± SD) 15 weeks (mean ± SD) P value^1^(within group) Baseline (mean ± SD) 15 weeks (mean ± SD) P value^1^(within group) P value^2^(between groups)  **Resting systolic BP**(mm Hg) 127.1 ± 13.4 128.4 ± 15.4 0.511 127.6 ± 15.8 127.8 ± 16.5 0.909 0.685  Male 125.4 ± 13.5 126.8 ± 14.6 0.612 129.0 ± 13.0 131.7 ± 12.6 0.244 0.691  Female 128.8 ± 13.3 130.0 ± 16.2 0.676 126.1 ± 18.7 123.3 ± 19.5 0.330 0.331  **Resting diastolic BP**(mm Hg) 73.8 ± 9.2 72.6 ± 9.1 0.299 75.0 ± 7.9 73.3 ± 9.4 0.159 0.746  Male 73.9 ± 9.3 71.7 ± 9.0 0.188 75.4 ± 8.0 73.9 ± 9.7 0.289 0.795  Female 73.6 ± 9.2 73.4 ± 9.3 0.893 74.6 ± 7.8 72.7 ± 9.2 0.357 0.516  **Resting pulse**(beats/minute) 69.4 ± 10.1 69.9 ± 9.5 0.657 66.5 ± 8.4 67.6 ± 10.1 0.394 0.725  Male 69.1 ± 11.8 69.8 ± 10.8 0.648 63.9 ± 7.8 66.1 ± 10.1 0.139 0.482  Female 69.7 ± 8.4 69.9 ± 8.2 0.872 69.6 ± 8.2 69.3 ± 9.6 0.865 0.812  **Weight**(kg) 70.3 ± 11.2 70.8 ± 11.6 0.033\* 73.0 ± 12.9 73.3 ± 13.3 0.209 0.622  Male 77.4 ± 9.6 78.0 ± 9.9 0.036\* 79.1 ± 11.1 79.5 ± 11.5 0.235 0.664  Female 63.7 ± 8.3 64.0 ± 8.7 0.359 66.1 ± 11.4 66.3 ± 11.9 0.640 0.746 ^1^Based on two-sided t test for dependent samples. ^2^Based on two-sided t test for independent samples \* Statistical significance was not maintained following Bonferonni correction In the treatment arm, 32 males and 33 females had blood pressure and pulse measurements at baseline and endpoint and 31 males and 33 females had weight measurements at baseline and endpoint. In the placebo arm, 34 males and 29 females had blood pressure and pulse measurements at baseline and endpoint and 33 males and 29 females had weight measurements at baseline and endpoint. ###### Physical parameters values of PS-DHA continuers\* at baseline and following 30 weeks of treatment Variable Baseline (mean ± SD) 30 weeks (mean ± SD) P value ---------------------------------- ---------------------- ---------------------- ---------  **Resting systolic BP**(mm Hg) 127.2 ± 13.5 126.3 ± 14.0 0.594  Male 125.7 ± 13.7 126.5 ± 14.3 0.716  Female 128.6 ± 13.4 126.0 ± 14.0 0.366  **Resting diastolic BP**(mm Hg) 73.8 ± 9.4 70.7 ± 9.8 0.006  Male 73.8 ± 9.7 71.6 ± 8.7 0.195  Female 73.8 ± 9.3 69.9 ± 10.7 0.010  **Resting pulse**(beats/minute) 69.4 ± 10.0 70.6 ± 11.9 0.368  Male 69.3 ± 11.6 71.1 ± 13.4 0.412  Female 69.5 ± 8.5 70.1 ± 10.6 0.687  **Weight**(kg) 70.0 ± 11.2 70.7 ± 11.4 0.015  Male 77.1 ± 9.6 77.7 ± 9.8 0.067  Female 63.7 ± 8.5 64.3 ± 8.8 0.104 \* include 29 males and 32 females P value based on two-sided t test for dependent samples Among PS-DHA naive participants, who received PS-DHA for 15 weeks, there were no significant changes in any of the tested physical parameters at the end of the open-label phase (data not shown). Biochemical parameters values during the double-blind phase are presented in Table [4](#T4){ref-type="table"}. No significant differences between groups were observed in the biochemical parameters, except for minor differences in sodium, calcium, chloride and females triglyceride levels. However, these differences did not survive Bonferroni correction and hence were rendered insignificant. In addition, during the course of the double-blind phase there were few parameters that showed a statistically significant change from baseline in the PS-DHA group, and other parameters in the placebo group. However again, the correction rendered them insignificant. ###### Biochemical parameters values at baseline and following 15 weeks of double-blind treatment PS-DHA Placebo -------------------------------- ---------------------- ---------------------- -------------------------- ---------------------- ---------------------- -------------------------- ---------------------------- **Variable** Baseline (mean ± SD) 15 weeks (mean ± SD) P value^1^(within group) Baseline (mean ± SD) 15 weeks (mean ± SD) P value^1^(within group) P value^2^(between groups)  **Glucose**(mg/dL) 106.6 ± 34.0 105.9 ± 37.1 0.885 103.4 ± 30.7 111.1 ± 47.3 0.082 0.175  Male 115.1 ± 42.0 110.4 ± 47.4 0.551 105.8 ± 33.6 118.3 ± 59.9 0.076 0.101  Female 98.0 ± 20.7 101.5 ± 22.5 0.330 100.6 ± 27.3 102.9 ± 25.6 0.662 0.847  **Sodium**(mmol/L) 140.6 ± 2.6 140.9 ± 2.7 0.316 140.9 ± 2.1 140.3 ± 2.3 0.014\* 0.017\*  Male 140.6 ± 2.6 140.9 ± 2.7 0.352 141.2 ± 2.0 140.6 ± 2.1 0.092 0.061  Female 140.5 ± 2.7 140.8 ± 2.7 0.582 140.6 ± 2.2 140.0 ± 2.5 0.079 0.136  **Calcium**(mg/dL) 9.6 ± 0.6 9.6 ± 0.5 0.452 9.4 ± 0.6 9.6 ± 0.5 0.012\* 0.038\*  Male 9.5 ± 0.4 9.5 ± 0.5 0.966 9.4 ± 0.3 9.5 ± 0.4 0.189 0.413  Female 9.8 ± 0.7 9.6 ± 0.5 0.371 9.5 ± 0.8 9.7 ± 0.5 0.032\* 0.051  **Phosphorus**(mg/dL) 3.4 ± 0.6 3.3 ± 0.5 0.103 3.4 ± 0.5 3.3 ± 0.5 0.346 0.807  Male 3.2 ± 0.6 3.1 ± 0.5 0.190 3.2 ± 0.5 3.2 ± 0.5 0.636 0.671  Female 3.7 ± 0.5 3.6 ± 0.5 0.326 3.5 ± 0.6 3.4 ± 0.5 0.397 0.932  **Chloride**(mmol/L) 101.6 ± 3.1 102.4 ± 3.0 0.010\* 101.9 ± 3.1 101.8 ± 2.3 0.880 0.047\*  Male 101.9 ± 2.8 102.6 ± 3.0 0.094 102.1 ± 3.0 102.1 ± 2.2 1.000 0.262  Female 101.2 ± 3.4 102.2 ± 2.9 0.053 101.6 ± 3.4 101.5 ± 2.4 0.819 0.104  **Potassium**(mmol/L) 4.6 ± 0.5 4.5 ± 0.3 0.031\* 4.6 ± 0.5 4.7 ± 0.7 0.703 0.128  Male 4.8 ± 0.7 4.6 ± 0.4 0.095 4.7 ± 0.4 4.7 ± 0.6 0.977 0.221  Female 4.5 ± 0.3 4.4 ± 0.3 0.173 4.6 ± 0.6 4.6 ± 0.8 0.638 0.345  **BUN**(mg/dL) 18.5 ± 4.9 18.9 ± 4.8 0.321 18.6 ± 5.2 19.7 ± 6.7 0.039\* 0.262  Male 19.6 ± 4.8 19.9 ± 4.9 0.679 20.5 ± 5.3 21.7 ± 7.2 0.101 0.249  Female\*\* 17.4 ± 4.9 18.0 ± 4.7 0.345 16.3 ± 4.1 17.3 ± 5.3 0.225 0.680  **Creatinine**(mg/dL) 0.9 ± 0.2 0.9 ± 0.2 0.223 0.9 ± 0.2 0.9 ± 0.2 0.416 0.953  Male 1.0 ± 0.2 1.0 ± 0.2 0.507 1.0 ± 0.3 1.0 ± 0.3 0.617 0.420  Female 0.8 ± 0.1 0.8 ± 0.2 0.237 0.7 ± 0.2 0.8 ± 0.2 0.041\* 0.265  **Alkaline Phosphatase**(U/L) 68.3 ± 22.5 67.3 ± 22.6 0.331 69.9 ± 30.0 69.4 ± 27.8 0.578 0.726  Male 69.6 ± 24.9 68.0 ± 25.7 0.380 73.3 ± 36.2 72.1 ± 33.0 0.286 0.842  Female 67.1 ± 20.1 66.7 ± 19.4 0.693 66.0 ± 20.8 66.3 ± 20.3 0.862 0.724  **ALT/SGPT**(U/L) 23.9 ± 26.4 20.7 ± 8.4 0.282 19.3 ± 6.4 19.5 ± 5.8 0.733 0.262  Male 28.9 ± 36.3 22.3 ± 9.7 0.273 19.2 ± 5.9 19.2 ± 5.8 1.000 0.276  Female 18.9 ± 6.8 19.0 ± 6.6 0.869 19.5 ± 7.1 19.9 ± 5.8 0.655 0.794  **AST/SGOT**(U/L) 25.2 ± 20.8 22.4 ± 6.1 0.250 21.9 ± 6.2 21.8 ± 5.3 0.879 0.280  Male 28.6 ± 28.6 23.3 ± 7.1 0.276 21.8 ± 5.7 21.3 ± 5.5 0.536 0.329  Female 21.9 ± 6.5 21.5 ± 5.0 0.685 21.9 ± 6.8 22.3 ± 5.0 0.704 0.577  **Total Bilirubin**(mg/dL) 0.6 ± 0.2 0.6 ± 0.3 0.361 0.6 ± 0.3 0.6 ± 0.3 0.673 0.747  Male 0.6 ± 0.3 0.6 ± 0.3 0.097 0.7 ± 0.3 0.7 ± 0.3 0.325 0.753  Female 0.5 ± 0.2 0.5 ± 0.2 0.699 0.5 ± 0.2 0.5 ± 0.2 0.423 0.826  **Total Protein**(g/dL) 7.3 ± 0.4 7.3 ± 0.4 0.190 7.3 ± 0.5 7.4 ± 0.5 0.188 0.936  Male 7.3 ± 0.4 7.3 ± 0.5 0.358 7.3 ± 0.4 7.3 ± 0.5 0.990 0.489  Female 7.3 ± 0.4 7.3 ± 0.4 0.362 7.4 ± 0.6 7.5 ± 0.4 0.091 0.443  **Triglycerides**(mg/dL) 141.6 ± 61.4 133.0 ± 70.0 0.133 128.9 ± 54.8 131.6 ± 58.5 0.677 0.186  Male 138.2 ± 68.1 133.8 ± 79.6 0.577 131.6 ± 54.9 121.7 ± 42.2 0.198 0.611  Female 145.1 ± 54.8 132.2 ± 60.0 0.132 125.9 ± 55.4 142.8 ± 71.9 0.103 0.025\*  **HDL**(mg/dL) 54.8 ± 12.4 56.4 ± 13.4 0.118 53.7 ± 14.1 54.4 ± 14.1 0.430 0.505  Male 48.8 ± 9.6 50.4 ± 10.5 0.044\* 48.3 ± 10.2 49.2 ± 11.2 0.348 0.604  Female 60.9 ± 12.1 62.4 ± 13.4 0.404 59.9 ± 15.4 60.3 ± 14.9 0.785 0.635  **LDL**(mg/dL) 97.6 ± 31.4 98.6 ± 31.7 0.745 98.9 ± 30.9 99.1 ± 25.6 0.940 0.853  Male 88.6 ± 30.4 91.0 ± 31.8 0.287 94.4 ± 28.9 92.9 ± 21.2 0.656 0.332  Female 106.6 ± 30.2 106.2 ± 30.2 0.946 104.1 ± 32.8 106.2 ± 28.5 0.672 0.743  **Total cholesterol**(mg/dL) 177.5 ± 38.2 181.1 ± 35.2 0.402 178.0 ± 37.2 179.4 ± 33.7 0.650 0.687  Male 164.6 ± 31.2 167.6 ± 33.1 0.283 168.5 ± 33.2 166.0 ± 26.3 0.487 0.226  Female 190.3 ± 40.6 194.5 ± 32.4 0.613 188.7 ± 39.0 194.6 ± 35.1 0.284 0.856 ^1^Based on two-sided t test for dependent samples. ^2^Based on two-sided t test for independent samples \* Statistical significance was not maintained following Bonferonni correction In the treatment arm, 33 males and 33 females had biochemical blood measurements at baseline and endpoint. In the placebo arm, 34 males and 30 females had biochemical blood measurements at baseline and endpoint. \*\* In the placebo arm 29 females had BUN measurements at baseline and endpoint. Hematological parameters values during the double-blind phase are presented in Table [5](#T5){ref-type="table"}. No significant differences between groups were observed, except for a minor difference in the neutrophils count of the female population. Again, this difference did not survive Bonferroni correction and hence was rendered insignificant. In addition, during the course of the double-blind phase there were few hematological parameters that showed slight changes from baseline in the PS-DHA group, and other parameters in the placebo group, however again, the correction rendered them insignificant. ###### Hematological parameters values at baseline and following 15 weeks of double-blind treatment ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- PS-DHA Placebo ---------------------------------------------- ---------------------- -------------- -------------------------- ---------------------- -------------- -------------------------- ---------------------------- **Variable** Baseline (mean ± SD) 15 weeks\ P value^1^(within group) Baseline (mean ± SD) 15 weeks\ P value^1^(within group) P value^2^(between groups) (mean ± SD) (mean ± SD) **White blood cell count**(×10^3^/cmm) 6.8 ± 1.8 6.5 ± 1.6 0.083 7.1 ± 2.0 6.7 ± 1.9 0.011\* 0.434 Male 6.8 ± 1.7 6.4 ± 1.5 0.020\* 7.3 ± 2.3 7.1 ± 2.2 0.298 0.467 Female 6.7 ± 1.9 6.7 ± 1.6 0.829 6.9 ± 1.4 6.3 ± 1.3 0.009\* 0.066 **Red blood cell count**(×10^6^/cmm) 4.6 ± 0.5 4.6 ± 0.5 0.441 4.6 ± 0.4 4.6 ± 0.4 0.332 0.765 Male 4.7 ± 0.5 4.7 ± 0.5 0.544 4.6 ± 0.4 4.6 ± 0.4 0.491 0.925 Female 4.5 ± 0.4 4.4 ± 0.4 0.622 4.5 ± 0.4 4.5 ± 0.4 0.488 0.760 **Hematocrit**(%) 41.2 ± 4.2 40.7 ± 4.3 0.113 41.6 ± 3.6 40.9 ± 3.6 0.045\* 0.504 Male 42.5 ± 4.3 42.1 ± 4.4 0.229 42.6 ± 3.6 42.0 ± 3.4 0.100 0.763 Female 39.8 ± 3.7 39.4 ± 3.8 0.309 40.5 ± 3.2 39.6 ± 3.5 0.193 0.548 **Hemoglobin**(g/dL) 13.7 ± 1.5 13.7 ± 1.5 0.434 13.7 ± 1.2 13.6 ± 1.2 0.213 0.669 Male 14.3 ± 1.4 14.3 ± 1.3 0.706 14.1 ± 1.2 13.9 ± 1.2 0.181 0.623 Female 13.1 ± 1.4 13.0 ± 1.4 0.481 13.3 ± 1.1 13.2 ± 1.1 0.494 0.856 **Platelet count**(×10^3^/cmm) 221.5 ± 62.7 226.8 ± 60.5 0.374 226.9 ± 73.7 230.8 ± 68.8 0.347 0.839 Male 200.1 ± 44.5 204.0 ± 44.3 0.479 202.8 ± 58.3 210.0 ± 61.1 0.128 0.648 Female 243.5 ± 71.3 250.2 ± 66.4 0.536 255.4 ± 80.6 255.3 ± 70.4 0.980 0.592 **Netrophils**(%) 62.8 ± 6.5 62.4 ± 6.4 0.514 63.8 ± 8.7 62.9 ± 8.6 0.207 0.543 Male 65.1 ± 6.0 64.4 ± 5.7 0.390 64.8 ± 9.0 64.8 ± 9.2 0.968 0.569 Female 60.4 ± 6.2 60.5 ± 6.7 0.969 62.6 ± 8.4 60.7 ± 7.5 0.097 0.147 **Netrophils Absolute**(×10^9^/L) 4.3 ± 1.3 4.1 ± 1.0 0.080 4.5 ± 1.4 4.2 ± 1.4 0.037\* 0.579 Male 4.5 ± 1.3 4.1 ± 1.1 0.027\* 4.7 ± 1.5 4.6 ± 1.7 0.675 0.265 Female 4.1 ± 1.2 4.0 ± 1.0 0.814 4.4 ± 1.2 3.8 ± 1.0 0.010\* 0.046\* **Lymphocytes**(%) 26.0 ± 6.1 26.1 ± 5.8 0.887 25.8 ± 8.7 26.4 ± 8.1 0.324 0.478 Male 23.8 ± 5.6 24.4 ± 5.1 0.453 24.5 ± 8.8 24.4 ± 8.7 0.893 0.529 Female 28.2 ± 5.8 27.8 ± 6.0 0.447 27.3 ± 8.4 28.8 ± 6.9 0.171 0.116 **Lymphocytes Absolute**(×10^9^/L) 1.8 ± 0.6 1.7 ± 0.5 0.167 1.9 ± 1.1 1.8 ± 0.8 0.051 0.449 Male 1.6 ± 0.5 1.6 ± 0.4 0.234 1.9 ± 1.4 1.7 ± 1.0 0.115 0.468 Female 1.9 ± 0.6 1.8 ± 0.6 0.458 1.8 ± 0.6 1.8 ± 0.5 0.238 0.819 **Monocytes**(%) 5.8 ± 1.4 6.0 ± 1.5 0.119 5.6 ± 1.0 5.9 ± 1.4 0.093 0.918 Male 5.9 ± 1.5 6.1 ± 1.4 0.343 5.7 ± 0.9 5.9 ± 1.3 0.299 0.976 Female 5.6 ± 1.4 5.9 ± 1.6 0.220 5.5 ± 1.1 5.8 ± 1.5 0.193 0.895 **Monocytes Absolute**(×10^9^/L) 0.4 ± 0.1 0.4 ± 0.1 0.731 0.4 ± 0.1 0.4 ± 0.1 0.565 0.516 Male 0.4 ± 0.1 0.4 ± 0.1 0.497 0.4 ± 0.1 0.4 ± 0.1 0.890 0.556 Female 0.4 ± 0.1 0.4 ± 0.1 0.284 0.4 ± 0.1 0.4 ± 0.1 0.359 0.159 **Eosinophils**(%) 3.0 ± 2.5 3.1 ± 3.3 0.365 2.5 ± 1.6 2.5 ± 1.5 0.844 0.540 Male 2.7 ± 1.5 2.8 ± 2.0 0.535 2.8 ± 1.8 2.7 ± 1.7 0.832 0.533 Female 3.3 ± 3.3 3.5 ± 4.2 0.513 2.2 ± 1.4 2.3 ± 1.2 0.660 0.796 **Eosinophils Absolute**(×10^9^/L) 0.2 ± 0.3 0.2 ± 0.3 0.853 0.2 ± 0.1 0.2 ± 0.1 0.227 0.400 Male 0.2 ± 0.2 0.2 ± 0.2 0.421 0.2 ± 0.1 0.2 ± 0.1 0.280 0.943 Female 0.2 ± 0.4 0.3 ± 0.5 0.382 0.1 ± 0.1 0.1 ± 0.1 0.513 0.284 **Basophils**(%) 0.7 ± 0.3 0.6 ± 0.3 0.041\* 0.7 ± 0.3 0.6 ± 0.3 0.101 0.684 Male 0.7 ± 0.3 0.6 ± 0.3 0.207 0.6 ± 0.3 0.6 ± 0.3 0.099 0.882 Female 0.8 ± 0.2 0.6 ± 0.3 0.113 0.7 ± 0.4 0.7 ± 0.3 0.515 0.482 **Basophils Absolute**(×10^9^/L) 0.1 ± 0.0 0.0 ± 0.0 0.006\* 0.1 ± 0.0 0.0 ± 0.0 0.019\* 0.854 Male 0.1 ± 0.0 0.0 ± 0.0 0.035\* 0.1 ± 0.0 0.0 ± 0.0 0.058 0.928 Female 0.1 ± 0.0 0.0 ± 0.0 0.077 0.1 ± 0.0 0.0 ± 0.0 0.174 0.701 **Large Unstained Cells**(%) 1.7 ± 0.6 1.7 ± 0.5 1.000 1.6 ± 0.6 1.7 ± 0.6 0.162 0.306 Male 1.8 ± 0.6 1.8 ± 0.5 0.616 1.6 ± 0.6 1.6 ± 0.6 0.623 0.483 Female 1.7 ± 0.7 1.7. ± 0.6 0.701 1.7 ± 0.5 1.9 ± 0.6 0.139 0.434 **Large Unstained Cells Absolute**(×10^9^/L) 0.1 ± 0.0 0.1 ± 0.0 0.652 0.1 ± 0.1 0.1 ± 0.1 0.571 0.948 Male 0.1 ± 0.0 0.1 ± 0.0 0.243 0.1 ± 0.1 0.1 ± 0.1 0.568 0.753 Female 0.1 ± 0.1 0.1 ± 0.0 0.784 0.1 ± 0.0 0.1 ± 0.0 0.847 0.742 **MCV**(μ^3^) 90.1 ± 4.1 89.6 ± 4.4 0.133 91.0 ± 4.3 90.1 ± 4.3 0.049\* 0.506 Male 90.7 ± 4.4 90.2 ± 4.4 0.204 92.2 ± 4.3 91.4 ± 4.6 0.211 0.780 Female 89.5 ± 3.9 89.1 ± 4.31 0.397 89.6 ± 3.9 88.7 ± 3.6 0.124 0.503 **MCH**(pg) 30.1 ± 2.1 30.1 ± 1.9 0.850 30.0 ± 1.5 30.0 ± 1.4 0.940 0.861 Male 30.6 ± 2.2 30.7 ± 2.0 0.651 30.4 ± 1.6 30.4 ± 1.5 0.842 0.658 Female 29.5 ± 1.8 29.5 ± 1.8 0.801 29.5 ± 1.1 29.5 ± 1.1 0.910 0.810 **MCHC**(%) 33.4 ± 1.6 33.6 ± 1.4 0.108 33.0 ± 1.2 33.3 ± 0.9 0.041\* 0.673 Male 33.8 ± 1.9 34.1 ± 1.7 0.116 33.0 ± 1.2 33.2 ± 1.1 0.267 0.794 Female 32.9 ± 1.1 33.1 ± 0.9 0.533 32.9 ± 1.1 33.3 ± 0.7 0.061 0.335 ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ^1^Based on two-sided t test for dependent samples. ^2^Based on two-sided t test for independent samples \* Statistical significance was not maintained following Bonferonni correction In the treatment arm, 33 males and 32 females had hematological blood measurements at baseline and endpoint. In the placebo arm, 33 males and 28 females had hematological blood measurements at baseline and endpoint. Adverse Events -------------- Adverse events reported during the course of the double-blind phase are presented in Table [6](#T6){ref-type="table"}. Twenty participants from the PS-DHA and 11 participants from the placebo group reported adverse events (29 and 15 adverse events, respectively). There were no significant differences between the study groups in the number of participants who reported an adverse event (P = 0.078) or in the number of reported adverse events (P = 0.087). Ten participants from the PS-DHA group and 8 participants from the placebo group were classified by the study physicians as suffering from related or probably related adverse events (16 and 11 adverse events, respectively). Again, there were no significant differences between the study groups in the number of participants who reported an adverse event (P = 0.637) or in the number of reported adverse events, classified as related or probably related to the study treatment (P = 0.472). ###### Adverse events reported during the course of the double-blind phase\* PS-DHA Placebo ------------------------------------------- ---------------------------------------------------------- ------------------------------------------------------------------ ---------------------------------------------------------- ------------------------------------------------------------------ **Adverse event** **Number of related or probably related adverse events** **Number of not-related or probably not related adverse events** **Number of related or probably related adverse events** **Number of not-related or probably not related adverse events** **Adverse events among study drop-outs** Gastrointestinal discomfort 4 1 Rush 1 Increased appetite and weight 1 Strange general feeling 1 Headache 2 dizziness 2 Mood swings 1 *SUM* *5 events in 5 participants* *8 events in 5 participants* **Adverse events among study completers** Gastrointestinal discomfort 9 3 1 2 Weight loss 1 1 Gastritis 1 Headache 1 1 Pneumonia 2 Hypertension 2 1 Hypothyroidism 1 Back pain 1 Leg wound 1 Redness in the mouth 1 Pruritus 1 Eyes inflammation 1 *SUM* *11 events in 5 participants* *13 events in 10 participants* *3 events in 3 participants* *4 events in 3 participants* \* Judged by the study physicians as related, probably related, not related or probably not related to the study treatment During the course of the double-blind phase, there were 5 severe adverse events (suspected acute diverticulitis, benign prostatic hyperplasia surgery, and 3 hospitalizations due to hyponatremia, bradycardia and abdominal discomfort) in 5 participants in the PS-DHA group and 2 severe adverse events (atrial fibrillation and epigastric pain leading to hospitalization) in one participant in the placebo group. All the severe adverse events were classified by the study physicians as not related or probably not related to the study treatment. Adverse events reported during the course of the open-label extension are presented in Table [7](#T7){ref-type="table"}. Ten participants from the PS-DHA continuers reported 12 adverse events of which 3 were classified by the study physicians as related or probably related to the study treatment. Six participants from the PS-DHA naive group reported 9 adverse events; none of which were classified as related or probably related to the study treatment. ###### Adverse events reported during the course of the open-label extension\* PS-DHA continuers PS-DHA naive\*\* ---------------------------------- ---------------------------------------------------------- ------------------------------------------------------------------ ------------------------------------------------------------------ **Adverse event** **Number of related or probably related adverse events** **Number of not-related or probably not related adverse events** **Number of not-related or probably not related adverse events** Gastrointestinal discomfort 1 3 Hypertension 1 3 Hypotension 1 Fall 1 Tenesmus 1 Amaurosis fugax 1 Mild bleeding due to Hemorrhoids 1 Platelets decrease 1 Hand Fracture 1 Hyperlipidemia 1 Dizziness 1 Urinal tract infection 1 Arm pigmentation 1 Headache 1 1 **SUM** **3 events in 3 participants** **9 events in 7 participants** **9 events in 6 participants** \* Judged by the study physicians as related, probably related, not related or probably not related to the study treatment \*\* There were no adverse events classified by the study physicians as related or probably related to the study treatment During the course of this phase, there were 4 severe adverse events (elective hospitalization for computed tomography urography, and chest pains, ear cholesteotoma and prostate surgery, leading to hospitalization), all classified by the study physicians as not related or probably not related to the study treatment. Discussion ========== PS is widely used as an over the counter (OTC) preparation in the aim of improving general health and particularly cognitive functions of elderly people. Traditionally, PS has been extracted from bovine brain; however, recently, the BSE epidemic necessitated the finding of alternative safe sources such as SB-PS. SB-PS, however, differs considerably from BC-PS mainly in the absence of DHA which is the predominant omega-3 LC-PUFA in the mammalian central nervous system. Observational and epidemiological studies have associated omega-3 LC-PUFA consumption with a reduced risk of impaired cognitive function in middle-aged population \[[@B19]\], and with a reduced risk of dementia \[[@B20],[@B21]\]. This association has been supported by interventional studies \[[@B22]-[@B24]\]. To gain the benefits of mammalian PS without the attendant risks, a safe-sourced PS with omega-3 LC-PUFA attached to its backbone (PS- omega-3) was developed. This compound was recently found to improve the symptoms of children with impaired visual sustained attention \[[@B15]\] and to protect middle-aged rats from scopolamine-induced deleterious effects \[[@B16]\]. The mechanism by which PS- omega-3 exerts its effects is not fully understood, however PS has been found to regulate key proteins in neuronal membranes, including sodium/calcium ATPase \[[@B25]\], protein kinase C \[[@B26]\] and Raf-1 protein kinase \[[@B27]\]. PS was also found to influence neurotransmitter activity, such as the release of acetylcholine, dopamine and noradrenaline \[[@B28]\]. In addition, PS- omega-3 was found to significantly increase DHA level in brains of middle-aged rats \[[@B16]\]. We have recently reported that a novel PS preparation (PS-DHA) may improve cognitive performance in non-demented elderly with memory complaints \[[@B17]\] and the safety results of this study are described herein. Participants who discontinued the double-blind phase due to adverse events were distributed equally over the two study arms. No significant findings were observed in the tested physical, hematological or biochemical parameters following 15 weeks of treatment and there was no significant difference in the frequency of adverse events between the groups. No subject discontinued the open-label phase due to an adverse event and there were no significant findings in the tested physical parameters. Overall, the treatment was well tolerated with the most frequent symptom related to gastrointestinal discomfort. These symptoms were considered mild by the subjects and the physicians, and were previously reported for other PS compounds \[[@B29]\]. No severe adverse events were classified by the study physicians as related or probably related to the study treatment. Conclusions =========== Taken in combination with the positive cognitive effects of the PS-DHA preparation reported previously \[[@B17]\], the study results support the safe use of PS-DHA for elderly people with memory complaints at a dosage of 300 mg PS/day for 15 weeks, or at a lower dose (100 mg PS/day) for 30 weeks. Competing interests =================== This study was funded by Enzymotec LTD, Israel. YR, TC and YH are employees of Enzymotec Ltd, and ADK served as consultant to the study. Authors\' contributions ======================= ADK, YR and YH interpreted the data and wrote the draft of this manuscript. The study protocol was designed by VV, ADK, YR and TC. VV and ADK were responsible for patient recruitment and management. All authors were involved in revising the manuscript and gave final approval for publication. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2377/11/79/prepub> Acknowledgements ================ We thank Dr. Ruta Verchovsky and Dr. Ilan Halperin from the Neurology Department of Tel-Aviv Sourasky Medical Center for monitoring the participants\' health and cognitive status throughout the study. We also thank Rachel Konopinsky-Link from the Neurology Department of Tel-Aviv Sourasky Medical Center for coordinating the study.

Introduction {#sec1} ============ The German chemist, Leopold Gmelin was perhaps the first person to coin the term "ester" probably as a contraction of the German Essigäther, "acetic ether".^[@ref1]^ In those days, esters were also called oxy-acid ethers. Generally, esters are compounds derived from acid (organic or inorganic) and alcohol.^[@ref2]^ Triglycerides, i.e., esters of fatty acids and glycerol, are one of the main classes of lipids and important compounds in biology since animal fats and vegetable oils are mainly triglycerides.^[@ref3]^ Polyesters are very important compounds found naturally, e.g., in the cuticle of plants to protect leaves, and their synthetic counterparts are the plastics used in bottles, resins, clothing, etc.^[@ref4],[@ref5]^ Many esters have a pleasant "sweet" odor, and they occur naturally in fruits and plants: isoamyl acetate (pear, banana), ethyl hexanoate (pineapple), octyl acetate (orange), and methyl butyrate (apple).^[@ref6],[@ref7]^ Esters are also used as solvents by the chemical industry, e.g., ethyl acetate is a common solvent with an annual global production of over 3.5 million tons in 2015.^[@ref8]^ DNA and RNA structures contain phosphodiester bonds in their structures. In fact, it would not be an exaggeration to state that esters are crucial for life on Earth.^[@ref9]^ There are numerous methods for the preparation of esters published in the literature. The most common and best known is the Fischer esterification, which is a reversible acid-catalyzed condensation reaction (see [Scheme [1](#sch1){ref-type="scheme"}](#sch1){ref-type="scheme"}), but there are many irreversible reactions catalyzed by one or more catalysts, such as the Mitsunobu \[activation by triphenylphosphine (PPh~3~) and diethyl azodicarboxylate\] and Steglich \[activation by carbodiimides like dicyclohexylcarbodiimide and dimethylaminopyridine\] esterifications.^[@ref10]^ Even though the literature is replete with methods for esterifications, novel approaches have been reported recently. Here are only a few examples: synthesis of ethyl lactate,^[@ref11]^ aryl carboxylates,^[@ref12]^ vinyl ether-containing esters,^[@ref13]^ and lipase-catalyzed alkyl valerates.^[@ref14]^ In addition, the use of triphenylphosphine oxide and oxalyl chloride \[(COCl)~2~\] coupling reagents to achieve esterification at room temperature has been reported as a novel method.^[@ref15]^ {#sch1} Generally, perhaps one of the easiest ways to prepare esters is the use of alcohol and a more reactive carboxylic anhydride or halogenide instead of acids (see [Scheme [1](#sch1){ref-type="scheme"}](#sch1){ref-type="scheme"}). These reactions are simple, high yielding, and mostly performed without any special catalysts; however, in most cases, a hydrogen halogenide (commonly HCl gas) scavenger is needed if carboxylic acid halogenide (most common is carboxylic acid chloride) is used.^[@ref10]^ From the point of view of the environment, the best way to prepare esters would be the case presented also in [Scheme [1](#sch1){ref-type="scheme"}](#sch1){ref-type="scheme"}, which is pretty close to the method we present in this paper. The use of only carboxylic acids and alcohols as starting materials instead of anhydrides or halogenides would be the ideal "green" procedure because, as generally known, acids and alcohols are so ubiquitous that they can be considered as "natural" compounds. Carboxylic acid anhydrides and halogenides (mostly chlorides) need to be prepared chemically (condensation of carboxylic acids to anhydrides or halogenation to halogenide) and that requires energy, catalyst(s), halogen source (like thionyl chloride, SOCl~2~), which is usually toxic, solvents, etc., i.e., these are not environmentally friendly compounds.^[@ref10]^ Thus, we have several fundamental problems to be resolved, such as global warming and plastic exposure; micro- and nanoplastics can be found almost everywhere around the planet.^[@ref16],[@ref17]^ It is evident that we need improved and more energy-efficient and environmentally friendly methods to prepare chemical compounds even if they are prepared on a small scale, according to the motto "great oaks from little acorns grow". In 2015, we published a "proof-of-concept" paper entitled, "A powerful tool for acid-catalyzed organic addition and substitution reactions", where we presented how extremely useful "tool" oven-dried Dowex H^+^ cation-exchange resin with NaI is.^[@ref18]^ By adopting the Dowex H^+^/NaI approach, one can prepare compounds like tertiary iodides, ethers, esters (also using R--CN as the acyl source and dimethylformamide as the formate source), biodiesel from used cooking oil (transesterification process), straight substitution of R--OH to R--I, cyclic ethers can be opened to useful building blocks,^[@ref19]^ iodine can be added to a compound with a triple bond to form an iodinated double bond, etc. (see refs^[@ref18],\ [@ref19]^). As mentioned, the method used in those papers is based on the use of oven-dried Dowex H^+^ cation-exchange resin with NaX (X = I or Br). It is known that Dowex-type H^+^ cation-exchange resins can be used as solid catalysts in esterification reactions;^[@ref20]−[@ref22]^ however, as far as we are aware, there are very few publications where the oven-dried Dowex H^+^ resin has been used by itself for esterifications, and no reports where it has been used with NaI. Here, we will present systematically how extremely useful and effective the dried Dowex H^+^/NaI approach is for esterification reactions and why the method can be considered as a green chemistry tool/method. In this paper, we will present several examples of how dried Dowex H^+^ can be used effectively with or (in some cases) without NaI in the synthesis of very sterically hindered esters using only carboxylic acids and alcohols as starting materials under relatively mild conditions, and in addition, examples of regioselective esterifications will be presented. In most of the cases, the isolation of the produced esters was very simple with no further purifications needed. Examples of the esterification of highly important natural amino acids will also be presented. It can be proposed that Dowex H^+^ cation-exchange resin is commonly used in all chemistry laboratories throughout the world, and therefore, it is widely available. The background describing how we initially discovered its potential for use in various chemical reactions can be found elsewhere.^[@ref18]^ Briefly, Dowex H^+^ needs to be dried before use (at 120 °C in oven for around 18--20 h; see [Figure [1](#fig1){ref-type="fig"}](#fig1){ref-type="fig"}), which makes it far more effective than undried resin (moisture content of regular Dowex H^+^ cation-exchange resin may be over 50%). In fact, the best way to make it as effective as possible is first to treat Dowex H^+^ with 2 M HCl, then wash several times with distilled water, and finally dry it (more detailed procedure can be found in [Experimental Section](#sec4){ref-type="other"}). {#fig1} Results and Discussion {#sec2} ====================== In [Table [1](#tbl1){ref-type="other"}](#tbl1){ref-type="other"}, we have collected 18 examples for the preparation of esters starting from carboxylic acids, and in [Table [2](#tbl2){ref-type="other"}](#tbl2){ref-type="other"}, we have presented six examples of acylation of selected alcohols by using dried Dowex H^+^ cation-exchange resin with or, in some cases (seven examples), without dried NaI. [Tables [1](#tbl1){ref-type="other"}](#tbl1){ref-type="other"} and [2](#tbl2){ref-type="other"} also describe the reaction conditions used (temperature and reaction time); detailed reaction conditions and isolation procedures can be found in [Experimental Section](#sec4){ref-type="other"}. ###### Esterification of Selected Carboxylic Acids by Dried Dowex H^+^/NaI Approach {#fx1} In the table, only the reaction time and temperature have been highlighted; detailed experimental procedures and conditions can be found in [Experimental Section](#sec4){ref-type="other"}. Isolated yields. Conversion according to ^1^H NMR spectrum in parentheses. Without NaI. Product co-evaporation with MeOH was observed; r.t. = room temperature, n.d. = not determined, n.i. = not isolated. ###### Acylation of Selected Alcohols by the Dried Dowex H^+^/NaI Approach {#fx2} In the table, only the reaction time and temperature have been highlighted; detailed experimental procedures and conditions can be found in [Experimental Section](#sec4){ref-type="other"}. Isolated yields. Conversion according to ^1^H NMR spectrum in parentheses; r.t. = room temperature, n.d. = not determined. Without NaI. One of the most interesting starting materials for esterification experiments presented here was tropic acid (**1**), which is a precursor of a very important natural compound atropine (**48**; see [Figure [2](#fig2){ref-type="fig"}](#fig2){ref-type="fig"}). Atropine (**48**) can be extracted from some plants like *Atropa belladonna*; it is an antagonist of the muscarinic cholinergic receptors, so it has a broad range of medical applications. Atropine (**48**) is also used as an antidote for poisoning from nerve gases (such as sarin and VX) and organophosphate insecticides.^[@ref23]^ In addition to being an important precursor of atropine (**48**), tropic acid (**1**) is an attractive compound having a sterically hindered secondary carboxylic acid group and a hydroxyl group (even though this is a primary one, it has a bulky phenyl and carboxylic acid group in the β-carbon) in the same molecule. Four different derivatives of tropic acid (**1**) were prepared: two esters (**22** and **23**; see [Table [1](#tbl1){ref-type="other"}](#tbl1){ref-type="other"}) and two acylated compounds (**41** and **42**; see [Table [2](#tbl2){ref-type="other"}](#tbl2){ref-type="other"}). {#fig2} Interestingly, according to SciFinder search, only eight examples of ways to prepare tropic acid methyl ester (**22**) from tropic acid (**1**) have been reported with poor yields (\<20%) or yields not even reported at all, and most of the esterifications were performed by basic Fischer esterification catalyzed by sulfuric acid.^[@ref24]^ There is one paper that describes the synthesis of tropic acid isopropyl ester (**23**) by chlorination of carboxylic acid group using thionyl chloride; however, yield was not mentioned.^[@ref25]^ As can be seen in [Table [1](#tbl1){ref-type="other"}](#tbl1){ref-type="other"}, by using our method, the isolated yields of tropic acid methyl (**22**) and isopropyl (**23**) esters were 83 and 72%, respectively, much higher than in the earlier reports. Only three examples to prepare acetylated tropic acid (**41**) from tropic acid (**1**) were found in the literature, according to SciFinder search, all utilizing acetyl chloride.^[@ref26]^ The isobutyl derivative (**42**) was totally unknown. By adopting the Dowex H^+^/NaI approach with tropic acid (**1**) and acetic or isobutyric acid, 99 and 44% isolated yields (99 and 88% conversions; see [Table [2](#tbl2){ref-type="other"}](#tbl2){ref-type="other"}) were achieved for compounds **41** and **42**, respectively. The usefulness of the Dowex H^+^/NaI method for esterifications of benzoic acid (**2**) and its 4-hydroxy (**3**) and 2-hydroxy (**4**, salicylic acid) derivatives was also tested. Methyl benzoate (**24**) was produced with a high isolated yield (82%) by refluxing **2** for 24 h in MeOH. 4-Hydroxybenzoic acid ethyl ester (**25**) was isolated with a reasonable yield (51%); however, to achieve 91% conversion of **3** to **25**, the reaction was refluxed for 72 h in EtOH. Only 50% conversion of salicylic acid (**4**) to its ethyl ester (**26**) was observed under the same conditions as for the synthesis of **25**. One can clearly discern the inactivation effect of the presence of a hydroxy group in the 2-position versus 4-position of the benzene ring for the synthesis of compounds **25** and **26** because the conversions were 91 and 50%, respectively, under the same conditions. We describe many examples of organic acids with different kinds of functional groups (**5**--**10**) to emphasize how effectively they can be esterified to compounds **27**--**32** by the Dowex H^+^/NaI approach under very mild reaction conditions, mostly by stirring at room temperature for 2--4 h \[compound **32**, which is an ethyl ester, not a methyl like the other compounds (**27**--**31**), needed stirring at 65 °C for 4 h to achieve 99% conversion\]. Methyl hex-5-ynoate (**28**) was stirred for 24 h to achieve 100% conversion (75% isolated yield); however, the synthesis was performed without NaI, so it was not unexpected that the time to reach 100% conversion would be longer. The reason why we tested the synthesis of **28** without NaI is the possibility of an addition reaction to the triple bond, which we have reported elsewhere.^[@ref18]^ This seems to happen even at room temperature, i.e., we observed this side reaction according to the ^1^H NMR spectrum in synthesis experiments to produce **28** with NaI after stirring for 24 h. Full 100% conversion of **6** to **28** can be achieved also with a catalytic amount of NaI (0.1 equiv) at room temperature with 4 h stirring; however, due to the slightly different isolation procedure, the isolated yield was 66% instead of 75%, with the procedure lacking NaI. We were not able to isolate compound **29** because of the previously observed co-evaporation with MeOH in the isolation procedure. It was also quite difficult to isolate compound **30** for the same reasons; however, we were able to obtain a 40% yield. Malic acid (**11**) was chosen to be one of the starting materials in our esterification experiments because it is a dicarboxylic acid and it has two different kinds of carboxylic acid groups, primary and secondary. The dimethyl ester of malic acid (**33**) was effectively synthesized by stirring at room temperature for 24 h. Amino acids are very important compounds for all life on Earth, but in addition, their structures were very interesting for the esterification experiments presented here because of α-amino groups and different side chains.^[@ref27]^ We selected four different kinds of amino acids (as their hydrochloride salts, **12**--**15**) for esterification experiments with the Dowex H^+^/NaI approach. According to a SciFinder search, by far the most commonly reported method to prepare compounds **34**--**37** has been the use of thionyl chloride to synthesize the acid chloride, followed by the addition of methanol, which is not an environmentally friendly approach. Surprisingly, very little attention has been paid to preparing amino acid methyl esters (**34**--**37**) from the corresponding amino acid hydrochlorides (**12**--**15**), and only 17 examples were found in the literature. We were able to prepare compounds **34**--**37** from **12**--**15** by using only dried Dowex H^+^ as the catalyst (see reaction conditions in [Table [1](#tbl1){ref-type="other"}](#tbl1){ref-type="other"}). The isolation of the products was extremely straightforward: only filtration and evaporation were needed and high yields (70--82%) were acquired to compounds **34**--**37** (see detailed procedures in [Experimental Section](#sec4){ref-type="other"}). The method reported here is a new, green, and straightforward way to prepare amino acid esters **34**--**37**, and most probably it can be applied to other amino acids as well. There were several reasons why citric acid (**16**) and iso-citric acid trisodium salt hydrate (**17**) were selected as starting materials for the esterification experiments with the Dowex H^+^/NaI method: (1) they are tricarboxylic acids; citric acid (**16**) has two primary carboxylic acid groups and one tertiary carboxylic acid group, and iso-citric acid trisodium salt hydrate (**17**) has one primary and two secondary carboxylic acid groups; (2) both are very important compounds and part of the citric acid cycle^[@ref28]^ to release stored energy by all aerobic organisms; (3) citric acid triethyl ester (**38**) is an important precursor in the synthesis of phosphocitric acid, which is a naturally occurring biomolecule found in cells;^[@ref29]^ (4) iso-citric acid triethyl ester (**39**) will become an important precursor for the synthesis of phospho-iso-citric acid, which has not yet been reported in the literature and it is not commercially available. The triethyl ester of citric acid (**38**) was prepared and isolated with quantitative yield (97% isolated yield; see [Table [1](#tbl1){ref-type="other"}](#tbl1){ref-type="other"}) by refluxing compound **16** for 24 h in EtOH. The isolated yield for triethyl ester of iso-citric acid (**39**) was rather low (29%); however, the procedure was very straightforward because we could use directly the commercially available salt form of isocitrate (**17**) with no need to first convert the salt to the acid. Actually, for our needs, the 29% yield was more than sufficient, so the optimization of the procedure did not seem necessary. It should be noted that, according to the ^1^H NMR spectrum measured from the sample taken in the reaction mixture of **39**, we were able to detect a mixture of triethyl ester (**39**) and diethyl esters of iso-citric acid. The presence of these other compounds and the isolation procedure of **39** (see [Experimental Section](#sec4){ref-type="other"}) were the main reasons for the low isolated yield (29%). As mentioned earlier, in [Table [2](#tbl2){ref-type="other"}](#tbl2){ref-type="other"}, we have collected six examples of acylation of selected compounds by using the method reported here. Compounds **41** and **42** have been discussed earlier in the text, and compound **40** is an example of the esterification of benzyl alcohol using only dried Dowex H^+^ without NaI at 80 °C for 24 h. Under those conditions, a high isolated yield (85%) was achieved. Mandelic acid (**19**) is an important compound, e.g., used as a precursor, in the pharmaceutical and cosmetic industries.^[@ref30]^ It has a hindered secondary hydroxyl group, which made it a valuable compound for our esterification experiments. In the literature, only two chemical methods to prepare acetylated mandelic acid (**43**) from **19** and acetic acid were found, catalyzed by zinc chloride^[@ref31]^ or silica sulfuric acid^[@ref32]^ with 72 and 92% reported yields, respectively. In the case when silica sulfuric acid was used as the catalyst, neither the experimental procedure nor the isolation procedure were reported, also NMR characterization data were missing. In our experiment, the acetylated mandelic acid (**43**) was produced at 90 °C after a 24 h reaction time with a rather good isolated yield (55%; 81% conversion was observed). This is a good alternative to these previously reported methods and demonstrates well that the Dowex H^+^/NaI approach is effective and useful also when highly steric ester structures need to be prepared using only alcohols and carboxylic acids as starting materials. In the preparation of octyl acetate (**44**) and but-3-en-1-yl acetate (**45**), we decided to test how effective the Dowex H^+^/NaI method would be if only 2 equiv of acetic acid (compared to alcohols **20** and **21**) were used in the reactions with no solvent and which temperature and reaction time would be optimal for good yields. Surprisingly, at 55 °C with a reaction time of 2 h, 97% conversion of octanol (**20**) to octyl acetate (**44**) was observed and a high isolated yield (83%) was achieved. After the reaction at the same 55 °C temperature for 18 h reaction time, but-3-en-1-yl acetate (**45**) was isolated with a reasonable yield (52%). In the isolation of **45**, we did not use any solvents, so the synthesis and isolation of **45** can be considered a very green procedure. According to a SciFinder search, there are only 18 references that are related to the synthesis of **45** from **21** using acetic acid as the acetylation reagent, 16 of which are patents. Perhaps the closest method to ours was catalyzed by sulfuric acid under the following conditions: 10 equiv acetic acid, 60 °C, 1 h; 72% isolated yield was reported.^[@ref33]^ But-3-en-1-yl acetate (**45**) is a very important precursor in the synthesis of pheromones.^[@ref33]^ According to the results reported in [Tables [1](#tbl1){ref-type="other"}](#tbl1){ref-type="other"} and [2](#tbl2){ref-type="other"}, we decided to conduct some experiments to clarify the selectivity of the Dowex H^+^/NaI method in esterification reactions. Mixtures (1:1) of phenol with octanol or isopropanol or cyclohexanol were reacted with 10 equiv of acetic acid under the conditions reported in [Scheme [2](#sch2){ref-type="scheme"}](#sch2){ref-type="scheme"}. All aliphatic alcohols were converted (100%) to their acetates over phenol. We also tested the combination of phenol, octanol, and cyclohexanol (1:1:1) under the same conditions at room temperature to determine if there was any selectivity of octanol over cyclohexanol; however, both were converted to their acetates (\>95%). {#sch2} Selective synthesis of citric acid dimethyl ester (**46**) starting from citric acid (**16**) with 60% isolated yield was obtained (see [Scheme [3](#sch3){ref-type="scheme"}](#sch3){ref-type="scheme"}). The whole synthesis procedure with isolation was rather straightforward, as shown in [Experimental Section](#sec4){ref-type="other"}. There are only a few reports in the literature to prepare compound **46**: one method is based on boric or boronic acid catalyzation,^[@ref34]^ and the other utilizes a Fischer esterification procedure and sulfuric acid catalyzation with 32% isolated yield;^[@ref35]^ another paper reported a 35% yield,^[@ref36]^ although with rather complicated isolation procedures compared to our method. We also conducted a straightforward test with benzoic acid (**2**) and phenylacetic acid (**5**) in the same pot; after 1.5 h stirring at room temperature, the conversion of **5** to **27** was observed to be 100% and benzoic acid did not react at all; so we separated and isolated both compounds (see [Scheme [4](#sch4){ref-type="scheme"}](#sch4){ref-type="scheme"} and the detailed procedure in [Experimental Section](#sec4){ref-type="other"}). This is not a surprise since the same kind of experiment has been described in the literature but with Amberlyst 15 as the catalyst.^[@ref37]^ Nonetheless, this is an example demonstrating that it is possible to use the Dowex H^+^/NaI method for esterification of less hindered carboxylic acids and separate them and then proceed with esterification of more hindered carboxylic acids like benzoic acid (**2**) or its derivatives at elevated temperatures (see [Table [1](#tbl1){ref-type="other"}](#tbl1){ref-type="other"}; preparation of compounds **24**--**26**). {#sch3} {#sch4} For the reasons mentioned earlier, [dl]{.smallcaps}-malic acid (**11**) was considered an interesting compound for selective esterification experiments. Malic acid is an important compound in biochemistry, produced by all living organisms, and it is part of the earlier mentioned citric acid cycle;^[@ref28]^ it also confers the sour taste to fruits.^[@ref38]^ Only one report for the synthesis of malic acid mono methyl ester (**47**) was found in the literature using enzymatic hydrolysis of dimethyl malic acid^[@ref39]^ (**33**), and therefore, this is the first chemical synthesis of **47** reported so far in the literature. The best and most adequate isolated yield (45%) was achieved when 500 mg of **11** was stirred with 500 mg of dried Dowex H^+^ resin without NaI for 2 h at room temperature (see [Scheme [5](#sch5){ref-type="scheme"}](#sch5){ref-type="scheme"}). The final product **47** could be readily separated from mixture containing diester **33** by crystallization (detailed procedure can be found in [Experimental Section](#sec4){ref-type="other"}). {#sch5} The effect of NaI in the reactions can be clearly seen in the results collected in [Table [3](#tbl3){ref-type="other"}](#tbl3){ref-type="other"}. The effect is much greater when preparing more steric esters (e.g., **42**) compared to their nonsteric counterparts (e.g., **44**). In addition, in the case of methyl benzoate (**24**), the use of 0.5 equiv of NaI instead 0.1 equiv gave better conversion, 92 and 83%, respectively. This was not the case with compound **42**; these results can be seen in [Table [4](#tbl4){ref-type="other"}](#tbl4){ref-type="other"} and will be discussed later. As a comparison, sulfuric acid (H~2~SO~4~) catalysis to prepare compounds **24** and **44** was tested and achieved conversions were rather similar to those with esterifications with the Dowex H^+^/NaI method. ###### Effect of NaI on Product Conversion in Selected Reactions {#fx3} Dried Dowex H^+^ with 0.5 equiv dried NaI. Only dried Dowex H^+^. Dried Dowex with 0.1 equiv dried NaI. H~2~SO~4~-catalyzed. ###### Effect of the Amount of Dried NaI Used in the Synthesis of the Highly Steric Compound **42** amount of NaI (equiv) conversion[a](#t4fn1){ref-type="table-fn"} (%) ----------------------- ------------------------------------------------ 0 19 0.1 **74** 0.2 66 0.5 25 50 °C, 24 h. We also tested the effect of the amount of NaI used in the synthesis of the highly steric ester **42**; the results are collected in [Table [4](#tbl4){ref-type="other"}](#tbl4){ref-type="other"}. The best conversion (74%) was achieved with 0.1 equiv of NaI; more NaI or its absence gave lower conversions. In fact, the amounts of NaI used in the esterification reactions reported in this paper have not been optimized, and according to the above result, it might be reasonable to test the optimal amount of NaI on a case-by-case basis to achieve the best conversion. Last but not least, we decided to test how effective and selective the Dowex H^+^/NaI method would be in the separation of valuable resin acids (**50**; see the general structure in [Figure [3](#fig3){ref-type="fig"}](#fig3){ref-type="fig"}) from a real tall oil sample.^[@ref40]^ Tall oil is a commercially valuable byproduct of softwood (pine or spruce) from pulp production with more than 1.6 M tonnes being produced annually. The major components of tall oil are unsaturated fatty acids and the more valuable resin acids (**50**), which are used in many industrial and consumer products. On an industrial scale, fatty acid components of the tall oil are separated from resin acids by high vacuum distillation.^[@ref41]^ However, the complete separation of the components by distillation methods is difficult and not energy-efficient. Tall oil esterification by the method described here, followed by extraction, leads to quantitative separation of fatty acid methyl esters (see [Scheme [6](#sch6){ref-type="scheme"}](#sch6){ref-type="scheme"}). The tertiary resin acids (**50**) remain intact in our method and are recovered from the process at a high yield (see procedure in [Experimental Section](#sec4){ref-type="other"}). {#fig3} {#sch6} The exact reaction mechanism is unclear, but the most probable mechanism for unhindered esters follows the traditional proton-catalyzed esterification mechanism. In this case, the H^+^ form solid resin is an excellent proton donor and carbonyl oxygen (C=O) is protonated, the nucleophilic alcohol attacks the positively charged carbon, and water is eliminated yielding the ester (e.g., see [Table [3](#tbl3){ref-type="other"}](#tbl3){ref-type="other"} product **24**). In the case of hindered esters (see [Table [4](#tbl4){ref-type="other"}](#tbl4){ref-type="other"}), the mechanism is the formation of R--I from R--OH and NaI in the presence of H^+^ from Dowex and/or the formation of HI, which is a better proton donor than Dowex since 19 and 74% conversions were obtained without and with 0.1 equiv of NaI, respectively. Generally, substitution of alcohol group to iodine requires either activation of OH group or elevated temperatures.^[@ref18],[@ref42]^ In the previous paper,^[@ref18]^ we have demonstrated and discussed the formation of HI based on an addition reaction to the double bond. At least under elevated temperatures, the R--I-based mechanism might be possible. Conclusions {#sec3} =========== The study clearly and unambiguously highlights the straightforward nature of these kinds of synthetic procedures, and usually products can be isolated without any further purifications (except two compounds, **42** and **43**, which required column chromatography purifications) using only carboxylic acids and alcohols as starting materials. Our method can be compared to the basic Fisher esterification, e.g., as catalyzed by sulfuric acid; however, the product isolation by using our method seems to be simpler than in Fischer esterifications (e.g., isolation of **34**--**37**) and no side reactions were observed under the conditions used; with sulfuric acid, side reactions are possible at least with unsaturated starting materials.^[@ref18]^ Generally, our conditions were mild, of course some elevated temperatures were needed in reactions when very steric esters like **42** and **43** were prepared or starting materials had unreactive or steric groups like **2** and **16**. We have demonstrated that the dried Dowex H^+^/NaI approach can be used for selective esterifications and there is clear evidence that NaI reduces the reaction energy needed in the esterification reactions, and in particular, it improves conversions when steric esters need to prepared; however, the amount of NaI may need to be determined on a test case-by-case basis to achieve the best results, as shown in [Table [4](#tbl4){ref-type="other"}](#tbl4){ref-type="other"}. In the experiments of the esterification of amino acids **12**--**15**, which were the last products we made, we showed that a sufficient amount of dried Dowex H^+^ was only 50 mg (for diacid **35**, more, i.e., 75 mg was needed) vs 100 mg of starting material (0.5:1 mass ratio) and the corresponding esters (**34**--**37**) were synthesized without NaI and isolated so simply that it could be used for industrial purposes (see procedures in [Experimental Section](#sec4){ref-type="other"}) with high yields (70--82%). We have also demonstrated how extremely useful the dried Dowex H^+^/NaI method is in the separation and isolation of highly valuable resin acids (**50**) from the mixture of acids (see [Scheme [6](#sch6){ref-type="scheme"}](#sch6){ref-type="scheme"}). The reusability of Dowex H^+^ resin was tested for the synthesis of **27** and **28**. In both cases, Dowex H^+^ resin was collected and reused two times (used for three times total), leading to ca. 100% conversion in all three times according to the ^1^H spectra (see the detailed procedure in [Experimental Section](#sec4){ref-type="other"}). And finally, in a nutshell, the presented method/procedure is extremely simple, very effective, and nontoxic, and Dowex H^+^ resin is regenerable,^[@ref18]^ reusable, and energy-efficient, which make it a green approach. Experimental Section {#sec4} ==================== General {#sec4.1} ------- ^1^H and ^13^C NMR spectra were recorded on a 600 MHz spectrometer operating at 600.2 and 150.9 MHz, respectively. The solvent residual peak was used as a standard for ^1^H and ^13^C measurements in CDCl~3~ and CD~3~OD (7.26 or 77.16 ppm for CDCl~3~ and 3.31 or 49.00 ppm for CD~3~OD, respectively),^[@ref43]^ in D~2~O 4.79 ppm in ^1^H measurements and added CD~3~OD in the ^13^C measurements (49.00 ppm). The *^n^J*~HH~ couplings were calculated from the proton spectra, and all *J* values are given in Hz. Appropriate two-dimensional NMR measurements were performed to confirm structures when needed. Mass spectra were recorded on a quadrupole time-of-flight mass spectrometer using electrospray ionization (ESI) with positive ionization mode for totally novel compound **42** and also for compounds **23**, **39**, and **47** (compounds for which chemical syntheses were not found in the literature). The purity of the products was determined from ^1^H spectrum and was ≥95% unless stated otherwise. All starting materials (**1--21**) and the Dowex H^+^ ion-exchange resin used in the study were commercially available. Example of the Preparation of Dried Dowex H^+^ Ion-Exchange Resin {#sec4.2} ----------------------------------------------------------------- Dowex ion-exchange resin (50Wx8 hydrogen form, 100--200 mesh; 25 g) was stirred for 0.5 h in 2 M HCl solution (50 mL) at room temperature before it was filtered, and washed with distilled H~2~O until the filtrate pH was neutral. Finally, Dowex was dried in an oven for 18--20 h at 120 °C and stored in a closed bottle. Preparation of 3-Hydroxy-2-phenylpropanoic Acid Methyl Ester (**22**) {#sec4.3} --------------------------------------------------------------------- Tropic acid (100 mg, 0.60 mmol), dried NaI (18 mg, 0.12 mmol, 0.2 equiv), dried Dowex 50W-X8 ion-exchange resin (H^+^-form, 200 mg), and dried MeOH (1 mL) were stirred overnight (around 18 h) at room temperature before Dowex was collected after filtration and washed with MeOH, and the reaction mixture was evaporated to dryness in vacuo. The residue was dissolved in dichloromethane (DCM) (5 mL), washed with 10% Na~2~S~2~O~3~ (sodium thiosulfate, 2--3 mL), dried over MgSO~4~, and evaporated to dryness in vacuo. 3-Hydroxy-2-phenylpropanoic acid methyl ester (90 mg, 83%) was obtained as a colorless syrup. ^1^H NMR (CDCl~3~): δ 7.37--7.32 (m, 2H), 7.32--7.28 (m, 1H), 7.28--7.25 (m, 1H), 4.16--4.11 (m, 1H), 3.88--3.84 (m, 1H), 3.84--3.80 (m, 1H), 3.71 (s, 3H), 2.28 (br, 1H, −O[H]{.ul}). ^13^C NMR (CDCl~3~): δ 173.8, 135.7, 129.0 (2C), 128.3 (2C), 127.9, 64.7, 54.0, 52.4. Preparation of 3-Hydroxy-2-phenylpropanoic Acid Isopropyl Ester (**23**) {#sec4.4} ------------------------------------------------------------------------ Tropic acid (100 mg, 0.60 mmol), dried NaI (45 mg, 0.30 mmol, 0.5 equiv), dried Dowex 50W-X8 ion-exchange resin (H^+^-form, 300 mg), and 2-propanol (1.5 mL) were refluxed for 24 h before Dowex was collected after filtration and washed with 2-propanol, and the reaction mixture was evaporated to dryness in vacuo. The residue was dissolved in DCM (5 mL), washed with 10% Na~2~S~2~O~3~ (sodium thiosulfate, 2--3 mL), dried over MgSO~4~, and evaporated to dryness in vacuo. 3-Hydroxy-2-phenylpropanoic acid isopropyl ester (90 mg, 72%) was obtained as a colorless syrup. ^1^H NMR (CDCl~3~): δ 7.35--7.31 (m, 2H), 7.30--7.24 (m, 3H), 5.08 (h, 1H, ^3^*J*~HH =~ 6.3), 4.13--4.07 (m, 1H), 3.83--3.78 (m, 2H), 2.32 (t, 1H, −O[H]{.ul}, ^3^*J*~HH~ = 6.4), 1.25 (d, 3H, ^3^*J*~HH~ = 6.3), 1.14 (d, 3H, ^3^*J*~HH~ = 6.3). ^13^C NMR (CDCl~3~): δ 172.9, 136.0, 128.9 (2C), 128.3 (2C), 127.7, 68.7, 64.9, 54.3, 21.9, 21.6. MS (ESI^+^) calcd for C~12~H~16~O~3~ \[M + H\]^+^ 209.1172, found: 209.1175. NMR spectra in [Supporting Information](http://pubs.acs.org/doi/suppl/10.1021/acsomega.9b00790/suppl_file/ao9b00790_si_001.pdf) p. S2. Preparation of Methyl Benzoate (**24**) {#sec4.5} --------------------------------------- Benzoic acid (100 mg, 0.82 mmol), dried NaI (61 mg, 0.41 mmol, 0.5 equiv), dried Dowex 50W-X8 ion-exchange resin (H^+^-form, 400 mg), and dried MeOH (3 mL) were refluxed for 24 h before Dowex was collected after filtration and washed with MeOH, and the reaction mixture was evaporated to dryness in vacuo. Methyl benzoate (91 mg, 82%) was obtained as a colorless oil. ^1^H NMR (CDCl~3~): δ 8.06--8.03 (m, 2H), 7.58--7.53 (m, 1H), 7.46--7.42 (m, 2H), 3.92 (s, 3H). ^13^C NMR (CDCl~3~): δ 167.3, 133.0, 130.3, 129.7 (2C), 128.5 (2C), 52.2. Preparation of Ethyl 4-Hydroxybenzoate (**25**) {#sec4.6} ----------------------------------------------- 4-Hydroxybenzoic acid (200 mg, 1.45 mmol), dried NaI (110 mg, 0.73 mmol, 0.5 equiv), dried Dowex 50W-X8 ion-exchange resin (H^+^-form, 1.0 g), and 94% (A) EtOH (8 mL) were refluxed for around 72 h before Dowex was collected after filtration and washed with EtOH, and the reaction mixture was evaporated to dryness in vacuo. The residue was dissolved in DCM (5 mL), washed twice with saturated NaHCO~3~ (3 mL), dried over MgSO~4~, and evaporated to dryness in vacuo. Ethyl 4-hydroxybenzoate (122 mg, 51%) was obtained as a white powder. ^1^H NMR (CD~3~OD): δ 7.88--7.85 (m, 2H), 6.83--6.80 (m, 2H), 4.31 (q, 2H, ^3^*J*~HH~ = 7.1), 3.92 (t, 3H, ^3^*J*~HH~ = 7.1). ^13^C NMR (CD~3~OD): δ 168.3, 163.6, 132.7 (2C), 122.5, 116.1 (2C), 61.7, 14.7. Preparation of Methyl 2-Phenylacetate (**27**) {#sec4.7} ---------------------------------------------- Phenylacetic acid (100 mg, 0.73 mmol), dried NaI (55 mg, 0.34 mmol, ca. 0.5 equiv), dried Dowex 50W-X8 ion-exchange resin (H^+^-form, 300 mg), and dried MeOH (3 mL) were stirred for 2 h at r.t. before Dowex was collected after filtration and washed with MeOH, and the reaction mixture was evaporated to dryness in vacuo. The residue was dissolved in DCM (8 mL), washed with 10% Na~2~S~2~O~3~ (sodium thiosulfate, 2--3 mL) and saturated NaHCO~3~ (2--3 mL), dried over MgSO~4~, and evaporated to dryness in vacuo. Methyl 2-phenylacetate (69 mg, 63%) was obtained as a colorless liquid. ^1^H NMR (CDCl~3~): δ 7.36--7.31 (m, 2H), 7.30--7.25 (m, 3H), 3.69 (s, 3H), 3.63 (s, 2H). ^13^C NMR (CDCl~3~): δ 172.2, 134.1, 129.4 (2C), 128.7 (2C), 127.2, 52.2, 41.3. Preparation of 5-Hexynoic Acid Methyl Ester (**28**) {#sec4.8} ---------------------------------------------------- 5-Hexynoic acid (300 mg, 2.68 mmol), dried Dowex 50W-X8 ion-exchange resin (H^+^-form, 250 mg), and MeOH (3 mL) were stirred for 24 h at room temperature before Dowex was collected after filtration and washed with a small portion of diethyl ether before the solvents were removed in vacuo. 5-Hexynoic acid methyl ester (254 mg, 75%) was obtained as a colorless liquid. ^1^H NMR (CDCl~3~): δ 3.68 (s, 3H), 2.46 (t, 2H, ^3^*J*~HH~ = 7.4), 2.26 (t + d, 2H, ^3^*J*~HH~ = 7.0, ^4^*J*~HH~ = 2.7), 1.96 (t, 1H, ^3^*J*~HH~ = 2.7), 1.85 (qv, 2H, ^3^*J*~HH~ = 7.2). ^13^C NMR (CDCl~3~): δ 173.7, 83.4, 69.2, 51.7, 32.8, 23.7, 18.0. Preparation of Hexanoic Acid Methyl Ester (**30**) {#sec4.9} -------------------------------------------------- Hexanoic acid (300 mg, 2.58 mmol), dried NaI (39 mg, 0.26 mmol, 0.1 equiv), dried Dowex 50W-X8 ion-exchange resin (H^+^-form, 250 mg), and MeOH (3 mL) were stirred for 4 h at room temperature before Dowex was collected after filtration and MeOH removed in vacuo. The residue was dissolved in diethyl ether (5 mL), washed with 10% Na~2~S~2~O~3~ (sodium thiosulfate, 0.5 mL), dried over MgSO~4~, and diethyl ether was removed in vacuo. Methyl hexanoate (133 mg, 40%) was obtained as a colorless liquid. ^1^H NMR (CDCl~3~): δ 3.66 (s, 3H), 2.29 (t, 2H, ^3^*J*~HH~ = 7.5), 1.62 (qv, 2H, ^3^*J*~HH~ = 7.5), 1.35--1.26 (m, 4H), 0.89 (t, 3H). ^13^C NMR (CDCl~3~): δ 174.5, 51.6, 34.2, 31.5, 24.8, 22.4, 14.0. Preparation of 10-Undecenoic Acid Methyl Ester (**31**) {#sec4.10} ------------------------------------------------------- 10-Undecenoic acid (300 mg, 1.63 mmol), dried NaI (24 mg, 0.16 mmol, 0.1 equiv), dried Dowex 50W-X8 ion-exchange resin (H^+^-form, 200 mg), and MeOH (3 mL) were stirred for 4 h at room temperature before Dowex was collected after filtration and washed with a small portion of MeOH, and the reaction mixture was evaporated to dryness in vacuo. The residue was dissolved in diethyl ether (4 mL), washed with 10% Na~2~S~2~O~3~ (sodium thiosulfate, 0.5 mL), dried over MgSO~4~, and diethyl ether was removed in vacuo. 10-Undecenoic acid methyl ester (249 mg, 77%) was obtained as a colorless oil. ^1^H NMR (CDCl~3~): δ 5.84--5.77 (m, 1H), 5.01--4.90 (m, 2H), 3.66 (s, 3H), 2.30 (t, 2H, ^3^*J*~HH~ = 7.5), 2.06--2.00 (m, 2H), 1.65--1.58 (m, 2H), 1.40--1.33 (m, 2H), 1.32--1.26 (m, 8H). ^13^C NMR (CDCl~3~): δ 174.5, 139.3, 114.3, 51.6, 34.2, 33.9, 29.41, 29.37, 29.26, 29.19, 29.0, 25.1. Preparation of 11-Bromoundecanoic Acid Ethyl Ester (**32**) {#sec4.11} ----------------------------------------------------------- 11-Bromoundecanoic acid (10 g, 37.7 mmol), dried NaI (280 mg, 1.87 mmol, 0.05 equiv), dried Dowex 50W-X8 ion-exchange resin (H^+^-form, 1 g), and absolute EtOH (35 mL) were stirred for 4 h at 65 °C before Dowex was collected after filtration and washed with EtOH, and the reaction mixture was evaporated to dryness in vacuo. The residue was dissolved in DCM (25 mL), washed with 10% Na~2~S~2~O~3~ (sodium thiosulfate, 5 mL), and dried over MgSO~4~ before DCM was removed in vacuo. 11-Bromoundecanoic acid ethyl ester (10.9 g, 99%) was obtained as a brown oil. ^1^H NMR (CDCl~3~): δ 4.11 (q, 2H, ^3^*J*~HH~ = 7.1), 3.39 (t, 2H, ^3^*J*~HH~ = 6.9), 2.77 (t, 2H, ^3^*J*~HH~ = 7.5), 1.87--1.81 (m, 2H), 1.65--1.56 (m, 2H), 1.45--1.37 (m, 2H), 1.33--1.25 (m, 10H), 1.24 (t, 3H, ^3^*J*~HH~ = 7.1). ^13^C NMR (CDCl~3~): δ 174.0, 60.3, 34.5, 34.1, 32.9, 29.5, 29.4, 29.3, 29.2, 28.9, 28.3, 25.1, 14.4. Preparation of Dimethyl 2-Hydroxysuccinate (**33**) {#sec4.12} --------------------------------------------------- [dl]{.smallcaps}-Malic acid (100 mg, 0.75 mmol), dried Dowex 50W-X8 ion-exchange resin (H^+^-form, 200 mg), and dried MeOH (3 mL) were stirred for 24 h at room temperature before Dowex was collected after filtration and washed with MeOH, and the reaction mixture was evaporated to dryness in vacuo. Dimethyl 2-hydroxysuccinate (92 mg, 76%) was obtained as a colorless syrup. ^1^H NMR (CDCl~3~): δ 4.52--4.49 (m, 1H), 3.82 (s, 3H), 3.72 (s, 3H), 2.90--2.85 (m, 1H), 2.83--2.78 (m, 1H). ^13^C NMR (CDCl~3~): δ 173.9, 171.1, 67.4, 53.0, 52.2, 38.6. Preparation of Methyl 2-Aminoacetate Hydrochloride (**34**) {#sec4.13} ----------------------------------------------------------- Glycine hydrochloride (100 mg, 0.90 mmol), dried Dowex 50W-X8 ion-exchange resin (H^+^-form, 50 mg), and MeOH (3 mL) were stirred for 24 h at 55 °C before Dowex was collected after filtration and washed with MeOH, and the reaction mixture was evaporated to dryness in vacuo. Methyl 2-aminoacetate hydrochloride (92 mg, 82%) was obtained as a white solid. ^1^H NMR (CD~3~OD): 3.85 (s, 2H), 3.84 (s, 3H). ^13^C NMR (CD~3~OD): δ 168.7, 53.7, 41.2. Preparation of (*S*)-Dimethyl 2-Aminosuccinate Hydrochloride (**35**) {#sec4.14} --------------------------------------------------------------------- [l]{.smallcaps}-Aspartic acid hydrochloride (100 mg, 0.59 mmol), dried Dowex 50W-X8 ion-exchange resin (H^+^-form, 75 mg), and MeOH (3 mL) were stirred for 24 h at 55 °C before Dowex was collected after filtration and washed with MeOH, and the reaction mixture was evaporated to dryness in vacuo. (*S*)-Dimethyl 2-aminosuccinate hydrochloride (81 mg, 70%) was obtained as a colorless solid. ^1^H NMR (CD~3~OD): δ 4.41 (t, 1H, ^3^*J*~HH~ = 5.5), 3.85 (s, 3H), 3.76 (s, 3H), 3.08 (d, 2H, ^3^*J*~HH~ = 5.5). ^13^C NMR (CDCl~3~): δ 171.4, 169.6, 54.0, 53.0, 50.4, 34.8. Preparation of (*S*)-Methyl 2-Amino-3-(4-hydroxyphenyl)propanoate Hydrochloride (**36**) {#sec4.15} ---------------------------------------------------------------------------------------- [l]{.smallcaps}-Tyrosine hydrochloride (100 mg, 0.46 mmol), dried Dowex 50W-X8 ion-exchange resin (H^+^-form, 50 mg), and MeOH (3 mL) were stirred for 24 h at 55 °C before Dowex was collected after filtration and washed with MeOH, and the reaction mixture was evaporated to dryness in vacuo. (*S*)-Methyl 2-amino-3-(4-hydroxyphenyl)propanoate hydrochloride (80 mg, 75%) was obtained as a white solid. ^1^H NMR (CD~3~OD): δ 7.09--7.04 (m, 2H), 6.80--6.76 (m, 2H), 4.26--4.22 (m, 1H), 3.81 (s, 3H), 3.19--3.14 (m, 1H), 3.10--3.04 (m, 1H). ^13^C NMR (CDCl~3~): δ 170.6, 158.4, 131.5 (2C), 125.6, 116.9 (2C), 55.4, 53.6, 36.6. Preparation of (*R*)-Methyl 2-Amino-3-mercaptopropanoate Hydrochloride (**37**) {#sec4.16} ------------------------------------------------------------------------------- [l]{.smallcaps}-Cysteine hydrochloride (100 mg, 0.63 mmol), dried Dowex 50W-X8 ion-exchange resin (H^+^-form, 50 mg), and MeOH (3 mL) were stirred for 24 h at 55 °C before Dowex was collected after filtration and washed with MeOH, and the reaction mixture was evaporated to dryness in vacuo. (*R*)-Methyl 2-amino-3-mercaptopropanoate hydrochloride (83 mg, 76%) was obtained as a white solid. ^1^H NMR (CD~3~OD): δ 4.34 (t, 1H, ^3^*J*~HH~ = 5.1), 3.87 (s, 3H), 3.09 (d, 2H, ^3^*J*~HH~ = 5.1). ^13^C NMR (CDCl~3~): δ 167.7, 54.3, 52.5, 23.8. Preparation of Triethyl 2-Hydroxypropane-1,2,3-tricarboxylate (Triethyl Citrate) (**38**) {#sec4.17} ----------------------------------------------------------------------------------------- Citric acid (200 mg, 1.04 mmol), dried NaI (234 mg, 1.56 mmol, 1.5 equiv), dried Dowex 50W-X8 ion-exchange resin (H^+^-form, 1.2 g), and abs. EtOH (10 mL) were refluxed for 24 h before Dowex was collected after filtration and washed with EtOH, and the reaction mixture was evaporated to dryness in vacuo. The residue was dissolved in DCM (10 mL), washed with 10% Na~2~S~2~O~3~ (sodium thiosulfate, 2--7 mL), dried over MgSO~4~, and evaporated to dryness in vacuo. Triethyl citrate (278 mg, 97%) was obtained as a colorless oil. ^1^H NMR (CDCl~3~): 4.28 (q, 2H, ^3^*J*~HH~ = 7.2), 4.14 (q, 4H, ^3^*J*~HH~ = 7.2), 2.90--2.86 (m, 2H), 2.80--2.76 (m, 2H), 1.30 (t, 3H, ^3^*J*~HH~ = 7.2), 1.25 (t, 6H, ^3^*J*~HH~ = 7.2). ^13^C NMR (CDCl~3~) δ 173.5, 169.9 (2C), 73.4, 62.5, 61.1 (2C), 43.5, 14.22 (2C), 14.17. Preparation of Triethyl 1-Hydroxypropane-1,2,3-tricarboxylate (Triethyl Isocitrate, Mixture of Isomers) (**39**) {#sec4.18} ---------------------------------------------------------------------------------------------------------------- Iso-citric acid trisodium salt hydrate (1 g, 3.87 mmol), dried NaI (1.21 g, 8.07 mmol, 2.2 equiv), dried Dowex 50W-X8 ion-exchange resin (H^+^-form, 10 g), and abs. EtOH (35 mL) were refluxed for 48 h before Dowex was collected after filtration and washed with EtOH prior to the reaction mixture being evaporated to dryness in vacuo. The residue was dissolved in diethyl ether (20 mL) and washed three times with 0.5 M NaOH (8 mL), dried over MgSO~4~, and evaporated to dryness in vacuo. Triethyl isocitrate (311 mg, 29%, calculated based on anhydrous starting material) was obtained as a colorless viscous oil. ^1^H NMR (CDCl~3~): 4.37--4.34 (m, 1H), 4.33--4.24 (m, 2H), 4.20--4.10 (m, 4H), 3.53--4.49 (m, 1H), 3.16 (d, 1H, −OH, ^3^*J*~HH~ = 5.4), 2.94--2.89 (m, 1H), 2.66--2.61 (m, 1H), 1.32 (t, 3H, ^3^*J*~HH~ = 7.2), 1.27 (t, 3H, ^3^*J*~HH~ = 7.2), 1.23 (t, 3H, ^3^*J*~HH~ = 7.2). ^13^C NMR (CDCl~3~) δ 173.2, 171.9, 170.9, 70.8, 62.4, 61.4, 61.0, 45.0, 32.6, 14.3, 14.2, 14.1. MS (ESI^+^) calcd for C~12~H~20~O~7~ \[M + H\]^+^ 277.1282, found: 277.1287. NMR spectra in [Supporting Information](http://pubs.acs.org/doi/suppl/10.1021/acsomega.9b00790/suppl_file/ao9b00790_si_001.pdf) p. S3. Preparation of Benzyl Acetate (**40**) {#sec4.19} -------------------------------------- Benzyl alcohol (209 mg, 200 μL, 1.93 mmol), dried Dowex 50W-X8 ion-exchange resin (H^+^-form, 200 mg), and glacial acetic acid (1.5 mL) were stirred for 24 h at 80 °C before Dowex was collected after filtration and washed with DCM, and then the reaction mixture was evaporated to dryness in vacuo. The residue was dissolved in DCM (8 mL), washed with saturated NaHCO~3~ (2--3 mL), dried over MgSO~4~, and evaporated to dryness in vacuo. Benzyl acetate (246 mg, 85%) was obtained as a colorless liquid. ^1^H NMR (CDCl~3~): δ 7.39--7.31 (m, 5H), 5.11 (s, 2H), 2.11 (s, 3H). ^13^C NMR (CDCl~3~): δ 171.1, 136.1, 128.7 (2C), 128.41 (2C), 128.40, 66.5, 21.2. Preparation of 3-(Acetyloxy)-2-phenylpropanoic Acid (**41**) {#sec4.20} ------------------------------------------------------------ Tropic acid (100 mg, 0.60 mmol), dried NaI (18 mg, 0.12 mmol, 0.2 equiv), dried Dowex 50W-X8 ion-exchange resin (H^+^-form, 200 mg), and glacial acetic acid (1 mL) were stirred overnight (around 18 h) at 50 °C before Dowex was collected after filtration and washed with DCM, and the reaction mixture was evaporated to dryness in vacuo. 3-(Acetyloxy)-2-phenylpropanoic acid (124 mg, 99%) was obtained as a slightly brownish syrup. ^1^H NMR (CDCl~3~): δ 7.38--7.30 (m, 5H), 4.61--4.56 (m, 1H), 4.38--4.33 (m, 1H), 3.99--3.95 (m, 1 H), 2.04 (s, 3H). ^13^C NMR (CDCl~3~): δ 177.5, 171.0, 134.3, 129.1 (2C), 128.5, 128.3 (2C), 64.9, 50.7, 21.0. Preparation of 3-(Isobutyryloxy)-2-phenylpropanoic Acid (**42**) {#sec4.21} ---------------------------------------------------------------- Tropic acid (200 mg, 1.20 mmol), dried NaI (18 mg, 0.12 mmol, 0.1 equiv), dried Dowex 50W-X8 ion-exchange resin (H^+^-form, 400 mg), and isobutyric acid (2 mL) were stirred for 24 h at 60 °C before Dowex was collected after filtration and washed with DCM, and then the reaction mixture was evaporated to dryness in vacuo. The residue was purified by silica column chromatography using hexane/ethyl acetate (1:1) as the eluent. 3-(Isobutyryloxy)-2-phenylpropanoic acid (124 mg, 44%) was obtained as an amorphous solid. ^1^H NMR (CDCl~3~): δ 7.38--7.29 (m, 5H), 4.59--4.53 (m, 1H), 4.40--4.35 (m, 1H), 2.51 (h, 1H, ^3^*J*~HH~ = 7.0), 1.103 (d, 3H, ^3^*J*~HH~ = 7.0), 1.100 (d, 3H, ^3^*J*~HH~ = 7.0). ^13^C NMR (CDCl~3~): δ 176.9, 176.7, 134.4, 129.1 (2C), 128.39, 128.38 (2C), 64.6, 50.6, 34.0 19.0 (2C). MS (ESI^+^) calcd for C~13~H~16~O~4~ \[M + H\]^+^ 237.1121, found: 237.1124. NMR spectra in [Supporting Information](http://pubs.acs.org/doi/suppl/10.1021/acsomega.9b00790/suppl_file/ao9b00790_si_001.pdf) p. S4. Preparation of 2-(Acetoxy)-2-phenylacetic Acid (**43**) {#sec4.22} ------------------------------------------------------- [dl]{.smallcaps}-Mandelic acid (100 mg, 0.66 mmol), dried NaI (10 mg, 0.067 mmol, 0.1 equiv), dried Dowex 50W-X8 ion-exchange resin (H^+^-form, 200 mg), and glacial acetic acid (1 mL) were stirred for 24 h at 90 °C before Dowex was collected after filtration and washed with DCM, and the reaction mixture was evaporated to dryness in vacuo. The residue was purified by silica column chromatography using ethyl acetate/MeOH (8:2) as the eluent (Rf. 0.8). 2-(Acetoxy)-2-phenylacetic acid (70 mg, 55%) was obtained as a white solid. ^1^H NMR (CD~3~OD): δ 7.51--7.46 (m, 2H), 7.40--7.32 (m, 3H), 5.86 (s, 1H), 2.14 (s, 3H). ^13^C NMR (CDCl~3~): δ 173.1 (very broad signal, hardly visible), 172.1, 136.3, 129.9, 129.6 (2C), 128.7 (2C), 76.6 (broad signal), 20.7. Preparation of Octyl Acetate (**44**) {#sec4.23} ------------------------------------- 1-Octanol (2.0 mL, 1.65 g, 12.65 mmol), glacial acetic acid (1.44 mL, 1.51 g, 25.15 mmol, 2.0 equiv), dried Dowex 50W-X8 ion-exchange resin (H^+^-form, 1.0 g), and dried NaI (190 mg, 1.27 mmol, 0.1 equiv) were stirred for 2 h at 55 °C before Dowex was collected after filtration and washed with hexane (15 mL). Hexane was washed with 10% Na~2~S~2~O~3~ (sodium thiosulfate, 5 mL) and saturated NaHCO~3~ (20 mL), dried over MgSO~4~, and the hexane was removed in vacuo. Octyl acetate (1.82 g, 83%) was obtained as a colorless oil. ^1^H NMR (CDCl~3~): δ 4.05 (t, 2H, ^3^*J*~HH~ = 6.8), 2.04 (s, 3H), 1.61 (qv, 2H), 1.38--1.22 (m, 10H), 0.88 (t, 3H). ^13^C NMR (CDCl~3~): δ 171.7, 64.8, 31.9, 29.35, 29.31, 28.7, 26.0, 22.8, 21.2, 14.2. Preparation of But-3-en-1-yl Acetate (**45**) {#sec4.24} --------------------------------------------- 3-Buten-1-ol (1.0 mL, 843 mg, 11.69 mmol), glacial acetic acid (1.35 mL, 1.42 g, 23.58 mmol, 2.0 equiv), dried Dowex 50W-X8 ion-exchange resin (H^+^-form, 350 mg), and dried NaI (75 mg, 0.50 mmol, 0.05 equiv) were stirred overnight (ca. 18 h) at 55 °C before Dowex was collected after filtration, and the reaction mixture was moved to a separation funnel. The reaction mixture was washed with saturated NaHCO~3~ (2--3 mL), the water layer (lower layer) was removed, and the ester phase was washed with a very small portion of 10% Na~2~S~2~O~3~ (sodium thiosulfate) and dried over MgSO~4~. But-3-en-1-yl acetate (698 mg, 52%) was obtained as a slightly yellow liquid. ^1^H NMR (CDCl~3~): δ 5.82--5.74 (m, 1H), 5.14--5.05 (m, 2H), 4.12 (t, 2H, ^3^*J*~HH~ = 6.8), 2.40--2.35 (m, 2H), 2.04 (s, 3H). ^13^C NMR (CDCl~3~): δ 171.3, 134.1, 117.3, 63.7, 33.2, 21.1. Preparation of 2-Hydroxy-4-methoxy-2-(2-methoxy-2-oxoethyl)-4-oxobutanoic Acid (Dimethyl Citrate) (**46**) {#sec4.25} ---------------------------------------------------------------------------------------------------------- Citric acid (200 mg, 1.04 mmol), dried NaI (48 mg, 0.32 mmol, 0.3 equiv), dried Dowex 50W-X8 ion-exchange resin (H^+^-form, 300 mg), and MeOH (3 mL) were stirred for 24 h at room temperature before Dowex was collected after filtration and washed with MeOH, and the reaction mixture was evaporated to dryness in vacuo. The solid residue was stirred in DCM (4 mL) for 2 h before centrifugation. Liquids were removed and the treatment was repeated once. Solids were dried in vacuo. Dimethyl citrate (138 mg, 60%) was obtained as a white powder. Mp 115--116 °C. ^1^H NMR (CDCl~3~): 3.67 (s, 6H), 2.97--2.93 (m, 2H), 2.84--2.80 (m, 2H). ^13^C NMR (CDCl~3~) δ 176.6, 172.1 (2C), 74.3, 52.5 (2C), 44.1 (2C). Example of the Selective Acetylation of Cyclohexanol over Phenol {#sec4.26} ---------------------------------------------------------------- Cyclohexanol (110 μL, 106 mg, 1.06 mmol, 1 equiv), phenol (100 mg, 1.06 mmol, 1 equiv), glacial acetic acid (610 μL, 640 mg, 10.66 mmol, 10 equiv), dried Dowex 50W-X8 ion-exchange resin (H^+^-form, 200 mg), and dried NaI (16 mg, 0.11 mmol, 0.1 equiv) were stirred for 24 h at 50 °C. An NMR sample was taken from the reaction mixture, and 100% conversion of cyclohexanol to acetylated cyclohexanol was observed according to the ^1^H NMR spectrum (phenol conversion was 0%). ^1^H NMR (CDCl~3~): δ 4.77--4.71 (m, 1H), 2.04 (s, 3H), 1.88--1.81 (m, 2H), 1.75--1.68 (m, 2H), 1.57--1.51 (m, 1H), 1.44--1.31 (m, 4H), 1.28--1.20 (m, 1H). Preparation of 2-Hydroxy-4-methoxy-4-oxobutanoic Acid (Malic Acid Monomethyl Ester, **47**) {#sec4.27} ------------------------------------------------------------------------------------------- [dl]{.smallcaps}-Malic acid (500 mg, 3.73 mmol), dried Dowex 50W-X8 ion-exchange resin (H^+^-form, 500 mg), and MeOH (5 mL) were stirred for 2 h at room temperature before Dowex was collected after filtration and washed with MeOH, and the reaction mixture was evaporated to dryness in vacuo. Residue was dissolved in ethyl acetate (3 mL) and added to a glass column filled with silica gel (2 cm diameter, 15 cm high) and eluted with 75 mL of ethyl acetate (this was done to remove unreacted starting material [dl]{.smallcaps}-malic acid from reaction mixture). Ethyl acetate was removed in vacuo and the residue was dissolved in chloroform (5 mL) and placed in the freezer overnight. The formed crystals were separated and dried in vacuo. 2-Hydroxy-4-methoxy-4-oxobutanoic acid (249 mg, 40%) was obtained as a white solid. Mp 76--77 °C. ^1^H NMR (CD~3~OD): δ 4.52--4.48 (m, 1H), 3.74 (s, 3H), 2.80--2.75 (m, 1H), 2.69--2.64 (m, 1H). ^13^C NMR (CD~3~OD): δ 175.2, 173.9, 68.6, 52.6, 39.9. MS (ESI^+^) calcd for C~5~H~8~O~5~ \[M + H\]^+^ 149.0445, found: 149.0449. NMR spectra in [Supporting Information](http://pubs.acs.org/doi/suppl/10.1021/acsomega.9b00790/suppl_file/ao9b00790_si_001.pdf) p. S5. Example of the Selective Esterification of Phenylacetic Acid over Benzoic Acid and Their Isolation from the Reaction Mixture {#sec4.28} ---------------------------------------------------------------------------------------------------------------------------- Phenylacetic acid (100 mg, 0.73 mmol), benzoic acid (100 mg, 0.82 mmol), dried NaI (25 mg, 0.17 mmol), dried Dowex 50W-X8 ion-exchange resin (H^+^-form, 300 mg), and dried MeOH (3 mL) were stirred for 1.5 h at r.t. before Dowex was collected after filtration and washed with MeOH, and the reaction mixture was evaporated to dryness in vacuo. The residue was dissolved in diethyl ether (5 mL), washed twice with 10% NaHCO~3~ (2 × 5 mL) and 10% Na~2~S~2~O~3~ (sodium thiosulfate, 0.5 mL), dried over MgSO~4~, and evaporated to dryness in vacuo. Methyl 2-phenylacetate (59 mg, 54%) was obtained as a colorless liquid. The combined NaHCO~3~ phases were carefully acidified with 2 M HCl (strong release of gases!) and extracted twice with diethyl ether (2 × 5 mL). The combined diethyl ether phases were dried over MgSO~4~ and evaporated to dryness in vacuo. Benzoic acid was obtained as a white solid. Methyl 2-phenylacetate: ^1^H NMR (CDCl~3~): δ 7.36--7.31 (m, 2H), 7.30--7.25 (m, 3H), 3.69 (s, 3H), 3.63 (s, 2H). Benzoic acid: 8.15--8.10 (m, 2H), 7.65--7.60 (m, 1H), 7.52--7.45 (m, 2H). Example of the Selective Esterification of Fatty Acids (**49**) over Resin Acids (**50**) and Their Isolation from the Reaction Mixture {#sec4.29} --------------------------------------------------------------------------------------------------------------------------------------- Prepurified, methanol-treated pine wood extract (40 g), dried NaI (2 g, 13 mmol), dried Dowex 50W-X8 ion-exchange resin (H^+^-form, 2.2 g), and dry MeOH (80 mL) were stirred at room temperature for 18 h before Dowex was collected after filtration and washed with MeOH, and the reaction mixture was evaporated to dryness in vacuo. The oily product was made basic with 1 M NaOH and washed three times with EtOAc (70 mL) to remove unsaturated fatty acid methyl esters. The remaining oily phase was acidified with 3 M HCl and extracted three times with EtOAc (50 mL). The organic fraction was evaporated in vacuo yielding a 31 g mixture of resin acids for further use (see the NMR spectra in the [Supporting Information](http://pubs.acs.org/doi/suppl/10.1021/acsomega.9b00790/suppl_file/ao9b00790_si_001.pdf) on pages S6--S7). Dowex H^+^ Resin Handling in the Reusable Experiments {#sec4.30} ----------------------------------------------------- Dowex was collected by a Bühner funnel, washed with a large volume of MeOH, and dried in an oven for 1 h at 120 °C before reuse in the synthesis of **27** and **28**. The Supporting Information is available free of charge on the [ACS Publications website](http://pubs.acs.org) at DOI: [10.1021/acsomega.9b00790](http://pubs.acs.org/doi/abs/10.1021/acsomega.9b00790).^1^H and ^13^C NMR spectra of all new compounds **23**, **39**, **42**, and **47**; ^1^H NMR spectra of resin acids (**50**) separation are presented just like the ^1^H NMR spectra of all other isolated esters ([PDF](http://pubs.acs.org/doi/suppl/10.1021/acsomega.9b00790/suppl_file/ao9b00790_si_001.pdf)) Supplementary Material ====================== ###### ao9b00790_si_001.pdf The authors declare no competing financial interest. This research was supported by the Jane and Aatos Erkko Foundation and by Business Finland, project ValueWoodChem 3280/31/2015. P.A.T. is grateful to Jane and Aatos Erkko Foundation's support. The authors thank Maritta Salminkoski for her expert technical assistance in the syntheses, and Dr Marko Lehtonen and Miia Reponen for MS measurements.

INTRODUCTION ============ Gastric cancer, the incidence of which became declining in many industrialized countries, is still one of the major causes of mortality from cancer death in the world.[@B1] Whereas the postoperative 5-year survival rates are 90% to 95% for early gastric cancer (EGC), only 20% to 40% of patients with advanced gastric cancer are expected to survive for 5 years or more.[@B2]-[@B4] Moreover, surgery is no more needed in a considerable fraction of patients who are diagnosed with EGC, owing to recent advances in endoscopic resection techniques and technologies. Therefore, early detection of EGC through the vigilant follow-up in the high risk groups is probably the effective strategy for improving survival and quality of life. The pathogenesis of gastric cancer, particularly the intestinal type, can be explained by a cascade from chronic gastritis through atrophic gastritis (AG), intestinal metaplasia (IM), and dysplasia to cancer. AG is characterized by chronic inflammatory processes of gastric mucosa that leads to the loss of glandular structure and a reduction of gastric secretory function. The presence of AG is known to be a risk of gastric cancer, which increases with the degree and extension of atrophy.[@B5] IM is defined as replacement of gastric columnar epithelial cells by cells of intestinal morphology with the presence of goblet cells, Paneth cells and absorptive cells.[@B5] Both AG and IM represent an obligatory transitional step in gastric carcinogenesis and are undisputed indicators of an increased risk for gastric cancer as compared with chronic gastritis in the absence of these lesions.[@B5]-[@B7] Therefore, the recognition of these lesions by endoscopy in a general population indicates the necessity of follow-up endoscopy, which helps the early detection of gastric cancer leading to a better prognosis and treatment by endoscopic resection. The prevalence of AG and IM vary among countries. That is, they are relatively high in countries with a higher prevalence of *Helicobacter pylori* infection and gastric cancer.[@B8] *H. pylori* infection is the major cause of chronic gastritis, which leads to AG and IM over a long period. However, despite a high rate of *H. pylori* infection, there are some regions with a low prevalence of precancerous lesions and gastric cancer.[@B9] Therefore, other factors such as host and environmental factors might play a role in these differences. In Western countries, AG and IM are generally observed in histological examination of random biopsies obtained during endoscopy, whereas in Asian countries including Korea, the presence and extension of AG and IM are frequently observed by endoscopy. A health check-up program designed to detect gastric cancer was implemented by the South Korean government in 2001 for biannual evaluation of Korean citizens over age 40. Thus, if we know the endoscopic prevalence of AG and IM in the general population and their role in the localization of high risk group of gastric cancer, it would be very useful for prevention of gastric cancer. From this background, the aims of this study were to evaluate the prevalence of endoscopic AG and IM, and to document the risk factors for the development of these precancerous lesions with the special reference to *H. pylori* infection, host and environmental factors in a Korean general population. MATERIALS AND METHODS ===================== 1. Study population ------------------- A total of 4,023 subjects who underwent screening endoscopy during a routine general check-up from January to December in 2011, were prospectively enrolled at eight nationwide healthcare centers in Korea. Subjects with a history of gastrointestinal surgery or with systemic disease requiring chronic medication except hypertension and diabetes mellitus were excluded. This study was reviewed and approved by the Institutional Review Board of the eight participating hospitals and written informed consent was obtained from all participating subjects. 2. Questionnaire ---------------- All subjects, who provided informed consents, underwent a clinical interview based on a structured questionnaire to assess personal and clinical data under the supervision of a well-trained interviewer before the endoscopy at the eight participating healthcare centers. The questionnaire included questions regarding demographic data, the presence of upper gastrointestinal symptoms, such as epigastric pain or discomfort, dyspepsia, epigastric soreness, and abdominal pain during the previous year, comorbid disease, history of *H. pylori* eradication, drug history including nonsteroidal anti-inflammatory drugs (NSAIDs) and antibiotics, alcohol consumption, smoking history, consumption of dairy products, and relatives of gastric cancer. 3. Endoscopic examination ------------------------- All of the endoscopic examinations were carried out and assessed by endoscopy experts of the eight participating healthcare centers. The endoscopic findings were examined, in a standardized manner, for typical macroscopic changes including erythema (diffuse, spotty, or linear), erosions (small superficial defects in the mucosa with flat edge and white/yellow color or small bleeding spots \[petechiae\]), absence of rugae, and visible blood vessels. Investigators did agree with simplification of endoscopic definition about AG and IM. Endoscopic AG was defined as thinning, whitish mucosal change or visible submucosal vascular patterns and endoscopic IM as white plaque-like elevation in antrum and corpus. Any overlapping findings were described. 4. Determination of *H. pylori* status -------------------------------------- Blood samples were obtained from each participant immediately after endoscopy. Isolated serum samples were neatly arranged in storage boxes and stored at -70℃. *H. pylori* infection was diagnosed by enzyme-linked immunosorbent assay (ELISA) for anti-*H. pylori* immunoglobulin G (IgG) using a Genedia kit (Genedia *H. pylori* ELISA; Green Cross Medical Science Corp., Eumseong, Korea), with duplicate determinations according to the manufacturer\'s guidelines. The Genedia kit used *H. pylori* antigen obtained from Korean *H. pylori* strains, with a sensitivity and specificity in Korean adults of 97.8% and 92.0%, respectively. The cutoff optical density (OD, 450 nm) of *H. pylori* IgG was 0.406. 5. Statistical analysis ----------------------- All statistical analyses were performed using the Stat View software package (SAS Institute, Cary, NC, USA). Continuous variables were analyzed by Student\'s t-test. Categorical variables were analyzed using chi-squared test or Fisher\'s exact test. Multivariate logistic regression was used for the analysis of risk factors, which were expressed as the odds ratio (OR) and 95% confidence intervals (CI). The p-values of less than 0.05 were considered statistically significant. RESULTS ======= 1. Baseline characteristics of subjects --------------------------------------- A total of 4,023 subjects were included in the study. This study group comprised 2,358 males (58.6%) and 1,665 females (41.4%). The mean age was 48.7±11.3 (mean±standard deviation \[SD\]) years with a range from 15 to 98 years. *H. pylori* IgG positivity was demonstrated in 2,407 subjects (59.8%). Baseline characteristics of subjects including height, weight, body mass index (BMI), cholesterol, triglyceride, fasting glucose, smoking, alcohol, education, income level, NSAID use, high salt diet, history of *H. pylori* eradication, gastrointestinal symptoms, consumption of dairy product, and relatives of gastric cancer are described in [Table 1](#T1){ref-type="table"}. 2. Prevalence rates of AG and IM -------------------------------- The prevalence rates of AG and IM, diagnosed by endoscopic findings, were 40.7% (1,638 of 4,023) and 12.5% (502 of 4,023), respectively. The prevalence rates of AG and IM in males were significantly higher than those in females (43.3% vs 37.1% and 15.4% vs 8.3%, respectively; p\<0.001) ([Tables 2](#T2){ref-type="table"} and [3](#T3){ref-type="table"}). The prevalence rate of AG increased with age (from 29.9% in those aged less than 40 years to 59.4% in those aged 60 years or older) ([Fig. 1A](#F1){ref-type="fig"}). The prevalence rate of IM also increased with age (from 3.4% in those less than 40 years of age to 17.4% in those 60 years or older; p\<0.001) ([Fig. 1B](#F1){ref-type="fig"}). 3. The identification of risk factors for AG by univariate and multivariate analysis ------------------------------------------------------------------------------------ In a univariate analysis of risk factors for AG, older age group of above 40 to 59, older age group of above 60 years, male gender, *H. pylori* IgG positivity, IM, and low education below college were found to be associated with a high risk of AG ([Table 2](#T2){ref-type="table"}). Multivariate analysis showed that the significant independent risk factors for AG were older age group of 40 to 59 (OR, 2.55; 95% CI, 2.05 to 3.18), above 60 years old (OR, 5.00; 95% CI, 3.71 to 6.74) compared with below 40, male gender (OR, 1.38; 95% CI, 1.17 to 1.64), *H. pylori* IgG positivity (OR, 1.41; 95% CI, 1.19 to 1.66), IM (OR, 4.29; 95% CI, 3.55 to 5.50), and low education below college (OR, 1.35; 95% CI, 1.01 to 1.79) ([Table 4](#T4){ref-type="table"}). 4. The identification of risk factors for IM by univariate and multivariate analysis ------------------------------------------------------------------------------------ In a univariate analysis of risk factors for IM, older age group of above 40 to 59, older age group of above 60 years, male gender, *H. pylori* IgG positivity, AG, BMI, triglyceride level, *H. pylori* eradication, relatives of gastric cancer, smoking, alcohol, low education, and consumption of dairy product were found to be associated with a high risk of IM ([Table 3](#T3){ref-type="table"}). Multivariate analysis showed that the significant independent risk factors for IM were older age group of 40 to 59 (OR, 3.16; 95% CI, 2.11 to 4.72), above 60 years old (OR, 3.25; 95% CI, 2.05 to 5.15), male gender (OR, 1.88; 95% CI, 1.39 to 2.54), *H. pylori* IgG positivity (OR, 2.17; 95% CI, 1.72 to 2.74), AG (OR, 3.68; 95% CI, 2.95 to 4.60), relatives of gastric cancer (OR, 1.48; 95% CI, 1.12 to 1.96), low education below college (OR, 1.47; 95% CI, 1.06 to 2.00), and consumption of dairy product at least five times per week (OR, 1.40; 95% CI, 1.12 to 1.76) ([Table 5](#T5){ref-type="table"}). DISCUSSION ========== The development of gastric cancer is generally accepted to be a multistep progression from *H. pylori*-related chronic inflammation of the gastric mucosa, to AG, IM, dysplasia, and, finally, intestinal-type gastric cancer. The risk of gastric cancer is closely related to the degree and extension of AG and IM.[@B5],[@B10],[@B11] Therefore, evaluating the prevalence and risk factors for these precancerous lesions such as AG and IM may be helpful to prevent the development of gastric cancer. However, it is nearly impossible to take biopsy from antrum and body in the general population without definite lesion, especially in a huge population as our study. Thus, it is necessary to evaluate prevalence rates of endoscopic AG and IM, especially where the health check-up endoscopy is popular for gastric cancer screening. In the present study, the prevalence of AG and IM were found to be 40.7% and 12.5%, respectively. In Western Europe, the overall prevalence rates of AG and IM were approximately one-third and one-fourth of subjects, respectively.[@B12]-[@B14] In contrast, the prevalence of AG and IM were 55.5% and 28.5%, respectively, in Japan.[@B15] Previously, we have shown that the prevalence rates of AG and IM, diagnosed by histology, were 42.5% and 28.6%, respectively in the antrum.[@B8] Considering histological diagnosis of IM is more accurate and sensitive than endoscopic diagnosis,[@B13],[@B16] the prevalence rate of endoscopic IM in the present study, 12.5%, is not very low. Another reason of lower prevalence of IM in this study could be explained by the population of the participating subjects: 21.5% was below 40 years old, and only 16.7% was ≥60. In our study, the prevalence of AG and IM increased significantly with age and this phenomenon could be explained by *H. pylori* infection. *H. pylori* infection usually occurs in the childhood, but AG and IM progress in the elderly population due to long infectious periods with *H. pylori*.[@B17] In addition, the prevalence of AG and IM in males were significantly higher than those in females. It is generally understood that the prevalence of *H. pylori* infection and gastric disorders in males is more common than that in females.[@B16],[@B17] Furthermore, seropositivity of *H. pylori* was found to be the risk factor of AG and IM in the present study. As it is well known, *H. pylori* infection is the most common risk factor of glandular atrophy and subsequent IM.[@B8] Our consistent results support that endoscopic diagnosis of AG and IM could be a useful tool for localization of high risk group of gastric cancer without histology. Interestingly, AG and IM were found to be a risk factor for each other in our study. Usually, these two lesions have been associated with chronic inflammatory processes such as *H. pylori* infection. Thus, these results might be originated from that atrophy and IM ensue sequentially over time after *H. pylori* infection.[@B18],[@B19] In addition, the degree of IM has been correlated with the degree of atrophy.[@B20] However, IM can also occur without atrophy.[@B5],[@B21] Furthermore, the risk factors for AG and IM were rather different.[@B8] Bacterial factors were found to be important risk factor for AG but host and environmental factors were more important for IM, suggesting that sometimes it does not go together.[@B8] In our study, low education below college was a risk factor for the development of AG and IM. Previous reports showed that lower education was associated with *H. pylori* infection[@B22] and the development of AG in *H. pylori* infected subjects.[@B23] As lower education is usually associated with lower socioeconomic status and poorer hygiene, it is identified as risk factors for *H. pylori* infection.[@B22],[@B23] *H. pylori* infection is likely to contribute to this association between low education and the high prevalence of AG and IM. The proportion of family history of gastric cancer was found to be 11% in this study, rather higher than that in the general population. This could be originated from more concern of the relatives of gastric cancer about their health they might frequently receive health check-up, compared general population. Interestingly, family history of gastric cancer was a risk factor for the development of IM, but not of AG in our study. First-degree relatives of gastric cancer were found to have a higher risk of developing gastric cancer.[@B24],[@B25] The possible causes of familial aggregation of gastric cancer are not only genetic factors but also environmental factors including *H. pylori* infection, excessive intake of salt and N-nitroso compound, and a deficiency of dietary antioxidants among gastric cancer patients and their family members.[@B26] Relatives of gastric cancer have an increased prevalence of precancerous lesions including AG and IM, and *H. pylori* plays an instrumental role in determining the risk of precancerous lesions among relatives of gastric cancer.[@B25],[@B27],[@B28] Consequently, prophylactic *H. pylori* eradication is advised to the relatives of gastric cancer in Korea for prevention of gastric cancer.[@B29] Similar to the family history of gastric cancer, consumption of dairy product at least five times per week was a risk factor for the development of IM but not AG. Previously, milk or yogurt prevents the development of AG and IM through its defense mechanism against the attachment of *H. pylori* to the gastric mucosa[@B30],[@B31] and also, consumption of fermented dairy products confers an enhanced therapeutic benefit on *H. pylori* eradication.[@B32] As our study showed different result further studies are necessary in the future. Our study has limitations. The diagnosis of AG and IM was made only by endoscopic findings and not confirmed by histology. Although histological examination is considered as the gold standard for diagnoses of AG and IM, interobserver variation for diagnoses of AG on biopsy specimens has been shown to be high,[@B33],[@B34] and sampling errors exist, especially when multifocal AG is present.[@B35],[@B36] Also, there are few reports concerning the high agreement between the endoscopic and histological atrophy scores.[@B37],[@B38] The image quality of conventional endoscopes has improved dramatically over the last decade and typical endoscopic findings are interpreted as signs of AG and IM. In addition, to prevent the interobserver variation in the diagnosis of AG and IM, only endoscopic experts participated in this study. Even though the significance of typical endoscopic findings in relation to histology is still uncertain,[@B16],[@B39] similar results regarding the prevalence and risk factor of endoscopic AG and IM to those from the histological AG and IM support clinical significance of our study. Also, the severity and location of endoscopic AG and IM were not classified mainly because there is no standardized grading system of atrophy and IM. In addition, as the endoscopy of this study was performed as one of health check-up it was very difficult to ask the participating doctors to fill up all of those things in detail. In conclusion, the prevalence of endoscopic AG and IM in the general population were not so high in Korea, especially in case of IM. The risk factors of AG and IM were similar but relatives of gastric cancer and consumption of dairy product were the risk factors of IM, but not of AG. Therefore, it is worthwhile to describe the degree and extent of AG and IM in detail during endoscopy, according to diagnostic criteria. This work was supported by grant no. 06-2011-130 from the Seoul National University Bundang Hospital research fund. No potential conflict of interest relevant to this article was reported. {#F1} ###### Baseline Characteristics of the 4,023 Subjects  Data are presented as mean±SD or number (%). NSAID, nonsteroidal anti-inflammatory drug; *H. pylori*, *Helicobacter pylori*; IgG, immunoglobulin G. ###### Univariate Analysis of the Risk Factors for Atrophic Gastritis  Data are presented as number (%). *H. pylori*, *Helicobacter pylori*; IgG, immunoglobulin G; BMI, body mass index; NSAID, nonsteroidal anti-inflammatory drug. ^\*^Some data were missing. Missing values were not included. ###### Univariate Analysis of the Risk Factors for Intestinal Metaplasia  Data are presented as number (%). IM, intestinal metaplasia; *H. pylori*, *Helicobacter pylori*; IgG, immunoglobulin G; BMI, body mass index; NSAID, nonsteroidal anti-inflammatory drug. ^\*^Some data were missing. Missing values were not included. ###### Multivariate Analysis of the Risk Factors for Atrophic Gastritis  B, estimate; SE, standard error; Exp(β), odds ratio; CI, confidence interval; *H. pylori*, *Helicobacter pylori*; IgG, immunoglobulin G. ###### Multivariate Analysis of the Risk Factors for Intestinal Metaplasia  B, estimate; SE, standard error; Exp(β), odds ratio; CI, confidence interval; *H. pylori*, *Helicobacter pylori*; IgG, immunoglobulin G; NS, not significant. [^1]: Young-Eun Joo and Hyun-Kyung Park contributed equally to this work as first authors.

INTRODUCTION ============ The first screening method of screening for trisomy 21 was based on the maternal age. Since the late 1980s, a test was universalized for three fetoplacental products (alpha-fetoprotein, unconjugate estriol, and total hCG) in the maternal serum at the second trimester of gestation. In addition, various sonographic markers such as nuchal fold thickness and femur length in second trimester were introduced, and the nuchal translucency (NT) became important as the early screening method for chromosomal abnormality ([@B1]-[@B3]). The test, using maternal serum biochemical markers, has a relatively fixed cut-off of 1:270, and the sensitivity is known to be about 60%. Many studies have reported that the thickened NT could identify about 75% of affected fetuses for a false positive rate of 5%, but various cut-offs were used in those studies ([@B4]-[@B7]). Although many studies have asserted that different cut-offs of NT according to the crown rump length (CRL) should be used for screening chromosomal aberration, there are few studies actually comparing the sensitivities and false positive rates of NT with different cut-offs within the same group. In this study, we report herein on those values of NT for chromosomal aberration with three different cut-offs, 2.5 mm, 3.0 mm and the 95th percentile for the CRL in a Korean population. MATERIALS AND METHODS ===================== This study was conducted with 2,570 pregnancies by searching our clinical database from January 2001 through December 2001. The inclusion criteria were singleton pregnancies who examined ultrasonographic fetal NT measurement between 11 weeks and 14 weeks of gestation at our institution and performed regular antenatal care during the pregnancy with known pregnancy outcome. Pregnancy outcomes were ascertained from the obstetric and neonatal medical records. This study was approved by the institutional review board. The ultrasonographic scans of fetal NT measurement were carried out by 5 sonographers and 2 obstetricians experienced first-trimester scans more than 5 yr. Fetal NT thickness was measured using the transabdominal ultrasonography (HDI 3000, ATL, Bothell, WA, U.S.A.) in a good mid-sagittal section of the fetus with magnification so that the fetus occupied at least 75% of the image ([@B8]). In cases of visualization proved difficult or suspicious findings, a transvaginal ultrasonography was also used. The NT was defined as the black area between the inner skin on the outline echo and the outer border of the soft tissue overlying the cervical spine. The maximal thickness of the black area was measured with a caliper placed on the lines to 0.1 mm when the sagittal section of the fetuses was obtained. At same time, the fetal CRL was also measured. The range of fetal CRL was from 45 mm to 84 mm. Great care was also taken to clearly distinguish between the fetal skin and the amniotic membrane. In some cases this could only be achieved by asking the mother to cough or by tapping the maternal abdomen. In some cases, it was helpful to use the cineloop function (a function that permits the replay of ultrasound images taken in the past few seconds) to visualize the fetus in a satisfactory position. At least three measurements were taken during the scan and the largest of the three measurements was recorded. We underwent fetal karyotyping in one hundred twenty five women with increased fetal NT (≥2.5 mm) by chorionic villus sampling or amniocentesis. Fetal karyotyping was also offered to women who were maternal age of 35 yr old or more at the time of delivery (n=158), had a positive result on the triple test (n=92), had the history of chromosomal abnormality in any previous pregnancy (n=8), had major fetal structural anomalies on ultrasound examination (n=12), and others (n=24). The karyotyping was also performed in Intrauterine fetal death (IUFD) (n=4) cases. Among the cases who had not been offered fetal karyotyping, we examined the neonates or stillborn for phenotypic abnormality indicating chromosomal aberrations. We excluded the cases (n=8) who had terminated their pregnancies because of major structural anomaly on antenatal ultrasound without fetal karyotyping. The frequencies of chromosomal aberrations were analyzed and the sensitivities and false positive rates were calculated according to three different cut-offs, namely 2.5 mm, 3.0 mm and the 95th percentile for each fetal CRL. We used the data about the 95th percentile of NT for each CRL of our institution ([@B9]). RESULTS ======= A total of 2,570 single pregnancies were included in our analysis. The mean maternal age was 29.9±3.3 yr old. The mean CRL was 60.1±9.1 mm. The frequencies of fetuses with increased NT according to cut-offs of 2.5 mm, 3.0 mm, and the 95th percentile of each CRL were 7.1%, 4.1% and 6.5% respectively. [Table 1](#T1){ref-type="table"} shows the types and frequencies of chromosomal aberrations according to the various cut-offs of NT. There were a total of 31 cases (1.2%) of chromosomal aberrations in this study. Among these cases, trisomy 21 was most frequent chromosomal aberration (38.7% \[12/31\]). The frequencies of chromosomal aberrations that ultrasonographic NT measurement was thickened than that of 2.5 mm, 3.0 mm, and the 95th percentile of NT measurement at each CRL were 11.5% (21/161), 15.9% (17/89), and 13.2% (22/146), respectively. We could detect one more case of trisomy 18 when we used the 95th percentile as the cut-off than the fixed cut-off of 2.5 mm. [Table 2](#T2){ref-type="table"} shows the sensitivities and false positive rates of NT with each cut-off for screening all the chromosomal aberrations and trisomy 21. The sensitivity for detection of chromosomal aberrations was high (70.9%) in the 95th percentile as the cut-off compared with the fixed cut-offs. The false positive rate was 3.5% when using 3.0 mm, but the sensitivity was 54.8%. When using the 95th percentile of NT, the sensitivity was higher and the false positive rate was lower than those of using 2.5 mm for detecting chromosomal aberrations. The sensitivities of the NT measurement to detect trisomy 21 with two cut-offs, 2.5 mm and the 95th percentile were same (75%), and the false positive rate of the 95th percentile (6.2%) was lower than that of 2.5 mm (6.8%). DISCUSSION ========== The various screening methods for detecting chromosomal aberration have been investigated and many studies are ongoing to search for a more efficient screening method. Among the various methods, we evaluated the sensitivities and false positive rates of the screening test of NT measurement alone (not including maternal age) in the first trimester for chromosomal aberrations by using three different cut-offs in a Korean population. The sensitivity of the NT measurement with 3.0 mm as a cut-off was 54.8% in this study. The early studies about NT measurement mainly used 3.0 mm as the cut-off and the reported various sensitivities were from 39% through 93% ([@B10]-[@B12]). This wide range of results may be due to the different study populations and/or to an inadequate gestational age at the time of NT measurement. Most of these studies were mainly targeted to a high risk population that was scheduled for fetal karyotyping ([@B11], [@B12]), so the sensitivity of detecting an affected fetus with an increased NT would be higher than that used in the general population. On the other hand, some early studies reported low sensitivity in their results ([@B13], [@B14]). In those studies, the adequate gestational age (in weeks) of the NT measurement was not established, so cases that measured earlier than 11 weeks of gestation were also included as a large proportion of the subjects. All the cases in which the translucency was not visible or measurable were included in the classification of less than 1.0 mm. The cases under 10 weeks of gestational age had the probability of having an immeasurable NT, and the studies that included these subjects would have results with low sensitivity. Currently, the adequate period of the NT measurement is known to be the time between 11 and 14 weeks of gestation ([@B8]). In this study, we included pregnant woman in this period. As extensive data is accumulated regarding the first-trimester NT screening of populations at high risk, attention has been drawn to those populations at low risk and general population ([@B15]-[@B17]). This data is of major importance because these populations represent the women for whom the vast majority of pregnancies with fetuses having abnormal chromosomes will occur. The results of the present study are similar to those results of Hafner et al. ([@B16]). They assessed what percentage of chromosomal anomalies could be detected by a NT measured in 4,233 unselected women. They diagnosed one additional Down syndrome case by using a sliding scale for the cut-off value with two standard deviations instead of the fixed value of 2.5 mm. So, they suggested that detection rate could probably be improved by raising the values with increasing gestational age. We used the 95th percentile of each CRL as the cut-off value in this study, and we could detect one additional case of trisomy 18. Recent studies have reported the sensitivities of NT according to various gestational age-dependent cut-offs for detecting chromosomal abnormality ([@B15], [@B17], [@B18]). One of the largest multicenter study ([@B18]) showed that the sensitivity of NT with gestational age-dependent cut-offs for trisomy 21 was 77% for a false positive rate of 5%. The frequency of trisomy in our study is higher than that of other studies. We presume the reason for high occurrence of trisomy in this study is that we included two cases of trisomy referred from private clinics with abnormal ultrasound findings in first trimester and another two cases of trisomy with IUFD before amniocentesis (The NT thickness of those cases were all increased over 3 mm). Although the sensitivities and false positive rates of NT for detecting of chromosomal aberration were changed after excluding those four cases (62.9%, 48.1%, 66.6% and 6.4%, 3.6%, 5.8% according to cut-offs 2.5 mm, 3.0 mm, 95th percentile), the superiority of 95th percentile NT value as the cut-off for screening test was not changed. This study is the first study to compare the sensitivities of nuchal translucency alone as a screening test for chromosomal aberration with different cut-offs in a Korean population. When using the 95th percentile of NT, we identified that the sensitivity was higher than that of fixed cut-off for detecting chromosomal aberrations. So, we can suggest that this cut-off should be used for the nuchal translucency measurement screening method in Korean population. ###### Types and frequencies of chromosomal aberrations according to different cut-offs of NT in 2,570 cases  Others: 47,XX+21/46,XX, de novo 45,XX t(13;14) (q10;q10), 47,XY+MAR/46,XY, 45,X/46,X+MAR, 46,XX/46,XY. ###### Sensitivity, specificity, positive predictive rate, false positive rate of each cut-off of NT for screening all chromosomal aberrations and trisomy 21  PPR, Positive predictive rate; FPR, False positive rate.

To the Editor: In January 2005, the chief surgeon in a squadron of French policemen reported a cluster of Plasmodium vivax malaria attacks in troops returning from a 108-day operation in French Guiana. We conducted a retrospective cohort study to describe the malaria attacks and determine factors related to them. A self-administered questionnaire was drawn up, with questions concerning operations in French Guiana (dates, locations) and preventive measures implemented against malaria. A malaria case was defined by the association of clinical signs and Plasmodium parasites in blood smears or quantitative buffy coat tests (per definition of military epidemiologic surveillance). The 40-person mission in French Guiana (Operation Anaconda) took place from July 26, 2004, to November 6, 2004 (108 days of exposure). This mission against clandestine gold panning was conducted in a deep-forest environment where the troops were temporarily housed in villages of Brazilian gold panners. Occasionally, they washed themselves late in the evening in stagnant water near the river and patrolled outside during maximum biting periods. All troops received a chemoprophylaxis (doxycycline 100 mg daily) during the mission and for 4 weeks afterward. From July 2004 through January 2005, 10 persons had [\>]{.ul}1 malaria attacks (attack rate 25%) for a total of 18 malaria attacks (incidence 13/100 person-months of exposure). P. vivax was isolated for 17 attacks and P. falciparum for 1 attack ([Figure](#F1){ref-type="fig"}). Five patients had 1 malaria attack, and 4 patients had up to 3 relapses. Six patients had a malaria attack while receiving doxycycline. {#F1} Regarding chemoprophylaxis compliance, 34% reported missing \<1 dose per week and 32% were fully compliant. The troops did not have permethrin-impregnated battlefield uniforms as do soldiers in the French Army. They had to impregnate their own uniforms with permethrin. Only 37% said they always wore clothing that fully covered them during the mission, and 86% reported having frequently used a repellent. All reported having slept under mosquito nets. No association was found between malaria attacks and regular chemoprophylaxis intake or use of repellents. Only 1 operation in French Guiana was associated with the risk of experiencing malaria attacks: 39% of troops located in Sikini had at least 1 malaria attack versus 7% of troops in other areas (relative risk: 5.9 \[95% confidence interval 0.8--41.7\]). The incidence rate for this study was 10 times higher than the maximum incidence rate observed for French troops deployed in Côte d\'Ivoire (1.3/100 troop-months in 2004). During an earlier Operation Anaconda, 37 of 62 persons deployed near the Sikini area had [\>]{.ul}1 malaria attacks (attack rate 61%). Of these, 30 had [\>]{.ul}1 attacks caused by P. vivax; occasionally an attack was associated with P. falciparum ([@R1]). Our results suggest that the Sikini area was the high-risk area for malaria transmission (although the large confidence interval reflects a lack of power in our analysis). The operation dates (15--28 September) are compatible with the duration of the first cases of malaria occurrence. French Guiana is the only French territory, except for Mayotte, where malaria is endemic, with nearly 5,000 cases per year, occurring mainly along the rivers bordering Suriname and Brazil ([@R2]). The highest frequencies of malaria appear during the dry season (September to December) in French Guiana ([@R3]), but no seasonality was described near the Brazilian border ([@R4]). The Sikini area is located near the Oyapock River (Brazilian border). The mean annual incidence in Amerindians there is 48.6%, mainly due to P. falciparum (incidence 24.8%) and P. vivax (incidence 25.9%) ([@R2]). P. vivax malaria incidence has increased in the Oyapock region, from 30% in 1987 to 50% in 2000--2004 ([@R2]*,*[@R4]*--*[@R7]). French troops were deployed in an area where parasite circulation was high. Troops had contacts with clandestine gold panners, mainly Brazilian illegal residents. This population, in which malaria incidence is almost impossible to evaluate, comes from Amapa State, where the incidence of malaria is increasing ([@R5]). In 2003, 60.9% of patients with malaria cases at Cayenne Hospital had a Brazilian name compared with 35.4% in 2000 ([@R6]). Also, the gold panners diverted the river and built basins where vectors could easily multiply ([@R7]). Initial malaria attacks were treated with chloroquine or quinine. Five patients experienced [\>]{.ul}1 relapses (maximum 3 relapses). The relapses were treated with 50-mg daily doses of primaquine for 4 patients and by chloroquine for the fifth patient. Two patients had relapses after receiving primaquine. Primaquine resistance information was not available. However, resistance to primaquine has emerged in P. vivax strains ([@R8]). We recommended that pre-impregnated battlefield uniforms be available for French policemen and chemoprophylaxis adherence be reinforced by directly observed intake by supervisory staff. Relapses of P. vivax malaria are a major therapeutic problem, particularly after primaquine therapy. *Suggested citation for this article*: Verret C, Cabianca B, Haus-Cheymol R, Lafille J-J, Loran-Haranqui G, Spiegel A. Malaria outbreak in troops returning from French Guiana \[letter\]. Emerg Infect Dis \[serial on the Internet\]. 2006 Nov \[*date cited*\]. <http://dx.doi.org/10.3201/eid1211.060530> We thank G. Debrabander for assistance with the preparation of this article.

A FUNDAMENTAL COMPONENT of moving the US health care system toward lower-cost, higher-quality, and more coordinated care is effectively managing clinically complex individuals, especially those with concurrent chronic conditions, also called multiple chronic conditions (MCC). Multiple chronic conditions is defined as having 2 or more chronic conditions that last more than 1 year and require ongoing medical attention or limit activities of daily living ([@R25]). Approximately 4 in 10 (42%) Americans have MCC, and that rate climbs to 81% for those aged 65 years and older ([@R5]). Care quality, outcomes, and quality of life decline as a person experiences more chronic conditions---including premature death, receipt of conflicting health advice, chance of hospitalizations, and poorer day-to-day functioning ([@R1]). Seventy-one percent of health care spending goes toward treating people with MCC ([@R11]) and those with MCCs account for 93% of total Medicare fee-for-service spending ([@R6]). Having concurrent chronic conditions expands the level of complexity in managing patient health care needs ([@R25]), particularly for populations experiencing socioeconomic factors that impede their ability to access and effectively use high-quality primary care. Many studies have shown that these groups have higher rates of MCC, including older populations ([@R5]); low-income populations ([@R20]); publicly insured populations ([@R2]; [@R26]), especially dually eligible populations ([@R6]) and nonelderly Medicaid beneficiaries ([@R20]; and some racial or ethnic minority groups ([@R10]; [@R20]). These populations are overwhelmingly served by the nation\'s primary health care safety net, yet little is known about the role of safety net providers in managing patients with MCC and, by extension, the role safety net providers may play in transforming the health care system toward greater value-driven care delivery. Community Health Centers (CHCs) are the largest network of safety net primary care providers, serving more than 28 million patients---roughly 1 in 12 residents---across the United States ([@R13]). Community Health Centers must provide a comprehensive set of services that compliment and improve access to primary care, serve everyone no matter their insurance status or ability to pay, be overseen by patient-majority governing boards to ensure responsiveness to community needs, and locate in or serve federally designated medically underserved areas or populations. In 2018, 91% of patients had incomes at or below 200% of the federal poverty level, nearly half (48%) were enrolled in Medicaid, 23% were uninsured, and 63% were members of racial/ethnic minority groups ([@R4]). That same year, CHCs served a diverse range of special populations, including 995 000 agricultural workers, 1.4 million patients with experiencing homelessness, 4.4 million patients living in or near public housing, and 6.7 million patients best served in a language other than English. Only 1 previous study ([@R24]) examined MCC across the CHC population and compared it with the general population utilizing private practices for ambulatory care, finding no significant difference in the average count of chronic conditions between care settings. This study, however, did not control for differences in patient populations and used data from 2006. Given the rapid growth in the Health Center Program following the Affordable Care Act of 2010, the patient population accessing CHCs and experiencing MCCs may look different today. This study seeks to provide a comprehensive and more up-to-date analysis of the rate of MCC within the CHC patient population in comparison with the general population receiving primary care at private practices. Because populations served by the primary care safety net may experience different mixes and rates of MCC, documenting the differences between the health care safety net and non--safety net providers can help inform policy makers, payers, and providers about the resources, capacity, infrastructure, and competencies necessary to effectively identify and manage these complex patients, improve outcomes and health equity, and control costs. This study may also inform risk stratification models, thereby assisting providers with targeting appropriate interventions such as care coordination for the most at-risk patient populations. It may also inform risk adjustment methodologies as payers increasingly turn to alternative payment models that place providers at some level of financial risk for population outcomes. METHODS ======= Data source ----------- Our analysis combined data from the CHC and private practice physician (PPPs) samples of the 2013 National Ambulatory Medical Care Survey (NAMCS), which is the most recent year available for CHC data, although data on PPPs are available for later years. National Ambulatory Medical Care Survey is an annual, nationally representative survey of ambulatory care visits to nonfederal, office-based physicians (PPPs) and includes a stand-alone national survey of CHCs using the same survey instrument and reporting period (December 24, 2012, through December 22, 2013), although the sampling methods for the PPP and the CHC NAMCS differ. The PPP NAMCS used a 2-stage probability sample that first selected physicians within specified geographies and then patient visits within practices. The PPP sampling frame drew from all physicians in the master files maintained by the American Medical Association and the American Osteopathic Association and sampled up to 30 physician visits within a randomly assigned week for each private practice. Of the 6999 in-scope physicians sampled, PPP NAMCS collected responses from 2879 physicians for a response rate of 41% and 54 873 visits. The CHC NAMCS utilized a 3-stage probability sample based on (1) selecting service delivery sites within specified geographies, (2) selecting providers within a site, and (3) sampling visits from providers. The sampling frame drew from a list of delivery sites provided by Health Resources and Services Administration, then sampled up to 3 physicians and nonphysician clinicians within service delivery sites, and finally collecting up to 30 visits within a randomly assigned week for each provider. The 2013 CHC NAMCS collected responses from 1340 services delivery sites and 2289 providers for a 2-stage response rate of 62% and 50 814 visits. Definitions ----------- We limit our analysis to visits drawn from primary care physicians and psychiatrists because (*a*) the CHC NAMCS also sampled nonphysician clinicians and (*b*) the PPP NAMCS also includes specialists who are uncommon in CHCs. Primary care physicians include general and family practice, internal medicine, pediatrics, and OB/GYNs. All results are interpreted in patient visits to primary care physicians and psychiatrists, rather than patients themselves. Data for chronic conditions were drawn from checkboxes (Yes/No) embedded in the NAMCS survey instrument, which captures 14 chronic conditions: arthritis, asthma, cancer, cerebrovascular disease, congestive heart failure, chronic obstructive pulmonary disease, chronic renal failure, depression, diabetes, hypertension, hyperlipidemia, ischemic heart disease, obesity, and osteoporosis. These data were extracted from the patient medical record, with a checked box indicating that a patient was diagnosed at some point previously and not necessarily at the current visit. We also created binary variables for any chronic condition (1 or more), 2 or more, and 3 or more chronic conditions. Analysis -------- The combined CHC and PPP NAMCS public use files contained 105 687 observations in all (50 814 in CHC NAMCS and 54 873 in PPP NAMCS). We removed 50 838 sampled visits (48% of total observations) that were not seen by a primary care physician or a psychiatrist as defined previously (23 440 \[46%\] for CHCs and 27 398 \[50%\] for PPPs). Next, we removed 1068 observations (2% of sampled visits to primary care physicians and psychiatrists) where checkboxes for chronic conditions were blank or unknown (421 \[2%\] for CHCs and 647 \[2%\] for PPPs). Our final sample after these exclusions was 53 781 visits (26 953 for CHCs and 26 828 for PPPs). The logistic regression models (described later) further excluded 3503 responses (7% of our final sample) where the expected source of payment was blank or unknown (1519 \[6%\] for CHCs and 1984 \[7%\] for PPPs). These exclusions were similarly proportionate across variables in the PPP and CHC sample of primary care providers and psychiatrists. We used 2 analyses to capture a comprehensive description of chronic conditions at patient visits to primary care physicians and psychiatrists at CHCs and PPPs. The first analysis measured overall presence of chronic conditions at CHCs and PPPs using χ^2^ tests for specific chronic conditions (eg, arthritis, asthma, etc) as well as for counts of chronic conditions (categorized as 1 or more, 2 or more, and 3 or more chronic conditions). In addition, we tested for differences in prevalence among age groups, including those aged 65 years and older, working age (18-64 years), and children (younger than 18 years). Second, we compared the odds of having any chronic condition, 2 or more chronic conditions, or 3 or more chronic conditions at visits to CHCs and PPPs while controlling for important covariates using logistic regression. In total, we ran 3 regression models---1 for each binary dependent variable (having any chronic condition, 2 or more, and 3 or more chronic conditions). Each model assumed a binomial distribution with a logit link function and adjusted for survey weights and the NAMCS complex sampling design. The independent variable of interest for each model was whether the visit occurred at a CHC or a PPP. We controlled for data collection methods, expected source of payment, and a limited number of patient characteristics; these include (*a*) the type of provider sampled for the visit (categorical); (*b*) whether the visit is from a new or established patient (categorical), which is important for how the chronic conditions checkbox data were collected; (*c*) 4 binary variables for the expected source of payment (private insurance, Medicare, Medicaid, or uninsured); (*d*) age (continuous), which was truncated at age 87 years for consistency between CHC and PPP samples; (*e*) sex (male/female); and (*f*) urban/rural geography, represented as inside or outside a metropolitan statistical area. The binary variable for no insurance was created by combining responses indicating that the expected source of payment was either self-pay or no charge/charity. The logistic regression models do not control for race/ethnicity. Given that our study is interested in describing the rates of chronic conditions for the CHC patient population, which represents a confluence of social factors associated with chronic conditions, we decided not to adjust for race/ethnicity. In addition, there is a large and uneven amount of imputed records for this variable. For PPP NAMCS, 37.9% of responses were missing race, ethnicity, or both ([@R19]). This was the case for 21.2% of responses to the CHC NAMCS ([@R21]). Similarly, we did not control for patient income because NAMCS stopped providing the poverty rate for patient zip codes in the public use files in 2012, and this measure was unavailable for our analyses. Our analyses were conducted using R, version 3.5.3, using the "srvyr" package ([@R8]). All estimates and standard errors account for the complex sampling design of NAMCS by incorporating visit-level survey weight, masked clustered stratum, and primary sampling unit (a masked service delivery site marker in CHC NAMCS and a physician marker in PPP NAMCS) variables provided in the NAMCS public use files. The NAMCS survey weights account for all sampling stages and adjust for nonresponse bias, allowing our weighted estimates to be nationally representative. Bonferroni corrections for multiple comparisons were applied to χ^2^ tests and logistic regression models based on a significance level of *P* ≤ .05. RESULTS ======= Table [1](#T1){ref-type="table"} shows weighted estimates of the variables included in our analyses. The age distribution in patient visits to CHCs and PPPs is very distinct for the youngest and oldest age groups after applying survey weights. Community Health Centers see a large proportion of children younger than 18 years (32.8%) compared with PPPs (23.6%). Conversely, PPPs see more than double the estimated proportion of visits from elderly patients (22.2%) than CHCs (10.3%). ###### Selected Characteristics of Visits to Primary Care Physicians and Psychiatrists in Community Health Centers Versus Private Practices, United States, 2013 ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ Variable Category Community Health Centers\ Private Practice Physicians\ All Visits\ N = 30,880 k[a](#tbl1-1){ref-type="table-fn"}\ N = 524,750 k[a](#tbl1-1){ref-type="table-fn"}\ N = 555,629 k[a](#tbl1-1){ref-type="table-fn"}\ Weighted Percent[b](#tbl1-2){ref-type="table-fn"}\ Weighted Percent[b](#tbl1-2){ref-type="table-fn"}\ Weighted Percent[b](#tbl1-2){ref-type="table-fn"}\ (95% CI) (95% CI) (95% CI) ------------------------------------------------------------- ------------------------------------------------------------- ---------------------------------------------------- ---------------------------------------------------- ---------------------------------------------------- Age (Binned), y \<18 32.8 (28.2-37.3) 23.6 (20.7-26.4) 24.1 (21.3-26.8) 18-34 18.5 (16.6-20.5) 16.7 (15-18.4) 16.8 (15.2-18.4) 35-44 11.3 (10.4-12.3) 10.2 (9.5-11) 10.3 (9.6-11) 45-64 27.1 (24.6-29.5) 27.3 (25.8-28.9) 27.3 (25.8-28.8) 65+ 10.3 (8.6-12) 22.2 (20.4-23.9) 21.5 (19.9-23.2) Sex Female 60.5 (58.9-62.1) 60.4 (58.9-62) 60.4 (59-61.9) Male 39.5 (37.9-41.1) 39.6 (38-41.1) 39.6 (38.1-41) Urban/rural MSA 86.3 (83.4-89.3) 90.4 (88.5-92.3) 90.2 (88.4-92) Non-MSA 13.7 (10.7-16.6) 9.6 (7.7-11.5) 9.8 (8-11.6) Physician specialty General and family practice 57.7 (51.5-63.9) 39.7 (35.5-43.8) 40.7 (36.8-44.6) Psychiatry 3[c](#tbl1-3){ref-type="table-fn"} (−0.2 to 6.3) 6.8 (5-8.6) 6.6 (4.9-8.3) Internal medicine 10.6 (7.1-14.1) 23.6 (19.7-27.4) 22.9 (19.2-26.5) Pediatrics 23.9 (18.7-29.2) 18.9 (16.1-21.7) 19.2 (16.5-21.9) OB/GYN 4.8 (3-6.5) 11 (8.7-13.4) 10.7 (8.4-12.9) New or established patient Established patient 85.2 (82.6-87.8) 89.5 (88.3-90.7) 89.3 (88.1-90.4) New patient 14.8 (12.2-17.4) 10.5 (9.3-11.7) 10.7 (9.6-11.9) Expected source of payment[d](#tbl1-4){ref-type="table-fn"} Private 18 (15.1-21) 59.7 (57.2-62.2) 57.4 (55-59.8) Medicare 13 (11-15.1) 21 (19.1-22.9) 20.6 (18.8-22.3) Medicaid 50 (46.4-53.5) 16.5 (14.5-18.4) 18.3 (16.4-20.2) Uninsured 11.9 (10.1-13.6) 4.2 (3.3-5.2) 4.7 (3.8-5.5) Unknown/blank observations[e](#tbl1-5){ref-type="table-fn"} 5.2 (3.8-6.7) 5.5 (4.4-6.7) 5.5 (4.4-6.6) ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ Abbreviations: CI, confidence interval; k, thousands; MSA, metropolitan statistical area. ^a^These figures reflect estimates of the total number of visits in each setting and are rounded to the nearest 1000 based on guidelines from the National Center for Health Statistics (NCHS). The estimated total visits to Community Health Center and Private Practice Physicians do not sum precisely to the "All Visits" estimate due to rounding. The unweighted number of observations used in this analysis is 53 781 (26 953 for Community Health Centers and 26 828 for private practice physicians). ^b^Survey weights used in the analysis are provided by NCHS in the National Ambulatory Medical Care Survey public use files and account for all sampling stages and adjust for nonresponse bias. ^c^Estimate considered unreliable due to a relative standard error greater than 30%. ^d^Percentages may not add up to 100% because visits may have more than 1 expected source of payment. ^e^Removed from sample for logistic regression models. These distinctive age demographics impacted the results from the χ^2^ analyses of the prevalence of MCC. There were no significant differences in the count of chronic conditions for visits from all patients, children, or the elderly. When looking at visits from working age patients, however, CHC patient visits had higher rates of any chronic condition (63%), 2 or more chronic conditions (34.6%), and 3 or more chronic conditions (16.9%) compared with PPP visits (56.4%, 27.7%, and 13.2%, respectively), as shown in Table [2](#T2){ref-type="table"}. ###### Prevalence of Any or Multiple Chronic Conditions at Visits to Primary Care Physicians and Psychiatrists in Community Health Centers Versus Private Practices by Age, United States, 2013 Age Group Total Chronic Conditions CHC (95% CI) PPP (95% CI) *P* -------------- -------------------------- -------------------------------------- -------------------------------------- ----------------------------------------- All ages ≥1 chronic conditions 51.6 (48.4-54.8) 54.1 (51.7-56.5) .227 ≥2 chronic conditions 27.8 (25.3-30.4) 30.5 (28.2-32.7) .131 ≥3 chronic conditions 13.9 (12.3-15.5) 16.9 (15.2-18.6) .012 Ages 65+ y ≥1 chronic conditions 90.3 (88.2-92.4) 89.4 (87.9-91) .522 ≥2 chronic conditions 71.5 (68.4-74.5) 68.2 (65.1-71.4) .144 ≥3 chronic conditions 40.8 (36.9-44.7) 43.8 (40.4-47.2) .261 Ages 18-64 y ≥1 chronic conditions 63 (60.1-65.9) 56.4 (53.9-58.9) \<.001[a](#tbl2-1){ref-type="table-fn"} ≥2 chronic conditions 34.6 (32.4-36.8) 27.7 (25.5-29.9) \<.001[a](#tbl2-1){ref-type="table-fn"} ≥3 chronic conditions 16.9 (15.4-18.5) 13.2 (11.6-14.9) .002[a](#tbl2-1){ref-type="table-fn"} \<18 y ≥1 chronic conditions 19.7 (16.5-22.9) 15.5 (13.2-17.9) .034 ≥2 chronic conditions 2.3 (1.4-3.3) 1.4 (1-1.8) .029 ≥3 chronic conditions n/a[b](#tbl2-2){ref-type="table-fn"} n/a[b](#tbl2-2){ref-type="table-fn"} n/a[b](#tbl2-2){ref-type="table-fn"} Abbreviations: CHC, Community Health Centers; CI, confidence interval; n/a, not applicable; PPP, private practice physicians. ^a^Significance under Bonferroni correction at 0.005, *m* = 11. ^b^Estimates were considered unreliable because they were based on fewer than 30 observations and had a relative standard error greater than 30%. Table [3](#T3){ref-type="table"} shows the estimated rate of specific chronic conditions by age group. We also provide a summary of chronic conditions with higher prevalence in CHCs or PPPs in Table [4](#T4){ref-type="table"}. Among all ages, CHCs had higher rates of obesity and asthma, while PPPs had higher rates of arthritis, cancer, cerebrovascular disease, hyperlipidemia, ischemic heart disease, and osteoporosis. Among elderly patient visits, CHCs had higher rates of diabetes, while PPPs had higher rates of cancer. Visits from working age patients to CHCs had higher rates of diabetes, hypertension, and obesity; there were no chronic conditions with significantly higher prevalence for these patients visiting PPPs. Children visiting CHCs had higher rates of obesity at CHCs, and no chronic conditions were more prevalent at PPP visits from children. ###### Prevalence of Selected Chronic Conditions at Visits to Primary Care Physicians and Psychiatrists in Community Health Centers Versus Private Practice by Age, United States, 2013 Age Group Chronic Condition CHC (95% CI) PPP (95% CI) *P* -------------- -------------------------- -------------------------------------- -------------------------------------- ----------------------------------------- All ages Arthritis 6.4 (5.6-7.2) 9.7 (8.6-10.7) \<.001[a](#tbl3-1){ref-type="table-fn"} Asthma 8.5 (7.5-9.5) 6.7 (6.1-7.3) .001[a](#tbl3-1){ref-type="table-fn"} Cancer 1.4 (1.1-1.7) 3.8 (3.1-4.6) \<.001[a](#tbl3-1){ref-type="table-fn"} CEBVD 0.9 (0.7-1.1) 1.5 (1.2-1.8) \<.001[a](#tbl3-1){ref-type="table-fn"} Chronic renal failure 1.1 (0.8-1.4) 1.4 (1-1.9) .225 Congestive heart failure 0.9 (0.7-1.1) 1.4 (1.1-1.7) .002 COPD 3.1 (2.7-3.6) 4.2 (3.7-4.8) .002 Depression 11.5 (9.3-13.8) 13.2 (11.8-14.5) .242 Diabetes 13.5 (11.8-15.1) 11.8 (10.7-12.8) .087 Hyperlipidemia 14.6 (12.7-16.4) 19.9 (18-21.8) \<.001[a](#tbl3-1){ref-type="table-fn"} Hypertension 23.8 (21.4-26.1) 26.8 (24.8-28.8) .054 Ischemic heart disease 1.3 (1.1-1.6) 2.5 (2-2.9) \<.001[a](#tbl3-1){ref-type="table-fn"} Obesity 13.5 (11.7-15.2) 8.1 (7.2-8.9) \<.001[a](#tbl3-1){ref-type="table-fn"} Osteoporosis 1.1 (0.8-1.4) 3 (2.5-3.5) \<.001[a](#tbl3-1){ref-type="table-fn"} Ages 65+ y Arthritis 16.1 (13.9-18.4) 20.5 (18.5-22.5) .005 Asthma 5.1 (3.6-6.6) 5.4 (4.2-6.5) .788 Cancer 5.1 (3.7-6.6) 10.4 (8.4-12.4) \<.001[a](#tbl3-1){ref-type="table-fn"} CEBVD 3.5 (2.6-4.4) 4.4 (3.5-5.2) .174 Chronic renal failure 5.4 (3.3-7.6) 4.8 (3.4-6.3) .648 Congestive heart failure 4.4 (3.2-5.6) 4.9 (3.8-5.9) .581 COPD 9.2 (6.8-11.6) 9.5 (8.2-10.8) .813 Depression 11.6 (9.2-14) 15 (12.8-17.3) .049 Diabetes 37.4 (33.6-41.2) 26.5 (24.3-28.7) \<.001[a](#tbl3-1){ref-type="table-fn"} Hyperlipidemia 39.4 (34.2-44.5) 44.1 (40.6-47.7) .137 Hypertension 67.1 (64.4-69.8) 63.9 (61-66.8) .111 Ischemic heart disease 6.5 (4.5-8.4) 8.1 (6.6-9.6) .216 Obesity 11.1 (8.9-13.2) 9.2 (7.5-10.9) .168 Osteoporosis 6.8 (4.5-9.2) 10.7 (9.1-12.3) .016 Ages 18-64 y Arthritis 8 (7.1-8.9) 9.1 (7.8-10.4) .156 Asthma 7.5 (6.6-8.4) 6.7 (6-7.4) .148 Cancer 1.5 (1.1-2) 2.8 (2.1-3.4) .002 CEBVD 1 (0.7-1.2) 0.9 (0.6-1.2) .868 Chronic renal failure 0.9 (0.6-1.1) 0.7 (0.4-0.9) .293 Congestive heart failure 0.7 (0.5-1) 0.6 (0.4-0.8) .292 COPD 3.4 (2.9-4) 3.3 (2.7-3.8) .662 Depression 17.4 (14.1-20.8) 16.6 (15-18.3) .659 Diabetes 16.8 (15.5-18.1) 10.7 (9.5-11.9) \<.001[a](#tbl3-1){ref-type="table-fn"} Hyperlipidemia 18.3 (16.1-20.4) 18.5 (16.4-20.5) .9 Hypertension 29.5 (27.5-31.5) 23.2 (21.4-25) \<.001[a](#tbl3-1){ref-type="table-fn"} Ischemic heart disease 1.2 (0.9-1.5) 1.3 (1-1.5) .687 Obesity 17.4 (15-19.8) 10.1 (9-11.3) \<.001[a](#tbl3-1){ref-type="table-fn"} Osteoporosis 0.7 (0.4-0.9) 1.1 (0.8-1.4) .026 \<18 y Arthritis 0.5 (0.3-0.7) 0.7 (0.4-0.9) .393 Asthma 11.2 (9.4-13.1) 7.9 (6.8-9) .002 Cancer n/a[b](#tbl3-2){ref-type="table-fn"} n/a[b](#tbl3-2){ref-type="table-fn"} n/a[b](#tbl3-2){ref-type="table-fn"} CEBVD n/a[b](#tbl3-2){ref-type="table-fn"} n/a[b](#tbl3-2){ref-type="table-fn"} n/a[b](#tbl3-2){ref-type="table-fn"} Chronic renal failure n/a[b](#tbl3-2){ref-type="table-fn"} n/a[b](#tbl3-2){ref-type="table-fn"} n/a[b](#tbl3-2){ref-type="table-fn"} Congestive heart failure n/a[b](#tbl3-2){ref-type="table-fn"} n/a[b](#tbl3-2){ref-type="table-fn"} n/a[b](#tbl3-2){ref-type="table-fn"} COPD n/a[b](#tbl3-2){ref-type="table-fn"} n/a[b](#tbl3-2){ref-type="table-fn"} n/a[b](#tbl3-2){ref-type="table-fn"} Depression n/a[b](#tbl3-2){ref-type="table-fn"} n/a[b](#tbl3-2){ref-type="table-fn"} n/a[b](#tbl3-2){ref-type="table-fn"} Diabetes n/a[b](#tbl3-2){ref-type="table-fn"} n/a[b](#tbl3-2){ref-type="table-fn"} n/a[b](#tbl3-2){ref-type="table-fn"} Hyperlipidemia n/a[b](#tbl3-2){ref-type="table-fn"} n/a[b](#tbl3-2){ref-type="table-fn"} n/a[b](#tbl3-2){ref-type="table-fn"} Hypertension n/a[b](#tbl3-2){ref-type="table-fn"} n/a[b](#tbl3-2){ref-type="table-fn"} n/a[b](#tbl3-2){ref-type="table-fn"} Ischemic heart disease n/a[b](#tbl3-2){ref-type="table-fn"} n/a[b](#tbl3-2){ref-type="table-fn"} n/a[b](#tbl3-2){ref-type="table-fn"} Obesity 7.4 (5.3-9.5) 2.3 (1.7-2.9) \<.001[a](#tbl3-1){ref-type="table-fn"} Osteoporosis n/a[b](#tbl3-2){ref-type="table-fn"} n/a[b](#tbl3-2){ref-type="table-fn"} n/a[b](#tbl3-2){ref-type="table-fn"} Abbreviations: CI, confidence interval; CEBVD, cerebrovascular disease; CHC, Community Health Centers; COPD, chronic obstructive pulmonary disease; n/a, not applicable. ^a^Significance under Bonferroni correction at 0.001, *m* = 45. ^b^Estimates for either CHCs or private practice providers were considered unreliable because they were based on fewer than 30 observations or had a relative standard error greater than 30%. ###### Chronic Conditions With a Higher Prevalence at Visits to Community Health Centers Versus Private Practice Physicians, United States, 2013[a](#tbl4-1){ref-type="table-fn"} Age Group Community Health Centers (CHC) Private Practice Physicians (PPP) ------------------------ -------------------------------- ----------------------------------- All ages Obesity Arthritis Asthma Cancer \... Hyperlipidemia \... Ischemic heart disease \... Osteoporosis \... CEBVD Ages 65 y and above Diabetes Cancer Working age (18--65 y) Diabetes \... Hypertension \... Obesity \... Children (\<18 y) Obesity \... Abbreviation: CEBVD, cerebrovascular disease. ^a^Inclusion of a chronic condition in the center column indicates that prevalence of this condition was higher among CHC patient visits; inclusion in the right column indicates higher prevalence among PPP patient visits. In our multivariable analyses using logistic regression (Table [5](#T5){ref-type="table"}) to control for data collection methods, expected source of payment, and a limited number of patient characteristics, CHCs had higher odds of having any chronic condition (odds ratio \[OR\] = 1.35) and 2 or more chronic conditions (OR = 1.306) but no difference for 3 or more chronic conditions. Medicaid-covered and uninsured patient visits were also significant predictors. Medicaid was associated with increased odds of having 2 or more chronic conditions (OR = 1.36) and 3 or more chronic conditions (OR = 1.57). Uninsured patient visits were less likely to have 1 or more (OR = 0.562), 2 or more (OR = 0.617), and 3 or more (OR = 0.557) chronic conditions. Age (in years) was also a significant predictor in all 3 logistic regression models with ORs ranging from 1.055 to 1.061 in each model. ###### Estimated Odds Ratios for 1+, 2+, and 3+ Chronic Conditions at Visits to Primary Care Physicians and Psychiatrists in Community Health Centers Versus Private Practices, United States, 2013 Variable Outcome: Any Chronic Condition Outcome: ≥2 Chronic Conditions Outcome: ≥3 Chronic Conditions --------------------------------------------------------- -------------------------------- ---------------------------------------- -------------------------------- ---------------------------------------- --------------------- ---------------------------------------- Setting of care (reference: private practice physician) Community health center 1.35 (1.18-1.545) \<.01[a](#tbl5-1){ref-type="table-fn"} 1.306 (1.102-1.548) \<.01[a](#tbl5-1){ref-type="table-fn"} 1.11 (0.906-1.36) .31 Data collection methods Provider type (reference: general and family practice) Psychiatrist 2.041 (1.382-3.014) \<.01[a](#tbl5-1){ref-type="table-fn"} 0.324 (0.19-0.555) \<.01[a](#tbl5-1){ref-type="table-fn"} 0.292 (0.157-0.541) \<.01 Internal medicine 1.403 (1.171-1.68) \<.01[a](#tbl5-1){ref-type="table-fn"} 1.356 (1.104-1.665) \<.01[a](#tbl5-1){ref-type="table-fn"} 1.26 (0.98-1.62) .07 Pediatrician 0.701 (0.562-0.875) \<.01[a](#tbl5-1){ref-type="table-fn"} 0.366 (0.259-0.516) \<.01[a](#tbl5-1){ref-type="table-fn"} 0.241 (0.125-0.465) \<.01[a](#tbl5-1){ref-type="table-fn"} OB/GYN 0.275 (0.223-0.339) \<.01[a](#tbl5-1){ref-type="table-fn"} 0.241 (0.186-0.311) \<.01[a](#tbl5-1){ref-type="table-fn"} 0.228 (0.165-0.315) \<.01[a](#tbl5-1){ref-type="table-fn"} Patient type (reference: established patient) New patient 0.653 (0.541-0.787) \<.01[a](#tbl5-1){ref-type="table-fn"} 0.607 (0.494-0.747) \<.01[a](#tbl5-1){ref-type="table-fn"} 0.626 (0.51-0.769) \<.01[a](#tbl5-1){ref-type="table-fn"} Expected source of payment Private insurance 0.871 (0.714-1.063) .17 0.83 (0.668-1.031) .09 0.793 (0.631-0.995) .05 Medicare 1.157 (0.936-1.429) .18 1.03 (0.826-1.285) .79 0.947 (0.762-1.178) .63 Medicaid 1.138 (0.92-1.407) .23 1.361 (1.085-1.707) \<.01[a](#tbl5-1){ref-type="table-fn"} 1.567 (1.237-1.983) \<.01[a](#tbl5-1){ref-type="table-fn"} Uninsured 0.562 (0.416-0.76) \<.01[a](#tbl5-1){ref-type="table-fn"} 0.617 (0.446-0.854) \<.01[a](#tbl5-1){ref-type="table-fn"} 0.557 (0.385-0.805) \<.01[a](#tbl5-1){ref-type="table-fn"} Patient characteristics Age (Years) 1.055 (1.051-1.06) \<.01[a](#tbl5-1){ref-type="table-fn"} 1.061 (1.056-1.066) \<.01[a](#tbl5-1){ref-type="table-fn"} 1.06 (1.055-1.065) \<.01[a](#tbl5-1){ref-type="table-fn"} Sex (reference: Female) Male 0.96 (0.872-1.057) .4 1.066 (0.955-1.189) .26 1.015 (0.89-1.157) .83 Rurality (reference: Urban) Rural 0.967 (0.756-1.235) .79 1.138 (0.891-1.454) .3 0.869 (0.682-1.107) .26 Abbreviations: CI, confidence interval; OR, odds ratio. ^a^Significance under Bonferroni correction at 0.017, *m* = 3. We conducted sensitivity analyses for the logistic regression models by adding race/ethnicity as a control variable and again using a recoded age variable with 5 distinct age groups. Most coefficients were similar in size and direction. Notably, Medicare as the expected source of payment became a significant and slightly more powerful predictor in each model when we used the recoded age variable as a predictor. We also performed a quasi-Poisson regression using the total count of chronic conditions as the outcome variable, finding again that all coefficients were roughly similar in direction and significance compared with the logistic regression models. DISCUSSION ========== This is the first study to provide a comprehensive description of the rates of MCC at patient visits to the nation\'s largest network of safety primary care providers. The results of this study indicate that CHCs provide a high volume of care for patients with chronic conditions, particularly MCCs, and that visits to CHCs have a higher rate of MCCs compared with PPPs. In particular, visits among the bulk of health center patients---those ages 18 to 64 years---are more likely to be for patients with 1 or more, 2 or more, and 3 or more chronic conditions compared with the same group at PPPs. After controlling for data collection methods, expected source of payment, and a limited number of patient characteristics, the average visit to CHCs has a 31% higher odds of having 2 or more chronic conditions and 35% higher odds of having any chronic condition compared with PPP patient visits. Multiple regression models indicate that Medicaid coverage, by far the dominant insurer across CHC patients nationally, was strongly associated with having 2 or more and 3 or more chronic conditions. These findings show that Medicaid is clearly a critical insurer for those patients experiencing MCC, even after adjusting for age and other covariates, and helps make it possible for patients to access care. The reverse is true for lack of insurance, where we see that visits for those without coverage are less likely to have been diagnosed with MCCs, possibly because individuals seek out coverage (whether from private insurance, Medicaid, or elsewhere) if they receive a diagnosis. Age is also a significant, leading, and positive predictor of having higher counts of chronic conditions. Our analysis of prevalence for specific chronic conditions between age groups illustrates this point. When all ages are included, the prevalence of obesity and asthma is greater at CHCs, whereas conditions such as arthritis, cancer, cerebrovascular disease, hyperlipidemia, ischemic heart disease, and osteoporosis are greater at PPPs. However, when looking at patients younger than 65 years, visits to CHCs showed a higher rate of costly diseases that disproportionately affect low-income and minority communities such as diabetes, hypertension ([@R15]), and obesity ([@R23]). Community Health Centers serve far more children, who tend to experience few of the studied chronic conditions overall, and far fewer elderly patients, who tend to have far more. Proportionately, both CHCs and PPPs serve roughly the same amount of working-age patients. This study demonstrates that CHCs improve access to care for medically underserved populations and patients with extensive and costly health care needs. Previous research shows that health centers also excel in quality standards and controlling the costs of care. For example, a previous study ([@R12]) compared 18 quality measures at CHCs and PPPs, finding that, after controlling for patient characteristics, CHCs performed better on 6 measures and no differently than private practices on 12 measures. Other studies have found cost savings for children ([@R3]) and Medicaid ([@R22]) and Medicare ([@R17]) beneficiaries utilizing CHCs for primary care compared with other providers. By serving a generally younger patient population than PPPs, health centers also play an important role in screening, treating, and managing common chronic conditions before they progress into more acute stages---or lead to new chronic conditions---especially as patients age into Medicare. As health centers treat more patients with costly, complex, and concurrent chronic conditions, they will play a larger role in bending the cost curve and improving health outcomes for the nation\'s medically underserved. Community Health Centers are rapidly growing to serve more underserved patients and communities, and the growth in the number of health center patients with chronic illnesses such as diabetes, depression, human immunodeficiency virus, substance use disorder, chronic obstructive pulmonary disease, and obesity has outpaced patient growth overall ([@R18]), and many of these conditions often co-occur. Effectively managing patients with chronic conditions---especially MCC---requires ongoing care coordination, patient support, and care integration. In many cases, patients with 1 or more chronic conditions may require access to specialty care to manage their conditions. Community health center patients have greater challenges accessing needed specialty care, especially those covered by Medicaid or who are uninsured ([@R7]; [@R9]; [@R14]). Federal statute governing the Health Center Program requires that CHCs offer a comprehensive set of services beyond primary care, such as behavioral, oral, pharmacy, and "enabling" services that facilitate access to and better use of health care services (examples of these "enabling" services often include transportation, case management, health education, and translation). Recent research demonstrates that enabling services help patients navigate the health care system and achieve greater access to needed care ([@R27]) and may even lead to improved health outcomes, although more research is needed on specific patient outcomes. Sustaining CHC financing is necessary to ensure that CHCs can continue to provide comprehensive, high-quality, and integrated primary care to treat patients with or at risk of chronic conditions. This is particularly important for patients with MCC whose complex health needs generally require a higher volume of health care services, including nonclinical enabling services that address nonclinically derived causes of poor health and higher costs. Health center financing is also important for delivering effective preventive care so that patients can be at lower risk of developing additional chronic illnesses. Nationally, the largest sources of CHC financing are Medicaid reimbursement and federal grants through Section 330 of the Public Health Services Act. Although Section 330 federal funding makes up the bulk of grant revenues, Medicaid makes up CHCs\' largest source of revenue overall (44%) ([@R4]). Health centers depend on 330 grants and adequate Medicaid reimbursement to remain viable and provide both insured and uninsured patients the full range of services necessary to prevent and treat chronic conditions, some of which, like enabling services, are often not reimbursable by third-party payers. Moreover, as Medicaid and other payers increasingly move toward value-based care, the cost of providing the full range of clinical and nonclinical services necessary for managing patients with MCC must be accounted for within new payment models, including the cost of needed face-to-face encounters. Future research is needed to explore the impact of recent CHC growth, particularly since implementation of the Affordable Care Act, which made possible rapid CHC expansion. Medicare is a growing source of coverage for CHC patients and CHC revenue, with many Medicare CHC patients dually enrolled in Medicaid, given their low incomes. As patients age at CHCs, we expect the volume of MCC to increase, especially as patients age into Medicare. Further research is also needed to guide how risk adjustment models can better account for the extent to which providers---particularly safety net providers---are successfully managing high-need, high-risk patients. This study has important limitations. The checkbox data for 2013 NAMCS are limited to 14 chronic conditions and exclude some important conditions, such as substance use disorder. National Ambulatory Medical Care Survey also captures only diagnosed chronic conditions and may undercount the true rate of MCC, especially among the uninsured. In addition, the most recent CHC NAMCS data were collected in 2013, prior to many states implementing Medicaid expansion, which increased access to health insurance and primary care for millions of Americans ([@R16])---many of which utilized CHCs. CONCLUSION ========== This study documents the different rates of MCC and mixes of chronic conditions at primary care and mental health visits to CHCs and PPPs. Well-coordinated, integrated, comprehensive, and continuous primary care is essential for managing MCC as well as the social determinants that exacerbate them, especially for vulnerable populations utilizing the health care safety net. Community Health Centers are well positioned to serve complex patients, given their unique model of care and mission to serve clinically and socially complex populations. The authors acknowledge Ron Yee, MD, MBA, FAAFP, for his assistance with this manuscript. The authors have disclosed that they have no significant relationships with, or financial interest in, any commercial companies pertaining to this article.

INTRODUCTION ============ With an aging population, cognitive decline and dementia are becoming major public health issues \[[@R1]\]. An estimated 35.6 million people were affected by dementia in 2010, and this number is projected to almost double every 20 years \[[@R2]\]. Cognitive decline may involve one or several cognitive domains, such as memory, attention, and language \[[@R3]\], as well as functional abilities, such as activities in daily living (ADLs) \[[@R4]\]. Cognitive decline commonly coexists with cardiovascular disease, which seems to play a major role in the development and progress of cognitive decline and dementia \[[@R5]--[@R9]\]. Cardiovascular disease is also the most common cause of death in people with vascular dementia and mixed dementia \[[@R10],[@R11]\]. In very old (age ≥ 80 years) individuals, cognitive impairment is associated with low systolic BP (SBP) \[[@R12]--[@R15]\]. As both low SBP and cognitive impairment are separately associated with increased mortality risk \[[@R16]--[@R21]\], the mortality risk associated with blood pressure (BP) may differ with respect to level of cognitive function. Systematic reviews of randomized controlled trials have investigated the effects of antihypertensive therapy in patients with dementia aged at least 65 years \[[@R22]\] and more than 80 years \[[@R23]\], but limited statistical power prevented the authors from drawing definitive conclusions. A population-based study of older people (age ≥ 75 years; mean age, 83 years) revealed an independent association between low BP and increased mortality risk in cognitively impaired participants \[[@R24]\], defined by mini-mental state examination (MMSE) scores less than 24. Other population-based studies with somewhat younger participants (age ≥ 65 years; mean ages, 74 and 76 years) showed associations between diastolic BP (DBP), but not SBP and mortality that differed with respect to MMSE score \[[@R25],[@R26]\]. This discrepancy may be because of age-related changes in SBP level \[[@R27],[@R28]\] and the type of dementia \[[@R29]--[@R31]\]. Owing to these processes, extrapolation of results from younger populations to the very old may be misleading. The aim of this study was to investigate whether the association of BP with mortality differs with respect to MMSE score in a representative sample of very old individuals, including those with severe cognitive impairment. METHODS ======= Setting ------- Baseline data for this study were taken from the population-based, prospective Umeå 85+/GErontological Regional DAtabase (GERDA) study, conducted by Umeå University, Sweden, in collaboration with Åbo Akademi University and the University of Vaasa, Finland. The Umeå85+/GERDA study has been described in detail elsewhere \[[@R32]\]. It was approved by the Regional Ethical Review Board in Umeå (§99--326, §05--063 mol/l, §09--178 mol/l, §2015-296-31) and the Ethics Committee of Vaasa Central Hospital (§05--87). The objective of the GERDA study is to increase knowledge of the living conditions of very old people, increase quality of life in this population, and provide data to support planning of future eldercare. Design ------ Every other inhabitant of eight municipalities in northern Sweden and western Finland aged 85 years, and all of those aged 90 and at least 95 years, as listed in national population and tax registers, were invited to participate. All participants provided informed oral consent and relatives also provided informed oral consent when appropriate. Baseline data for Swedish participants in the present study were collected in three rounds, commencing in the years 2000, 2005, and 2010, and those for Finnish participants were collected in one round commencing in 2005. Trained assessors with medical education collected data using a standardized questionnaire and assessments at participants' homes or care facilities and medical records. Data were also collected from care personnel and relatives of cognitively impaired individuals as proxy respondents when possible. The questionnaire included subjective and objective questions and covers sociodemographic, socioeconomic, and medical factors; quality of life; and attitudes toward aging, participants' current situation, and the current situation with eldercare. Participants ------------ Participants in the Umeå85+/GERDA study from whom SBP measurements and MMSE scores were collected during home visits were included in the present study. For those who participated in more than one round of data collection, the first sets of data from home visits were included in this study. All participants were followed to determine survival status after 2 years. Measures -------- The study outcome was all-cause mortality within 2 years. Dates of death were collected from death certificates, population registers, and medical records. Information on cohabitation, education, and smoking status was collected from the respondent. Information on diagnoses, medical conditions, and drug prescriptions collected from the respondent was verified and complemented using medical records from hospitals, general practitioners, and care facilities. A specialist in geriatric medicine verified and complemented diagnoses using all collected data, including assessments, tests, pharmacological treatments, and medical records. Dementia and depression were diagnosed according to the 'Diagnostic and Statistical Manual of Mental Disorders,' fourth edition, text revision \[[@R33]\] criteria. Participants' height and weight were measured to calculate BMI (kg/m^2^). The Barthel ADL index was used to assess dependency in daily activities on a scale of 0--20, with a score of 20 indicating total independence in personal ADLs \[[@R34]\]. BP was measured using a calibrated manual sphygmomanometer and stethoscope after 5 min rest in a supine position. Pulse pressure was calculated by subtracting DBP from SBP. To allow for detection of nonlinear associations with mortality, SBP, and pulse pressure were classified in quintiles (≤125, 126--139, 140--149, 150--164, and ≥165 mmHg and ≤55, 56--65, 66--75, 76--90, and \>90 mmHg, respectively). DBP was classified in quartiles because of its narrower distribution (\<70, 70--74, 75--80, and \>80 mmHg). Cognitive impairment was assessed using the MMSE, a validated and commonly used screening tool for cognitive impairment \[[@R35],[@R36]\]. In community and primary care settings, the MMSE has 82--88% sensitivity for dementia and 86% specificity \[[@R37],[@R38]\] in pooled analyses (most common cutoff value, 23/24). MMSE scores range from 0 to 30, with higher scores indicating better cognitive function. Participants were classified into four subcohorts based on MMSE score (0--10, 11--17, 18--23, and 24--30). MMSE score cutoff values of 17 or less and at least 24 are commonly used to differentiate severe, mild, and no cognitive impairment, respectively \[[@R36]\]. To investigate differences within the MMSE score = 0--17 group, an additional cutoff value of 10 or less designating very severe cognitive impairment was used; this threshold has been used previously to indicate severe dementia \[[@R39],[@R40]\]. Statistical analysis -------------------- Associations between all-cause mortality and categorized BP variables were analyzed using Cox proportional-hazard regression models. The BP category with the lowest risk was used as reference. Two models were developed: a basic model, including only age, sex, and BP (model 1) and a fully adjusted model, including age, sex, baseline characteristics associated with death, and BP (model 2). BP variables (SBP, DBP, and pulse pressure) were entered separately into each model. Baseline characteristics potentially associated with death (Table [1](#T1){ref-type="table"}) were identified from previous research. Associations between baseline characteristics and 2-year mortality were tested using Student\'s *t*-test and Pearson\'s χ^2^ test. Bivariate correlations were tested between all baseline characteristics associated with death, and the dementia and antidepressants variables were removed because of strong correlations (*r* \> 0.6) with MMSE score and depression, respectively. The BMI variable was removed because of missing values. Remaining baseline characteristics associated with death at a significance level of *P* ≤ 0.15 were entered in a multivariate analysis using Cox proportional-hazard regression models. To reduce the number of variables \[[@R41]\], only baseline characteristics associated with death at a significance level of *P* ≤ 0.05 in the multivariate analysis were entered into model 2, together with BP. The Schoenfeld residuals test was used to identify time-dependent variables \[[@R42]\]. Interaction effects between BP and MMSE score were tested by entering an interaction term into each model. These analyses were performed using SPSS Statistics software (version 22.0; IBM Corporation, Armonk, New York, USA). Associations of mortality with SBP and MMSE score in the total sample were explored graphically with flexible parametric models (knots in default position) using STATA statistical software (release 12; StataCorp 2011, College Station, Texas, USA). All analyses were two tailed and *P* \< 0.05 was considered to be significant. RESULTS ======= The flow of study participation is shown in Fig. [1](#F1){ref-type="fig"}. In total, 1115 participants were included in the present study population (participation rate, 66% of those who received invitations), of whom 293 (26%) participants died within 2 years (mean ± SD, 1.73 ± 0.53 years). The numbers of deaths were about equally distributed among MMSE score subcohorts, but larger proportions of participants in the lower than in the higher MMSE score subcohorts died (mortality rates: MMSE = 0--10, 0.593; MMSE = 11--17, 0.506; MMSE = 18--23, 0.239; MMSE = 24--30, 0.129). Compared with the study population, individuals who declined participation or from whom no SBP measurement or MMSE score was obtained did not differ significantly in age (89.8 ± 4.8 vs. 89.4 ± 4.6 years, *P* = 0.064), but the proportion of women was larger (75.0 vs. 66.5%, *P* \< 0.001). {#F1} Table [1](#T1){ref-type="table"} shows the baseline characteristics of the study population according to MMSE score. Age and prevalence of dementia showed increasing trends with decreasing MMSE score, whereas BP, pulse pressure, and Barthel ADL index showed decreasing trends. Individuals with MMSE scores of 0--10 showed the highest prevalence of cerebrovascular disease, hip fracture, and angina pectoris and the lowest prevalence of atrial fibrillation. Individuals with MMSE scores of 11--17 had the highest prevalence of depression, atrial fibrillation, and congestive heart failure, and the highest mean number of prescribed drugs. Figure [2](#F2){ref-type="fig"} shows survival curves based on flexible parametric models according to SBP and MMSE score in the total sample. As can be seen, mortality risk was increased primarily among individuals with low SBP (Fig. [2](#F2){ref-type="fig"}a) or MMSE score less than 21 (population mean, Fig. [2](#F2){ref-type="fig"}b). The survival curve according to MMSE score had a sigmoidal shape; maximum risk was approximately 3.4, among individuals with MMSE scores less than 5. {#F2} Model 2 included age, sex, atrial fibrillation, depression, and Barthel ADL index. No time-dependent variables were identified. Interaction effects between MMSE score subcohorts and SBP categories were borderline significant in models 1 (*P* = 0.069), and 2 (*P* = 0.068). Interaction effects between MMSE score subcohorts and DBP and pulse pressure categories respectively were not significant in model 1 or 2 (data not shown). Table [2](#T2){ref-type="table"} presents hazard ratios for death according to SBP category and MMSE score subcohort. Survival curves based on Cox proportional-hazard regression models are shown in Fig. [3](#F3){ref-type="fig"}. Among participants with MMSE scores of 0--10, SBP at least 165 and 125 mmHg or less were associated with increased mortality risk \[model 2: hazard ratio 4.54, 95% confidence interval (CI) = 1.52--13.60 and hazard ratio 2.23, 95% CI = 1.12--4.45, respectively\], compared with SBP of 126--139 mmHg in the fully adjusted model. In model 1, SBP 125 mmHg or less was associated with increased mortality risk also among participants with MMSE scores of 18--23 and 24--30, compared with SBP at least 165 mmHg (hazard ratio 2.81, 95% CI = 1.20--6.54 and hazard ratio 2.19, 95% CI = 1.09--4.40, respectively). For DBP and pulse pressure, the only significant associations were with MMSE scores of 18--23 in model 1; increased mortality risk was observed for DBP less than 70 mmHg compared with 75--80 mmHg (hazard ratio 1.96, 95% CI = 1.04--3.70) and pulse pressure 55 mmHg or less compared with more than 90 mmHg (hazard ratio 2.84, 95% CI = 1.08--7.45). {#F3} DISCUSSION ========== In this study of 1115 very old individuals, the association between SBP and mortality differed with respect to MMSE score. BP was not associated independently with mortality risk, except among participants with MMSE scores of 0--10. Among these participants, high (≥165 mmHg) and low (≤125 mmHg) SBP were associated independently with mortality risk, compared with intermediate (126--139 mmHg) SBP. In age and sex-adjusted analyses, low SBP was also associated with increased mortality risk in participants with MMSE scores at least 18. To our knowledge, the association of BP and mortality has not previously been analyzed separately in individuals with MMSE scores 10 or less. In contrast to those with higher cognitive function, these individuals seem to be vulnerable to the harmful effects of high and low SBP, indicating a nonlinear association. Larger studies are needed to confirm the risks associated with high and low SBP in people with very severe cognitive impairment, but this finding indicates that BP research in general very old populations may not be extrapolated to those with very severe cognitive impairment. Previously, an independent association of low SBP with 5-year mortality was shown to pertain exclusively to very old individuals with MMSE scores less than 24, as opposed to those with scores at least 24 \[[@R24]\]. The results of the present study indicate that the association of SBP with mortality pertains to individuals with much lower cognitive function than previously known, and that BP may not be an individual risk factor for 2-year mortality in individuals with MMSE scores more than 10, which is the great majority of the population. The previous findings of independent associations between low DBP and mortality risk that differed with respect to MMSE score \[[@R24]--[@R26]\] were not replicated in the present study. Low SBP was a marker for increased mortality risk in individuals with MMSE scores at least 18. These results are consistent with some previous BP research among very old individuals, the majority of which have no or mild cognitive impairment \[[@R19]--[@R21]\]. Specifically, atrial fibrillation, depression, and dependency in ADLs seem to account for mortality risk associated with low SBP in the present study, as the associations were not robust against adjustments for these factors. Notably, low SBP was not a marker or risk factor for mortality risk in individuals with MMSE scores of 11--17, in contrast to those with higher and lower scores. This inconsistency across MMSE score subcohorts may imply that different mechanisms underlie the mortality risk associated with low SBP in very old individuals with MMSE scores less than 11 and more than 17. The increased mortality risk with high and low SBP in individuals with very severe cognitive impairment may be mediated by cerebrovascular disease, such as microinfarctions, hemorrhages, and white matter hyperintensities. Cerebrovascular disease accounts for 12--17% of cardiovascular causes of death in people with vascular or mixed dementia \[[@R11]\]. Individuals with very severe cognitive impairment are more likely to have reduced cerebral autoregulation of BP because of severe dementia \[[@R43],[@R44]\]. This reduced autoregulation may predispose individuals to further cerebrovascular damage, caused by dissociations in cerebral blood flow and demand \[[@R10],[@R45]--[@R49]\]. Individuals with high or low SBP in combination with reduced cerebral autoregulation of BP may be at greater risk of such cerebrovascular damage \[[@R49],[@R50]\], particularly in the presence of arterial stiffness \[[@R51]\]. The strengths of the present study include the use of specially trained assessors, who maintained the high quality of data collection and home visitation to minimize healthy user bias. The classification of BP values and MMSE scores using multiple categories allowed for interpretation of nonlinear associations, but resulted in potentially limited statistical power of some analyses and the inability to examine the impacts of age, sex, and BP treatment. The study has other limitations. Single measurement of MMSE score and BP did not capture variability in these measures over time, but was likely sufficient for group-level statistical analysis. A 24-h BP monitoring could have given valuable information about BP variability and episodic hypotension \[[@R52]\]. BP was measured with participants in a supine position, which may have yielded higher values relative to a seated position because of the high prevalence of orthostatic hypotension in very old people \[[@R53]\]. The proportion of women was smaller in the study sample compared with individuals who declined participation or from whom no SBP measurement or MMSE score was obtained, which may impede the generalizability of the results. The underrepresentation of women in the study sample has probable consequences for the generalizability of the results, such as the lower-than-average prevalence of conditions with female predominance (e.g. dementia) \[[@R54]\]. Because the MMSE has higher sensitivity for severe than for mild cognitive impairment, people with mild cognitive impairment may have been misclassified \[[@R36]\], potentially interfering with comparability among subcohorts. The validity of MMSE scores in terms of cognitive impairment is affected by several factors that may be relevant for our study population, such as age, length of education, verbal fluency, sensory impairment, cultural background, institutional living, and depression \[[@R36],[@R55],[@R56]\]. In conclusion, in very old individuals, the association between SBP and mortality appears to differ with level of cognitive function. Very old individuals with very severe cognitive impairment and low or high SBP may have increased mortality risk. Low SBP may be a marker for increased mortality risk in individuals with no or mild cognitive impairment. Larger studies are needed to confirm these results, but the results of BP research in the general very old population should not be extrapolated to those with very severe cognitive impairment. To better target groups potentially benefitting from intervention, we need to define easily distinguishable groups within this heterogeneous population with and without increased risk for negative outcomes of high or low BP. The issue of treatment individualization is particularly important in very old people, who are especially susceptible to adverse drug reactions and polypharmacy. ACKNOWLEDGEMENTS ================ Part of this work was previously presented at the 8th International Association of Gerontology and Geriatrics European Region Congress in Dublin, Ireland, April 2015; and at the Clinical Osteoporosis Research School State Of The Art Symposium: Falls and Fractures in the Oldest Old in Stockholm, Sweden, May 2015. Sources of funding and support: The study was supported by the Swedish Research Council (grant no. K2014-99X-22610-01-6), a regional agreement between Umeå University and Västerbotten County Council on cooperation in the fields of Medicine, Odontology and Health, the Research Foundation of the Faculty of Medicine and Odontology at Umeå University, the Detlof Research Foundation, the Swedish Dementia Association, and the European Union and the Regional Development Fund: the Interreg IIIA Mitt-Scandia and the Bothnia-Atlantica Program. Conflicts of interest --------------------- There are no conflicts of interest. Abbreviations: ADLs, activities in daily living; BP, blood pressure; CI, confidence interval; GERDA, gerontological regional database; MMSE, mini-mental state examination Reviewer\'s Summary Evaluation ============================== Reviewer 2 ---------- This observational study aimed at investigating the association of BP with all cause mortality in subjects over 85 years with different MMSE scores (four subgroups). Subjects were a representative sample of the population in Sweden and Finland (GERDA study), including 1115 individuals. After 2 years follow-up 26% of subjects had died. The higher mortality risk was observed among those with the most severe dementia. Mortality risk in subjects with MMSE 10 or less was not linear and increased in patients with SBP of 125 mmHg or less, or at least 165 mmHg compared with intermediate SBP (126--139 mmHg) values following a 'J'-shaped curve. Despite several limitations (BP measured in supine position overestimates BP values; underrepresentation of women; MMSE much more sensitive for severe cognitive decline than for mild cognitive impairment), the nonlinearity relationship between SBP and mortality in very old subjects with dementia has implications for treatment of these patients. ###### Baseline sociodemographic and clinical characteristics of the study population MMSE score ----------------------------------------------- -------------- -------------- -------------- -------------- -------------- Age (years) 89.4 ± 4.6 92.5 ± 4.8 91.1 ± 4.9 89.8 ± 4.4 87.9 ± 3.9 Sex (female) 742 (67) 99 (84) 113 (68) 184 (64) 346 (64) Living alone (*n* = 1112) 875 (79) 108 (92) 142 (86) 229 (80) 396 (73) Education \< 8 years (*n* = 1086) 791 (73) 88 (85) 120 (77) 230 (80) 353 (65) Current smoker (*n* = 1109) 35 (3) 1 (1) 2 (1) 10 (4) 22 (4) Former smoker (*n* = 1109) 324 (29) 16 (14) 44 (27) 84 (29) 180 (33) Diagnoses and medical conditions  Diabetes 178 (16) 18 (15) 29 (18) 46 (16) 85 (16)  Congestive heart failure (*n* = 1114) 348 (31) 46 (39) 69 (42) 89 (31) 144 (27)  Atrial fibrillation 270 (24) 22 (19) 56 (34) 60 (21) 132 (24)  MI in the previous year 31 (3) 2 (2) 4 (2) 9 (3) 16 (3)  Angina pectoris 471 (42) 58 (49) 78 (47) 123 (43) 212 (39)  Cerebrovascular disease 243 (22) 33 (28) 41 (25) 62 (22) 107 (20)  Cancer in the previous 5 years (*n* = 1113) 141 (13) 11 (9) 15 (9) 33 (11) 82 (15)  Dementia 389 (35) 116 (98) 153 (92) 104 (36) 16 (3)  Hip fracture 201 (18) 43 (36) 36 (22) 45 (16) 77 (14)  COPD (*n* = 1114) 196 (18) 15 (13) 28 (17) 55 (19) 98 (18)  Depression 385 (35) 54 (46) 85 (51) 118 (41) 128 (24)  Rheumatic disorders 154 (14) 11 (9) 25 (15) 38 (13) 80 (15) Routine prescription medications  ACE inhibitors 224 (20) 17 (14) 43 (26) 69 (24) 95 (18)  β blockers 478 (43) 35 (30) 62 (37) 128 (44) 253 (47)  Calcium channel blockers 189 (17) 11 (9) 23 (14) 49 (17) 106 (20)  Diuretics 604 (54) 61 (52) 102 (61) 161 (56) 280 (52)  Benzodiazepines 287 (26) 33 (28) 48 (29) 81 (28) 125 (23)  Antidepressants 198 (18) 42 (36) 54 (33) 51 (18) 51 (9)  ASA 469 (42) 40 (34) 69 (42) 126 (44) 234 (43)  Neuroleptics 132 (12) 43 (36) 32 (19) 32 (11) 25 (5)  Warfarin 102 (9) 5 (4) 14 (8) 22 (8) 61 (11)  Opioids (*n* = 1114) 144 (13) 21 (18) 22 (13) 46 (16) 55 (10)  NSAIDs 68 (6) 4 (3) 5 (3) 20 (7) 39 (7)  Paracetamol 385 (35) 81 (69) 87 (52) 98 (34) 119 (22)  Statins 129 (12) 2 (2) 12 (7) 30 (10) 85 (16)  No. of prescribed drugs (*n* = 1108) 6.5 ± 4.0 7.5 ± 3.4 7.6 ± 4.2 6.8 ± 4.4 5.8 ± 3.7 Assessments  BMI (*n* = 1080) 25.4 ± 4.4 23.5 ± 4.8 25.2 ± 4.8 25.8 ± 4.2 25.6 ± 4.2  MMSE score (range, 0--30) 21.1 ± 7.6 4.1 ± 3.9 14.5 ± 1.9 20.9 ± 1.7 26.8 ± 1.9  Barthel ADL index (range, 0--20; *n* = 1113) 16.4 ± 5.5 6.7 ± 5.9 12.3 ± 6.1 17.4 ± 3.7 19.1 ± 2.0  SBP (mmHg) 146.1 ± 23.4 131.3 ± 19.4 139.3 ± 22.5 145.0 ± 22.9 152.0 ± 22.7  DBP (mmHg; *n* = 1110) 74.1 ± 11.7 68.3 ± 12.9 73.3 ± 11.6 74.2 ± 11.5 75.6 ± 11.1  Pulse pressure (mmHg; *n* = 1110) 72.0 ± 20.0 63.0 ± 17.7 66.1 ± 19.0 70.8 ± 18.8 76.4 ± 20.3 Data are presented as *n* (%) or mean ± standard deviation. Total number is presented in parentheses after characteristics with some missing data. ACE, angiotensin-converting enzyme; ADL, activities in daily living; ASA, acetylsalicylic acid; BP, blood pressure; COPD, chronic obstructive pulmonary disease; MI, myocardial infarction; MMSE, mini-mental state examination. ###### Hazard ratios for death according to SBP and Mini-Mental State Examination score subcohort^a^ MMSE 0--10 MMSE 11--17 MMSE 18--23 MMSE 24--30 ---------- ------------ -------------------- ------------- ------------- ------------------- ------- ---------- ------------------- ------- ---------- ------------------- ------- Model 1 118 (70) 166 (84) 289 (69) 542 (70) ≤125 2.41 (1.23--4.72) 0.011 1.45 (0.62--3.37) 0.389 2.81 (1.20--6.54) 0.017 2.19 (1.09--4.40) 0.029 126--139 1 1.41 (0.58--3.41) 0.449 1.97 (0.76--5.10) 0.161 1.23 (0.56--2.73) 0.604 140--149 1.73 (0.74--4.02) 0.203 1.48 (0.62--3.51) 0.379 1.87 (0.74--4.76) 0.187 1.06 (0.48--2.34) 0.883 150--164 1.65 (0.71--3.85) 0.242 1.23 (0.51--2.98) 0.645 1.42 (0.58--3.48) 0.446 1.28 (0.65--2.53) 0.480 ≥165 4.48 (1.51--13.23) 0.007 1 1 1 Model 2 117 (69) 165 (83) 289 (69) 542 (70) ≤125 2.23 (1.12--4.45) 0.023 1.18 (0.47--2.93) 0.729 1.99 (0.84--4.73) 0.121 1.60 (0.72--3.53) 0.247 126--139 1 1.33 (0.52--3.39) 0.549 1.55 (0.59--4.10) 0.378 1.13 (0.47--2.72) 0.788 140--149 2.25 (0.91--5.57) 0.081 1.45 (0.57--3.69) 0.433 1.47 (0.58--3.77) 0.420 1 150--164 1.63 (0.68--3.87) 0.272 1.35 (0.53--3.45) 0.528 1.36 (0.55--3.34) 0.507 1.16 (0.53--2.54) 0.714 ≥165 4.54 (1.52--13.60) 0.007 1 1 1.05 (0.47--2.31) 0.911 Model 1 was adjusted for age and sex. Model 2 was adjusted for age, sex, atrial fibrillation, depression, and Barthel Activities of Daily Living index. MMSE, mini-mental state examination. ^a^Calculated using Cox proportional-hazard regression models.

Background ========== The prevalence of symptomatic peripheral arterial disease in the adult population ranges between 0.6% and 9.2% and increases with age \[[@B1],[@B2]\]. Patients with peripheral arterial disease have an increased risk of cardiovascular morbidity and mortality, showing a similar risk factor profile to patients with other atherosclerotic diseases. In the non-pharmacological treatment of symptomatic peripheral arterial disease, peripheral vascular surgical interventions such as bypass grafting and endarterectomy play an important role \[[@B3]\]. The long-term aim of surgical interventions is to prevent amputation of the limb and to reduce its resulting disability. According to current guidelines, surgical interventions are indicated for individuals with symptomatic disease (claudication), significant functional disability, resistance to exercise or pharmacotherapy, and a reasonable likelihood of symptomatic improvement \[[@B4]\]. Whereas endarterectomy is an option in strictly localised disease, bypass grafting is generally used to circumvent severely stenosed sections of the peripheral arteries. Different materials can be used for bypass grafting including autologous and homologous grafts from the saphenous vein or the human umbilical vein as well as prosthetic graft materials such as polytetrafluoroethylene (PTFE) or polyester (Dacron^®^) grafts. Most studies so far have shown that autologous vein is superior to prosthetic graft materials in bypass surgery \[[@B5]-[@B7]\]. A recent review comparing venous and PTFE bypass procedures reported 5-year primary patency rates of 74% and 39%, respectively \[[@B8]\]. However, almost a third of patients eligible for peripheral bypass procedures do not have suitable veins, making the use of prosthetic materials necessary \[[@B9]\]. Also, due to the high prevalence of cardiovascular co-morbidity, it may be required to keep suitable autologous veins for potential future use in coronary artery bypass grafting. The objective of our systematic review was, therefore, to identify available evidence and compare the effectiveness of the prosthetic bypass materials Dacron^®^and PTFE in peripheral vascular bypass surgery and to perform meta-analyses, if possible. Methods ======= Literature search ----------------- A trained librarian performed a comprehensive systematic literature search for relevant publications using the following databases: AMED, BIOSIS Previews, CAB Abstracts, CATFILEplus (CATLINE), Cochrane Library -- CDSR, Cochrane Library -- CENTRAL, Elsevier BIOBASE, EMBASE, EMBASE Alert, ETHMED, GeroLit, GLOBAL Health, HECLINET, IPA, MEDLINE Alert, MEDLINE, NHS-CRD-DARE, NHS-CRD-HTA (INAHTA), NHS-EED, SciSearch, and SOMED. The search terms included \"bypass\", \"revascularization\", \"artery reconstruction\", \"graft\", \"prosthesis\", and \"material\". The search was performed in February 2005 with an update search performed in MEDLINE and CENTRAL (Cochrane Central Register of Controlled Trials) for publications until August 2008. The systematic database search was supplemented by manual search of reference lists of included articles. The inclusion criteria of studies were: (i) randomized controlled trial (RCT) as study design; (ii) comparison of polytetrafluoroethylene (PTFE) or polyester (Dacron^®^) grafts for peripheral vascular bypass surgery; (iii) publication in English or German; and (iv) publication from 1999 to date. We focused on publications of English language to cover the most important and qualitatively high trials therewith (we additionally searched for articles in German for a potential adaption to the situation in Germany). We included studies published in and after 1999 as the purpose of this review was to provide an overview focusing on the present evidence from more recent trials. Excluded were studies due to the following criteria: (i) case series; (ii) retrospective studies; (iii) studies comparing venous vs. prosthetic graft materials. Methodological assessment and endpoints --------------------------------------- The methodological quality of relevant publications was assessed using standardized checklists developed by the German Scientific Working Group \"Technology Assessment for Health Care\". \[[@B10]\] evaluating the selection process of patients, randomization procedure, assessment of outcomes, drop-out rates, and adequate statistical methods. The primary outcome for this review was primary patency, as defined by the authors. According to Rutherford et al. primary patency should be assessed by objective methods such as vascular imaging techniques, palpable pulse, biphasic or triphasic Doppler, segmental limb pressure index, etc. \[[@B11]\]. We here present primary patency rates as defined and reported in each trial. Secondary outcomes were secondary patency, graft infection rates, limb salvage or amputation rates, and perioperative (\< 30 days) mortality as presented in each study. Statistical analysis -------------------- To perform a meta-analysis, study data from accumulated life tables was extracted from each study where available. If failure data was given for one or three months intervals it was combined to yield 6-months interval data for all studies. Time of event (graft failure) or censoring (withdrawals) was thus assigned to the end of each 6-months interval. A Cox proportional hazard model was used to calculate hazard ratios with their 95% confidence intervals (CI) and standard errors (SE) for the independent variable \"material\". This was done for each study separately. In a random-effect meta-analysis these hazard ratios were then combined with weights according to their standard error (inverse variance method), yielding an overall hazard ratio (with 95% CI). Heterogeneity was tested using the chi-squared Q-statistic and inconsistency was quantified by I^2^. Life table data was also pooled to provide Kaplan-Meier graphs on graft failure events. All analyses were performed with SAS V9.1.3 (SAS Institute Inc., Cary, NC, USA) and Review Manager Version 5.0 (Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, 2008). Results ======= We identified a total of 4421 publications in the search process. Excluding duplicates and non-relevant publications based on their title and/or abstract resulted in 419 publications for further quality assessment (Fig. [1](#F1){ref-type="fig"}). Of these, 9 publications were found to be relevant to the research question and fulfilled all inclusion criteria \[[@B12]-[@B20]\]. {#F1} All 9 RCT included in our analysis used primary patency as their main outcome measure. All studies provided a definition for the term patency in their methods section, with primary patency usually meaning unassisted/uninterrupted patency with no follow-up procedures on the bypass assessed by objective methods; one study, however, defined primary patency as assisted primary patency \[[@B12]\]. Seven studies also reported results on secondary patency rates. Five studies present limb salvage or amputation rates, 4 perioperative mortality (\<30 days), and 6 graft infection as further outcomes. Most of these endpoints were not defined in detail. Table [1](#T1){ref-type="table"} gives an overview of study characteristics and main results of all RCT included in our analysis. ###### Overview of included studies of Dacron^®^vs. PTFE as bypass materials in peripheral vascular surgery ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- **Author** **Year** **Indication** **Site of bypass** **Intervention** **Additional intervention** **Follow-up**\ **N\*** **Primary patency**\ **Superior primary patency** **Funding** **(years)** **(N = number of patients or grafts under risk)** ---------------------------------- ---------- ---------------------------------------------------------------------- ----------------------------------------------- ---------------------------------------------------------------- -------------------------------------------------------------------------------- ---------------- --------- --------------------------------------------------- ------------------------------------------------ ------------------------ ----------------------------------- Johnson and Lee \[[@B12]\] 1999 disabling claudication, rest pain or tissue necrosis femorofemoral, axillofemoral, axillobifemoral Dacron^®^vs. PTFE Aspirin 650 mg/d 5 419 47 At 1, 3, 5 years:\ n.s.\ n.r. - Dacron^®^: 79%, 63%, 50% (N = 125, 73, 32)\ (p-value not reported) - PTFE: 77%, 62%, 47% (N = 103, 53, 15) Robinson et al. \[[@B13]\] 1999 disabling claudication, rest pain or tissue necrosis femoropopliteal (above-knee and below-knee) gelatine-sealed Dacron^®^vs. PTFE Cephalothin, Heparin, Aspirin 3 108 19 At 1, 2, 3 years:\ n.s.\ n.r. - Dacron^®^: 70%, 56%, 47% (N = 27, 18, 9)\ (p = 0.87) - PTFE: 72%, 52%, 52% (N = 33, 16, 10) Green et al. \[[@B14]\] 2000 superficial femoral artery occlusion femoropopliteal (above-knee) collagen-impregnated Dacron^®^vs. ePTFE n.r. 5 240 10 At 1, 3, 5 years::\ n.s.\ manufacturer of Dacron^®^ - Dacron^®^: 78%, 65%, 45% (N = 65, 25, 5)\ (p-value not reported) - PTFE: 80%, 63%, 43% (N = 66, 21, 5) Post et al. \[[@B15]\] 2001 indication for artificial graft of at least 20 cm length femoropopliteal (above-knee and below-knee) unsealed Dacron^®^vs. PTFE Anti-platelet drugs, Heparin or Coumadin 3 194 50 At 3 years (N = grafts under risk):\ n.s.\ manufacturer of Dacron^®^and PTFE - Dacron^®^: 64% (95%-CI \[55%;74%\], N = 28)\ (p = 0.89) - PTFE: 61% (95%-CI \[49%;72%\], N = 22) Prager et al. \[[@B16]\] 2003 aortoiliac occlusive disease aortoiliac gelatine-coated Dacron^®^vs. collagen-coated Dacron^®^vs. PTFE Antibiotics, Heparin 70 IU/kg, Fraxiparine 100 mg/kg/d (bid for patients with\ 8 149 35 At 5, 8 years:\ n.s.\ n.r. anastomoses) - C-Dacron^®^: 89%, 78% (N = 24, 11)\ (p \> 0.8) - G-Dacron^®^: 92%, 77% (N = 26, 11)\ - PTFE: 88%, 79% (N = 29, 13) Robinson and Fletcher \[[@B17]\] 2003 disabling claudication, rest pain or tissue loss femoropopliteal (above-knee and below-knee) fluoropolymer-coated Dacron^®^vs. PTFE Cephalothin, Heparin, Aspirin 2 129 21 At 6, 12, 24 month:\ PTFE\ manufacturer of Dacron^®^and PTFE - Dacron^®^: 50%, 36%, 36% (N = 27, 17, 9)\ (p = 0.002) - PTFE: 71%, 56%, 47% (N = 43, 28, 12) Devine and McCollum \[[@B18]\] 2004 occlusive arterial disease (superficial femoral or popliteal artery) femoropopliteal (above-knee and below-knee) collagen-coated, heparin-bonded Dacron^®^vs. PTFE Aspirin 300 mg/d 5 209 45 At 1, 3, 5 years:\ n.s.\ n.r. - Dacron^®^: 71%, 54%, 46% (N = 70, 45, 20)\ (p = 0.05) - PTFE: 62%, 44%, 35% (N = 62, 42, 25) Eiberg et al. \[[@B19]\] 2006 uni-ilia occlusive disease femorofemoral fluoropassivated, gelatine-sealed Dacron^®^\ n.r. 2 198 136 At 1, 2 years:\ n.s.\ manufacturer of Dacron^®^and PTFE vs. ePTFE - PTFE: 94%, 93% (N = 74, 63)\ (p = 0.350) - Dacron^®^: 92%, 87% (N = 87, 73) Jensen et al. \[[@B20]\] 2007 chronic lower limb ischaemia femoropopliteal (above-knee) gelatine-coated Dacron^®^\ n.r. 2 413 150 At 2 years:\ Dacron^®^\ manufacturer of Dacron^®^ vs. PTFE - Dacron^®^: 70% (N = 78)\ (p = 0.002) - PTFE: 57% (N = 72) ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- PTFE: Polytetrafluoroethylene. ePTFE: expanded Polytetrafluoroethylene. IU: international units. d: day. bid: bis in die. n.s.: not significant. n.r.: not reported. CI: confidence interval. \* Number of patients or grafts under risk Of the 9 included studies, 7 studies showed no significant differences between Dacron^®^and PTFE regarding primary patency, 1 study showed significantly higher patency rates for Dacron^®^after 2 years \[[@B20]\], and 1 study showed significantly higher patency rates for PTFE after 2 years \[[@B17]\]. The 7 RCT reporting results on secondary patency yielded results similar to primary patency rates, e.g. no significant differences between the prosthetic materials were found in 6 studies, and 1 study showed significantly higher secondary patency after 2 years with Dacron^®^grafts. Of the studies reporting results regarding limb salvage or amputation rates, 4 showed no significant difference, whereas 1 showed a significantly better rate for Dacron^®^grafts. None of the studies found significant differences in terms of graft infection or perioperative mortality (\<30 days). Meta-analysis ------------- Meta-analysis was performed for the comparison of Dacron^®^vs. PTFE on primary patency for the 5 studies where adequate data was available \[[@B12]-[@B14],[@B17],[@B18]\]. Two studies provided also data on secondary patency \[[@B13],[@B17]\]. The combined hazard ratios for PTFE vs. Dacron^®^were 1.04 (95% CI \[0.85;1.28\]) for primary patency (Fig. [2](#F2){ref-type="fig"}) and 1.02 (95% CI \[0.65;1.62\]) for secondary patency (Fig. [3](#F3){ref-type="fig"}). There was no significant heterogeneity between the studies (p = 0.32 and p = 0.24). The Kaplan-Meier curves of the pooled data reflect the similar efficacy of the materials regarding both primary and secondary patency (Fig. [4](#F4){ref-type="fig"} and Fig. [5](#F5){ref-type="fig"}). {#F2} {#F3} ![**Survival curves for Dacron^®^and PTFE on primary patency of pooled data**\[[@B12]-[@B14],[@B17],[@B18]\]**(PTFE: polytetrafluoroethylene)**.](1471-2482-8-22-4){#F4} ![**Survival curves for Dacron^®^and PTFE on secondary patency of pooled data**\[[@B13],[@B17]\]**(PTFE: polytetrafluoroethylene)**.](1471-2482-8-22-5){#F5} Discussion ========== The present meta-analysis indicated that there are no major differences in primary and secondary patency rates between the two prosthetic graft materials Dacron^®^and PTFE. Of the 9 included studies, one study showed statistically significant higher patency rates for Dacron^®^after 2 years \[[@B20]\], 1 study showed statistically significant higher patency rates for PTFE after 2 years \[[@B17]\], while 7 studies showed no statistically significant differences between the two materials regarding primary patency \[[@B12]-[@B16],[@B18],[@B19]\]. Our study complements a systematic Cochrane review on femoro-popliteal bypass surgery by Mamode and Scott comparing saphenous vein graft with PTFE or Dacron^®^, human umbilical vein with PTFE, and PTFE vs. Dacron^®^\[[@B21]\]. However, only one study comparing PTFE with Dacron^®^in above-knee popliteal grafting was identified \[[@B22]\]. This study by Abbott et al. was published before the search period of our review. It did not show a significant difference regarding primary or secondary patency rates at the 3-year follow-up. One of the studies in the present review compared gelatine-coated Dacron^®^vs. collagen-coated Dacron^®^showing no differences between the study groups \[[@B16]\]. In another study heparin-bonded Dacron^®^instead of bare Dacron^®^graft material was used. The issue of different coatings might in itself affect patency outcomes of Dacron^®^graft. However this question cannot adequately be assessed at the moment due to the lack of data. Only 5 studies presented data in adequate life table format for performing a meta-analysis. Results of the meta-analysis might change, if more study data could be included, especially data of the trial by Jensen et al. \[[@B20]\] as this was the largest trial to date comparing PTFE vs. Dacron^®^. However, it should be noted that patient recruitment and surgeries of that trial were performed up to 14 years before publication of the results (2 year follow up). For the meta-analysis no raw data with actual times of event or censoring could be used, as only aggregated life table data was available. This will overestimate event and censoring times in both groups, but should not have influenced the comparison of the two graft materials. In the context of clinical routine care, physicians regularly face decisions as to which alternative treatment strategies to recommend and use. Their decisions should be guided by best possible evidence of previous studies. Hence, systematic reviews of good quality RCT have evolved as important tool of decision support. It is important to note, however, that the RCT included in our systematic review were limited by a number of methodological limitations, such as rather small sample sizes, different methods for determining patency rates, a lack of consideration paid to additional factors that might affect outcomes such as baseline differences between groups, and inadequate interpretations of non-significant results. Reporting of existing baseline differences between the groups in each trial was heterogeneous. Only 4 studies provided information about adjustment for baseline differences as potentially confounding factors \[[@B13],[@B14],[@B18],[@B19]\]. Group differences at baseline may thus have biased the results. In 4 trials adequate sample size calculations were reported \[[@B15],[@B18]-[@B20]\]. The other trials may have been too small to detect any differences in the graft materials. In addition, no trial used equivalence testing to show that both graft types were similarly effective. All studies described how primary patency was assessed and defined, and most studies used recommended objective standard methods \[[@B11]\]. However, differences in patency rates between studies might have been affected by unequal assessments, while differences in patency between PTFE or Dacron^®^grafts within studies should not be affected (assuming same assessment standards for patients receiving PTFE or Dacron^®^grafts within the trial). Sources of funding were not specified in 4 trials, with the remaining trials being funded by or having received grants from bypass graft manufacturers, which might introduce bias to the results \[[@B23],[@B24]\]. No explicitly independent/non-manufacturer sponsored trials were found. In this review only trials published in English or German were considered for inclusion. Even though we feel to cover the most important and qualitatively higher trials with this strategy, this might result in relevant articles and evidence published in other languages being ignored. Results might change if a body of evidence from manuscripts in other languages would be available. Since 4 of the 9 includes trials were conducted in non-English speaking countries (Austria, Germany, and Scandinavian countries), but were published in English, we assume this might not be the case. Conclusion ========== Even in the light of methodological limitations of the included trials, the present meta-analysis and systematic review might offer a basis for clinical decision making in individual patients requiring peripheral vascular surgery. Between the two prosthetic materials PTFE and Dacron^®^no clear advantage of one over the other could be seen. Further independently funded studies should address the issue of heparin-bonded grafts as well as the identification of subgroups of patients, in which there might be a benefit of one material. Studies should be sufficiently powered to be able to detect differences or equivalence of PTFE and Dacron^®^grafts. Competing interests =================== The authors declare that they have no competing interests. Authors\' contributions ======================= SR and DE were responsible for the methodological design of the review, the literature selection and the quality assessment. SR carried out the data extraction, the summary of the findings, the meta-analysis and has participated in the writing of the manuscript. JMN has participated in the writing of the manuscript. TK and WG have revised the manuscript critically. HS advised the review as a clinical expert and has revised the manuscript critically. SW had the overall scientific responsibility and has participated in the writing of the manuscript. All authors read and approved the final manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2482/8/22/prepub> Acknowledgements ================ Financial support: DIMDI (German Institute of Medical Documentation and Information)

Introduction {#sec1} ============ Radiation therapy (RT) is an integral component of modern oncology care, spanning a broad range of indications from palliative to definitive intent therapy. Historically, radiation has been viewed almost exclusively as a local modality. From early radiobiological studies, the major mechanism of action of radiation has been found to be mediated by DNA damage, leading to the death of irradiated cells mostly at the time of cell division. However, a growing body of evidence in both the preclinical and clinical settings have yielded important insights on other radiation effects that can be sensed by both the innate and adaptive immune system. In some cases, radiation-induced antitumor effects contribute to cross priming and succeed at eliciting an immune response against the tumor. From classical radiobiology, it has been well established that RT exerts different effects when given at different dose-fractionation schemes, an observation that is summarized in the principles termed the 4 Rs: repair, reassortment, reoxygenation, and repopulation. These principles have provided the rationale that underlies most dose-fractionation strategies used in radiation oncology today. In addition, there is now recognition that cells of different origins respond differently to radiation even if all other variables are the same, pointing to an intrinsic property of the cell described as the fifth R: radiosensitivity. All of these properties are neatly encapsulated in the α/β ratio, which describes the curvature of the cell survival curve and directly provides an assessment of how sensitive a tumor (or target tissue) is to radiation fractionation. Tumors with a low α/β ratio are considered relatively resistant to low doses of radiation per fraction, thus implying that hypofractionated radiation (ie, fewer fractions at larger doses per fraction) would be more effective at achieving cell killing and tumor control. Conversely, normal tissues generally exhibit a high α/β ratio for acute side effects and thus are sensitive to relatively low doses of radiation per fraction, rendering a standard (conventionally fractionated) or hyperfractionated strategy more appropriate for normal tissue sparing. Studies to optimize radiation dose and fractionation have explored a variety of altered fractionation regimens with the goal of improving the therapeutic ratio. In settings when the α/β ratio is estimated to be low (eg, prostate or breast carcinomas), hypofractionated radiation has been investigated. The advent of advanced image guidance as an integral part of radiation treatment delivery has facilitated the adoption of hypofractionated, high-dose radiation regimens in the form of stereotactic body RT (SBRT). Whether cell killing and tumor control using SBRT can be adequately described by the classical linear quadratic model, particularly when \<3 fractions are used, has been a matter of debate. Regardless of the exact radiobiological principles at hand, multiple lines of evidence now demonstrate that excellent local control rates are achieved with SBRT across a broad range of clinical settings, leading to its widespread adoption. SBRT has also been tested in selected patients with oligometastatic disease, in whom it has occasionally resulted in durable control with long progression-free intervals.[@bib1] Another interesting application of SBRT in the metastatic setting of cancer is its combination with modern immunotherapy. Out-of-field effects of RT are a rare phenomenon, originally defined by Mole et al[@bib2] and known as the abscopal effect, and illustrate the generation of a clinically significant response at a distant metastatic site. A thorough review of reports of abscopal effects from radiation identified a total of 35 cases over 45 years.[@bib3] The immunologic nature underlying the abscopal effect has been reported in the preclinical setting.[@bib4], [@bib5], [@bib6], [@bib7], [@bib8] A plausible explanation of the rarity of abscopal response despite the demonstrated proimmunogenic effects of focal RT is the strong immunosuppressive microenvironment that characterizes established cancers.[@bib9] The availability of immune checkpoint blocking agents has enabled their testing in combination with RT, helping to potentiate tumor-directed immune responses in a clinically significant manner. One of the central issues in optimizing radiation and immunotherapy combinations is how to identify the best radiation dose-fractionation regimen. In contrast to clinical experience suggesting that a higher biologically effective dose (BED) may improve clinical outcomes (local control) across multiple tumor types,[@bib10] when RT was combined with immune checkpoint blockade to induce a systemic effect, a threshold dose to the generation of effective antitumor immunity was identified in the preclinical setting.[@bib11] In this review article we discuss the preclinical and clinical evidence regarding dose-fractionation considerations for RT and the influence of these variables on the generation of effective antitumor immunity. Radiation-induced immune responses {#sec1.1} ---------------------------------- Several lines of investigation have provided a greater understanding that not only does radiation directly influence tumor immunity, it also exerts its effects via a series of distinct mechanisms. These mechanisms include triggering immunogenic cell death,[@bib12] generating neoantigens and enhancing antigen processing and cross presentation,[@bib13] decreasing immunosuppression in the tumor microenvironment,[@bib14], [@bib15] overcoming T-cell exclusion from the tumor microenvironment,[@bib16] and increasing tumor cell recognition by the immune system.[@bib13] RT has been found to directly increase the immunogenicity of tumor cells by increasing the translocation of calreticulin to the tumor cell surface, the extracellular release of high mobility group protein 1, and the extracellular release of adenosine triphosphate, leading to immunogenic cell death.[@bib12] Additionally, the use of radiation has been found to generate neoantigens and could enhance antigen processing.[@bib13] Third, radiation has been found to reduce the degree of immunosuppression in the tumor microenvironment, in part because of the production of cytokines such as type I interferon.[@bib17], [@bib18] In addition, radiation can decrease immunosuppressive components in the tumor microenvironment,[@bib15] including T regulatory cells, myeloid-derived suppressor cells, and tolerogenic dendritic cells. It is also known that radiation can reprogram tumor-associated macrophages to convert an immunosuppressive tumor microenvironment to one that is more favorable for tumor immunity.[@bib14] Fourth, radiation treatment may help to overcome T-cell exclusion from the tumor microenvironment by normalizing the vasculature and increasing production of immune-attractant chemokines such as CXCL16.[@bib16], [@bib19] Lastly, radiation treatment leads to greater recognition of the tumor by the immune system. Radiation has been found to lead to downregulation of CD47 on tumor cells[@bib20] and upregulation of MHC class I[@bib18], [@bib21] with enhanced degradation of existing proteins, enhanced peptide production, and increased antigen presentation.[@bib13] Preclinical evidence {#sec2} ==================== In vitro data {#sec2.1} ------------- Radiation treatment can increase the immunogenicity of tumor cells by increasing their surface expression of MHC class I complexes, and this effect has been found to be dose dependent.[@bib13] Across a single-fraction dose range of 1 to 25 Gy, Reits et al[@bib13] found that higher radiation dose leads to increased degradation of proteins, leading to an increased intracellular peptide pool. In addition, higher radiation dose leads to increased activation of the mammalian target of rapamycin pathway, triggering increased peptide production and antigen presentation. Lastly, higher radiation dose also leads to the generation of novel proteins as a consequence of DNA damage, thus allowing the formation of neoantigens and presentation of neoepitopes on MHC class I molecules. As a proof of concept, these investigators found in their MC38 murine colorectal adenocarcinoma model that neither adoptive transfer of gp70-specific cytotoxic T lymphocytes nor radiation (10 Gy) alone was sufficient to cure mice bearing MC38 tumors, but radiation followed by adoptive transfer of cytotoxic T lymphocytes led to significant regression of all tumors, including cure in 63% of treated mice. Although a dose titration was not performed in their in vivo tumor experiment, it nevertheless was consistent with the role of radiation in increasing tumor immunogenicity. Subsequently it was reported that increasing radiation dose increases the induction of immunogenic cell death.[@bib22] In TSA murine mammary carcinoma cells radiated with single-fraction doses ranging from 2 to 20 Gy, there was a dose-dependent increase in the release of adenosine triphosphate into the extracellular matrix, translocation of calreticulin to the cell surface, and release of high mobility group protein 1 from dying tumor cells, all of which are key components of immunogenic cell death.[@bib12] Animal models {#sec2.2} ------------- The preclinical evidence supporting a radiation-induced systemic immune response falls mostly to the evolving story of the abscopal phenomenon in immunocompetent mouse models.[@bib4], [@bib5] In general there has been significant heterogeneity in how preclinical studies are conducted, and relatively few include a rigorous comparison of outcomes after different dose-fractionation regimens.[@bib23] A wide range of murine tumors have been tested, including mammary carcinomas (TUBO, FM3A, TSA, 4T1, 67NR), colon carcinomas (MCA38, HCT116, Colon 26), sarcomas (MethA, T241, MCA205), lung carcinoma (LLC), squamous cell carcinoma (VII), cervical carcinoma (C3), and melanoma (D5). Most of these tumor models are implanted heterotopically in a flank location, allowing for partial body irradiation and monitoring of the tumor response at a nonirradiated site. Another distant (nonirradiated) tumor readout used in a few studies is the incidence of distant metastases, often in the lung or liver. In an early experiment an abscopal response was identified against 67NR murine mammary carcinomas when radiation was given with Flt3 ligand.[@bib4] In this study there was no significant difference in the identified abscopal response at a single-fraction dose of 2 Gy or 6 Gy, suggesting no significant difference in the generation of systemic immune responses against this tumor across this single-fraction dose range in the presence of Flt3 ligand. When looking at single-fractionation strategies in the preclinical setting, prior studies have reported abscopal effects with single-fraction doses up to 60 Gy.[@bib24] However, no dose titration was specified in most of these studies, and thus it remains unclear whether there is a significant dose response to justify the extremely high radiation doses. Subsequently, different groups of investigators have approached preclinical studies from different angles. There have been conflicting preclinical results regarding whether a high-dose, single-fraction approach is superior in generating an abscopal response compared with a moderate- or low-dose, multiple-fraction approach. In one study, multiple dose-fractionation regimens (20 Gy × 1, 8 Gy × 3, and 6 Gy × 5) were compared for abscopal effects against TSA mammary carcinomas.[@bib7] Although all 3 radiation approaches limited primary tumor growth to a similar extent, the fractionated regimens were able to generate a greater abscopal response in the nonirradiated tumor. A parallel experiment found that the multiple-fraction approach of 8 Gy × 3 (in combination with anti-CTLA-4 mAb) was able to generate a greater abscopal response against MCA38 colorectal adenocarcinomas than the single-fraction approach of 20 Gy × 1. However, other investigators have found that a single-fraction approach was superior for the generation of systemic antitumor responses. In one study there was a greater degree of immune activation in tumor-draining lymph nodes when mice with B16-F0 tumors were treated with 15 Gy × 1 rather than 5 Gy × 3, even though both reportedly had a marked effect on tumor growth.[@bib25] In another study it was reported that an ablative radiation regimen of 20 Gy × 1 was superior to 5 Gy × 4 (given over 2 weeks) in terms of generating an antitumor immune response in a B16-SIY melanoma model.[@bib8] One of the problems with this comparison is the radiobiological inferiority of 5 Gy × 3 to 4; equivalence to a single dose of 15 to 20 Gy is usually achieved with 8 to 9 Gy × 3 or 6 Gy × 5 given within 1 week. In addition, these studies were conducted in the setting of strongly immunogenic tumor rejection antigens (OVA and SIY, respectively), which may not recapitulate what would typically occur with weakly immunogenic self-antigens in the clinical setting. Most importantly, neither of these studies used an immune-modulating agent (eg, CTLA-4 mAb) alongside RT. Another preclinical report compared 2 different fractionated regimens, observing that the abscopal antitumor response was significantly more potent when mice bearing LLC tumors were treated with 10 Gy × 5 rather than 2 Gy × 12.[@bib26] Again, these radiation dose-fractionation schemes differ significantly not only in terms of their total dose but also because of the BED. Thus it is not clear if the enhanced abscopal effect from the 10 Gy × 5 regimen simply reflected greater BED, greater total dose, greater dose per fraction, or some combination of these factors. Elucidating the cellular mechanisms underlying the abscopal response {#sec2.3} -------------------------------------------------------------------- Preclinical insights culminated recently in a direct comparison of subablative hypofractionated radiation versus ablative single-fraction SBRT. This study identified a threshold dose of \>10 to 12 Gy, beyond which there was an increase in immunosuppression and loss of the abscopal effect.[@bib11] In a variety of tumor models, including the murine TSA and MCA38 models, the use of a single large radiation dose of 20 to 30 Gy, in combination with anti--CTLA-4 blockade, led to reduced tumor immunogenicity and loss of the abscopal effect compared with a moderate subablative dose of 8 Gy × 3.[@bib11] It was noted that at very high radiation doses (\>10-12 Gy) per fraction, tumor cells release greater quantities of double-stranded DNA into the cytosol, triggering upregulation of the DNA exonuclease Trex1. Greater Trex1 expression then leads to increased clearance of double-stranded DNA from the cytosol. Decreased cytosolic DNA levels lead to decreased binding to cyclic guanosine monophosphate--adenosine monophosphate (GMP-AMP) synthase and decreased production of cyclic GMP-AMP. As a consequence, there is decreased binding of cyclic GMP-AMP to the protein stimulator of interferon genes, leading to decreased interferon regulatory factor 3 phosphorylation and nuclear translocation, thus decreasing the transcription of inflammatory genes such as type I interferon. Radiation effects on the tumor microenvironment {#sec2.4} ----------------------------------------------- In addition to the influence of radiation dose-fractionation on tumor cells, there is also preclinical evidence that the radiation dose used may generate different effects in the tumor microenvironment. It is known that low radiation doses of 2 Gy can promote inducible nitric oxide synthase expression by tumor-associated macrophages, suggesting that a proimmunogenic environment can be induced by radiation treatment at a low dose.[@bib14] In contrast, higher radiation doses have been found to result in tumor infiltration by protumorigenic macrophages,[@bib27] suggesting that there may be a window of radiation dosing that is most effective in supporting tumor immunity. There is also evidence that single radiation doses of 5 to 10 Gy cause relatively mild vascular changes, but doses \>10 Gy cause significant vascular damage and reduce vascular flow as a result of endothelial cell death, leading to reduced effector T-cell recruitment to the tumor.[@bib28] Lessons learned from preclinical evidence {#sec2.5} ----------------------------------------- The preclinical evidence to date suggests that abscopal effects are generally more prominent at larger fraction sizes, thus supporting hypofractionation, although there is a demonstrated dose threshold beyond which the frequency of such effects declines, and conflicting evidence exists on whether single or multiple fractions are more effective. Lessons learned from preclinical evidence have indicated that the optimal dose-fractionation regimen must take into account not only the best strategy to elicit an immunogenic cell death[@bib12] but also the optimal strategy to establish a proimmunogenic tumor microenvironment. Clinical evidence {#sec3} ================= Clinical investigations of abscopal response {#sec3.1} -------------------------------------------- In the clinical setting there has been a constellation of radiation dose-fractionation schema either in conjunction with immunotherapy or with radiation delivered as a single modality.[@bib29], [@bib30], [@bib31], [@bib32], [@bib33], [@bib34], [@bib35], [@bib36] A recent review found a total of 46 abscopal cases reported in 31 articles with a median dose of 31 Gy (range, 0.45-60.75 Gy) and a median dose per fraction of 3 Gy (range, 0.15-26 Gy).[@bib3] Although the primary mechanism remained unclear in the setting of early case reports, the immune system as primary mediator of abscopal response has subsequently been proven in clinical investigations. To date, no clinical studies have directly compared different dose-fraction regimens in their ability to elucidate optimal abscopal responses. Thus, from a clinical perspective, there has been a divergence between ablative versus subablative radiation dose fractionation approaches, and the resultant immune responses are likely to be varied. Early in the clinical investigations of immune-mediated tumor regression, it became clear that the likelihood of achieving abscopal response would be greatest with the combination of an immune adjuvant---a conclusion based in the hypothesis that a bolstered immune response mounts tumor responses both locally and distantly. In a recently published post hoc analysis of patients with advanced non-small cell lung carcinoma (NSCLC) treated on KEYNOTE-001, Shaverdian et al[@bib37] found that patients who received any RT before pembrolizumab had a significant improvement in median overall survival (OS) compared with those who never received RT (10.7 months vs 5.3 months; hazard ratio \[HR\], 0.58; *P* = .026).[@bib37] Additionally, patients who received prior RT had a significantly longer median progression-free survival (PFS) compared with those who never received RT (4.4 months vs 2.1 months; HR, 0.56; *P* = .019). However, this analysis could not account for whether different dose-fractionation options led to different immune or clinical effects in this patient population. Although interpretation of the findings is limited by the post hoc nature of this analysis and the lack of immune correlative studies, it does raise the interesting possibility that one mechanism to explain the benefit of RT before pembrolizumab is the generation of effective antitumor immunity by RT; pembrolizumab then potentiates this response in a clinically significant manner, thus leading to improvements in PFS and OS endpoints in this patient cohort. Two landmark case reports in 2012 and 2013, led by Postow et al[@bib38] and Golden et al,[@bib39] respectively, reported the clinical achievement of abscopal response when immunotherapy was combined with radiation. Both cases used anti-CTLA-4 therapy (ipilimumab) and subablative hypofractionated dose regimens (9.5 Gy × 3 [@bib38] and 6 Gy × 5 [@bib39]). Providing further evidence, 2 subsequent trials by Seung et al[@bib40] and Golden et al[@bib41] combined immunotherapy with differing dose-regimens to assess abscopal regression. In the ablative study by Seung et al,[@bib40] a phase 1 trial of interleukin 2 combined with 1 to 3 fractions of 20 Gy was tested in the setting of metastatic renal cell carcinoma and melanoma. Five of eight patients with melanoma (62.5%) experienced either complete or partial response in nonirradiated lesions, whereas both patients with renal cell carcinoma experienced a partial response, a response rate superior to that achieved with interleukin 2 alone.[@bib40] A "proof of principle" study led by Golden et al[@bib41] delivered subablative dosing to a metastasis in patients with metastatic solid tumors (breast, NSCLC, thymic cancer) in combination with granulocyte-macrophage colony-stimulating factor and achieved similarly successful abscopal results. In subsequent studies a variety of different radiation fractionations and dose regimens were explored in combination with immunotherapy, leading to a range of immunologic responses. These include investigations by Hiniker et al,[@bib42] who used a mixed range of both moderately hypofractionated (2.5-3 Gy × 10-15 fractions) and subablative SBRT doses in combination with anti-CTLA-4 therapy for patients with metastatic melanoma with an overall response (complete and partial) of 27% in unirradiated lesions. A similar rate of abscopal response (27%) was found in the recently reported iMOSART (pembrolizumab immunotherapy and multi--organ site ablative SBRT in patients with advanced solid tumors) phase 1 study.[@bib43] Tang et al[@bib44] evaluated multiple ablative SBRT dose fractionation regimens (BED ranging from 96 to \>100) for lung and liver metastases and found partial responses in unirradiated tumors in 10% of patients but no complete responses.[@bib44] A recently published study by Luke et al[@bib45] evaluated the feasibility of pembrolizumab and ablative SBRT in patients with metastasis from various solid tumors. Any tumor volume was potentially eligible, but for larger tumors only a portion of the gross tumor volume (up to 65 mL) was targeted with ablative SBRT. Despite the use of an "ablative" dose of SBRT, only 1 of the 68 evaluable patients achieved an in-field CR. The ablative dosing used by Luke et al[@bib45] resulted in 1 patient achieving a complete response in unirradiated tumor. The authors note the overall response rate (ORR) reported is comparable to that achieved in earlier preclinical models evaluating for abscopal response[@bib46], [@bib47] and the 26.9% ORR from the Golden et al[@bib41] "proof of principle" trial in the nonirradiated lesion per Response Evaluation Criteria in Solid Tumors. However, it should be noted that pembrolizumab monotherapy has about a 10% response rate, whereas immune adjuvants such as granulocyte-macrophage colony-stimulating factor are associated with no clinically significant responses when given as monotherapy in solid tumors. There are likely further dimensions to this finding, given the heterogeneous nature of radiation treatment planning: It is unknown whether portions of the target lesion in the subablative dose range exhibited a greater immune response compared with those falling within the ablative dose range. This study suggests a need for prospective comparisons of ablative versus subablative dose regimens to define how effectively they achieve abscopal responses. It also invites a reassessment of the dose ranges considered "ablative" in heavily pretreated patients and whether "ablation" is the appropriate descriptor in these patients. Most recently, investigators from the Netherlands Cancer Institute presented results from their PEMBRO-RT study (NCT02492568), a phase 2 trial of patients with advanced NSCLC after at least 1 prior line of systemic therapy who were randomly assigned to receive SBRT (8 Gy × 3) to 1 lesion followed by pembrolizumab within 7 days or to receive pembrolizumab alone.[@bib48] At the time of their presentation, 72 patients had been enrolled and 64 were evaluable for the primary endpoint of ORR. Compared with an ORR of 19% in the control arm (pembrolizumab monotherapy), the use of SBRT to 1 lesion increased ORR to 41%; median PFS was 1.8 months in the control arm and 6.4 months in the experimental arm (*P* = .04). Significantly, there was no increase in treatment toxicities: grade 3+ toxicities were noted in 22% of control patients and 17% of experimental patients. The findings from this study lend further support to the premise that a subablative SBRT regimen in combination with pembrolizumab can successfully elicit clinically significant abscopal effects, leading to a doubling of response rates in nonirradiated lesions. Although highly promising, there was one cautionary note from these investigators\' results: In post hoc analysis, it was noted that the proportion of PD-L1--positive tumors was different between treatment groups because this variable was not used for initial patient selection or stratification. Because strongly PD-L1--positive tumors are expected to respond more robustly to immunotherapy,[@bib49], [@bib50], [@bib51] this confounding variable could potentially account for at least some of the reported ORR differences. Lessons from a negative trial of RT and immunotherapy {#sec4} ===================================================== It should be acknowledged that there is also evidence that RT combined with immune checkpoint blockade may not be beneficial across all clinical scenarios and endpoints. In the phase 3 CA184-043 trial of patients with metastatic castration-resistant prostate cancer who received bone-directed RT (8 Gy × 1, up to 5 lesions) and then were randomly assigned 1:1 to either ipilimumab (10 mg/kg every 3 weeks for 4 doses) or placebo, the addition of ipilimumab to RT did not result in a significant improvement in OS as the primary endpoint.[@bib52] The OS curves were noted to cross, a violation of the proportional hazards model and implying that OS may not be an ideal clinical endpoint or a useful surrogate for the generation of potentially clinically significant tumor-directed immune responses. In support of this possibility, the addition of ipilimumab to RT did result in significantly prolonged PFS (median of 4.0 months vs 3.1 months; HR, 0.70; *P* \< .0001) and led to a higher frequency of posttreatment prostate-specific antigen reductions (13.1% vs 5.2%). In addition, post hoc analysis indicated that there was an improvement in OS with RT + ipilimumab in patients with good prognostic features (alkaline phosphatase \<1.5 times upper limit of normal, hemoglobin \>11 g/dL, and no visceral metastases), suggesting that there are subsets of favorable patients who may still benefit in the primary endpoint of OS and that measures of tumor burden and/or performance status are important to select patient populations that may benefit from RT + immunotherapy combinations. A high tumor burden may be inherently immunosuppressive, and even if an immune response could be successfully generated in this setting, it may not be able to overcome the high tumor burden to achieve clinically meaningful endpoints (eg, survival endpoints). The radiation target may also be an important consideration: Perhaps the bone microenvironment (or bone metastases specifically) is not as conducive to immune priming. Lastly, the overall negative findings from this trial could be interpreted as reinforcing the critical importance of dose and fractionation: Here patients received only 8 Gy × 1, a palliative dose known to be inferior for palliation control.[@bib53] No additional dose-fractionation regimens were tested in this trial. One possibility is that the dose per fraction could be sufficient, but the total dose may be inadequate for the generation of robust tumor-directed immune responses. Future directions {#sec5} ================= Emerging preclinical and clinical data continue to support radiation as a critical component in an immunotherapeutic regimen. Experimental and early clinical models have yet to confirm the optimal dose and fractionation scheme necessary to induce an abscopal response. However, both preclinical and clinical evidence continue to support hypofractionation, with an upper threshold to the fractional dose beyond which abscopal responses are less likely to occur. As further exploration in this area is performed, several considerations should be integrated, including the selection of the best endpoint (eg, local vs distant tumor control, or survival outcomes) as well as the ideal immunotherapy combination. Additionally, with further research the various implications of tumor type, tumor mutational burden, and host immune context (including prior and/or concurrent treatments) will become an increasingly important points of focus. The optimal radiation type (eg, photon vs charged particles) will need further exploration, as will the local and systemic implications of dose rate and frequency of delivery (daily, every other day, or once a week). Ultimately our current interpretation of what may be considered the "best" dose fractionation schemes will be limited until dedicated randomized studies comparing these concepts in combination with immunotherapy are pursued. Sources of support: This work had no specific funding. Conflicts of interest: Dr Formenti reports grants from BMS, Varian, Janssen, Regeneron, Eisai, Merck, other from BMS, Varian, Elekta, Janssen, Regeneron, GSK, Eisai, Dynavax, AstraZeneca, and Merck, outside the submitted work. Drs Ko and Benjamin report no funding or disclosures.

 INTRODUCTION {#sec1-1} ============ Postneurosurgical infection (PNSI) is a major problem and frequently requiring high dose and prolonged antibiotic therapy.\[[@ref16]\] Methicillin-resistant *Staphylococcus* (MRS) is the major Gram-positive organism causing PNSI and its incidence is increasing. Vancomycin is considered the treatment of choice in MRS PNSI; however, it has limited penetration into the cerebrospinal fluid (CSF) that depends on both the integrity of the blood--brain barrier and the inflammatory meningeal status.\[[@ref21]\] Linezolid is a bacteriostatic oxazolidinone antibiotic with a highly activity against Gram-positive cocci resistant to methicillin.\[[@ref6]\] The CSF penetration of linezolid is good and a CSF: plasma ratio of 0.7:1.6 *in vivo*.\[[@ref25],[@ref28]\] In addition, the side effects of this antibiotic are rare and it was generally well tolerated by patients.\[[@ref24]\] The purpose of this study is to evaluate the efficacy of linezolid in the treatment of PNSI caused by MRS. METHODS {#sec1-2} ======= An observational, noncomparative, prospective study was performed at our center, tertiary care teaching hospital, between June 2017 and December 2018. The hospital has a 36-bed neurosurgery ward and six of these beds are in an intensive care unit. Our protocol was approved by the ethical and scientific committee of traumatology and severe Burns Center, Tunisia. Patients were included for the study if they were at least 14 years of age and diagnosed with MRS PNSI. Empirical therapy, consists of vancomycin (60 mg/kg/24 h) in combination with cefotaxime (200--300 mg/kg/24 h), was administered as soon as the infection was suspected and before microbiologic testing result. Linezolid was administered in the standard dosage of 2 × 600 mg/day, after the identification and susceptibility testing of MRS. If the duration of the treatment exceeds 14 days, linezolid intravenous administration is relayed by the oral route at the same dose (600 mg × 2/day). Demographic, clinical, and laboratory information were collected prospectively. Clinical information included procedure surgery, antibiotic combination, duration of treatment, days of hospitalization, and outcomes. Definition of PNSI {#sec2-2_1} ------------------ PNSI was diagnosed according to the criteria of the Centers for Disease Control and Prevention,\[[@ref10]\] adapted to the local protocol, and was classified into meningitis/ventriculitis and brain abscess/empyema. MRS meningitis/ventriculitis was determined as follows: (1) a positive MRS CSF culture and (2) increased white cells (CSF white blood cell count ≥100/mm^3^) and decreased glucose (CSF glucose \<2.5 mmol/L or ratio of CSF glucose to blood glucose \<0.4) and increased lactate (CSF lactate \>4 mmol/l) associated with fever and/ or meningeal signs. MRS brain abscess/empyema was determined as follows: (1) MRS cultured from brain tissue or an abscess seen during a surgical operation and (2) fever, altered mental status, and/or focal neurologic deficits and suggestive computed tomography (CT) scans. Microbiologic success was defined, in the case of meningitis or ventriculitis, as the clearance of MRS from CSF on day 5 of treatment with linezolid. Brain abscess/empyema was defined as cured if CT showed no residual lesion and good neurological status. Laboratory methods {#sec2-2_2} ------------------ CSF was obtained either by lumbar puncture or from extraventricular drainage reservoirs. All CSF samples were analyzed for leukocyte count, glucose, and lactate level and cultivated in aerobic media and anaerobic media. Identification of the etiological agents and susceptibilities to antibiotics was determined by Clinical Laboratory Standards Institute (CLSI) methods.\[[@ref15]\] Statistical analysis {#sec2-2_3} -------------------- A descriptive analysis was performed using SPSS software version 24. Quantitative variables were expressed as the median. RESULTS {#sec1-3} ======= Among 240 patients operated during the study period, 19 had PNSI yielding a total incidence of 7.9%. A total of 10 cases of MRS PNSI were diagnosed. The median age was 48 years (range, 23--71) and 50% of patients were male. Meningitis was the most frequent infection, diagnosed in 6 cases (60%). This was followed by subdural empyema (20%) and ventriculitis (20%). Among the patients with postoperative meningitis, four had undergone surgery for supratentorial tumors, one for posterior fossa tumor, and one for Chiari I decompression. Two cases of ventriculitis occurred after the insertion of external ventricular drain for postoperative hydrocephalus and the patients with subdural empyema had undergone surgery for subdural hematoma. Only four patients received perioperative steroids (patient 2, 3, 4, and 6). Patient demographic data and type of PNSI are summarized in [Table 1](#T1){ref-type="table"}. ###### Clinical, microbiological features, treatment modalities, and outcome of patients.  Laboratory data {#sec2-3_1} --------------- The most common CSF abnormalities, for meningitis and ventriculitis, were pleocytosis (100%), hypoglycorrhachia (100%), and elevated lactate level (87%). The median CSF leukocyte count was 1224 cells/uL and the median of CSF lactate level was 9.8 mmol/l. CSF Gram stain was positive in 10% and culture was positive in all cases. MRS isolated were *Staphylococcus aureus* (MRSA) in seven cases and *Staphylococcus epidermidis* (MRSE) in three cases. Among the patient with PNSI caused by MRSE, two had an external ventricular drain and one had a CSF leakage. All isolated microorganisms were susceptible to vancomycin (minimum inhibitory concentration \[MIC\] = 2 mg/L), teicoplanin (MIC = 2 mg/L), and linezolid (MIC = 1) according to CLSI criteria \[[Table 2](#T2){ref-type="table"}\]. ###### Sensitivity pattern.  Antimicrobial treatment {#sec2-3_2} ----------------------- All patients had been given vancomycin and cefotaxime before receiving linezolid. The median duration of vancomycin administration was 3 days. The median duration of antimicrobial treatment was 18 days (range, 14--42). Patients with MRS meningitis and ventriculitis had received intravenously linezolid for an average of 14 days. Into the two cases of subdural empyema, patients had received intravenously linezolid for 14 days then related by oral linezolid for 14--28 days. During linezolid therapy, two patients had received additional antibiotics for nosocomial pneumonia (patient 2 had received ceftazidime for *Pseudomonas aeruginosa* pneumonia and patient 4 had received colistin for *Acinetobacter baumannii* pneumonia) \[[Table 1](#T1){ref-type="table"}\]. Clinical and microbiologic efficacy {#sec2-3_3} ----------------------------------- All patients with meningitis or ventriculitis had clearance of MRS from the CSF by day 5 of linezolid therapy. In patients with external ventricular catheter (patients 1 and 2), catheter was removed and followed by immediate replacement. In patients with subdural empyema, focal infection had improved on 14 days for patient 8 and on 18 days of linezolid therapy. There were no in-hospital mortalities in our series and all patients were cured at the end of treatment. There were no severe hematologic, renal, or hepatic toxicity during treatment with linezolid. DISCUSSION {#sec1-4} ========== Our study has shown that treatment with linezolid is a safe and effective alternative to vancomycin in patient with postneurosurgical ventriculitis, meningitis, and subdural empyemas caused by MRS. MRS is the most common isolated organism in the PNSI.\[[@ref13],[@ref18],[@ref29]\] In our study, the most commonly identified MRS was MRSA followed by MRSE. Often, the diagnosis of PNSI caused by MRSE is difficult and this is due to the contamination of the microbiological samples.\[[@ref12],[@ref19]\] In the present study, MRSE was associated with shunt infection or CSF leakage. The "gold standard" antimicrobial treatment of PNSI caused by MRS is vancomycin;\[[@ref2]\] however, the penetration of vancomycin into CSF is poor with concentrations equal or less than 10% of those measured in plasma.\[[@ref4]\] This concentration increases in the case of inflamed meninges.\[[@ref20]\] In addition, recently, studies have shown that the incidence of MRS with elevated MIC is increasing and was associated with higher mortality.\[[@ref5],[@ref11]\] In our study, all MRS isolated had a vancomycin MIC of 2 mg/L. Linezolid, bacteriostatic oxazolidinone antibiotic, has a good penetration into the CSF, with a median CSF/plasma ratio of 0.77 and CFS concentration exceeded the MIC of the Gram-positive bacteria that cause PNSI.\[[@ref4],[@ref14]\] Linezolid is recommended for the treatment of pneumonia and skin infection,\[[@ref7]\] and currently, it is increasingly indicated in the treatment of PNSI. The European Federation of Neurological Sciences guideline on the management of community-acquired bacterial meningitis recommends linezolid for the treatment of methicillin-resistant staphylococcal meningitis.\[[@ref3]\] Sipahi *et al*. reviewed in a retrospective cohort study 17 cases of methicillin-resistant staphylococcal postneurosurgical meningitis treated with linezolid (600 mg × 2). Microbiological efficacy rate was 88% by day 5 of linezolid treatment and there were no adverse events.\[[@ref26]\] The same author compared in a retrospective study vancomycin with linezolid in the treatment of MRSA meningitis. Microbiologic success rates on day 5 were superior with linezolid (*P* = 0.044) and a vancomycin MIC of 2 mg/L was found in five strains of MRSA.\[[@ref27]\] In our study, microbiologic success rates of linezolid in patients with meningitis or ventriculitis were 100%. There are few cases in literature that have examined the efficacy of linezolid in the treatment of subdural empyemas. Maure *et al*. have treated successfully two cases of MRSA subdural empyema with linezolid as an adjunct to surgical therapy.\[[@ref17]\] Bahubali *et al*. reviewed retrospectively 21 cases of MRSA intracerebral abscess treated with vancomycin or linezolid and have demonstrated that failure rate was lower with linezolid (25%) compared with vancomycin (43%).\[[@ref1]\] In our series, the two cases of subdural empyema have evolved well with linezolid treatment. Due to its good bioavailability, linezolid is also effective when taken orally.\[[@ref4]\] Martín-Gandul *et al*. evaluated the efficacy of oral linezolid in 77 patients with PNSI. In this study, stable patients were discharged with oral linezolid after a period of intravenous antimicrobial treatment. Seventy-two (93.5%) patients were cured at the end of treatment.\[[@ref16]\] In the present study, patients 8 and 9 received oral linezolid for 14 and 28 days and were cured without recurrence. The most common side effects after linezolid administration are gastrointestinal effects (nausea and vomiting) followed by hematological effects (anemia and thrombocytopenia) and lactic acidosis.\[[@ref8],[@ref9],[@ref22]\] No adverse effects were detected in our patients. The high consumption of linezolid in the intensive care unit is causing the increase of linezolid-resistant bacteria worldwide. Recently, Rodríguez-Lucas *et al*. were detected five cases of nosocomial ventriculitis by linezolid-resistant *S. epidermidis* in a Spanish hospital between 2013 and 2016.\[[@ref23]\] A targeted prescription of linezolid, especially in PNSI, will preserve its effectiveness and avoid the emergence of more resistance. Our study has some limitations. First, it was monocentric noncomparative observational study that included various PNSIs. Second, the small number of patients can limit our results on the efficacy of linezolid for the treatment of PNSI caused by MRS as well as underestimate the incidence of side effects. CONCLUSION {#sec1-5} ========== Our results suggest that linezolid as an alternative to vancomycin for the treatment of PNSI caused by MRS with a high rate of efficacy. Orally linezolid may be a very interesting option for early discharge of patients. **How to cite this article:** Rebai L, Fitouhi N, Daghmouri MA, Bahri K. Linezolid for the treatment of postneurosurgical infection caused by methicillin-resistant *Staphylococcus*. Surg Neurol Int 2019;10:215. Declaration of patient consent {#sec1-6} ============================== The authors certify that they have obtained all appropriate patient consent forms. Financial support and sponsorship {#sec1-7} ================================= Nil. Conflicts of interest {#sec1-8} ===================== There are no conflicts of interest.

The content published in Cureus is the result of clinical experience and/or research by independent individuals or organizations. Cureus is not responsible for the scientific accuracy or reliability of data or conclusions published herein. All content published within Cureus is intended only for educational, research and reference purposes. Additionally, articles published within Cureus should not be deemed a suitable substitute for the advice of a qualified health care professional. Do not disregard or avoid professional medical advice due to content published within Cureus. Introduction ============ The storage lesion refers to the set of biochemical and structural changes that occur during the storage of red blood cells (RBCs) \[[@REF1]\]. The transfusion of RBCs after a prolonged storage period leads to increased RBC lysis, exaggerated inflammatory response, and nitric oxide (NO) scavenging from free hemoglobin and microparticles \[[@REF2]\]. The deleterious effects of the storage lesion at a molecular level are well-established but the potential clinical relevance is unclear. Earlier studies reported increased mortality associated with transfusing older blood, but recent trials have reported no such effect of the storage lesion in critically ill patients \[[@REF3]-[@REF4]\]. Multiple trials have looked at the effect of the storage lesion on mortality and morbidity but little is known about its impact on the RBCs ability to raise post-transfusion hemoglobin. Blood products that have accumulated the storage lesion are more prone to hemolysis after transfusion and, hence, may impact the post-transfusion rise in hemoglobin. We hypothesize that if a relatively new pure red blood cell (PRBC) unit can achieve a higher rise in hemoglobin after transfusion, we can selectively transfuse new blood to patients requiring multiple transfusions and potentially limit the total number of transfusions required to reach a target hemoglobin. Materials and methods ===================== We used the blood bank order report to identify 723 consecutive patients who received PRBC transfusions over a three-month period (from October 2017 to December 2017) at a community teaching hospital. We excluded patients who received more than one unit in a 24-hour period, patients with active overt bleeding within 48 hours of blood transfusion, medical history, and/or laboratory evidence of hemolytic anemia, patients who had a major transfusion reaction, or patients who received an intravenous fluid bolus on the day of transfusion, the latter to negate its dilutional effect. Patients who did not have hemoglobin and hematocrit checked before and after the transfusion within a 24-hour period were also excluded. A thorough retrospective chart review of all PRBC transfusion orders was done by five internal medicine residents and 198 orders (patients) were included in the final analysis. The storage lesion was estimated by calculating the number of days to expiration each PRBC unit had on the day of transfusion. The median number of days to expiration on the day of transfusion was 11 days. We divided the patients into two groups based on the number of days to expiration of the PRBC unit each patient received. Patients who received blood close to its expiration date and, hence, relatively old blood (days to expiration from 0 to 11) were included in group A (n=99). Patients who received blood that was relatively new (days to expiration from 11 to 38) were included in group B (n=99). Baseline characteristics, including age, gender, height, weight, relevant blood count indices, and medical history, were compared. We calculated the mean pre-transfusion and the mean post-transfusion hemoglobin, hematocrit, and red blood cell count of all patients. To determine the effect of the storage lesion on efficacy, we compared the mean rise in hemoglobin, hematocrit, and RBC count between the two groups using the one-tailed t-test. All data were analyzed using SPSS 25.0 (SPSS Inc., Chicago, Illinois, US). Results ======= The baseline characteristics of patients in both groups were similar (Table [1](#TAB1){ref-type="table"}). ###### Patient characteristics SD - standard deviation; RBC - red blood cell count -------------------------------------------------------------- ------------------ ------------------ --------- Patient characteristics Old blood (n=99) New blood (n=99) p-value Age - mean (SD)  65.59 ± 18.8  65.46 ± 16.4  0.961  Male gender (n) 35  36 0.885  Height - mean (SD)  161.64 ± 19.52  165.95 ± 12.7  0.067  Weight - mean (SD)  72.53 ± 26.9  76.43 ± 19.7  0.246  Pre-transfusion hemoglobin -mean (SD)  7.41 ± 0.85  7.39 ± 0.74  0.844  Pre-transfusion hematocrit - mean (SD)  23.18 ± 2.79  23.09 ± 2.66  0.833  Pre-transfusion RBC count - mean (SD)  2.62 ± 0.44  2.67 ± 0.52  0.424  Post-transfusion hemoglobin - mean (SD)  8.39 ± 0.91  8.37 ± 0.85  0.846  Post transfusion hematocrit - mean (SD)  26.54 ± 2.96  26.34 ± 2.93  1.000 Post transfusion RBC count - mean (SD)  3.24 ± 0.41  3.09 ± 0.54  0.571  Hemoglobin recheck time after transfusion-mean hours (range) 9 (3-17) 8.47 (4-23) 0.387 Mean corpuscular volume (MCV) 90.01 ± 9.2  88.11 ± 10.26  0.177  Red blood cell distribution width (RDW) 18.36 ± 4.05  19.82 ± 5.32  0.034  Mean corpuscular hemoglobin concentration (MCHC) 31.71 ± 2.7  31.5 ± 1.8  0.556  Coronary artery disease n (%) 22 (22.2) 18 (18.2) 0.479  Chronic kidney disease n (%) 23 (23.2) 16 (16.2) 0.211  Cancer n (%) 36 (36.4) 48 (48.5) 0.084  Bone marrow suppression n (%) 7 (7.1) 11 (11.1) 0.322  Chronic inflammatory state n (%) 9 (9.1) 7 (7.1) 0.602  Infection n (%) 5 (5.1) 6 (6.1) 0.756  Number of transfusions in the last six months - mean (SD) 2.06 ± 3.95 3.33 ± 4.5 0.058 -------------------------------------------------------------- ------------------ ------------------ --------- The mean pre-transfusion hemoglobin in the group that received old blood was 7.41 ± 0.85 gm/dl as compared to 7.39 ± 0.74 gm/dl with no statistically significant difference (p-value 0.844). Similarly, there was no significant difference in the mean pre-transfusion hematocrit (23.18 ± 2.79 % vs 23.09 ± 2.66 % - p-value 0.833) and RBC count (2.62 ± 0.44 X10^6^/µl vs 2.67 ± 0.52 X10^6^/µl - p-value 0.833) between the two groups. The mean post-transplant hemoglobin (8.39 ± 0.91 gm/dl vs 8.37 ± 0.85 gm/dl - p-value 0.85), hematocrit (26.54 ± 2.96 % vs 26.34 ± 2.93 % - p-value 1), and RBC count (3.24 ± 0.41 X10^6^/µl vs 3.09 ± 0.54 X10^6^/µl - p-value 0.571) were comparable in group A vs B, respectively. The median days to expiration of the PRBC unit was four in group A as compared to 16 in group B. There was no statistically significant difference in the mean time lapse between transfusion and rechecking hemoglobin between the two groups, i.e., nine hours (range: 3-17) in group A vs 8.47 hours (range:4-23) in group B. Patients who received new blood had a higher mean red blood cell distribution width (RDW) at the time of transfusion as compared to the group that received old blood (19.82 ± 5.32 % vs 18.36 ± 4.05 % - p-value 0.034). There was no difference in the prevalence of coronary artery disease (22 vs 18 - p-value 0.479), chronic kidney disease (23 vs 16 - p-value 0.211), cancer (36 vs 48 - p-value 0.084), infections (5 vs 6 - p-value 0.756) in the group that received old blood vs. new blood, respectively. Group A received a mean of 2.06 ± 3.95 blood transfusions in the previous six months as compared to 3.33 ± 4.5 in group B with no statistically significant difference. The mean rise in hemoglobin after transfusing old blood was 1.01 gm/dl with no statistically significant difference from the mean rise in hemoglobin of patients who received new blood, i.e., 1.08 gm/dl - p-value 0.298. The mean rise in hematocrit and RBC count were also similar in both groups (3.37 % vs 3.61 % - p-value 0.249) and (0.42 X10^6^/µl vs 0.44 X10^6^/µl - p-value 0.097), respectively (Table [2](#TAB2){ref-type="table"}). ###### Post-transfusion rise in hemoglobin, hematocrit, and red blood cell count ----------------------------------- ------------ ------------ --------- Variable Old blood  New blood  p-value Mean rise in hemoglobin 1.01 1.08 0.2984 Mean rise in hematocrit 3.37 3.61 0.2498 Mean rise in red blood cell count 0.42 0.44 0.0979 ----------------------------------- ------------ ------------ --------- Discussion ========== To the best of our knowledge, this study is the first attempt to evaluate the effectiveness of the storage lesion on the post-transfusion rise of hemoglobin. Our study shows that patients who were transfused PRBCs with a shorter storage period did not have a higher rise in their hemoglobin, hematocrit, or RBC count within 24 hours post-transfusion compared to patients who received blood with a prolonged storage period. The storage lesion affects the post-transfusion viability of RBCs through various mechanisms. As the blood products are stored, the intracellular glucose level drops, resulting in the reduced production of adenosine triphosphate (ATP) and other important chemical mediators such as 2-3 diphosphoglycerate (2-3 DPG). These biochemical changes in the RBCs lead to an increased susceptibility to oxidative stress compromising membrane integrity and early hemolysis \[[@REF5]-[@REF6]\]. Moreover, the stored RBCs also undergo early exposure and the activation of removal signals present on the cell membranes that are identical to the physiological aging antigens. The activation of the removal signals leads to early immune recognition and the removal of red blood cells after transfusion \[[@REF7]-[@REF8]\]. Laboratory studies have shown evidence of increased in vivo hemolysis of RBCs that are transfused after an increased storage time. In a study by Luten et al., post-transfusion recovery (PTR), as measured by flow cytometry, was significantly higher when PRBC units with a shorter storage period (mean 5+/-2) were transfused as compared to units with a longer storage period (30 +/-3) in the same patient. The difference was attributed to the damage already incurred during the storage period, as most of the removal of RBC was during the first hour \[[@REF9]\]. In a separate study by Hod et al., there was a significantly higher rise in total bilirubin in patients who received older blood whereas there was no difference in the other parameters of hemolysis (i.e., lactate dehydrogenase and haptoglobin) \[[@REF10]\]. The biochemical changes during the storage period appear to be more pronounced than the normal aging of red blood cells in vivo, as evident by the increased hemolysis in the immediate post-transfusion period. Studies have shown that between 5% and 10% of RBCs disappear in the first 24 hours after transfusion followed by a linear disappearance curve \[[@REF11]\]. The average lifespan of RBCs that are stored for 42-45 days is around 90 days as compared to 120 days in normal RBCs produced by the individual's own hematopoietic stem cells \[[@REF12]\]. The results from our study seem counterintuitive, considering data from multiple laboratory studies suggesting the shorter survival of red blood cells with a prolonged storage period. Storage lesion-associated increased hemolysis is probably not severe enough to cause a significant difference in the post-transfusion hemoglobin that can be detected by standard laboratory techniques. It is noteworthy that we excluded patients with evidence of hemolytic anemia to limit its role as a confounding factor. The immune system of patients with hemolytic anemia is already primed to mount an immune response to the red blood cell. Transfusing old blood that is immunologically susceptible can lead to exaggerated hemolysis in such patients. Our study has limitations. This was a retrospective study with a limited patient population. Both groups were similar in patient characteristics but there is still a possibility of confounding factors that we may not have taken into account. Although we excluded patients who received intravenous fluid bolus, we did not account for the maintenance intravenous fluids that can potentially effect hemoglobin level. We cannot extrapolate our findings to all patients needing blood transfusions, as we excluded patients who received multiple transfusions and patients with evidence of hemolysis or bleeding. We only looked at hemoglobin levels within 24 hours of transfusion and cannot exclude the possibility that there could be a difference in the hemoglobin level at an extended follow-up period. This is, however, unlikely since most of the post-transfusion hemolysis occurs within 24 hours. We propose that a prospective clinical trial in controlled settings is required to further validate our findings. In future trials, we recommend following patients\' hemoglobin for an extended period of time and comparing the PRBC requirement in patients who received new blood versus old blood. Conclusions =========== In conclusion, although most laboratory studies have shown the storage lesion to be associated with shorter survival and increased hemolysis after transfusion, we found that it does not translate into a significant difference in the hemoglobin level after transfusing new blood versus old blood. We do not recommend preferentially transfusing blood with a short storage period based on clinical condition or need for multiple transfusions. The authors have declared that no competing interests exist. Consent was obtained by all participants in this study. Monmouth Medical Center IRB Committee issued approval 00002013. Dr. Louis Zinterhofer, Chairman IRB, performed an expedited review on this study on 1/15/2018. He determined this protocol is exempt from further IRB continuing review. The full IRB will be notified of this protocol at the next convened meeting and will be reflected in the minutes. Reason Exempt- 45 CFR 46.101(4). **Animal subjects:** All authors have confirmed that this study did not involve animal subjects or tissue.

Introduction {#sec1-1} ============ In recent years, great interest has been generated in determining the usefulness of teeth for sex determination in different species and populations. Identification of human remains during mass disasters is mainly carried out on bones and teeth as of hindered state of the soft tissues.\[[@ref1]\] Forensic dentistry has played crucial, often key role in the identification of victim of mass disasters.\[[@ref2]\] In massive tragedies and disasters caused by nature, the dentition is most often preserved, even when the bony structures of the body destroyed as of its physical characteristics.\[[@ref3]\] Because teeth are protected by jaw bones so it has the ability to resist better than any skeletal structures, the destructive action of the medium in which they are found.\[[@ref4]\] In post mortem destruction and fragmentation, the use of dental morphology to determine sexual dimorphism, a procedure established in anthropological and biological studies have played a significant role.\[[@ref5]\] Research on sexual dimorphism and its application to human identification have been few when compared with those of bones. Yet many forensic scientists frequently confront isolated teeth and are asked to determine the sex.\[[@ref6]\] Therefore teeth are of great importance because each individual\'s tooth may represent an opportunity to determine sex on its own unlike a long bone or skull from which a limited number of assessments can be made. The purpose of study is to investigate tooth size differences between the sexes to compare it by odontometry in mandibular canine and mandibular first molar. According to Boaz *et al*. (2009), teeth are known to have sexual dimorphism as is the systematic difference in form (shape, size and color) between different genders in same species.\[[@ref7]\] Sex dimorphism in tooth size and the accuracy of odontometrics, sex prediction is found to vary in different region and researchers have advocated the need of population specific data.\[[@ref8]\] Accordingly, estimation of sex does not represent a problem when a complete skeleton is found. Nevertheless, if only the mandibular bone along with the teeth is found or mandible fragments or even the teeth by themselves are available in the site, so estimation of sex may be performed with the help of teeth dimension.\[[@ref9]\] Mandibular canines have a mean age of eruption of 10.87 years and are least affected than any other teeth by periodontal diseases. They are last teeth to be extracted with respect to age. So, mandibular canine consider to have high degree of sexual dimorphism. But anterior teeth including canine are more prone to be fractured as compared to posterior teeth in trauma such as air disaster, hurricanes, and conflagration or road traffic accidents. So need arise to find sexual dimorphism for mandibular posterior teeth.\[[@ref10]\] Thus, comparing sexual dimorphism of permanent mandibular canine with mandibular first molar by odontometry need arises. Materials and Methods {#sec1-2} ===================== Total 100 participants aged between 17-25 years were included as this age is appropriate for a study as such because teeth do not show remodelling as in bone and therefore remain unchanged other than attrition and other dental diseases. Blinding of sampling was done and participants' identification number, gender and date of birth were recorded in computer. Intra-oral examination was carried out according to inclusion and exclusion criteria. Inclusion criteria {#sec2-1} ------------------ Individuals aged between 17-25 years having fully erupted dentition upto first permanent mandibular molar, restorations free, caries free healthy bilateral permanent mandibular first molars and mandibular canine with healthy periodontium. Exclusion criteria {#sec2-2} ------------------ Individual having any oral habits and teeth with orthodontic wire, attrition, abrasion or erosion, restored or carious adjacent tooth, malposition teeth and developmental anomalies were not included. Procedure {#sec1-3} ========= Impression of the mandibular arch was made with alginate (irreversible hydrocolloid impression material) in perforated trays (no 2, 3, 4). Written consents were obtained from the students who underwent examinations and or impression making. Alginate impressions were poured immediately with type IV dental stone to minimize dimensional changes. Dental cast were trimmed to remove excess dental stone. Buccolingual and mesiodistal dimension of the permanent mandibular canine and mandibular first molar were measured by using digital vernier caliper (resolution 0.01 mm). Crown was measured bucco-lingual by measuring greatest distance between facial and lingual surfaces of the crown and parallel to long axis of tooth from both sides' right and left and crown mesio-distal between its contact points \[Figures [1](#F1){ref-type="fig"}--[4](#F4){ref-type="fig"}\]. {#F1} {#F2} {#F3} {#F4} MD diameter of crown:- This measurement is the greatest mesio-distal dimensions between contact points of teeth on either side of jaw. BL diameter of crown:- This measurement is the greatest distance between buccal and lingual surfaces of crown taken at right angle to the plane in which mesio-distal diameter is taken. The measurements were performed by one person and all values were taken to two decimal places. Intra-observer error was assessed by using vernier caliper on 50 male and 50 female student study casts at a different time by the same observer. Good quality study casts were made. The mean value of MD and BL dimensions of male and female were subjected to following formula to calculate sexual dimorphism. Percentage of sexual dimorphism = \[(X~m~/X~f~)-1\] ×100 Where X~m~ = mean male tooth dimension, X~f~ = mean female tooth dimension. Results {#sec1-4} ======= The following parameters were measured and compared on study casts for MD and BL dimensions for right and left mandibular canine and mandibular first molar. It was observed that the mean values of MD diameter showed statistically significant differences between male and female with *P* \< 0.05 measured both in mandibular canine and mandibular first molar teeth when compared to bucco-lingual dimensionsThe mean values of the parameters were greater for mesio-distal width on the left side than on right side when it is measured for male mandibular canine than male mandibular first molarThe student *t* test showed that the difference in mean values of the parameters between the right and left side of mandibular canine were statistically significant with *P* \< 0.001 when compared to mandibular first molarSexual dimorphism was found to be 10.94% and 12.66% in MD dimensions of right and left mandibular canine respectively \[[Table 1](#T1){ref-type="table"}\]. When compared to 6.96% and 0.824% of right and left mandibular first molar \[[Table 2](#T2){ref-type="table"}\]Sexual dimorphism was found 8.33% and 12.65% in BL dimensions of the right and left mandibular canine respectively when compared to 6.62% and 6.75% of right and left mandibular first molar \[Tables [1](#T1){ref-type="table"} and [2](#T2){ref-type="table"}\]Among the mandibular first molar right side was found to exhibit more dimorphism 6.96% in terms of mesio-distal dimensionsMandibular first molar on left side has less dimorphism 6.75% in terms of bucco-lingual dimensionsWhen compared, MD dimensions of mandibular canine exhibit greater dimorphism with mandibular first molarThe present study is an attempt to establish sex differentiation through simple odontometrics technique. This study focused on bucco-lingual and mesio-distal measurements for male and female on permanent mandibular canine and mandibular first molar. Thus, bucco-lingual and mesio-distal diameter of right and left of mandibular canine and mandibular first molar in male and female were measured on study casts. The comparison of mean values of parameters measured between male and female showed highly significant differences of *P* \< 0.005. According to statistical values mandibular canine showed more significance in Garn\'s ratio of MD dimensions when compared to mandibular first molar \[[Table 3](#T3){ref-type="table"}\]Student *t* test showed MD of mandibular canine 4.85 more than BL 3.61 and mandibular first molar MD dimension 3.901 \> than BL dimensions 3.145 \[Tables [4](#T4){ref-type="table"} and [5](#T5){ref-type="table"}\]The mean MD of right and left mandibular canine are significantly different in male and female. Student *t* test for mesiodistal width of left mandibular canine is more than the right mandibular canine \[[Table 6](#T6){ref-type="table"}\]. which is supported by Vishwakarma and Guha\[[@ref11]\] studyApplication of student *t* test and *P* value suggest that mandibular canine showed high degree of sexual dimorphism over mandibular first molar in terms of MD dimension \[[Table 6](#T6){ref-type="table"}\]. ###### Comparison of dimensions (mean) for male and female in mandibular canine  ###### Comparison of dimensions (mean) for male and female in mandibular first molar  ###### Garn\'s ratio and sexual dimorphism in mandibular canine and mandibular first molar  ###### Student *t*-test and descriptive statistics in mandibular canine  ###### Student *t*-test and descriptive statistics in mandibular first molar  ###### "*t*" test and *P* value  Discussion {#sec1-5} ========== Accurate sex prediction is perhaps the most important step in post-mortem reconstructive identification of skeletal remains since it excludes approximately half of the population. This allows investigators and law enforcers to undertake a more focused search of the missing persons' files, and a potentially swift recovery of ante-mortem records. Biological analysis of hard tissues is shown to produce virtually 100% accurate sex identification.\[[@ref12][@ref13]\] However, it is not uncommon for investigative agencies to advice against invasive procedures that result in destruction of evidentiary material, thus necessitating the use of anthroposcopic and/or anthropometric parameters. Sexual dimorphism in tooth measurements has been evaluated for decades, with published reports on male and female odontometric differences available from various countries and diverse population groups. The most dimorphic dimension was bucco-lingual cervical diameter followed by buccolingual crown diameter. European population groups presented the highest degree in sexual dimorphism in teeth whereas native South Americans the lowest.\[[@ref14]\] Surprisingly, though, studies that have gauged sex differences in tooth size in South Asians in general, and Indians in particular are few and recent ones. The fact that most teeth complete development before skeletal maturation makes the dentition a valuable sex indicator particularly in young individuals.\[[@ref15]\] Univariate analysis of the study showed that Mesio-distal dimensions of male dentition are greater than those of female which is in accordance to previous studies. Studies on tooth morphology have in the past been conducted using either intraoral measurements or measurements on casts. Garn *et al*. (1967) and Nair *et al*. (1999) have found that the mandibular canines to exhibit the greatest sexual dimorphism among all teeth. Dahlberg consider mandibular canines as the 'key teeth' for personal identification.\[[@ref16][@ref17][@ref18]\] Hashim and Murshid (1993) conducted a study on Saudi males and females in the age group of 13-20 years to determine the highest likelihood of dimorphism and found that only the canines in both the jaws exhibited a significant sexual difference while the other teeth did not. They also concluded that there was no statistically significant difference between the left and right sides suggesting that measurements of teeth on one side could be truly representative when the corresponding measurements on other side were unavailable. A study by Kaushal *et al*. found a statistically significant sexual dimorphism in mandibular canines in 60 subjects of North Indian population and the mandibular left canine was seen to exhibit greater sexual dimorphism.\[[@ref10][@ref19]\] Schield *et al*. observed sexual difference in tooth size among American black, European and Mongoloid populations. The degree of sexual dimorphism of mandibular canine width was more in Ohio Caucasians and Australian aborigines than in Pima Indians and Tristanite population.\[[@ref20]\] The present study also states that the sexual dimorphism in mandibular canines. Only two studies were reported where maxillary canines were studied. Mohd. Abdulla reported the difference in Saudi population but with a low degree of sexual dimorphism (not statistically significant). Latest study reported by Sharma and Gorea on North Indian population (Patiala) supported our findings that statistically significance sexual dimorphism in present in case of canine. Similarly low degree of sexual dimorphism was reported by Al Rifaiy *et al*. in Saudi Arabian population and by a study of human fossil excavated at Ra\'s Al-Hamra, Eastern Arabian Coast, which showed a general low degree of sexual dimorphism of mandibular canine teeth.\[[@ref21][@ref22]\] Acharya and Mainalli found dimorphism in the mesio-distal dimension of mandibular second premolar in Nepalese population. The finding could be attributed to evolution resulting in a reduction in sexual dimorphism, causing an overlap of tooth dimension in modern males and females.\[[@ref23]\] In the present study, mean mesio-distal dimensions of males are found to be larger than those of females for mandibular canine and mandibular first molar \[Tables [1](#T1){ref-type="table"} and [2](#T2){ref-type="table"}\] Significant differences are observed between sexes of teeth consistent with Garn *et al*.\[[@ref16]\] who indicated that teeth of males were larger than females. According to Moss it is because of the greater thickness of enamel inn males due to the long period of amelogenesis as compared to females. However, in females the completion of calcification of crown occurs earlier in both deciduous and permanent dentition as quoted by de Vito. According to Pratiba *et al*.\[[@ref24]\] sex chromosomes cause different effects on tooth size. The Y chromosome influences the timing and rate of body development, thus producing slower male maturation and acts additively to a greater extent than the X chromosomes. Conclusion {#sec1-6} ========== Our study establishes the fact about permanent mandibular canine with mandibular first molar by odontometrics can be used as adjunct with other parameters for the comparison of sexual dimorphism for a limited value as in cases of highly damaged bodies where only teeth are available for sex determination. It is concluded that comparison of MD in mandibular canine and mandibular first molar showed a greater sexual dimorphism in males. It is recommended to consider the entity for sex determination along with odontometric and skeletal traits as it has shown moderate magnitude of dimorphism. **Source of Support:** Nil **Conflict of Interest:** The authors declare that they have no conflict of interest in the preparation andpublication of this article.

{#sp1 .441}

The neuropathogenesis of HIV is commonly characterized by chronic neuroinflammation in people living with HIV infection (PLHIV) who are aging and virally suppressed on combination antiretroviral treatment (cART).^[@R1]^ Both age and HIV-associated chronic inflammation may be associated with a higher risk of developing neurodegenerative disease such as Alzheimer disease (AD),^[@R1]^ especially as PLHIV are entering age where dementia starts to rise in the general population.^[@R2]^ AD and other forms of dementia and predementia syndromes are often characterized by abnormal deposition of cortical amyloid.^[@R3],[@R4]^ This neuropathologic process may be used as a way to identify aging PLHIV at risk of dementia and may represent one of the endpoints of HIV-related chronic neuroinflammation. One of the most efficient detection methods for brain amyloid deposition in the living brain is the β-amyloid radioligand ^11^C-labeled Pittsburgh compound B (^11^C-PiB) positron emission tomography (PET).^[@R5]^ To date, there is no clear evidence that middle-aged PLHIV are more at risk of developing AD than their HIV-negative (HIV−) counterparts. In particular, it remains unclear whether HIV leads to AD or to an AD-like neurodegeneration for the following reasons. Evidence from the literature remains limited and inconsistent. Achim et al.^[@R6]^ studied 40--65-year-old HIV+ participants using the molecular imaging probe 2-(1-{6-\[(2-\[F-18\]fluoroethyl)(methyl) amino\]-2-naphthyl} ethylidene)malononitrile (FDDNP)-PET and found increased binding of β-amyloid and tau protein aggregates in the posterior cingulate gyrus and the parietotemporal areas in three cases with mild neurocognitive disorder (MND). Furthermore, immunohistochemistry of the neocortex, subcortical white matter, and basal ganglia in HIV+ patients on cART found that 34 of 35 of these cases had intraneuronal deposition of β-amyloid. Tau and α-synuclein were also abundant with the distribution following axonal tracts. APOE genotyping, the major genetic AD risk factor, was not performed, and there were no concomitant neurocognitive data. Another study by Ances et al.^[@R7]^ that included 10 cognitively unimpaired HIV+ individuals (mean age 52 years, age range 46--58 years, 50% men) did not show an overall increased amyloid deposition using ^11^C-PiB/PET compared with 20 HIV− controls (age range 41--53 years, 80% men). This was despite 4 of 10 participants having low CSF amyloid beta (Aβ) 42 levels. The same research group led by Ances et al.^[@R8]^ also found no significant evidence of pathologic amyloid deposition with ^11^C-PiB/PET when compared with HIV− community age-comparable controls (88% men) in 5 HIV-associated neurocognitive disorder (HAND) cases (82% men) and 11 cognitively unimpaired HIV+ patients (100% men) aged in their 40s. The HIV+ groups had a significantly lower amyloid deposition than 11 AD cases aged in their 70s (55% men). APOE genotyping was conducted in \>90% of the HIV+ participants, and 45%--50% has at least one APOE 4 allele, but this was not associated with increased amyloid deposition. Analyses of CSF-AD markers were conducted in 77% of the HIV+ sample, with 5 HIV+ participants found to have low CSF-Aβ42 levels (\<500 pg/mL). CSF-Aβ42 levels were not found to be associated with increased amyloid deposition. A larger study of 26 HIV+ adults aged 26--67 years old, reported as part of a review, found no difference in ^11^C-PiB PET amyloid deposition and CSF-Aβ42 levels between the HIV+ and the HIV− patients aged 32--62 years old.^[@R9]^ This study did not provide APOE genotyping, and levels of cognitive performance were not reported. Finally, a case study of an HIV-infected (HIV+) 71-year-old man with an undetectable plasma viral load found a combined diagnosis of HAND and probable AD.^[@R10]^ This was based on a multidisciplinary examination including neuropsychological testing, brain imaging, CSF-AD markers, fluorodeoxyglucose PET, and florbetaben PET. Overall, ^11^C-PiB PET was the most commonly used method for detecting brain amyloid deposition. All studies conducted have focused on middle-aged PLHIV, mostly without HAND. All studies but one^[@R9]^ reported that most patients were on cART, although the degree of viral suppression was variable or not clearly reported. This represents a major caveat because detectable viral load has been found to be associated with increased Aβ-plaque in the hippocampus in a recent National NeuroAIDS Tissue Consortium (NNTC) analysis.^[@R11]^ APOE genotyping was conducted in only one study.^[@R8]^ While APOE4 status has been associated with HAND in some studies^[@R12],[@R13]^ this has not always been replicated,^[@R14]^ although this result may be influenced by cohorts that included PLHIV of relatively young age, who were not virally suppressed and clinically stable.^[@R14]^ Nevertheless, APOE4 status is a major risk factor for AD and other dementias^[@R14]^ and, thus, remains relevant in aging PLHIV cohorts that could be at higher risk of coincidental AD. Also, APOE4 is a strong modulator of ^11^C-PiB PET data. In a cohort of 497 cognitively normal (NL) middle- and older aged participants with ^11^C-PiB PET,^[@R15]^ a greater proportion of APOE4 carriers developed amyloid-β pathology, at an earlier age, and with faster amyloid-β accumulation. Furthermore, a recent and large neuropathologic study from the NNTC showed that APOE4 and age were strong independent predictors of the deposition of 4G8-Aβ plaques in the frontal neocortex, evident in 29% of a selected sample. ^11^C-PiB PET has high affinity and selectivity for fibrillar Αβ in plaques and extracellular amyloid deposits.^[@R5]^ It may thus lead to lower binding in PLHIV and patients with HAND for whom plaques have been found to be more often diffuse and intracellular.^[@R16][@R17][@R18]^ However, these neuropathologic findings were derived from cohorts that included relatively young patients, active AIDS, as well as many without full viral control and sometimes significant alcohol and drug use comorbidity.^[@R17][@R18][@R19]^ In addition, the detection of abnormal amyloid deposition in the oldest PLHIV with HAND using ^11^C-PiB PET is important, as AD may be coincidental. On the other hand, the diffuse type of plaques may only reflect the early phase of the disease. This study aimed to compare brain amyloid deposition quantified by ^11^C-PiB PET in 10 formally diagnosed HAND cases who were part of the Australian HIV and Aging Cohort. HAND cases were screened for hepatitis C infection, non-HIV neurologic conditions, alcohol and drug use to minimize non--HIV-related neuroinflammation, and other forms of brain pathology.^[@R20]^ These were compared with 39 NL cases, 7 cases with mild cognitive impairment (MCI), and 11 AD cases aged 55--74 years old, all from the Australian Imaging, Biomarkers and Lifestyle (AIBL) cohort.^[@R5],[@R21]^ Because of the relatively small sample size and heterogeneous pattern of amyloid deposition in the HIV+ group, we also conducted a series of case analyses. Finally, we collected long-term health outcomes 8--10 years after the end of the PET data collection to determine dementia progression. Methods {#s1} ======= The University of New South Wales, St. Vincent\'s Hospital, and Austin Health Human Research Ethics Committees approved this research protocol (HREC/08/096). Informed written consent or the guardian consent when appropriate was obtained from all participants. Participants {#s1-1} ------------ Ten HIV+ men formally diagnosed with HAND ([table 1](#T1){ref-type="table"}) were enrolled in a study to quantify brain amyloid deposition using the β-amyloid radioligand ^11^C-PiB as part of the Australian HIV and Brain Aging Cohort Study between 2009 and 2011. The inclusion/exclusion criteria have been described in detail previously.^[@R22]^ All were on cART with undetectable plasma and CSF viral load (\<50 copies/mL). Seven of 10 underwent a lumbar puncture and blood tests to determine CSF-AD markers and HIV markers, respectively, and 8 of 10 had APOE genotyping. All HAND cases underwent neuropsychological testing, and a high-resolution T1-weighted MRI scan was obtained for each person; these methods have been reported in detail elsewhere.^[@R13],[@R22],[@R23]^ Five cases had mild HAND, including 4 cases with asymptomatic neurocognitive impairment (ANI) and 1 case with MND. The 5 other cases had HIV-associated dementia (HAD) along its clinical spectrum (Memorial Sloan Kettering \[MSK\] scale^[@R24]^), including 1 case with mild HAD (MSK1), 2 with moderate HAD (MSK2), and 2 with severe HAD (MSK3). Note that on the MSK scale, MND is classified as MSK1. ###### Demographic and clinical characteristics of the study samples  AIBL data {#s1-2} --------- A request was made to access the summed digital imaging and communications in medicine PET AIBL data on August 28, 2018, at [aibl.csiro.au/research/support/](https://aibl.csiro.au/research/support/) and was approved on August 30, 2018. These data were collected by the AIBL study group following a published protocol.^[@R25]^ The AIBL study methodology has also been reported previously.^[@R5],[@R21]^ AIBL subjects aged between 55 and 74 years with a PET scanning protocol closest to that of the HIV+ sample were retained (that is PET scan with a duration of 1,200 seconds: 57/103). Demographic and clinical characteristics of the retained AIBL cohort are presented in [table 1](#T1){ref-type="table"}. The AIBL sample has comparable demographics with the HIV sample except for more women, which is typical of an elderly cohort, because women have a higher survival rate and are more represented in AD cases^[@R21]^; HIV in Australia remains most prevalent in men who have sex with men.^[@R22]^ It is important that there was no significant differences between sexes in the AIBL PET data (*t* test *p* values \>0.50). Although HIV infection was not formally diagnosed in those controls, the AIBL exclusion criteria were stringent and included a wide range of neurologic, psychiatric, and medical conditions in addition to laboratory test abnormalities.^[@R5],[@R21]^ Evidence of immune compromise would therefore have been detected. Furthermore, the age and demographics of this cohort limits the risk of HIV infection because older age and late diagnosis are more common in non--Australian-born people.^[@R26]^ See examples of AIBL PET data in figure e-1 and e-3, [links.lww.com/NXI/A248](http://links.lww.com/NXI/A248), in supplemental material. PET data acquisition in the HIV+ sample {#s1-3} --------------------------------------- All PET scanning took place at the Austin Health PET Centre in Melbourne (VIC, Australia). Patients were flown from Sydney and back, with their carer as appropriate. They were requested to fast for at least 6 hours before scanning. PET scans for the β-amyloid radioligand ^11^C-PiB were acquired using a Philips Allegro PET camera with a resolution of 5.0  ×  5.0  ×  6.5 mm^3^ (*x y z*). A transmission scan was performed for attenuation correction. PET images were reconstructed using a 3D row action maximum likelihood algorithm with a voxel size of 2.0  ×  2.0  ×  2.0 mm^3^ (*x y z*). Summed images from the 40- to 70-minute time frame were used in each study. Each participant received 370-MBq ^11^C-PiB IV over 1 minute. A 30-minute acquisition (6 × 5-minute frames) in a 3D mode starting 40 minutes after injection of ^11^C-PiB was performed (total PET scan duration of 1,800 ms). PET processing and analyses in HIV and AIBL data {#s1-4} ------------------------------------------------ All imaging data sets were processed using the PETPVE12 toolbox^[@R27]^ for SPM12 (v6906; Wellcome Trust Centre for Neuroimaging, London, UK; [fil.ion.ucl.ac.uk/spm](http://www.fil.ion.ucl.ac.uk/spm)) in Matlab r2012b (Mathworks Inc., Sherborn, MA), which has improved functions to handle amyloid PET partial volume effects (PVEs) compared with previous versions. First, all T1-weighted images were segmented and skull stripped using default parameters. Second, all PET images were coregistered to their associated T1 images using the toolbox\'s default parameters. PVE correction was performed using the Müller-Gärtner method,^[@R28]^ and the intensity of the PVE-corrected PET images was normalized to the average intensity of all cerebellar regions (cerebellum standardized uptake value ratio \[SUVRc\]) from the Automated Anatomical Labeling (AAL) atlas.^[@R29]^ The inverse deformation parameters from the segmentation step were used to transform all AAL regions from the Montreal Neurological Institute to native space, separately for all participants. Finally, SUVRc values were extracted for each region of the AAL atlas in native space. To reduce the total number of regions of interest (ROIs), the 116 regions from the AAL atlas were arranged based on a previous atlas nomenclature^[@R29]^ and averaged out to result in 22 final ROIs. Twenty-six regions related to the cerebellum and the vermis were used as the cerebellum reference. The remaining 90 regions were collapsed into the 22 ROIs. The algorithm used to collapse these regions was developed by M.S.K., neuroanatomist (see supplementary file table e-5, [links.lww.com/NXI/A248](http://links.lww.com/NXI/A248)). Statistical analyses {#s1-5} -------------------- ### Amyloid uptake and deposition distributions in the HIV+ sample {#s1-5-1} Two HIV cases had abnormal ^11^C-PiB uptake (figure e-4, [links.lww.com/NXI/A248](http://links.lww.com/NXI/A248)) in which the PiB uptake was higher in the skull and some parts of the white matter compared with cortical regions. These were subsequently excluded based on the recommendation of the Austin PET team. As a result, 8 HIV cases were considered in the following analyses. The SUVRc data were not normally distributed for any of the study groups. Therefore, all analyses were based on nonparametric statistics. ### Group analyses {#s1-5-2} Study groups (HIV, NL, MCI, and AD) were compared using the Kruskal-Wallis test followed by the Steel-Dwass method to compare all pairs. There were no covariates entered in the analyses. Although the sex ratio did not significantly differ in the AIBL data (*p* \> 0.50, and showed a slightly wider variance when women were included, [table 2](#T2){ref-type="table"}), the analyses were repeated excluding only the female participants in the AIBL NL group (N = 19) compared with the HIV sample, to account for a residual effect of sex. Owing to the limited sample size, this was not performed in the MCI (male = 4) and AD (male = 6) groups. These statistical analyses were conducted using the statistical package JMP 13 (SAS Institute Inc). ### Case analyses {#s1-5-3} Case 2, on visual inspection, clearly exhibited high amyloid deposition across the entire brain. To quantify how significant the amyloid deposition was in this case, we ran case analysis statistics using the program Singlims_ES.exe^[@R30][@R31][@R32]^ against the AD, MCI, and NL groups (AIBL women included). Case 1 had increased amyloid deposition in the hippocampus on visual inspection. Based on this, a single-case analysis was run for this ROI against the AD, MCI, and NL groups. Two other cases (cases 3 and 5) had a noticeable *decrease* in amyloid deposition across the basal ganglia. To quantify any significant differences, case statistics were run for these 2 cases for the cingulum, caudate, pallidum, putamen, and thalamus against the AD, MCI, and NL groups. ### Long-term clinical outcomes in the HIV+ sample {#s1-5-4} We collected long-term clinical information from medical records located at St. Vincent\'s Hospital (Sydney, NSW, Australia) and by contacting the participant\'s primary physicians. This allowed determination as to which patients had progressive dementia and other relevant comorbid conditions 8--10 years after the PET data collection. Data availability {#s1-6} ----------------- The HIV sample data are available on the UNSW Australia repository: [ros.unsw.edu.au/](https://ros.unsw.edu.au/) (search by author "Cysique" and title of the current study). The AIBL data are available upon request to the AIBL group at [aibl.csiro.au](http://www.aibl.csiro.au/). Results {#s2} ======= Group analyses {#s2-1} -------------- The HIV group did not differ statistically from the entire AIBL NL group on any comparisons. The HIV group showed a statistically lower amyloid deposition than the entire MCI group in the caudate and pallidum nuclei. Furthermore, the HIV group obtained a statistically lower amyloid deposition than the entire AD group in all regions, except for the hippocampi, amygdala, temporal pole, and thalamus. On all ROIs, the AD groups exhibited higher amyloid deposition than the NL group. The MCI group obtained higher amyloid deposition than the NL group on all regions, except for the thalami, hippocampi, occipital, parietal, and superior temporal regions ([table 2](#T2){ref-type="table"} and [figure 1](#F1){ref-type="fig"}). The analyses including only male participants produced similar results ([table 2](#T2){ref-type="table"}). Although because of a slightly more restricted variance in the AIBL NL male subgroup, we found a significantly lower amyloid deposition in the HIV sample in the medial frontal and superior parietal cortex. {#F1} Case analyses {#s2-2} ------------- On individual inspection, case 2 (aged 66 years) had high amyloid deposition consistent with AD (whole-brain SUVR~c~ = 2.83) ([figures 1 and 2](#F1 F2){ref-type="fig"}). On quantitative single-case analysis, case 2 showed a significantly increased deposition compared with the NL group, but no difference when compared with the MCI and AD cases. When including only men in the analyses, significant differences were again found between case 2 and the NL and AD groups, but not compared with the MCI group. These results are compiled in [table 3](#T3){ref-type="table"}, and individual ^11^C-PiB PET images are provided ([figure 2](#F2){ref-type="fig"}) and show that case 2 had the highest amyloid deposition compared with all the other groups. As described in [table 3](#T3){ref-type="table"}, case 2 was APOE ε4 homozygous with a CSF profile, based on standard cutoff values, consistent with AD. A detailed clinical history and neuropsychological profile for case 2 is presented in [table 2](#T2){ref-type="table"}. It shows a mix of mild and moderate impairment in cognitive domains affected by HIV-related brain injury (attention/working memory and speed of information processing), but not for motor functions. The memory profile was consistent with probable AD with moderate to severe impairment. . MCI = mild cognitive impairment; NL = cognitively normal.](NEURIMMINFL2019024604f2){#F2} ###### Comparisons of amyloid deposition on the study groups  ###### Case 2 with a probable AD and MND neuropsychological profile concomitant to the PET examination and clinical history  Case 1 (aged 62 years) had elevated amyloid deposition in both hippocampi (table e-1, [links.lww.com/NXI/A248](http://links.lww.com/NXI/A248)). Single-case analysis showed that there was no significant increase in deposition compared with the MCI and AD groups, but there was a trend compared with the NL group (*p* = 0.06; [table 2](#T2){ref-type="table"} and [figure 2](#F2){ref-type="fig"}). This individual was also APOE ε4 homozygous; however, CSF biomarkers were not consistent with an AD profile (CSF Aβ-42 = 168.21 pg/mL; t-tau = 132.18 pg/mL; and p-tau = 36.23 pg/mL). Neuropsychological deficits included mildly impaired speed of information processing and working memory and moderately impaired motor functions with preserved verbal fluency and memory. Case analysis statistics, as well as demographic, neuropsychological, HIV, CSF, and APOE data for case 3 (aged 52 years) and case 5 (aged 54 years) are presented in table e-2, [links.lww.com/NXI/A248](http://links.lww.com/NXI/A248), and e-3, [links.lww.com/NXI/A248](http://links.lww.com/NXI/A248), respectively, and in [figure 2](#F2){ref-type="fig"}. There was a significant *decrease* in amyloid deposition for both cases in all areas tested (range *p* \< 0.04--*p* \< 0.0001), except for the cingulum and putamen when compared with the NL group (*p* = 0.15--0.07). Furthermore, case 5 was identified as having an AD-like CSF profile. Both were cases with severe HAD. Long-term follow-up {#s2-3} ------------------- The 2 patients who originally had clinically severe HAD (MSK3) and 1 who had clinically moderate HAD (MSK2) and had normal amyloid died because of other comorbidities (see table e-4, [links.lww.com/NXI/A248](http://links.lww.com/NXI/A248)). Case 2 who had clinically moderate HAD + AD (MSK2) and significantly elevated amyloid also died after progression to severe dementia. Case 1 who had ANI, with elevated amyloid in the hippocampi, developed a progressive supranuclear palsy (PSP)-like illness, which has remained static. Case 1 also showed microinfarcts and hyperintensities of the white matter consistent with vascular injury on MRI and evidence of cognitive decline reaching an MND (MSK1) diagnosis within 3 years after study. Other cases, which had normal amyloid, included 1 ANI case (case 9), who did not progress clinically according to their physicians and are currently living independently. Case 6, who had MND (MSK1) at entry, has been cognitively stable despite a series of cancers, which have been successfully treated. Case 7, who had clinically mild HAD (MSK1) at entry, has been mostly stable except for progressing peripheral neuropathy. Of the 10 cases, 2 cases (cases 8 and 10) showed abnormal PiB PET uptake (figure e-4, [links.lww.com/NXI/A248](http://links.lww.com/NXI/A248)). Case 8 had ANI at entry that progressed clinically displaying worsening mood changes and behavioral problems with subsequent cognitive decline consistent with MND (MSK1). Case 10 also progressed clinically by reporting increasing cognitive difficulties in line with MND (MSK1). These 2 cases have not reached the HAD stage and still live independently. Discussion {#s3} ========== Consistent with previous studies,^[@R7][@R8][@R9]^ there was no statistical difference in amyloid deposition between the HIV group and NL cases (combining men and women) from the AIBL. However, the NL cases, as part of the AIBL study, were older than those in previous referenced studies. When repeating the analyses with only NL male participants from the AIBL, we found abnormally low amyloid deposition in the HIV group in the medial frontal and superior parietal cortex. This indicates that HIV may be associated with abnormally low amyloid deposition when compared with an older NL cohort. Our study has a small HIV sample size, yet it was a data-rich sample composed of well-characterized HIV+ cases aged 55.5 (46--68) years old, with formally diagnosed HAND, long-term HIV disease clinical stability, viral suppression, and low neurologic and psychiatric confounds. Critically, the sample had long-term follow-up of the participants over years (table e-4, [links.lww.com/NXI/A248](http://links.lww.com/NXI/A248)). This is unique compared with previous HIV PET studies. The small HIV sample size was amenable to a case analysis strategy, which showed abnormal increases of amyloid deposition in 2 HAND cases. Case 2 had an overall deposition of amyloid consistent with AD supported on single-case analysis. The typical AD-like CSF profile and progressive cognitive impairment with a mixed picture of typical AD and HAND deficits supported a probable dominant AD diagnosis. It is important that case 2 was APOE E4 homozygous and was the second oldest participant in the sample (66 years of age). This is consistent with potential coexistence of HAND and AD in an advanced HIV+ patient, as previously described.^[@R10]^ From an AD prevalence perspective, it is logical that some HIV+ patients will develop AD based on genetic and other predisposing factors. It is, however, not clear whether the start of neurodegeneration may be earlier because of HIV. It is also uncertain whether amyloid deposition may have a faster deposition rate,^[@R33]^ although this is certainly possible given that this case had the highest amount of amyloid deposition across all our comparisons. Also, we noted that the cognitive profile was unusual compared with HAND (sparing of motor functions and long-term stability of psychomotor speed) and AD profiles (restricted memory deficits with no agnosia, apraxia, or language deficits except toward the end stage).^[@R34]^ From a clinical perspective, it has been argued that HAND and AD can be distinguished to some degree.^[@R2]^ However, an extensive test battery with the inclusion of agnosia, apraxia, and language would be needed in addition to the usual cognitive domains assessed in HAND to better address this question. It has been noted that ^11^C-PiB PET, which has high affinity and selectivity for fibrillar Αβ in plaques and extracellular amyloid deposits,^[@R5]^ may lead to lower binding in PLHIV and patients with HAND for whom plaques have been found to be more often diffuse and intracellular.^[@R16][@R17][@R18]^ In this study, case 2 shows that such an interpretation may not be correct in aging PLHIV with a possible coincidental AD pathology and profile. A role for HIV in the dysregulation of brain amyloid is biologically plausible. Hypotheses include recent research into the antimicrobial activity of amyloid^[@R35],[@R36]^ in which increased amyloid may reflect a long-term response to the presence of HIV infection in the brain (potentially in brain cellular reservoirs). Supporting this, Bourgade et al.^[@R37],[@R38]^ suggest that β-amyloid can inhibit herpes simplex virus 1 and is associated with overproduction of amyloid more generally in infection-related neurologic diseases. Although case 1 showed no statistical overall increase in amyloid deposition, a trend toward abnormal binding in the hippocampus was observed. The CSF biomarker profile was also not consistent with an AD-like profile. Recent literature on AD suggests that, even subthreshold increases in amyloid levels can result in memory deficits, and in some cases it is the rate of amyloid accumulation, rather than the overall amyloid level, that more accurately predicts worsening memory.^[@R39],[@R40]^ However, case 1 had a PSP-like illness, the pathologic underpinning of which in the context of HIV is unknown at present. Nonetheless, it is interesting that he was homozygous for APOE E4. There is evidence that progression into a PSP-like illness is associated with a homozygous APOE E4 genotype.^[@R41]^ In this study, 2 of the 3 cases (cases 3 and 5) with severe HAD (pre-cART) who survived on cART showed low amyloid deposition. These cases were the most vulnerable and died within 8--10 years of the study\'s long-term outcomes collection. All were younger than 65 years with long HIV duration (\>25 years). Cases 3 and 5 exhibited major subcortical atrophy, consistent with severe HIV-related brain injury.^[@R42][@R43][@R45]^ This level of subcortical atrophy potentially biased the measurement of amyloid in the basal ganglia. Nevertheless, there is at least 1 plausible mechanism for the abnormally low level of amyloid. Microglia and astrocytes primed by HIV may play important roles in engulfing and purging Aβ plaques from the brain.^[@R46]^ In severe HAD with a disrupted blood-brain barrier, amyloid clearance may be enhanced in some individuals. Also, in treated PLHIV, amyloid clearance may occur through HIV-1 Tat priming and activation of microglia.^[@R47]^ Case 5 had a CSF biomarker profile that supported AD (T-tau \>350 Ab42/P-tau \<6.5), but the PET scan did not confirm increased amyloid deposition suggesting caution in the clinical use of CSF for AD diagnosis in patients with HIV, especially those with HAND. Finally, we cannot exclude that in these younger cases, the plaques may have been diffuse rather than fibrillar, and this may have contributed to a lower ^11^C-PiB PET signal.^[@R16][@R17][@R18]^ Although our study does not provide statistical evidence of an overall increase in amyloid in PLHIV with HAND, it highlights the importance of interindividual variations in patients with chronic HIV and HAND and the need to focus the research on favorable and unfavorable phenotypic and/or biotypic clusters.^[@R48]^ In this small sample, we were able to detect cases with high, focused (albeit trending) and very low brain amyloid. Such complex phenotypes if already apparent in a pilot study suggest a need to focus on long-term individual phenotypic analyses. Future longitudinal studies monitoring amyloid deposition as PLHIV age are crucial in understanding the relationship between abnormal amyloid levels and HAND as well as other comorbidities. It is noteworthy that all the HIV+ patients in this study also had a large number of medical comorbidities (See table e-4). How exactly such comorbidities played a role in their amyloid pattern and then clinical progression or lack thereof remains to be fully elucidated. Although we used a PET protocol that retained the individual neuroanatomy, it would be important that larger studies use a multimodal imaging strategy because abnormal brain aging in HIV can be detected by using volumetric MRI.^[@R49]^ Our study had some limitations. The controls used in this study were obtained from AIBL and included a mix of male and female participants differing from the HIV+ sample who were all men. The sex difference had a minimal effect on the cases with AD + HAD, when specifically investigated. We also had missing data for the APOE genotyping and CSF profiles and recognize this, just as for previous studies have, as a limitation. Although 50% of the HAND cases had HAD as per Frascati criteria at study entry, it is important to note that the MSK clinical staging was well represented as well as those with mild HAND as per Frascati criteria. Finally, our pilot sample size and case series analyses yield indicative rather than definitive findings. It remains unclear why 2 HAND cases had no cortical amyloid uptake. Despite a thorough search, we could not find reports of similar issues in the HIV or non-HIV literature. There was no PET scan technical issue reported by the Austin Health PET Centre examiners. In conclusion, relative to NL older controls, brain amyloid burden does not differ in virally suppressed HAND at the group level. However, individual analyses show that abnormally high and low amyloid burden occur. These results indicate that focus on individual phenotypes in longitudinal analyses in the HIV-aging cohort is needed. The authors thank the participants for their time on the study and dedicate this work to the patients whom they cared for and who have since died. They thank the physicians of the HIV+ patients for responding in a timely and detailed fashion to their queries. They thank Dr. Jones, Prof. Villemagne, and Prof. Rowe from the Centre for PET, Austin Health, Melbourne, Australia, for their support in interpreting the data acquisition validity. Study funding ============= This work was supported by the NHMRC project grants (APP568746; PI Cysique, APP1105808; PI Brew, and APP1045400; PI Cysique), Peter Duncan Neuroscience Research Unit at St. Vincent\'s Hospital Centre for Applied Medical Research (Director: Prof. Brew), and a UNSW Australia Faculty of Medicine development. Disclosure ========== B.J. Brew reports personal fees from AbbVie, personal fees from ViiV, grants and personal fees from Biogen Idec, personal fees from Merck Sharp and Dohme, grants from the National Health and Medical Research Council, and grants from the National Institutes of Health, outside the submitted work. C.D. Rae reports personal fees from Philips Healthcare, personal fees from Springer, and personal fees from Elsevier, outside the submitted work. Other authors report no declaration of interest. Go to [Neurology.org/NN](https://nn.neurology.org/content/7/4/e739/tab-article-info) for full disclosures. AAL : Automated Anatomical Labeling Aβ : amyloid beta AD : Alzheimer disease AIBL : Australian Imaging, Biomarkers and Lifestyle ANI : asymptomatic neurocognitive impairment APOE : apolipoprotein E ^11^C-PiB : ^11^C-labeled Pittsburgh compound B cART : combination antiretroviral therapy HAD : HIV-associated dementia HAND : HIV-associated neurocognitive disorder MCI : mild cognitive impairment MND : mild neurocognitive disorder MSK : Memorial Sloan Kettering NNTC : National NeuroAIDS Tissue Consortium PLHIV : people living with HIV infection PSP : progressive supranuclear palsy PVE : partial volume effect ROI : region of interest SUVR : standardized uptake value ratio [^1]: Go to [Neurology.org/NN](https://nn.neurology.org/content/7/4/e739/tab-article-info) for full disclosures. Funding information is provided at the end of the article. [^2]: The Article Processing Charge was funded by the authors. [^3]: All authors are aware of the terms of the AIBL Data Use Agreement. Data used in the preparation of this article were obtained from the Australian Imaging, Biomarkers and Lifestyle Flagship Study of Ageing (AIBL) funded by the Commonwealth Scientific and Industrial Research Organisation (CSIRO), which was made available at the ADNI database ([loni.usc.edu/ADNI](http://www.loni.usc.edu/ADNI)). The AIBL researchers contributed data but did not participate in analysis or writing of this report. See [aibl.csiro.au](http://www.aibl.csiro.au) for further details.

Study Highlights {#cts12502-sec-0020} ================ **WHAT IS THE CURRENT KNOWLEDGE ON THE TOPIC?** ✓ Determining the optimal dose for the treatment of rheumatoid arthritis (RA) demands several large studies conducted over at least 12 weeks of treatment or placebo control. More recently, the US guidance for drug development for RA treatments includes suggestions that pharmacokinetics/pharmacodynamics (PK/PD) studies assessing response between 2 and 8 weeks may be more informative in distinguishing doses. **WHAT QUESTION DID THIS STUDY ADDRESS?** ✓ Can a small study including assessment of PK/PD over a wide range of doses over a 4‐week period provide sufficient information to inform the design of subsequent studies, minimizing the number of patients and length of time required to determine the optimal dose or a narrow range of doses suitable for further study? **WHAT THIS STUDY ADDS TO OUR KNOWLEDGE** ✓ Empiric application of an Emax model to observations collected over a wide range of concentrations can provide insight about potentially effective dosing/concentration ranges to explore in larger phase IIb and III studies. **HOW THIS MIGHT CHANGE CLINICAL PHARMACOLOGY OR TRANSLATIONAL SCIENCE** ✓ It is important to extend the dosing range as wide as possible in order to gain the most insight into a possible concentration--response range. Limiting the length of the study to a very short interval may allow doses that would not be reasonable for a longer study, due to patient safety or lack of effective treatment. The successful registration of a new drug requires completion of at least two randomized, controlled clinical trials demonstrating safety and efficacy of the drug at a defined dose in the intended patient population. Translation of the preliminary assessment of safety and pharmacokinetics (PK) observed in first‐in‐human studies to patients, and of evidence of efficacy from nonclinical models to patients, are two of the most important tasks in drug development. Rheumatoid arthritis (RA) is the third‐most common type of arthritis, after osteoarthritis and gout.[1](#cts12502-bib-0001){ref-type="ref"} The prevalence of RA is believed to range from 0.4--1.3% internationally, and an estimated 1.5 million (0.6%) adults aged 18 years and older in the United States were reported to have had RA in 2005.[2](#cts12502-bib-0002){ref-type="ref"} In addition to morbidity and decreased quality of life due to pain and joint damage, mortality is increased among those who have RA compared with those who do not.[2](#cts12502-bib-0002){ref-type="ref"} The development of effective treatments is clearly a significant public health concern. The first US regulatory guidance for the development of drug products for the treatment of RA was finalized in 1999,[3](#cts12502-bib-0003){ref-type="ref"} and a revised draft guidance was published in 2013.[4](#cts12502-bib-0004){ref-type="ref"} Key differences between the two guidance documents, which reflect the advances in clinical research and drug development of treatments of RA, are emphasis on the length of clinical trials and the role of pharmacokinetic (PK) and pharmacodynamic (PD) assessments. In the first guidance document, it is recommended that clinical trials continue for at least 6 months, due to the long duration of RA and the importance of demonstrating the durability of the treatment, particularly for biological agents where the development of antidrug antibodies could potentially interfere with clinical efficacy.[3](#cts12502-bib-0003){ref-type="ref"} In the second guidance, earlier evaluation of clinical response, i.e., between 2 and 8 weeks, prior to the likely plateau of clinical response, was suggested as likely to be more informative and more likely to distinguish between doses.[4](#cts12502-bib-0004){ref-type="ref"} A second difference between the two guidance documents is a reflection of the usefulness of PK and PD assessments of different dosage regimens in optimizing treatment. The first guidance document recommended assessment of PK primarily to understand the absorption, distribution, metabolism, and elimination (ADME) characteristics of the drug and the potential impact of immunologic response on ADME of biological therapies,[3](#cts12502-bib-0003){ref-type="ref"} whereas the second guidance advocates linking drug‐concentration results to measures of efficacy and safety, thus providing support for the recommended dosage.[4](#cts12502-bib-0004){ref-type="ref"} Etanercept (Enbrel, Pfizer, New York, NY) is a biological disease‐modifying antirheumatic drug (DMARD) that inhibits tumor necrosis factor (TNF), and thus decreases inflammation. The first clinical trials of etanercept in patients with RA began in 1993^5^ and it was approved for reducing the signs and symptoms of RA in 1998 in the United States.[6](#cts12502-bib-0006){ref-type="ref"} Subsequent studies were performed to expand the approved RA indications to include inducing major clinical response, inhibiting the progression of structural damage, and improving physical function. Development of etanercept occurred concurrently with the advancement of science, which resulted in the differences between the two guidance reports. The goal herein is to report the results of a small, previously unpublished, dose‐finding study that incorporated PK/PD analyses performed during the development of etanercept, thus providing a case study of the advances in clinical trial design for RA that encompasses the differences between the two guidance documents. This may offer a helpful example of study design for dose‐finding studies in other diseases, especially those with small patient populations. METHODS {#cts12502-sec-0030} ======= All available dose‐finding studies conducted during the development of etanercept for the treatment of RA were identified. Information on the doses administered, number of patients enrolled, length of study, outcome measures (including PK assessment), and study conclusions was collated and reviewed. Five of the six studies have been published previously (**Table** [1](#cts12502-tbl-0001){ref-type="table-wrap"}).[5](#cts12502-bib-0005){ref-type="ref"}, [7](#cts12502-bib-0007){ref-type="ref"}, [8](#cts12502-bib-0008){ref-type="ref"}, [9](#cts12502-bib-0009){ref-type="ref"}, [10](#cts12502-bib-0010){ref-type="ref"} The remaining study is described below. ###### Dose‐ranging studies of etanercept in patients with rheumatoid arthritis Study *N* Dose Duration (months) Outcomes assessed Conclusions ---------------------------- ----- -------------------------------------------------------------------- ------------------- --------------------------------------- ------------------------------------------------------------------- Moreland *et al*., 1996^5^ 22 4, 8, 16, 32 mg/m^2^ i.v. then 2, 4, 8, 16 mg/m^2^ SQ twice‐weekly 1 Various timings; efficacy, safety, PK No clear dose response; serum concentrations proportional to dose Moreland *et al*., 1997^7^ 180 0.25, 2, 16 mg/m^2^ SQ twice‐weekly vs. placebo 3 Every 2 wk for 3 mo; efficacy, safety 16 mg/m^2^ most efficacious, no dose‐limiting toxicity Moreland *et al*., 1999^8^ 234 10, 25 mg SQ twice‐weekly vs. placebo 6 2 wk, 3 mo; efficacy, safety 25 mg more rapid and better response, no dose‐limiting toxicity Bathon *et al*., 2000^9^ 632 10, 25 mg SQ twice‐weekly vs. MTX 12 3, 6, 9, 12 mo; efficacy and safety 25 mg improved efficacy compared with MTX Johnsen *et al*., 2006^10^ 77 25, 50 mg SQ twice‐weekly 6 4, 8, 12, 24 wk; efficacy, safety No difference in efficacy or safety i.v., intravenous; MTX, methotrexate, PK, pharmacokinetics; SQ, subcutaneous. John Wiley & Sons, Ltd. Patients {#cts12502-sec-0040} -------- Eligible patients aged 18 to 75 years were enrolled from September 1997 to August 1998. All patients met the 1987 American Rheumatism Association (ARA) criteria for RA and belonged to ARA functional class I to III. Their onset of disease had to have been after 16 years of age and disease duration had to be \<10 years. They had to have failed to respond to at least one previous DMARD and have had active RA, defined by the presence of ≥ six swollen and ≥12 tender joints and at least two of the following three criteria: erythrocyte sedimentation rate (ESR) ≥28 mm/h, C‐reactive protein (CRP) ≥20 mg/L, and morning stiffness ≥45 minutes. Prior treatment with an anti‐TNF monoclonal antibody or with a soluble TNF receptor were exclusion criteria. The study was conducted in accordance with the Declaration of Helsinki and independent Ethics Committee approval was obtained for the protocol and all amendments. Written informed consent was obtained from all patients at the time of enrollment. Study design {#cts12502-sec-0050} ------------ This randomized, double‐blind, placebo‐controlled, prospective study included a DMARD washout period, a treatment period, and a follow‐up period. Etanercept was supplied in vials of sterile lyophilized powder containing 10‐ or 25‐mg doses that were reconstituted with 1 mL of water for injection. Patients were randomly assigned in block sizes of 5 in a 2:2:1 ratio (two for each active group, one for placebo) to receive subcutaneous injections of etanercept (10 mg once weekly, 10 mg twice weekly, 50 mg once every 2 weeks, 50 mg once weekly, or 50 mg twice weekly) or matching placebo for 4 weeks. The doses and regimens were chosen to provide a large range of etanercept serum concentrations to examine potential concentration--response relationships for various markers of biologic activity. Prior to this study, doses up to 25 mg twice weekly had been evaluated. The dose of 50 mg twice weekly was added to this study as a preliminary assessment of further benefit and safety. Although the short duration of treatment would not result in optimal response, it was thought to be long enough to elicit some change in disease activity evaluations. All DMARDs had to be discontinued at least 4 weeks prior to dosing of study medication and the follow‐up period after treatment was 4 weeks. Assessments {#cts12502-sec-0060} ----------- Prestudy screening assessments included medical history, physical examination, vital signs, electrocardiogram, joint assessment, pain (visual analog scale (VAS)), duration of morning stiffness, CRP, ESR, hematology, blood chemistry, serum human chorionic gonadotropin test for women who entered menopause less than 2 years before screening visit, and urinalysis. Additional assessments taken at baseline included rheumatoid factor, etanercept concentration, interleukin (IL)‐6, IL‐1 receptor antagonist (IL‐1Ra), matrix metalloproteinase (MMP)‐3, and testing for antibodies (antinuclear, antidouble‐strand DNA, anticardiolipin, antietanercept). Etanercept serum concentration, number of swollen/painful joints, ESR, IL‐6, and MMP‐3 were assessed at baseline, on days 8, 15, 22, and 29 of treatment, and during weeks 1, 2, and 4 of follow‐up. Blood samples for etanercept concentration measurements were collected at the same time as efficacy assessments. Etanercept concentrations were determined by a validated enzyme‐linked immunosorbent assay with a limit of quantitation of 0.781 ng/mL based on a 1:5 minimum sample dilution. All reported adverse events were recorded. Data analysis {#cts12502-sec-0070} ------------- Sample size and power were not applicable and no formal hypothesis testing was done. There was no formal between‐group statistical evaluation of efficacy. The last observation carried forward imputation method was used for missing data. For the PK analysis, serum concentration measurements were combined with dose administration information and fit with a one‐compartment PK model with first‐order absorption using WinNonlin (v. 1.5, Pharsight, Mountain View, CA) for each individual patient. Population PK methods were not utilized because of the small number of observations. Predose etanercept concentrations above the lower limit of quantification, caused by crossreactive analytes in the serum, were not subtracted from the postdose concentrations because they were very low. If a one‐compartment model did not fit well for an individual, noncompartmental methods were used. The first‐order elimination rate constant (λ~z~) was determined using log linear regression of etanercept concentrations in the terminal portion of the elimination phase. Apparent clearance (Cl/F) was estimated by computing a ratio of dosing rate (mg/h) to concentration at steady‐state (C~ss~). Apparent volume of distribution (V/F) was calculated as a ratio of Cl/F to λ~z~. Half‐life was calculated as 0.693/λ~z~. An analysis of variance ([anova]{.smallcaps}) was performed to test differences in PK parameters (absorption rate constant, λ~z~, area under the serum concentration--time curve (AUC), Cl/F, and V/F) between dose groups. AUC values were normalized for dose prior to performance of ANOVA. There was considerable noise in the PD observations because of a small number of sparsely timed PD observations, so a naïve pooled‐estimate process was used to fit PD observations to pooled C~ss~ data as follows. Etanercept C~ss~ for each patient as measured on day 29 was used, with the exception of patients who received 50 mg once every 2 weeks, for whom the day 22 C~ss~ was used. The low concentrations of etanercept reported in patients treated with placebo were used for their day 29 C~ss~. Next, to provide even weighting across the concentration spectrum, the range of observed C~ss~ values was broken into six intervals so that each interval of C~ss~ values contained at least five observations. The intervals were uneven in length, being shorter where there were more observations. A mean C~ss~ was calculated for each interval and the mean of the PD observations was calculated using each patient\'s corresponding observation. PD analysis was performed by fitting a sigmoidal maximal effect (E~max~) model to mean C~ss~ for each interval and mean change for each interval from baseline to 29 days in the following: number of painful joints, number of swollen joints, pain (VAS), morning stiffness, ESR, IL‐6, IL‐1Ra, and MMP‐3. The E~max~, half‐maximal effect (EC~50~), and sigmoidicity factors (ς) were estimated using WinNonlin. PD assessment of adverse events was not attempted. RESULTS {#cts12502-sec-0080} ======= Patient disposition {#cts12502-sec-0090} ------------------- Of the 61 patients enrolled, 49 received etanercept and 12 received placebo. PK parameters were determined in 43 patients. Demographic and baseline characteristics are shown in **Table** [2](#cts12502-tbl-0002){ref-type="table-wrap"}. ###### Demographic and baseline characteristics ------------------------------------------------------------------------------------------------------------------------------------------------- **Etanercept** -------------------------- -------------------------- -------------------------- --------------- ---------------- --------------- --------------- Age, y 51.5 ± 12.4 52.8 ± 14.7 51.6 ± 13.9 49.3 ± 12.5 55.6 ± 12.8 53.0 ± 15.3 Female, *n* (%) 11 (100) 10 (91) 6 (67) 7 (78) 7 (78) 11 (92) Weight, kg 78.7 ± 11.6 64.3 ± 8.9 72.9 ± 11.4 74.9 ± 9.8 68.1 ± 12.2 68.5 ± 9.7 Disease duration, y 4.2 ± 3.1 6.5 ± 3.3 3.5 ± 2.9 4.2 ± 3.4 5.0 ± 2.6 5.0 ± 3.6 RF positive, U/L 83.3 ± 103.9 (*n =* 10) 48.1 ± 65.6 (*n =* 10) 34.7 ± 43.1 37.6 ± 22.9 79.9 ± 77.0 97.2 ± 93.2 NSAIDs, *n* (%) 8 (73) 7 (64) 8 (89) 7 (78) 7 (78) 8 (67) Corticosteroids, *n* (%) 5 (45) 8 (73) 1 (11) 3 (33) 7 (78) 7 (58) ≥1 prior DMARD, *n* (%) 2 (18) 1 (9) 5 (56) 6 (67) 3 (33) 4 (33) Pain (VAS) 67.9 ± 16.5 67.0 ± 22.6 61.6 ± 17.0 62.7 ± 34.6 58.8 ± 15.6 60.8 ± 19.6 Painful joints 29.8 ± 15.0 35.4 ± 15.9 28.6 ± 12.2 27.9 ± 13.1 26.2 ± 10.7 32.5 ± 11.8 Swollen joints 20.3 ± 12.7 28.7 ± 12.9 20.8 ± 6.6 21.7 ± 8.4 24.6 ± 9.6 23.5 ± 7.2 Morning stiffness, min 106.8 ± 75.7 177.3 ± 193.6 80.6 ± 47.2 142.5 ± 125.0\ 142.2 ± 81.5 172.5 ± 126.7 (*n =* 8) CRP, mg/L 52.8 ± 30.1 (*n =* 9) 61.8 ± 32.6 (*n =* 10) 58.8 ± 28.8 72.2 ± 27.3 37.2 ± 27.7 47.8 ± 18.4 ESR, mm/h 31.7 ± 22.0 (*n =* 10) 64.7 ± 46.3 (*n =* 10) 42.4 ± 26.5 50.4 ± 21.6 35.9 ± 35.5 45.0 ± 34.7 IL‐1Ra, pg/mL 1433 ± 690.1 (*n =* 10) 1165 ± 763.7 (*n =* 10) 1337 ± 465.3 1216 ± 832.9 1367 ± 1313 1090 ± 671.7 IL‐6, pg/mL 30.6 ± 26.8 (*n =* 10) 46.3 ± 28.8 (*n =* 10) 86.9 ± 125.1 55.8 ± 35.0 30.9 ± 30.9 43.3 ± 45.7 MMP‐3, ng/mL 109.3 ± 102.0 (*n =* 10) 192.3 ± 115.7 (*n =* 10) 236.3 ± 136.6 296.7 ± 154.7 128.8 ± 114.8 154.9 ± 127.0 ------------------------------------------------------------------------------------------------------------------------------------------------- Values are mean ± standard deviation unless otherwise stated. CRP, C‐reactive protein; DMARD, disease‐modifying antirheumatic drug; ESR, erythrocyte sedimentation rate; IL‐1Ra, interleukin‐1 receptor antagonist; IL‐6, interleukin‐6; MMP‐3, matrix metalloprotease‐3; NSAID, nonsteroidal antiinflammatory drug; RF, rheumatoid factor; VAS, visual analog scale. John Wiley & Sons, Ltd. Safety {#cts12502-sec-0100} ------ Etanercept appeared to be generally well tolerated by patients. Injection site reactions, rhinitis, and upper respiratory infections were the most common treatment‐emergent adverse events for most etanercept‐treated patients. Two patients in the etanercept 10 mg twice‐weekly group and one patient in the placebo group withdrew from the study due to adverse events: one patient in the 10 mg twice‐weekly group experienced lower extremity vascular disorder (not related to etanercept treatment); a second patient receiving 10 mg twice‐weekly had facial edema and pruritus after the first injection and mild dyspnea occurred after the second injection; a patient in the placebo group had a broken tibia. These three patients were excluded from the PK analysis. PK analysis {#cts12502-sec-0110} ----------- The mean etanercept concentrations by dose are shown in **Figure** [1](#cts12502-fig-0001){ref-type="fig"}. Exposure with the 10 mg once‐weekly dose was clearly lower than the other etanercept doses and the 50 mg twice‐weekly dose was clearly higher, whereas the 10 mg twice‐weekly, 50 mg every 2 weeks, and 50 mg once‐weekly doses provided similar exposure to each other, consistent with the respective C~ss~ values. PK parameters are shown in **Table** [3](#cts12502-tbl-0003){ref-type="table-wrap"}. The PK parameters for seven patients (*n =* 1 each for all dosing groups with the exception of the 50 mg once‐weekly group, *n =* 2 in the 50 mg once‐weekly group) were assessed using noncompartmental analysis methods, because of inadequate characterization of the first‐order absorption rate constant. {#cts12502-fig-0001} ###### Steady‐state etanercept PK parameters in patients with RA **Dose (*n*)** **C~max~ (mg/L)** **T~max~ (h)** **T~1/2~ (h)** **Cl/F (mL/h)** **AUC (mg·h/L)** **C~ss~ (ng/mL)** **V/F (L)** ------------------------------ ------------------- ---------------- ---------------- ----------------- ------------------ ------------------- ------------- 10 mg once‐weekly (11) 0.40 ± 0.17 65 ± 39 68.2 ± 27.4 168 ± 82 68 ± 26 300 ± 135 16 ± 11 10 mg twice‐weekly (9) 0.57 ± 0.42 58 ± 27 61.0 ± 12.8 203 ± 181 74 ± 39 806 ± 424 18 ± 16 50 mg once every 2 weeks (9) 2.21 ± 1.58 43 ± 50 62.9 ± 23.1 158 ± 108 294 ± 202 915 ± 621 33 ± 55 50 mg once‐weekly (5) 1.56 ± 0.73 54 ± 64 58.1 ± 19.0 234 ± 108 181 ± 132 1170 ± 660 32 ± 32 50 mg twice‐weekly (9) 3.26 ± 1.66 45 ± 30 57.8 ± 26.1 143 ± 85 412 ± 162 3888 ± 1247 12 ± 7.6 Overall mean ND 53 ± 40 62 ± 22 175 ± 116 ND ND 21 ± 29 Values are mean ± standard deviation. AUC, area under the concentration--time curve; C~max~, maximum serum concentration; Cl/F, apparent total clearance; C~ss~, concentration at steady‐state; ND, not determined; T~½~, elimination half‐life; T~max~, time to reach maximum concentration; V/F, volume of distribution. John Wiley & Sons, Ltd. PD analysis {#cts12502-sec-0120} ----------- There were between five and 14 concentration--effect pairs in each of the six C~ss~ ranges. The means of the pooled etanercept concentrations and PD observations used for the modeling procedure are shown in **Table** [4](#cts12502-tbl-0004){ref-type="table-wrap"}. Attempts to fit a model to pain (VAS), duration of morning stiffness, and IL1‐Ra were not successful. The model did not fit or was inappropriate for various reasons, including failure to demonstrate a relationship and/or lack of a clear maximum effect. However, PD measures for swollen joints, painful joints, ESR, IL‐6, and MMP‐3 and corresponding mean estimates of C~ss~ were successfully fit with a sigmoidal E~max~ model. The estimates for PD model parameters are shown in **Table** [5](#cts12502-tbl-0005){ref-type="table-wrap"}. The plots of the predicted vs. observed effect for swollen joints and painful joints, representative of those for other PD measures, are shown in **Figure** [2](#cts12502-fig-0002){ref-type="fig"} **a,b**, respectively. These figures both show a plateau at ∼2,000 ng/mL, with no increase in percent change from baseline in either swollen joints (**Figure** [2](#cts12502-fig-0002){ref-type="fig"} **a**) or painful joints (**Figure** [2](#cts12502-fig-0002){ref-type="fig"} **b**), despite increased etanercept concentrations. ###### Mean concentrations and pharmacodynamic measurements (% change from baseline) **Concentration range (ng/mL)** ---------------------------------------------------- --------------------------------- ---------- --------------------------------------------------- -------------- -------------- --------------------------------------------------- *n* 7 14 7 9 5 6 Concentration range, ng/mL 0.96--90.2 124--422 514--881 1,012--1,985 2,100--2,985 3,185--6,080 Mean concentration, ng/mL 20.0 296 710 1,214 2,429 4,510 Swollen joints 1.00 13.83 40.99 33.37 51.83 54.16 Painful joints 15.62 14.92 31.35 52.26 62.21 69.00 Pain (VAS) 9.38 28.73 30.95[a](#cts12502-tbl4-note-0002){ref-type="fn"} 42.27 58.78 80.04 Morning stiffness 20.24 22.44 39.88 45.31 72.92 86.11 ESR 9.38 16.67 25.82 37.25 22.35 37.24 IL‐6[b](#cts12502-tbl4-note-0003){ref-type="fn"} --124.3 25.15 22.91 32.27 --0.49 57.41 IL‐1Ra[c](#cts12502-tbl4-note-0004){ref-type="fn"} 5.16 --0.07 12.53 11.09 --0.24 13.98 MMP‐3[d](#cts12502-tbl4-note-0005){ref-type="fn"} --28.61 12.86 15.98 28.22 25.83 28.23[e](#cts12502-tbl4-note-0006){ref-type="fn"} Patients receiving placebo were included in the pharmacodynamic analysis. One patient was removed from the analysis due to an outlier reading of percentage change from baseline of --933%. Values were scaled prior to analysis by adding 125 to score to avoid negative numbers. Values were scaled prior to analysis by adding 1 to score to avoid negative numbers. Values were scaled prior to analysis by adding 30 to score to avoid negative numbers. One patient was removed from the analysis due to an outlier reading of percentage change from baseline of --248%. ESR, erythrocyte sedimentation rate; IL‐6, interleukin‐6; IL‐1Ra, interleukin‐1 receptor antagonist; MMP‐3, matrix metalloprotease‐3; VAS, visual analog scale. John Wiley & Sons, Ltd. ###### PD parameters from E~max~ models for clinical and biological efficacy variables after 1 month of etanercept treatment in patients with RA **E~max~ (%)** **EC~50~ (ng/mL)** **Sigmoidicity factor (ς)** ---------------- ---------------- -------------------- ----------------------------- Swollen joints 56.97 567 1.33 Painful joints 66.82 573 1.67 ESR 46.00 465 0.48 IL‐6 28.11 87 3.13 MMP‐3 27.88 159 1.39 ESR, erythrocyte sedimentation rate; IL‐6, interleukin‐6; MMP‐3, matrix metalloprotease‐3. John Wiley & Sons, Ltd. {#cts12502-fig-0002} DISCUSSION {#cts12502-sec-0130} ========== The goal of this article was to present this previously unpublished study that includes two important changes in the guidelines for the development of medications to treat RA: i) make early assessments of clinical response; ii) to use PK/PD assessments to plan further studies. The analyses conducted may not reflect all of the advances that have occurred in the 20 years since this study was conducted, but they are the ones that were used at the time and were used to inform subsequent work. Although conditions were not optimal for assessment of PK parameters, due to the small number of observations and the approximate timing of collection with respect to dose, a reasonable approximation was made. The PK observations in this study were similar to those reported elsewhere.[11](#cts12502-bib-0011){ref-type="ref"} Exposure was proportional to dose received. In this study, the mean C~ss~ was 1,170 ng/mL in patients receiving 50 mg once‐weekly, which is consistent with the AUC calculation of 143,600 ng·h/mL after a 25‐mg dose (∼965 ng/mL) that was reported in a study of 25 patients with RA who received 25 mg twice‐weekly for 6 months and had PK assessed at the beginning and end of the study.[11](#cts12502-bib-0011){ref-type="ref"} In an early study of patients with RA treated with etanercept by both intravenous (i.v.) and subcutaneous (s.c.) administration, Moreland *et al*.[5](#cts12502-bib-0005){ref-type="ref"} reported they were unable to discriminate between the doses administered. Their study was similar in design to the current study; however, it included only 22 patients, of whom 18 received active drug. The administration of i.v. doses prior to starting s.c. twice weekly doses may have obscured the difference between dose groups. Additionally, the range of doses was lower than in the current study, with only the highest dose group (16 mg/m^2^/week (∼50 mg/week)) in the range of what has been shown to be effective. Although etanercept concentrations were collected during that study, they were not used for PK/PD analysis. The current study utilized all of the available data from all patients, even those receiving placebo. Lacking enough data to support a population PK/PD analysis, an E~max~ model was empirically applied to the broad range of observations collected in the small group of patients studied. An effort to weight values somewhat evenly across the range was made by permitting uneven interval lengths that included similar numbers of observations. This study showed that after 4 weeks, further improvement in PD measures was not observed, as concentrations increased above ∼2,000 ng/mL. Studies of longer duration might be expected to have different results. However, after 1 month of treatment, the EC~50~ values for swollen and painful joints and ESR were similar (range 465--573 ng/mL), suggesting that doses must achieve at least these concentrations. Thus the 10 mg once‐weekly dose, which resulted in a C~ss~ of 300 ± 135 ng/mL (**Table** [3](#cts12502-tbl-0003){ref-type="table-wrap"}), would not be effective in most patients, and a 50 mg twice‐weekly dose would be excessive; observations that were confirmed in the longer studies by Moreland *et al*.[7](#cts12502-bib-0007){ref-type="ref"}, [8](#cts12502-bib-0008){ref-type="ref"} and Johnsen *et al*.[10](#cts12502-bib-0010){ref-type="ref"} The approved dosage for the treatment of RA is 50 mg per week administered in a single 50‐mg dose or as divided doses of 25 mg twice‐weekly; therefore, the results of the optimal concentration range in this dose‐finding study are consistent with what was ultimately shown to be the effective dose. C~ss~ was the PK parameter used in this PK/PD analysis. Other parameters, such as C~max~ or AUC, were not explored and may represent a limitation. The limitation, particularly in a short study such as the current study, is mitigated by the observation that etanercept exhibits linear PK and so C~max~ and AUC would be expected to vary in direct relation with C~ss~.[11](#cts12502-bib-0011){ref-type="ref"} An exposure--effect relationship conducted by Lee *et al*., which explored the probability of achieving ACR20 after 6 months of treatment in 182 patients with RA, used cumulative AUC, calculated from the first dose, as the measure of etanercept exposure.[12](#cts12502-bib-0012){ref-type="ref"} These investigators suggested that cumulative AUC was a better measure of exposure than C~ss~ because it better reflected the entire 6‐month drug experience rather than the more limited C~ss~ observed at 6 months. However, trough concentrations have successfully been used by other investigators to evaluate clinical response, as will be discussed further below. The duration of this study was short compared with most PK/PD studies. However, on the basis of better dose discrimination, the revised US guideline encourages assessment of treatment in the 2‐ to 8‐week window after treatment initiation, as early responses point to more effective therapy.[4](#cts12502-bib-0004){ref-type="ref"} At the end of 4 weeks of treatment, it was reasonable to assume that steady‐state etanercept concentrations had been achieved and that patients would have achieved some measure of relief of their symptoms. Patients in other dose‐finding studies reported improvement in their symptoms as early as 2 weeks after starting twice‐weekly treatment.[7](#cts12502-bib-0007){ref-type="ref"}, [8](#cts12502-bib-0008){ref-type="ref"} The range of doses and concentrations for studies of shorter duration, such as the current study, can be broader than is possible for many conventional clinical studies because patients receiving inadequate treatment in those studies will tend to discontinue the study drug and/or safety concerns may be higher for studies of longer duration. This study and discussion focused primarily on the translation of initial PK/PD information into a dosing range that would be reasonable to evaluate in larger phase III clinical trials. It was performed across a wide range of doses, in patients who were treated for only a short period of time, and was able to obtain PD assessments for painful and swollen joints, ESR, IL‐6, and MMP‐3. Thus, this study design may offer useful information prior to the larger and longer studies that are necessary, given the relatively large number of patients suffering from the disease and the duration of their expected treatment. The dose‐finding studies conducted during etanercept development as well as the PK/PD analysis conducted in the current study supported the use of 50 mg weekly and concluded that further dose escalation was unlikely to be helpful; however, patients differ in both their PK parameters and their response to treatment. Early dose‐finding studies were largely based on clinical findings, but further investigation of PK may be warranted. Two more recent studies in TNF‐inhibitor--naïve patients with RA treated with etanercept 50 mg weekly (as either a single dose or two divided doses) reported that higher serum concentrations are associated with better clinical outcomes. In a study of 19 patients with active RA reported by Daïen *et al*., 3‐month etanercept concentrations were found to be lower in six nonresponders (median 1,750 ng/mL) than in 13 responders (median 3,700 ng/mL, *P* = 0.03) after 6 months of treatment and correlated significantly with change in disease activity score in 28 joints between baseline and 6 months (*r* = --0.62, *P* = 0.006).[13](#cts12502-bib-0013){ref-type="ref"} Although all of the patients in the study were treated with 50 mg weekly and there was no change in the dose administered, the authors suggested that the absence of an optimal clinical response to etanercept may have been due to low etanercept concentrations and they postulated that patients with low etanercept levels may benefit from either an increased treatment dose or earlier change of treatment. The concentrations observed by these investigators are higher on average than those observed in the current study, and may reflect differences in administration technique, compliance, sampling time, and assay. There may be other differences between the patients that could explain their differences in response, such as differing from each other in their disease activity, but considering increasing doses rather than changing medications may be a reasonable choice for some patients. Results from a study by Jamnitski *et al*. in 292 patients with RA also demonstrated that lower etanercept levels were associated with a lack of response to therapy.[14](#cts12502-bib-0014){ref-type="ref"} Among patients who were treated with 50 mg weekly, administered in either one or two doses, etanercept levels were significantly higher (*P* \< 0.05) in patients with good European League Against Rheumatism (EULAR) responses (median 3,780 ng/mL; interquartile range (IQR) 2,530−5,170 ng/mL) than in those with moderate EULAR responses (median 3,100 ng/mL; IQR 2,120−4,470 ng/mL) or nonresponders (median 2,800 ng/mL; IQR 1,270−3,930 ng/mL) following 6 months of therapy. The concentration ranges overlap and were higher than observed for the same dose in the current study. The authors suggested that therapeutic drug monitoring and adjusted dosing regimens might be needed in certain groups of patients to obtain optimal response. Extrapolating from the results of the Daïen *et al*.[13](#cts12502-bib-0013){ref-type="ref"} and Jamnitski *et al*.[14](#cts12502-bib-0014){ref-type="ref"} studies, the relationship between etanercept concentrations and effect needs additional exploration. A study demonstrating improvement in clinical response with changes in dosing and etanercept concentration would fill a gap in our current knowledge, as would better understanding of individual patient characteristics associated with less than optimal response. The target range of etanercept concentrations may indeed be higher than anticipated from the current study for some populations of patients. Although an exhaustive review of the many studies on drugs in development for the treatment of RA is outside the scope of this study; there are other examples of small (fewer than 100 patients), short (no longer than 1 month) studies with a wide range of doses (greater than fivefold) that include PK assessment to permit further exploration of response information collected.[15](#cts12502-bib-0015){ref-type="ref"} This approach, a small, parallel‐dose study of patients with active disease of very short duration, may also be appropriate in other disease areas. In conclusion, sigmoidal E~max~ models were successfully applied to mean PD observations and steady‐state etanercept concentration data collected from a small number of patients with RA administered doses that ranged 10‐fold, from 10 mg once‐weekly to 50 mg twice‐weekly. The EC~50~ values suggest that doses ranging between 10 mg twice‐weekly and 50 mg once‐weekly would provide concentrations associated with clinical response; however, doses outside the range would either be ineffective or would add no additional benefit. This study was funded by Wyeth, which was acquired by Pfizer in October 2009: study number 0881A1100 ‐ B1801312 (trial conducted prior to ClinicalTrials.gov registration requirements). The authors thank all patients who participated in this study and the medical staff of all participating centers. Medical writing support was provided by Rina Vekaria Passmore of Engage Scientific Solutions and was funded by Pfizer. J.K‐B., F.B., H.J., and K.P. wrote the article; J.K‐B. and F.B. designed the research; J.K‐B. and F.B. performed the research; J.K‐B., F.B., H.J., and K.P. analyzed the data. All authors had support from Pfizer for the submitted work; F.C.B. was the principal investigator of the study supported by Wyeth, which was acquired by Pfizer; J.K.B. and H.J. are currently employees of Pfizer and K.P. is a former employee of Pfizer.

Since the first report in 1954 outlining the *in vivo* use of a double-beam spectrophotometer to measure the absorption of near-infrared light by various light absorbing pigments (or chromophores) involved in the mitochondrial respiratory chain (Chance, [@B5]), substantive advancements have been made in the use of near-infrared spectroscopy (NIRS) to monitor the oxidative state of various human tissues. Indeed, when Chance first reported the measurement of cytochrome c oxidase in yeast cells (Chance, [@B5]), he likely had little inclination that this would one day lead to the development of the modern day cerebral oximeter, an increasingly utilized monitor aimed at optimizing patient outcomes in a range of surgical settings. However, it was almost 25 years later that Jöbsis published a series of investigations in animals (and in human volunteers) demonstrating that blood flow related changes in brain oxygenation could be monitored non-invasively. These studies established that NIRS could be used to monitor regional brain oxygen saturation (rSO~2~) in a potentially clinically useful manner (Jobsis, [@B15]). However, as is the case with many clinical developments, it took decades of further advancements before the first cerebral oximeter was approved by the United States Food and Drug Administration (Widman, [@B23]). Indeed, now more than 20 years since its first approval (and 60 years after the seminal work by Chance), we are arguably at a cross roads in the further development and understanding of how to use cerebral oximetry in clinical practice. Although substantial potential exists for NIRS-associated advancements in patient monitoring (and related improvements in outcome) in a wide variety of clinical situations, there are a number of uncertainties that could prevent these devices from reaching their full potential. Although there are hundreds of reports outlining numerous plausible uses for cerebral oximetry, notably in the operating room and intensive care unit (Grocott et al., [@B12]; Ghosh et al., [@B9]; Zheng et al., [@B24]), we are still years from having large scale, randomized controlled clinical trials definitively examining the full potential clinical benefit of this device. The results of smaller trials, if they are to be confidently believed, need to be replicated and corroborated on a much larger scale. Indeed, there are a number of reasons for the considerable uncertainty regarding the capability of cerebral oximeters, despite substantive promise. Inter-device variability, variability in oxygen saturation targets and thresholds, issues related to absolute vs. relative saturation changes, along with studies limited by their observational nature, have all contributed to this current lack of confidence. Understanding the true clinical value of this device is further obscured by wide inter-study variability. With respect to the variability in technology amongst the individual devices, it is not entirely certain whether information gained from one manufacturer\'s device can be appropriately compared to that of another. The differences in technology, including the choice of light source (i.e., laser vs. light emitting diodes) and specific wavelengths, the various proprietary absorption and processing algorithms, as well as the validation sequences used, all contribute to questions of their broader comparability (Ghosh et al., [@B9]). Variability might also exist with respect to how the devices account for differences in skin pigmentation (Bickler et al., [@B2]). Furthermore, the manner in which these devices handle issues related to spatial resolution could have the largest impact on whether rSO~2~ measurements are affected by extracranial contamination (Davie and Grocott, [@B6]). That is, how much non-cerebral saturation signal from the superficial tissues, contaminating the deeper cerebral saturation signal, does each of these devices contain? Indeed, we recently determined, using an experimental design that allowed the separation of scalp oxygen saturation from that of cerebral oxygen saturation, that the scalp contamination can lead to as much as a 17% change in the NIRS signal. Importantly, newer generation devices were much less prone to the contamination. Comparative studies using more than one technology will be needed to determine if the differences in technology are clinically relevant. Optimizing outcome is predicated on understanding what brain oxygen saturation thresholds are worrisome enough to clinically warrant intervention. Reports thus far have substantive inconsistencies as to what the optimal cerebral saturation target should be---which leads us to the question: what is the desaturation threshold, that when crossed, increases the risk of adverse outcomes? (Murkin et al., [@B19]; Hemmerling et al., [@B13]; Kazan et al., [@B16]; Fischer et al., [@B8]; Heringlake et al., [@B14]; Tang et al., [@B22]). Specifically, is there an *absolute* saturation threshold at which there is an increased risk of adverse clinical outcomes, such has been demonstrated with jugular bulb desaturation in carotid endarterectomy surgery (Moritz et al., [@B17]), or is there a *relative* change (from a pre-determined baseline saturation) that is more important to signal impending danger? Furthermore, if it is a relative change, how much of a relative change? Defining desaturation is further complicated by the influence of oxygen supplementation on relative changes in cerebral oxygenation. For example, breathing 100% oxygen can elevate baseline rSO~2~ and artificially widen the difference between baseline rSO~2~ and the lowest rSO~2~ value during a surgical procedure (Bussieres et al., [@B4]). Finally, it is uncertain as to whether there is a "dose response" with respect to cerebral desaturation. That is, does a brief but substantial desaturation portend a worse outcome compared to a lesser degree of desaturation occurring for a more prolonged period of time? Early data examining the relationship between the area under the curve for various rSO~2~ thresholds suggests the latter may be the case (Fischer et al., [@B8]). Whether it is an absolute or relative change in rSO~2~ that is important will only be determined by analyzing both variables in adequately sized observational trials (see below). The interventional algorithms used to optimize saturation are also incompletely described with respect to which of the commonly used interventions (FiO~2~, PaCO~2~, blood pressure, cardiac output, hemoglobin) are considered best to optimize rSO~2~ (Denault et al., [@B7]). Each one of these interventions could conceivably improve outcome, or make it worse. Increasing the PaCO~2~, blood pressure and cardiac output could increase cerebral blood flow while augmenting the FiO~2~ and hemoglobin (with transfusion) could also increase the oxygen content. Combined, this would increase overall cerebral oxygen delivery. Conversely, there is the potential for substantive morbidity related to inappropriate transfusion, targeted hypercapnia and/or hypertension. For example, erroneously increasing the PaCO~2~ above the normal physiologic level in the setting of cardiopulmonary bypass (CPB) could lead to increases of cerebral blood flow in excess of what is needed for cerebral metabolism, with a subsequent increased delivery of CPB-related air bubbles and particulate debris. This is similar to the use of pH-stat blood gas management, where the addition of CO~2~ to the CPB fresh gas flow (thus correcting the CO~2~ for hypothermic temperature) has been associated with worse postoperative cognitive outcomes (Patel et al., [@B21]). Ultimately, the safety of various interventional algorithms will be borne out from prospective randomized controlled trials of large size and multi-center nature. Arguably, one of the greatest limitations thus far in our understanding of cerebral oximetry has been the failure to confidently delineate which clinical outcomes have relevance to rSO~2~ measurements. That is, one would intuitively expect that because it is a direct cerebral oximeter, neurologic outcomes themselves would have the best relationship to brain oxygen saturation. However, a cogent argument can easily be made regarding the biologic plausibility that non-cerebral outcomes could be better related to measurements of cerebral saturation. Murkin has logically argued that the brain serves as an "index organ" for other tissue saturation (Murkin, [@B18]). However, the brain\'s inherent protective mechanisms (such as metabolic and pressure related autoregulation) suggest that if brain desaturation does occur, it is likely that other tissues have long since desaturated (Boston et al., [@B3]; Grocott, [@B10]). That is, the brain in effect is the last organ to be compromised, and in some respects is not an early warning system such as a canary in the coalmine (Grocott, [@B11]), but just the opposite (i.e., a late indicator of trouble). Determining which endpoints to target in outcome studies employing cerebral oximetry-guided intervention could be the largest roadblock to the clinical progress of cerebral oximetry. Our understanding of how oximetry could impact patient outcomes has been limited by the variable endpoints outlined in multiple, but mostly small, observational studies. These outcomes have been so diverse that there is considerable uncertainty as to which ones could confidently be the focus in larger trials. We would argue that there is still a desperate need for a large prospective observational trial involving at least 1000 patients, if not more (an arbitrary number, but seemingly large enough study), to determine which outcomes may be related to desaturation. Only then would it be possible to design a randomized controlled trial to investigate the ability of an interventional algorithm, aimed at restoring cerebral saturation, to modify the previously determined target outcomes. This could also serve to delineate a potential saturation threshold to maintain (such as a relative or absolute change). A good example of this is the recent work that we performed in a small observational study (*n* = 109) in cardiac surgery (Arenson et al., [@B1]). Firstly, we demonstrated that a somewhat conservative threshold (of an absolute saturation of less than 50%) showed a far stronger relationship to adverse outcomes \[such as acute kidney injury (AKI)\] than any of the relative changes from baseline. Furthermore, and as previously hypothesized, it was the non-neurologic outcomes that had substantive relationship to desaturation, with renal dysfunction demonstrating the strongest relationship. This is consistent with other recent work demonstrating that when patients were managed at low blood pressures (i.e., those below cerebral autoregulatory thresholds as defined by cerebral oximetry), they had an increased incidence of AKI (Ono et al., [@B20]). Thus, the field of cerebral oximetry research is still relatively young and contains as many questions as it does certainty. In order to move forward in an expedited and confident fashion, taking advantage of the last 60 years of research in cerebral NIRS, we will need to reconsider our approach regarding further clinical development. Otherwise, we risk wasting this potentially valuable technology, and most importantly, depriving our patients the benefit of improved clinical outcomes with its universal use. [^1]: This article was submitted to Integrative Physiology, a section of the journal Frontiers in Physiology [^2]: Edited by: Patrice Brassard, Laval University, Canada

1. Introduction {#s0005} =============== Marketing and management research and marketing and management practice have been acknowledged as losing touch with each other ([@bb0170]; [@bb0215]). Universities reward academics for publications while the broader industry domain questions the benefits of such narrow pursuits. Such is the frustration of research outcomes coming at the expense of impact and relevance ([@bb0215]) that many may be satisfied if academics did publish *and* perish. Gimmicky marketing campaigns like being in "the top 50 youngest universities" potentially alienate more discerning industry stakeholders. While challenging the value of universities internationally is not a new concern, few could deny a growing intensity around business schools having 'lost their way' ([@bb0030]), with ongoing tensions around academic rigor versus professional relevance ([@bb0160]). Management academics generating superfluous theories missing societal and professional challenges adds frustratingly to a global phenomenon of constant questioning of the value of universities. This overall relevance (or lack thereof) of universities and their business and management schools satisfies [@bb0210] definition of a 'perplexing situation' -- a 'principled disagreement about what counts.' There are several alternative conceptions of what management research versus management educators and society see as valuable ([@bb0035]). Adding to this value perplexity are anecdotes of prominent businesspeople dropping out of university. Jobs, Gates and Zuckerberg are common examples. Overlooked is that the latter two made it to Harvard, and all three were extraordinarily capable. For all the anecdotes, there are multiple counterpoints and interesting stories about the value derived from education. Steve Jobs made a conscious decision to save his adoptee parents the educational costs of completing his degree because he could not see any real benefit. However, he still attended courses like calligraphy which he noted later as serendipitously fundamental in Apple\'s success. Ray Dalio of Bridgewater fame in *Principles* ([@bb0050], p.ix) discusses his own academic trials and tribulations: "I\'m a 'dumb shit'" is how he puts it. His transformation came via a growing interest at college in stocks. Eventually he made it to Harvard and now advocates for case studies, the benefits of a prestigious alumni, and continuous learning. Jack Ma of Alibaba fame never deviated from his early passion for education. It may come as a surprise to some that Ma sat the Chinese university entrance exam multiple times before acceptance into teacher training. This revelation is matched only by the 2018 shock announcement that he intended to step down at Alibaba to return to teaching. Warren Buffet\'s Harvard rejection led fortuitously to an opportunity to study under Benjamin Graham at Columbia Business School. Graham\'s book *The Intelligent Investor* was crucial in Buffet\'s development with Graham going on to employ his prodigy. Academics like Graham are often under-rated, but he provides just one example of the value of effective teaching and diffusion of research. With total student debt now estimated in the US to be beyond US\$1.5 trillion it is understandable why stories of dropping out resonate, with the opposite getting little attention. Bok\'s Law (former President of Harvard) is worth remembering: "If you think education is expensive --- try ignorance." The OECD *Education at a Glance 2018* report highlights that technological change will exacerbate the difference between higher-educated "haves and have-nots" ([@bb0175], p. 11), 'Those who have attained only upper secondary education will earn 65% as much as a tertiary graduate, on average.' The broader societal value of education turns the attention to the real focus of this paper: addressing a shrinking social-research-practice commons. We accordingly believe it\'s time the impact of *teaching* and *learning* ([@bb0260]) was even more vigorously added to the debate about research versus practice ([@bb0215]). In doing so we challenge disciplines like marketing and management to increasingly "practise what they preach". Public commentary on the marketing academy now challenges not only the research relevance but the value of the discipline more generally ([@bb0130]). Too often we preach accelerated change and disruption but are conservative in responding to such trends. In an age of Twitter, blogs, podcasts, sound bites and Ted talks, a "stand and deliver" andragogy no longer suffices. In continually discussing whether research conforms to science i.e. 'pure basic', 'user-inspired', 'tinkering' or 'pure applied' ([@bb0215], p.291), we are often falling into our own potential production-oriented myopia. Timely, better, and more contemporary results are what our *consumers* want. Many urge early career academics to publish or perish and have done so since the first use of this term in *The Academic Man: A Study in the Sociology of a Profession* in 1942 ([@bb0165]). However, the education market demands much more. Teaching, executive training, book and blog publishing may be low priorities in academia compared to peer-reviewed papers, but are effective ways for researchers to gain practitioner and societal relevance. Our contemporary knowledge consumers see rigor as an order-qualifier while relevance is their order-winner. Deeper understanding of rigor *and* relevance is key here ([@bb0255]; [@bb0270]), as is research-related academic-practitioner engaged scholarship ([@bb0265]). A key proposition of this paper is that research alone overlooks an array of interactive educational opportunities in our armoury, and that innovative teaching practice, processes and programs can be used to help bridge the research-practice gap. This paper outlines the causes of the shrinking commons and likely consequences of leaving this unchecked, before discussing potential solutions with reference to contemporary Australian cases. These cases outline interesting variants that tap into an educational sweet spot: *teaching-practice*, highlighting that teaching---leading to the ultimate goal of learning---is an important way of realigning stakeholder interests and strengthening the commons ([@bb0035]). Extending [@bb0260] intervention in executive education, we conclude that *teaching-practice* (whether undergraduate, post-graduate, MBA or executive level) has important intrinsic and extrinsic value for improving *research-practice*. We suggest a-priori that our professionally related disciplines of marketing and management need to acknowledge that teaching is far from an *add-on* to a researcher\'s cause. 2. University and practice: descent or ascent with modification {#s0010} =============================================================== Identifying that we are all part of life\'s descent with modification is something for which we can thank Charles Darwin. Existentially many of us are hoping to leave a greater legacy than descending simply into this adaptation pathway. Fortunately, academics share a privileged position where passing on knowledge and extending a legacy is possible. Potentially academics, as researchers and teachers, are therefore offered a different type of advantage: ascent with modification. In the increasing hurly-burly of publishing, teaching and service, it is easy to forget the academic vocation offers an important chance to add to the stock of ideas. There have been few better times for sharing knowledge than the current era of servitisation, digitisation, computerisation, virtual worlds and AI. Inevitably the role of universities and academics is changing and for traditionalists, not all will be to their liking. Concerns about *what ought to be* largely go beyond the scope here. However, in fields like management and marketing a truism remains: we must be relevant to our respective professions ([@bb0030]). Avoiding such responsibilities ([@bb0130]) is not a solution as [@bb0190] identified in the *Journal of Marketing*, '...it is our responsibility to work on relevant problems, make a difference, and push for institutional changes.' The aphorism often misattributed to Lewin, \"there is nothing more practical than good theory\" ([@bb0025]) resonates here. Irrespective of the misattribution, Lewin was a great action researcher who truly understood the value offered by combining practice and theory. For management and marketing theory to ascend---to improve, advance and remain useful---we need to have authentic and embedded dialogue and more effective communication between stakeholders. The last few decades have seen the opposite with management education and research caught up in a rigor v. relevance debate and false premise that to 'gain more of one, we must lose some of the other, in an ongoing zero-sum game' ([@bb0110], p. 777). 2.1. A shrinking commons {#s0015} ------------------------ Arguments about whether business schools should focus on pure research as opposed to writing texts and teaching are not new. Germany debated such elements in the 18th and 19th centuries, with textbooks often regarded as inferior by-products aimed at codifying and simplifying theoretical concepts ([@bb0290]). Research rigor versus practice became hotly contested in the US around the mid twentieth century ([@bb0110]) with 'physics envy' ([@bb0225]) finding its way into Western business schools. If business schools wanted to be taken seriously academically, it was thought that they needed to reduce practicality for more purist pursuits. The challenge of satisfying a "commons" between the need for (a) developing rigorous research, while (b) helping business and practice, but also (c) fulfilling societal needs, was emerging. [Fig. 1](#f0005){ref-type="fig"} illustrates the ideal where social, researcher and practice stakeholders collaborate and balance interests to advance the commons. It provides an adaptation of the 'triple helix' parameters of university-industry-government put forward by [@bb0080]. Social stakeholders include the work of governments that fund universities (main providers of research and tertiary education in Australia), regional communities with a stake in general economic prosperity, taxpayers and employees. Research stakeholders include all activities conducted by tertiary institutions including research and higher education. Practice stakeholders include all activities undertaken by management practitioners, namely deploying and managing assets and resources on behalf of corporate entities both large and small.Fig. 1Stakeholders in a management research and education commons.Fig. 1 Each domain represents a complex network of actors joined by national interest: social stakeholders have a stake in how both research and practice advance social good; research has a stake in social stakeholders as subjects of inquiry and as funders of their efforts; practitioners have a stake in social stakeholders as markets (consumers), as resources (employees), as regulators (government), and as arbiters of corporate conduct (public interest). Although the commons is broader than an Industrial Marketing setting, the intersection aligns with the marketing configuration of co-creation of value ([@bb0275]). We support the [@bb0105] view of *value* in the commons (improvement i.e. closer circles mean users are better off). The commons is different from firm-customer relationships and we also agree ([@bb0105], p.290) 'From a value creation point of view, the fact that interactions do not include two parallel processes but one merged coordinated interactive process is key.' The researcher circle or sphere, like a firm, is responsible for sharing in the *creation*. A fully functioning triple or quadruple helix (with the added sphere of community) sees complementary interactions between stakeholders leading to a temporal expansion and strengthening of shared components ([@bb0145]). In management education at present, the reverse seems to be happening. Parties pursue divergent goals and the common ground is shrinking ([@bb0020]; [@bb0100]). The co-creation of value between research and practice is ironically often neglected by marketing and management academics. [Fig. 1](#f0005){ref-type="fig"} implies that research incorporates teaching, yet unfortunately instruction is often treated as a chore, whereby academics are trading-off between research and the business schools\' lower priority of teaching ([@bb0010]; [@bb0030]). None of this is helping the management research and education commons. What has largely been lost in this often US dominated debate is that stakeholders are not uniform. Each country has its own unique characteristics, impacting how each country\'s commons operates. The embeddedness between university and industry in Germany, for example, with institutes such as Fraunhofer ([@bb0015]) has placed a greater emphasis on engagement over international university rankings. Cambridge in the UK, and MIT in the US have vibrant research-teaching-practice cultures with embedded commercial institutions built into their ecosystems. China has weathered a Cultural Revolution and reveres further education. Their research-education commons is currently more open with businesses often spinning out of university ([@bb0150]), an interesting by-product of more relaxed IP laws and fewer university-commercialisation caveats. Australia regularly ranks highly among university systems for publications (ranking 10th in the 2017--18 Global Competitiveness Index) but has a woeful record for translating research into practice. In terms of university-industry collaboration in R&D, it ranks 31st, placed below Indonesia, Tajikistan and Kenya. In this commons discussion it therefore is important not to see social-research-practice as an international "one-size-fits-all". Denigration of researchers and teachers in western cultures has become a common pastime, irrespective of global rankings. The teaching-only path has been tried with moderate success but is tantamount to oblivion for aspiring long-term academics ([@bb0215]). Nevertheless, like [@bb0110] we are hoping for a middle ground, avoiding trade-offs between relevance versus rigor and critically *teacher* versus *researcher*. Gulati discusses the academic 'boundary spanner' in helping bridge such 'tribalism'. Supporting [@bb0260] we see interactive education as an underestimated, but fundamental process for diffusing and disseminating more of our most relevant research outcomes. Improving incentives to help with such causes have been spasmodic. While the research-education-practice gap remains and indeed grows, the problem for a management research and education commons exacerbates. Stakeholders in the commons may be surprised by the size of investment in *academics-as-researchers* over *academics-as-teachers*. Funding for many universities emanates from student numbers but rewarding academics for publications dominates, dependent on each university\'s and country\'s policies. Popular writer Malcolm Gladwell has been scathing of wealthy universities in the US seeking philanthropic endowments while struggling universities rely solely on public funding. Finding common ground and shared value, linking the three stakeholders (social-research-practice) in an ideal system, seems increasingly difficult. No-one appears to be winning. Signalling to job markets that you have a degree has been described as a lottery ([@bb0205]) with graduand employment failing to align with skills sought by employers. In many countries, students pay high fees for the privilege of tertiary education, so little wonder there is dissatisfaction with outcomes. This more compromised commons process, with divergent pressures drawing away from shared value (worse off and declining co-creation), is illustrated in [Fig. 2](#f0010){ref-type="fig"} .Fig. 2Divergent pressures shrink the commons.Fig. 2 Social stakeholders in this shrinking commons complain, like others, that research is not relevant and also that practitioners and researchers need to place more emphasis on community and social goals beyond economic ones. Social stakeholders are cognisant of increased complexity and are looking to corporate practitioners to share the burden through taxation and pro-social actions, and to researchers for better solutions to such *wicked* issues. The perplexing situation and commons dilemma does align well to the criteria of wicked ([@bb0005]). Researchers complain that social stakeholders are not investing sufficiently in research and development; at the same time, they complain that practitioners are satisfied with quick-fix fads and ignore fundamentally sound theories. Institutions compete for global prestige and resources using their prominence in elite journals as their main quantum, with a limited and closed community as principal audience and gatekeepers. As a result, what is produced offers diminishing relevance to social stakeholders and practitioners. For their part, practitioners complain that social stakeholders over-regulate the domain with researchers providing theoretical answers, out of touch with contemporary exigencies. Clearly few parties are winning in what seems a race to the bottom. All parties are justified in their concerns that a shock to the existing system (loss of international student funds, reductions in government support, further decline in confidence in the sector) is likely to create a tipping point with all commons stakeholders losing out. The coronavirus COVID-19 offers one such example with Australia particularly adversely impacted due to disproportionately high Chinese student numbers. Open innovation ([@bb0125]) is an important concept relating to a shared commons. Supplementing organisational ideas through research and development exchanges whether outside-in (OI), inside-out (IO) or coupled (OI and IO) makes increasing sense. However, again failing to *practise* what they *preach*, universities and researchers have been laggards themselves in harnessing the benefits of more open innovation. The level of siloing within and across universities is embarrassing and not new. Redressing the institutional and cultural logics that have developed over decades is complex, requiring a trans-disciplinary approach. Ultimately for a management research and education commons to thrive, we need to find and nurture more common ground. Apart from [@bb0260] exposé of executive education, the role of teaching seems too-often missing from such discussions. We suggest it is time to rethink the role of teaching in order to redress the shrinking commons. 2.2. Reclaiming the common ground {#s0020} --------------------------------- The research versus practice debate barely mentions education and teaching. In [@bb0040] and [@bb0215], the word 'education' appears only in the bibliography, in journal titles. [@bb0020] only mention 'education' five times. It is interesting that researchers and educators appear to abdicate responsibility for knowledge propagation. The role of educators to profess has meant different things in different settings. However, when faculty members think that research and ensuing publications are the only route to passing on ideas and knowledge, then a new Dark Age may have dawned. Little wonder practitioners have largely turned their backs on higher-order, more abstract research and are only interested in skills. In a report on the future of education by Australia\'s peak business lobby group, the Business Council of Australia ([@bb0095]), the word 'skills' appears 68 times (omitting page headers and the proper names of specific skill development programs) while the word 'research' appears only three times. Practitioners no longer believe in it. As an example, one of the authors of this study was instructed in a government project not to mention the word 'research'; it was taboo. Practice and social stakeholders increasingly juxtapose problems in the research domain with poor teaching quality and outcomes. Indeed teaching has its own perceived failings, exacerbated by growing expectations of universities as vocational training centres. The authors of this paper importantly have no bias against deriving *knowledge-through-theory.* We believe basic theory does have value and some concepts, even in business, can be derived through more abstract pursuits. We also concur with increasing calls for utilising the abductive paradigm, and acknowledge that it offers an important form of middle ground ([@bb0045]; [@bb0065]). Proponents of more participatory academic intervention are not new and include Kurt Lewin, Reg Revans and more recently Evert Gummesson. Gummesson\'s standing in marketing theory and advocacy of techniques such as action research has been fundamental in shining a light on the critical nexus between practice, research and teaching ([@bb0115]). He understands that methodologies bringing researchers closer to marketing experience are likely also to better reflect the real "truth". Falsifying social or business-related phenomena is at best transitory with the likes of Popper acknowledging that social sciences are human-centred with people-related systems dynamic and constantly changing ([@bb0230]). [@bb0085] explains the difference between the theoretical researcher and expert practitioner, articulating the role of technical skills. Flyvbjerg draws on the Dreyfus model ([@bb0060]) of skills development incorporating: (1) novices (programmed in basic tasks and operational rules); (2) advance beginners (understand context specific operations needed to function adequately); (3) competent performers (programmed in varying situational aspects but simplify complexity to function most effectively); (4) proficient performers (tackle more advanced positive and negative situational challenges while still growing in competency); and (5) experts (advanced understanding and can sense, seize and adjust at a skill level). One could argue that *master* be added as a sixth level, where experts take on a sage-like capacity to train others, such as Master Black Belts in Six Sigma. [@bb0085], drawing on Aristotle, identifies that this mastery of knowledge in academic and professional pursuits requires phronesis or practical wisdom. The late polymath Peter Drucker is an example of more practical wisdom, rarely relying on logical empiricism. As a more rare counterpoint, Joseph Schumpeter created theoretical accuracy by embedding himself in extant and historical literature. However, equally it could be argued that Schumpeter\'s underperformance as Austrian Finance Minister and in banking was in part due to his lack of previous practical exposure. On that level he was at best a competent performer. The late Clayton Christensen made the combination of good theory and good practice part of his forte, and is one of the exemplars for others to follow. Reclaiming the research centre in [Fig. 1](#f0005){ref-type="fig"} requires researchers and practitioners to agree on a 'composite object' or shared goal that unites interests and acts as a bridge ([@bb0035]). More interactive education and teaching provides an important andragogical-logic in the context of management research and practice that could help find such common ground. Bennis and O\'Toole assert, as described in [Fig. 2](#f0010){ref-type="fig"}, that academics have given up important elements: 'Businesspeople are starting to sense that individuals in the academy are not engaged in the same profession they practice' (2005, p.6). Management education and its knowledge-creating interactions, such as [@bb0265] engaged research scholar, can enable innovative business strategies and sometimes even generate breakthrough theories. For the sake of the commons, and our social, practice and research stakeholders, let\'s hope that there is an opportunity to rekindle such value (a win-win-win). However, getting institutions to change takes effort and time and requires a major rethinking of policies and incentives. There is no doubt that we will need more than chance to secure such positive outcomes. 2.3. A key role for the academic synthesiser and communicator {#s0025} ------------------------------------------------------------- Academics, whether researchers and/or teachers, have always had an important knowledge diffusion and exchange role. Defining what makes a great teacher versus researcher is beyond the scope of the current article, but it is recognised that some individuals are quality teachers while others have strength in research and publications. Some institutions like London Business School and Harvard Business School value both. Ask a past or present marketing student to identify a prominent academic and Philip Kotler is commonly nominated. Kotler influenced marketing theory and practice through teaching and effective text-book communication. Communicating complex research into easily understood interactive prose and visuals is probably easier when your teacher (as in Kotler\'s case) is Paul Samuelson. Undervalued by many academic peers is Kotler\'s impact on transferring complex marketing concepts to the masses. The best texts and academic manuscripts are often derived from a synthesis of higher-order research. Nobel winner Samuelson in economics was a standout. Kotler will no doubt be remembered long after most marketing theorists have been forgotten and is an example of an educator capable of extracting theory that he made useful and more practical. [@bb0295] in *Consilience* suggested the 21st century would be focused on those who can synthesise. Kotler managed this incredibly well (over 315,000 citations on Google Scholar). Drucker, as the father of management, is similarly recognised. His McKinsey Award at the age of 95 for the best article published in *Harvard Business Review* in 2004, was exceptional. Explaining complex theory effectively and simply to students and managers is too often underrated. It\'s why journals like HBR and California Management Review focus increasingly on communication of ideas over complex explanations of detailed empiricism. Teaching is currently experiencing similar change with the advent of MOOCs, block teaching and more online and blended learning. [@bb0285] identified: 'We also ring an alarm bell for educators. Media...and private sector web-based organisations are gaining rapid influence by creating programmes and content that inform and inspire. It will not take long before these forms of media are able to replace what is currently offered by educators.' Some of our more applied universities are now taking up this challenge with executive training direct to the workplace. Single-loop learning (understanding more of the what and how), double-loop learning (understanding more of the how and why) and importantly sharing and transferring triple-loop learning (going more to a metatheoretical why) ([@bb0250]) is aided in such contexts. \"Ah-ha\" theoretical insights often derive from these latter teaching-practice *why* moments yet this contribution often goes unrecognised. 3. Interactive educational management and marketing cases that are up close and personal {#s0030} ======================================================================================== The four co-authors of this paper are researchers and educators. Like their colleagues in management and business, they are mindful of KPIs that are prevalent in the university domain and within business and management. All four have taught at undergraduate and post-graduate levels at the Australian National University (ANU), have been practitioners, consult to industry, are keen to see students succeed and are academically and vocationally driven. Their specialty areas include marketing, entrepreneurship, innovation, leadership and change management. Located in the Australian capital Canberra, ANU is part of a Group of Eight (Go8) leading Australian universities. The four cases illustrated below (Cases A to D) are built around the combined authors\' educational teaching and training praxis. Praxis is an apt description ([@bb0045]) as it emanates from the Greek for a theory or lesson being enacted, and reflects that these four cases are direct lessons from academics engaging and interacting in their field. Most of the cases remain operational with continuous adjustments, refinements and recalibrations. Student evaluation, peer reviews and feedback from practice and social domains inform improvements. Cases A to D represent multiple offerings from undergraduate through to executive and industry educational programs and training. One of the most important attributes of the ANU management school teaching approach is the introduction of evidence-based management (EBM) which puts emphasis on problem/solutions from multiple perspectives including scientific literature, organisations, stakeholders and practitioners. EBM ([@bb0055]) acknowledges that better business decisions require accurate data, facts and triangulation. Such material can be derived from secondary sources, surveys, qualitative material, cases and/or business operations. Ultimately EBM relies on a combination of theory and practice---with [@bb0135] Systems 2 or more reflective thinking (as opposed to Systems 1, more automatic thinking)---illustrative of this approach ([@bb0235]). The four cases described are customised to the ANU management school philosophy but are examples of initiatives increasingly offered by a range of Australian universities in programs encouraging more interactive theory-practice work integrated learning (WIL). The cases are supported by peer reviews and student feedback but for the purposes of this article the material is built on the diaries, memos, observations and reflective discussions of the academics involved ([@bb0145]). Two of the authors have developed programs applying action learning and action research for more participatory approaches with Cases B, C and D offering examples closer to [@bb0260] executive and industry training workshops. This latter case (D) has been subject to participant entry and exit surveys with some of the findings published in *R&D Management*. 3.1. Case A - going live with problem-based learning {#s0035} ---------------------------------------------------- Since 2003, ANU post-graduate management programs have required final-year students to complete, as a capstone project, a live consultancy brief for a local company, focused on an innovative growth initiative and development. By 2018, one author had mentored 83 of these projects for 62 local client organisations. The strategy challenges are posed as unstructured, complex problems requiring participants to deploy and adapt multiple frameworks learned across their entire management program. The projects have yielded rich research insights into the growth strategies of local companies across multiple industries as well as accelerating their development in practice by embedding novel strategic frameworks into the local business ecosystem. Another problem-based interactive example is ANU\'s global marketing course which incorporates an International Business Plan Competition (IBPC). This brings together business, industry and government partners to create export-ready outcomes for local companies (approximately 7 to 10 partnerships per annum). Student teams develop global marketing plans for the respective companies with each finalist pitching their outcomes to an expert panel of industry practitioners. Companies benefit from well-developed internationalisation plans, students gain real-world insights and faculty members gain access to in-situ business activity and research data. The program has two offerings: one at undergraduate and one at post graduate level. The recruitment of companies is undertaken in close collaboration with industry and government stakeholders and the program has been responsible for multiple international export successes. An induction of companies into international marketing theory and process kick-starts the IBPC, conveying expectations and providing recruited SMEs with a snapshot of relevant theory. Tools and modules have been developed to assist students in quickly understanding and applying theory to the live businesses, with standout students selected by the SMEs to enter internships. ANU\'s College of Business and Economics (CBE) operates a growing and extensive internship program with approximately 70 undergraduate and postgraduate students working in businesses each semester. Two of the authors of this paper provide academic supervision, ensuring students apply suitable theoretical elements to their experience. Projects are research related and monitored by academic supervisors for adequate rigor. Successful interns are consistently hired by participating companies. This internship program now supports international students in acquiring positions in their home market. Internships are an important way of assimilating and disseminating research concepts into the field, and are increasingly common in university settings. In 2008, the ANU launched InnovationACT, a network-based program encouraging students, staff, academics and industry mentors to collaborate in local venture initiatives. The program has gradually expanded to include other educational institutions in Canberra. Over the past four years, the program has awarded over AUD\$200,000 in seed funding and other resources. In 2017, InnovationACT\'s online platform was viewed over 30,000 times and a record thirty teams completed the program which includes a series of practice-oriented workshops based on the ANU\'s research into new venture processes ([@bb0185]). In 2014, the ACT Government launched the Canberra Innovation Network (CBRIN) to further consolidate collaborations between local government, innovators and education and research institutions. InnovationACT has built strong ties to CBRIN, the Griffin Accelerator which supports local high-growth firms, and the federal intellectual property agency---IP Australia---linking social, research and practice stakeholders ([@bb0185]). ANU\'s College of Business and Economics has now also piloted a Venture Lab for further integration of entrepreneurial ventures created by students, including the promotion of joint innovation efforts. Student-centred activities linked to design thinking are central to this initiative. The start-up entrepreneurial portfolio is managed and developed by one of the co-authors who is currently undertaking research into entrepreneurial opportunities and design. 3.2. Case B - research-practitioners as program co-creators {#s0040} ----------------------------------------------------------- Two authors have delivered a pilot executive education offering at the ANU in partnership with a professional association of transformational change practitioners. This program has combined academic theory, EBM, and practical expertise in transformational change. The project-based course has built on action-learning approaches to accelerated executive learning ([@bb0240]). The intention from the outset was to use the educational context to go beyond current theory and practice to co-create new knowledge applicable to both domains. The course had initially been organized around three face-to-face modules of four days each, with a final module of two days, delivered over a nine-month period, with time between modules for application and practice. The first module entitled 'Understanding' introduced participants to the current state of research in transformational change. Two external program sponsors then posed wicked transformational challenges, on which participants worked as members of a team, for the entire program. The second module, 'Activation', presented a theoretical lens for interpreting the wicked challenges and focused on scoping the project work. Module 3, 'Implementing and Adjusting', deepened participants\' understanding of transformational change by drawing upon previous insights of transformational leader-practitioners. Dialectic debate through Modules 1 and 2 surfaced conflicting cognitive frames between research and practice. In Module 3, intensifying deadline pressure exposed participants to their assumptions and blind spots, which brought frustrations into the open as they grappled with tensions between theory and practice. As the date for delivery loomed, pressure to articulate a strategy resolved the tension into synthesis. Module 4 concluded the program with presentations of findings to sponsors and a broader audience of practitioners, and sharing of team and individual reflections. Along the way, the ANU and executive-based teams experienced their own version of a "pressure-cooker" challenge. What seems clear and straightforward at the outset creates unexpected tensions, taking learning in new directions. A shared goal is to collaborate on a common educational platform to co-create knowledge and tools to benefit both research and practice through transformational change. Although there is general agreement on overall goals, strategies for achieving these vary significantly between teachers (who prioritise theory and learning assuming that practitioners find theory valuable) and practitioner students (who emphasise tools and solutions, believing academics are unaware of current practice and unable to respond quickly enough to emerging opportunities). Much time in this program has therefore been spent on building a common language. The most useful tools are theoretically based but practically useful. For example, causal loop diagrams based upon systems theory are an abstract, yet applicable lens for understanding complex dynamics associated with transformational change; an EBM framework provides a practical context for methodological rigor in data gathering, critical appraisal and reflection. Participants learn transformational skills while academics find new ways to explain value propositions. A new MBA program has also been launched at ANU with one of the authors responsible for developing an Entrepreneurship and Innovation course for this program. An important component of the program is the inclusion of weekly readings of key journal articles on boundary spanning, innovation champions, ambidexterity, opportunity recognition and bricolage. The major assignment is the development of an innovation plan and strategy for implementation in each MBA student\'s business or government department. Many of the executive participants acknowledge that they are implementing the innovation plans in their workplace. For example, a state innovation award and funding was recently received by one of the participating MBA students based on their respective innovation project. Course objectives are adjusted in early weeks, depending on the skill, knowledge and learning needs of the cohort to ensure students gain significant practical and theoretical benefits. 3.3. Case C - internationalising programs with co-creation across business schools {#s0045} ---------------------------------------------------------------------------------- Two of the authors have been key contributors to a Master of Management (MoM), a jointly accredited program taught in Mandarin at Tsinghua University in Beijing, currently in its 16th year. The authors have taught approximately 60 students per cohort in the final course, New Venture Creation. Participants are senior executives and leaders from various provinces across China. The course focuses on business model and business plan development for new ventures, including potential spinouts and spinoffs. A recent addition to the program has been the incorporation of tools for applying innovation championing to enterprises. The MoM has witnessed many changes in China with enrolling participants now often already successful independent entrepreneurs. In earlier cohorts, most participants came from government or State Owned Enterprises (SOEs). Increasingly, the Chinese students/executives in the program are now often creating ventures pitched at service opportunities and social innovations. The New Venture Creation course has provided important innovation skills and encouraged student teams to create and kick-start real ventures built around team capabilities, analysis of combined resources and clarification of a feasible opportunity. A number of these new ventures have successfully transitioned from the classroom to commercial reality. A key aim is to equip all students with a theoretical and practical capacity to kick-start a venture but there are important additional objectives. Advancing understanding of teamwork, applying knowledge gained from the suite of earlier management courses, and sharing vignettes of existing Chinese entrepreneurs (including those in the class) are other key elements of the course. The opportunity for Australian academics to immerse themselves in Beijing with executives from an array of China\'s leading businesses, including consulting, finance and government enterprises, has obvious two-way benefits. Beijing\'s Zhongguancun district, where Tsinghua is located, is renowned as China\'s Silicon Valley and is a key centre for enterprise growth and innovation. Tsinghua was recently ranked 16th on the QS rankings (2020) which places it close to the top in Asia. Tsinghua cohorts now visit Australia both prior to commencing the MoM and at its completion, and academics have utilised these opportunities to introduce students to Canberra businesses to advance first-hand understanding of Australian businesses and policy environments. One of the authors is now teaching a new course on Innovation to the MoM cohorts and already the results are highlighting interesting additional theory-practice anomolies and insights for those taught (Chinese executives) as well as those teaching (Australian instructors). 3.4. Case D - societal programs that assist place-based co-construction {#s0050} ----------------------------------------------------------------------- One of the authors has had the opportunity to incorporate theory in educational training in regional research and practitioner interventions. These stemmed from a range of quadruple helix research undertakings mostly supported by grants from industry and policy stakeholders. The author has been responsible for several place-based interventions at enterprise, cluster and regional levels and has trained and supported two significant clusters in New South Wales (NSW): Hunternet (170 businesses as cooperative members) and Central Coast Industry Connect (a 200 plus database). Activities with these member bodies have led to training and leadership initiatives, for example the development and implementation of training programs for China, Japan and Korea as part of the development of Australia\'s recent Free Trade Agreements. Additional work by this author with place-based initiatives has extended to federal government grants and initiatives. These opportunities have led to benefits like the development of an Innovation Champions Program (with over 30 participants, approximately 10 per session) and two follow-up regional innovation management (RIM) training programs (North East and North West Tasmania). The latter were supported by Skills Tasmania with several findings showcased at peak academic/industry international conferences. The author supports one of these international conferences (International Society for Professional Innovation Management - ISPIM) with wicked problem training programs (specific for each conference destination) and action research/learning skills. The wicked problem sessions attract over 40 participants (policy, practitioners, academics), with action research/learning sessions limited to approximately 12 participants (practitioners and researchers). The success of the abovementioned quadruple-helix inspired endeavours has led to closer university-industry-government-community linkages. These interventions also have international implications, extending to markets like Germany. The Australian-German initiatives are aimed at advancing regional development and business growth in Australia. A potential programs has been underway in Bendigo, Victoria, with training likely modelled on a combination of successful domestic RIM approaches and regional transition lessons from Fraunhofer Institute in the Kaiserslautern district of Germany. The Bendigo-Fraunhofer program aims to link business, academics, students and communities in what could be described as a potential international cluster of innovation ([@bb0075]). Workshops incorporating industry, policy and regional stakeholders were conducted in 2018 as part of this initiative. Another project that emulates this much deeper exchange between educators and the business environment is an action learning intervention on the Central Coast, NSW. Nine companies were involved in an endeavour to improve industry outcomes by upskilling participants, building family business leadership skills and helping small businesses unlock growth opportunities. Reg Revan\'s action learning process of combining the right mix of formal programmed knowledge (P) with questioning and actions (Q) has been inspirational in these developments. The author\'s role has been to develop and facilitate the initial program (now in its 3rd year) which is an example of [@bb0265] engaged scholarship. Bringing out the thoughts and reflections of industry peers is paramount here. Indeed, these programs go beyond simply researchers crossing the rigor and relevance divide and acknowledge that engaged scholarship is increasingly possible through targeted educative initiatives. The latter cases suggest that academic boundary spanners are becoming harder to distinguish from practitioners. Nevertheless, it is important to highlight that all four authors of this study have always been keen to keep their theoretical toolkits close-at-hand. 4. Encouraging a co-created common ground {#s0055} ========================================= These illustrative interactive educational cases provide narratives of academics crossing the threshold between research and practice: success relies on academics\' business, research and practitioner strengths, combined with advanced educational and facilitation capacity. Common to the first three cases (A, B and C) are the following four characteristics that appear necessary to increase the common ground.1.Each case uses overarching educational goals and processes to integrate mindsets, interests and approaches of social, research and practice stakeholders, bridging and advancing understanding across the three domains of research, practice and social stakeholder while delivering valuable outcomes to each. In so doing, a common basis for value is established ([@bb0210]).2.Relevant processes are multidirectional. They do more than apply established theoretical frameworks derived from management research into the social and practice domains; they integrate the value of practitioners\' and social stakeholders\' experiences as contributors to new knowledge. In doing so, they stimulate researchers and students to generate unique adaptations of theory that stretch beyond established foundations, reframing them into unique client-specific outcomes. In the process, they also challenge social and practice stakeholders to engage with theory in hands-on interactions that deepen their portfolio of mental models.3.They require a core team of research-practitioners willing to experiment across the three domains ([@bb0180]). This is often at personal cost to individual careers because efforts devoted to the goals of 'other' domains, or leading initiatives that bridge domains, attracts little career recognition. But as the value of shared outcomes becomes more visible, bridging these domains is recognised as a solution to the individual pressures each domain faces, rather than as a competing resource demand. Engagement is increasingly regarded as a key process for reigniting relevance, for generating research funds, and for enhancing graduate employability.4.Multidirectional interactions in management education experiments generate knowledge that propagates and extends new insights beyond direct participants to other domain members through network interactions during and after programs. The experience of business innovation ([@bb0280]) suggests that researchers who expose their emerging theories to early testing in an educational context will generate more novel and robust theories faster than those who try perfecting their work in research silos. Over time, these illustrative experiments help catalyse the development of additional change agents across domains in a minor social movement that gradually influences others to shape common ground. Case D is a good example of more extreme testing of theory within the regional field. This has considerable benefits for both educator(s) and beneficiaries and is important in countering the divide between university, business and society. Case D has also initiated multiple research spinoffs and has significantly influenced regional outcomes. Practitioners have participated in such programs as have government field workers, helping diffuse benefits of these interventions more broadly. Regions targeted through the interventions have often underperformed in educational achievement, with many participants actually gaining exposure to higher levels of education for the first time. Cases B and C have similarly high impacts albeit all lessons are conducted within a more contained educational environment. Case A targets undergraduate students transitioning into industry. Theory passed on to them has immediate impacts and longer-term diffusion advantages particularly when lessons around such theory are also incorporated by employing businesses. Graduands who have learned to adapt theories and create unique client-specific frameworks rapidly create networks of advocates across surrounding business ecosystems. Research-practice experiments in management education are thus a highly effective means of disseminating new knowledge more rapidly than via the conventional means of academic publication. 4.1. Identifying the interactive educational value of each stakeholder offering {#s0060} ------------------------------------------------------------------------------- [Table 1](#t0005){ref-type="table"} provides a synthesis and cross-case analysis and lists benefits of these educational interactions. The table reconfigures cases to align educational gain with the level of offering: undergraduates (2nd and 3rd year), postgraduate (minimal work experience), MBA (prior extensive work experience), executive leadership (CEOs), Chinese MoM (executive level), and regional industry training workshops (extensive regional business and industry stakeholder expertise). The table outlines social, practice and research gains from educational interactions, with the final column focused on the shared commons and benefits of academics engaging deeply with external stakeholders.Table 1Gains from educational interaction for various stakeholders.Table 1Domain program typeSocial gainPractice gainResearch gainEducation gainUndergraduate-Increase in social awareness of graduand benefits\ - Increased internships into various networks including not-for-profits\ -Adds considerably to successful stock of start-ups-Improved business performance and export plans\ -Opportunities to mentor and train\ -Stimulus of economic activity through new and creative ideas\ -Interns expose industry to benefits of quality research projects-Dissemination of research ideas to students and businesses\ -Experience of enacting theories in the workplace\ -Identification of potential higher degree research students-Maintain currency of skills needed to service operational businesses\ - Identify gaps between practice needs and what is being taught\ - Inform management colleagues about educational currencyPostgraduate-Australian and international links\ - Broadening of intercultural understanding\ -Funding source for universities\ -Accelerating development by embedding frameworks into local business ecosystem-Exposure to international ideas\ -Stimulus of economic activity through new and creative ideas\ -Links to international markets\ -Well-developed internationalisation plans\ -Businesses can trial interns prior to employment-Dissemination of theory\ -Research opportunities in new markets\ -Research insights into growth strategies of local companies across multiple industries-Understand more fully cross-cultural nuances and benefits of engaging internationals with stakeholders\ - Breeds goodwill and engagement among international partners with marketing and management schoolsMBA-Community advocacy of value of education\ -Disseminates new knowledge to community-Application of new techniques and knowledge\ -Improvement in quality of training and mentorship through experience of train-the-trainer\ -Advances leadership and management literacy and practice-Testing of co-created new knowledge\ -Networks foster collaboration and open innovation\ -Inspires "why" questions and triple loop challenges-Informs educators more directly about participant and agency needs\ -Keeps programs in touch with needs of aspiring executives\ -Sharpens programs and teaching styles to higher-order needsChinese MoM-Links Australia with China\ -Opens options for China community engagement\ -Exposes Chinese to Australian opportunities-Advances Chinese business practices\ -Encourages international partnerships\ -Increases creativity and innovation\ -New venture creation\ -Encourages teamwork-Opportunities for cross-cultural research\ -Potential funding sources for research\ -Access to business settings in China, including China\'s Silicon Valley-Advances skills in different styles of behaviours\ -Opens cultures to varying educational practices, norms and thinking styles\ -Keeps marketing and management programs up-to-date with international business activity and casesExecutive Leadership-Advances leadership quality\ - Advocacy for lifelong learning\ -Positive influences on key decision makers-Improved financial returns and leadership skills\ -Increased ability to manage rapid change\ -Immersion in EBM\ -Complex problem solving through "wicked" methods-Access to research in top companies\ -Alumni networks\ - On-going research grant opportunities\ -Increases currency in research and contributions-Immediate feedback mechanism for educators to industry leaders\ -Keeps educators savvy with higher-order needs of top management teams\ -Adds executive insights to management and marketing schoolsRegional Industry Training-Broaden community exposure to education\ -Adds value to regional systems and clusters\ -Lifts quality of regional thinking-Increased enterprise and regional value\ -Growth for individual enterprises\ -Experience of action learning to improve business outcomes\ -Familiarisation with new frameworks-Opportunities for action research\ -Improved likelihood of successful Linkage and ARC partnership grants\ -Opportunity to test current theory in practice-Ensures needs of regions inform educators, management and marketing schools and universities\ -Builds two-way links and bonds across respective domains 4.2. Understanding education\'s contribution to the commons {#s0065} ----------------------------------------------------------- The illustrative cases and educational praxis demonstrate that co-creating new knowledge with practitioners can result in significant achievements. The contribution that each program type can make to the domains of the commons varies, consequently instruction and teaching processes need to reflect these variations. The needs of participants in an MBA class, and the commensurate value for the commons, will be different to that of an undergraduate program. When training industry partners or international postgraduate students, the context is different again. [@bb0235] notion of the role of a \'choice architect\' is particularly vital here. Educators accordingly need to be cognisant of the choices they are making with what they include/exclude in relation to guiding a cohort. The amount of programmed instruction with higher-order research is dependent on the levels and context, as illustrated in the range of illustrative cases. Alternatively, choices may be made to embed more action-based inquiry to ensure more practical translation is occuring. The benefit for the commons is the architectural "osmosis" which encourages co-creation and a constant "drip-feed" of new knowledge through *learning by doing,* and an appropriate \"dose\" of *programmed instruction* and *research*. This is not limited to executive training but can be facilitated across all domains as the examples within Case A highlight. This paper asserts that *teaching* is an important bridge between research and practice, with benefits across all domains within the commons. As outlined in the Introduction---the more we isolate management researchers toward publishing, and either-or pursuits around rigor and away from social phenomena---the more we threaten the relevance and quality of what management and marketing schools produce. Knowledge is most effectively created through the continuous interaction of ideas, theories, schemas and empirical data in a continual dialogue, not only among researchers but also through engagement with others in the social, communal and corporate domains where human capital interacts. Exclusivity of the peer-review audience toward higher-ranked research publications makes research increasingly inward-looking and inaccessible. For those in business practice, prior experience, rules-of-thumb and intuition provide poor guidance for addressing uncertainty and wicked-style problems. Linking theories to practice through educational exchanges, as detailed in the cases and summarised in [Table 1](#t0005){ref-type="table"}, is a mutually beneficial pathway for management and marketing. One characteristic common to all the cases is a willingness on the part of each of the authors to experiment, applying a spirit of inquiry, in the educational setting. [Fig. 3](#f0015){ref-type="fig"} identifies the gains attributed to education and teaching becoming more aligned to others. The education *sweet-spot* derived from the discussion is where we expect research and practice to increasingly intersect. The calls by Storbacka for more abductive studies and Van de Ven for engaged scholarship are platforms for increased researcher-derived relevance. However, similar to [@bb0260], we believe that the potential and scope for educators to make a difference is understudied. [Fig. 3](#f0015){ref-type="fig"} highlights an important implication for research arms of business schools. To remain relevant to practice and social stakeholders it will be beneficial to offer significantly increased value through effective teaching programs. [Table 1](#t0005){ref-type="table"} highlights benefits of more applied teaching to both practice and social stakeholders. Direct benefits ensue through theory being disseminated in ways respective users *can* and *want* to learn. Educators and their institutions should not be limited by student evaluations but be bolder in encouraging teaching that incorporates more interesting theory-practice pursuits; they have an important role in leading the commons as theoretical plus teaching experts. Applied, relevant, purposeful, action-oriented learning, undertaken in a variety of ways (formal learning for executives and MBAs, internships, Masters level studies, problem-based learning, industry-based competitions) is a clear way to achieve this. Increasing the understanding of social stakeholders will mean better outcomes and less resistance as depicted in [Fig. 3](#f0015){ref-type="fig"}.Fig. 3Co-creating value through teaching and education: growing the commons.Fig. 3 One caveat in the process is the quality of the teacher. While beyond the scope of this paper, it must be recognised that not all researchers are capable of this type of instruction. Drucker highlighted the importance of playing to strengths. Inappropriately pushing researchers beyond their specialisation into training and facilitation roles may not result in ideal outcomes. However, there are ample educators and practitioners with skills to fill the sweet-spot void. Professors of Practice are one avenue, but universities and faculties should also look for additional boundary spanners and to team academics with practitioners to further bridge domains and diffuse the research benefits of the commons. As Tushman et al. indicated and we have further developed, the domains of the commons can only come closer if genuine efforts to change behaviours are taken beyond rhetoric to real incentives and actions. Rewarding academics for successful strategies around WIL needs to be matched in workload adjustments and additional rewards. *Practise*, *publish* and *prosper* is more representative of what we academics should be aspiring to achieve rather than the *publish or perish* mantra. 5. Discussion {#s0070} ============= All three domains (research, practice and social) need to reinvigorate their commitment to the pursuit of understanding, recognising that the complexity of the commons has taken us well beyond the capacity of any one domain to achieve its goals without interaction with others. [Fig. 3](#f0015){ref-type="fig"} identifies that education provides one bridge to effective interaction. However, multiple perspectives have moved the commons beyond taming such dilemmas with simplistic linear solutions. Incentives and rewards in all three domains are complicated and there will be many trial-and-error interactions without necessarily guaranteeing immediate and measurable benefits. Young career academics are under increasing scrutiny to perform well on student evaluations even though teaching is often treated as a given, and not celebrated like high level publication. These younger academics increasingly game the teaching system and avoid risk in pursuit of a simple "tick" on their education-related tenure evaluations. This only exacerbates distrust of students and stakeholders. We cannot predict which experiments will lead to which outcomes ([@bb0200]) but our more interactive educational cases show that there are plenty of opportunities and plenty *do* work. 5.1. Theoretical and case implications {#s0075} -------------------------------------- The message of this paper is clear: we cannot continue to overlook the role of *teaching* in strengthening relevance and building connections between research and practice. The authors are cautious in using terms like \"pracademic\". If this implies a boundary spanner gifted in both *practice* and *academia* then we are supportive, knowing that improved translation across our stakeholder groups is critical ([@bb0155]). We have identified that one-size-does-not-fit-all and have offered a range of examples where early and mid-career researchers can fine-tune their educative abilities. Like researchers in search of better methods to bring them closer to practice, *teaching* and *learning* need similar strategies. However, instruction through practice alone is insufficient, and evidence and rigor must be maintained. We suggest that well-designed action research and action learning are effective but currently underutilised ways of co-creating new knowledge within practical environments. Embedded field-work teaching and instruction definitely has merit in leading to win-win-win outcomes. Cases B and D are representative of typologies where academics not only fine-tune their craft but gain immediate and critical feedback from the market. Sometimes the response can be blunt but that is what reflective inquiry is all about. Anne Cunliffe, Andrew Van de Ven and the late Donald Schon, Chris Argyris and Reg Revans should all be on the reading lists of aspiring theory-practice reflexive academics. Fortunately, as the authors of this paper and their ANU colleagues in design science research (DSR) are showing, such action-based reflexive studies *can* be published. The commons could draw insights from the Industrial Marketing and Purchasing (IMP) Group\'s actors-activities-resources model (AAR) in terms of delving more deeply into how bonds, ties and links ([@bb0120]) are stimulated across our three domains. The approach outlined in this paper correlates with a more applied philosophy of improving education/research processes in the broader social sciences ([@bb0085]). The five stages in Dreyfus\' model of skills development of *novice*, *advanced beginner*, *competent performer*, *proficient performer*, and *expert* is important. How many proficient performers or true experts have we got in the disciplines of business? Senior academics never practising their trade in pursuit of purist academic pathways, are not likely to pass the test. Our purists are needed but they have a limited market with often long term horizons. Turning to our graduands, the best skill level we can probably aspire to regarding teaching and learning is a *novice* with some reaching *advanced beginners*. This is reliant on practice-based exposure, and if we deprive our students of such opportunities, we have forsaken their chance of attaining applied skills readiness. This is an important gap that practice and social stakeholders are asking us to fill ([@bb0130]). MBAs and executives are calling for a different type of training. They increasingly want more advanced capabilities that add to their unique value and individual brand. This means complementing industry experience and helping them advance to *proficient performers* or *experts*. Cases B and C illustrate these aspirations. Even a *competent performing* management academic would struggle at this level given they are often lacking exposure to practice. It is something that medical professions and universities like Stanford and Harvard do well. Micro-credentials and advances that support a growth mindset ([@bb0070]) are increasingly fundamental at such levels, as are the benefits of sharing experiences with peers and expert lecturers. Only those with equivalent understanding of "real world" practices are likely to be able to teach in such environments. University settings not willing to invest in such innovations are likely to become increasingly redundant. The benefits of getting closer to practice and social partners cannot be underestimated but whatever tools we use as academics, it is important that we draw from quality evidence to make decisions. If we stand still it will only be a matter of time before the market and burgeoning industry competitors leave us behind. [@bb0220] in his *Confessions of a Pracademic* investigates in conceptual detail the Theory of Engagement, starting with a practical or perceived problem, followed by analysis then conceptualisation. Such spirals of inquiry focus on a deeper understanding, new knowledge and ultimately improved outcomes. Much of the criticism of theory development in marketing and management is about providing post priori results and in essence informing practitioners of what they already know. As our praxis cases identify, this can be overcome when stakeholders are more intertwined and co-creating together. Susskind is proud of his engagement process that he suggests outsmarts the current academic system. This starts with documenting issues before theory building, which is then followed like many of our examples, with exchange through teaching, training and active educational partnerships. Our regional industry training approach (Case D) is closest to gaining direct societal and industry-wide impacts. While currently an exception for universities, such approaches are slowly becoming more popular. Swinburne University of Technology in Australia with their approach to Industry 4.0 and partners like Siemens follow this style of engaged WIL interactivity. FIRenze SmarT (FIRST) Working Lab is a European attempt trying similarly to bridge the teaching-practice gap. Incentivising professors to champion industry projects for students sounds easy, but is difficult when schools do not seriously acknowledge the individual academic\'s publishing trade-off and resource outlay. Unlike business innovation, where there is a 'fuzzy front-end' ([@bb0140]), experiments in management education seem to have a 'fuzzy middle'. The 'fuzzy middle' makes direct measures of cause-effect relations between programs and outcome inconclusive. For example, entrepreneurship education programs often measure the number of new ventures *launched* by participants yet in our experience few of the ventures designed in undergraduate entrepreneurship programs move beyond prototyping. Nevertheless, anecdotally, the same individuals who met, interacted, and learned together, are occasionally discovered two or three years later still connected and launching completely different ventures. Skills development, as in the Dreyfus model, do take time to perfect with good doses of experiential learning critical for graduand development. We agree with [@bb0195] that a blend of 'programmed learning' and dose of real 'action' is important for growing into a profession. Novice students are not totally dissimilar to entrepreneurial start-ups with wide and varied skills that effectuate before they narrow their capability and competences. This is different to the modern more causal MBA student (more competent and proficient), who is more like the incumbent firm trying to reconfigure resources, perfect their craft, and exploit new opportunities. The experienced executives in Cases B and C were increasingly wanting more than simply business smarts as they pursue high-order expertise and a rare combination of practical and theoretical wisdom. Our executives are increasingly striving for currency and are conscious of the up-and-comers with their newly acquired MBAs. Success in our digital AI world is likely to be built around continuous learning and constant program improvement. We need better measures of the longer-term impacts of educational experiments when outcomes are mediated by indeterminate 'fuzzy middle' processes. A key aspect of this study is promoting education as a fillip for reversing what [@bb0215] and others see as a continual research relevance decline in the commons. Narrowing the gap between those that *teach* and/or those that *research* should be high on the agenda of university administrators. The *sweet spot* for a win-win-win scenario has potential to extrapolate the co-created value harvestable from engaged scholarship that combines not only Van de Ven\'s better *research-practice* but emphasises *teaching-practice* as well. 5.2. Practical implications {#s0080} --------------------------- Several practical implications stem from this research. Firstly, a person with a PhD with extensive real-world business expertise is a rare find in Australia yet this research identifies that transdisciplinary and boundary-spanning academics are critically important in helping overcome the current divide between practice and research and a shrinking commons. We should start with PhDs as they enter the field, and train them in the science of teaching and communicating and not simply the science of research. Pracademics crossing these boundaries are ideal but we need to be careful that we don\'t lose research credibility in such a process. *Practise, publish and prosper* has merit and for marketing and management academics this means communicating theory more effectively while simultaneously investigating practical and management benefits. Marketing and management are not alone here but it is ironic that we don\'t *practise* our communication skills as well as we *preach* them in theory. Osterwalder and his co-designers (*Business Model Generation* and *Value Proposition Design*) are typical of a new breed of consultants turning academic tools upside-down. Like Eric Ries, this popularist new breed is simplifying complex theoretical phenomena and making it more user-friendly ([@bb0090]). Their refinements are now commonly seen in the curricula. Adam Grant from organisational psychology is particularly astute at using different mediums to communicate difficult concepts simply and humorously. It may be time to reconsider reward structures for academics and, as indicated by [@bb0080], appoint more industry-capable academics committed to mentoring researchers and teachers. An academic balanced scorecard and workload model that properly rewards research, teaching and practice would be a great start. This stream should not simply support Professors of Practice but also extend more commonly to Lecturers of Practice and Associate Professors of Practice. A second practical implication is to carefully design programs around a strong theoretical base, cognisant of industry and practical elements. The programs discussed in our cases were offered at later undergraduate years or pitched at postgraduates, executives and external regional stakeholders. The practice elements of these courses are built around students having multidisciplinary and comprehensive theoretical foundations. The transfer of higher-order theory occurred as a result of these foundations not despite them, with EBM and rigor still central. As David Epstein in *Range* points out, we need to be careful to promote critical and diverse abstract thinking and not limit our students to simplistic over-structured thinking. This study has only touched upon opportunities for universities and a limited array of options. There is a suite of possibilities where educators can work with practitioners for similar benefit without falling into the potential trap of becoming vocational providers. We should not lose sight of the importance of being deep and unique in our research skills and offering that as a key point of difference to external practitioners and consultants. ANU is a university particularly clear in its support of Humboldtian research-teaching traditions. Once again, one-size does not-fit-all, and there are many universities that should also pursue their own visions and traditions. The key is to ensure the commons is growing and strengthening. More can and needs to be done to reduce the dissonance and to build a shared overarching logic for valuing management research and practice that delivers benefits to all stakeholders. Globally we must recognise the value of tertiary institutions. Some countries are better than others with China respecting with reverence the role of professor or teacher. It is important that the current dialogue changes and policy supports the virtues of on-going learning. Industry shares in this responsibility. Where wins occur, these should be celebrated and publicly acknowledged to build positive momentum. Australia is poorly performing on translating an excellent per capita patent and publication performance into commercial outcomes. Australia has therefore had to recently introduce a national policy to lift engagement and impact. Europe and the UK are designing similar policies. Significant funds are now being allocated at the national top-down level to incentivise universities to pull change up through the system. This is an interesting nudge to a national system. Internationally this could be supported by global and business school rankings that additionally reward universities with PhDs possessing and nurturing strong practice capability. We mentioned earlier that the commons aligned well to a wicked problem or challenge. These multiple stakeholder societal problems are messy, intractable, confusing and incorporate countervailing opinions ([@bb0005]). Taming such challenges requires transformational insight and ultimately more *wicked innovation*. Changing top-down national and international systems is complicated. The *AMS Review* has added a more robust *theory* and *practice* section which is one example of changing international efforts. Many of the more effective instruments for positive change are likely to be tactical and bottom-up; i.e. at your own business school, faculty and university level. A simple start for each business school, at the regional commons and quadruple helix level, is more effective engagement with industry, government and community on a place-based level. Some do this well, but most don\'t. The IMP AAR model combines well with regional innovation systems models ([@bb0245]) as a way of mapping and understanding each local helix and commons stakeholders. Key stakeholders can then be selected as part of an advisory board to help set KPIs and monitor a new academic balanced scorecard. The advisory board should measure aspects like the number of interns and the number of courses where WIL is operational and effective. Rewards for practice around educational programs and impact need to at least match workload rewards for publishing. Advisory boards should incorporate alumni and executives that have been involved in educational programs. Monitoring skill performance of graduands for key capabilities around immediate employability could also be introduced as a broader regional commons initiative. Instead of simply surveying students on their satisfaction, an omnibus of the region should measure effectiveness. This should incorporate international partners to assess global impact. Rewards for educational excellence, including high quality theory, could be applied and then monitored by the advisory board. Reviewing research quantum and other measures to ensure the commons remains connected should also form part of the brief. As suggested, this process is not new to some business schools, but its broader diffusion would add some competitive tension to the sector. Signalling is indeed a two-way activity. Knowing that a university ranks highly in skills development and employability signals to the best management and marketing students where to enrol. 5.3. Limitations and future research directions {#s0085} ----------------------------------------------- Variations across cultures and across universities are noted in the literature review; such variations are likely to have strong implications. The commons portrayed through our cases in the Australian context may be different to that of other countries. This requires further exploration and offers a strong opportunity for future research. Additionally, the discussion in this study is limited to one university. It would be valuable to audit programs in other institutions both in and beyond Australia to gather a fuller picture of the status quo. This links closely to our identification of the need for more research to measure impacts of effective teaching-practices and interventions. Shining a light on teaching-practice as a way of bridging the growing gap in research-practice is a key aim of the paper. Unfortunately, this coincides with what some are suggesting is a sector on the edge of disruption. Improving teaching and education in business schools, as a *science* and as a *practice*, has therefore never been more pressing. Fortunately, there are many ways forward. Gamification, virtual reality, artificial intelligence and social media are all potentially ways to improve the effectiveness of teaching-practice outcomes. These instruments are already in use in many of our institutions and adapted appropriately, can help turn teaching into *learning* with andragogical rigor. A detailed study of these tools and their role within the commons is beyond this article\'s scope but such investigations will be fascinating going forward. What our senior executive praxis cases highlight is that wisdom seldom emanates solely from technology enablers. Case D was all about tactile and sensory peer interactions and in these settings, like in Cases B and C, even the educator becomes an "actor" in a virtuous knowledge "theatre". Learning how to facilitate such theatres is an important horizon for academics traditionally used to standing and delivering. The evidence highlights that marketing and management is not lacking in theory-driven research and researchers. Ironically a major research challenge going forward is to find ways of attracting and incentivising better *communicators* and more interactive *educators* to research ways of turning our current *researchers* into better *teachers*. We also need to realise that there is extraordinary scope for us as educators to diffuse our research into the broader community via a range of innovative teaching practices. Ultimately, in the traditions of Wilhelm Von Humboldt, we should not lose sight of the value of *research and education* combining to build the *moral character* of our students in what should be a *better commons*. Maybe after all our *perplexing situation* is that we have forgotten the roots of why universities are actually here.

{#sp1 .567} {#sp2 .568} {#f1 .568}

Background ========== Computational prediction of transcription factor binding sites (TFBS) from co-expressed/co-regulated genes is an important step towards deciphering complex gene regulatory networks and understanding gene functions. Given the promoter sequences of a set of co-expressed/co-regulated genes, the goal is to find short DNA sequences (\"motifs\") whose occurrences (with allowed mismatches) in the sequences cannot be explained by a background model. An accurate identification of such motifs is computationally challenging, as they are typically very short (8-15 bases) compared to the promoter sequences (hundreds to thousands bases). Furthermore, there is often a great variability among the binding sites of any given TF, and the biological nature of the variability is not yet well understood. Finally, in many cases, the TFBS may appear only in a subset of the putatively co-regulated genes. Despite the challenge, many computational methods have been developed and have been proven useful in predicting real binding sites \[[@B1]\]. The existing algorithms can be roughly classified into two broad categories according to the motif representations: those based on position-specific weight matrices (PWMs), and those based on consensus sequences. Examples of the former include well-known programs such as MEME \[[@B2]\], AlignACE \[[@B3]\], GibbsSampler \[[@B4]\], and BioProspector \[[@B5]\]. The latter category includes Weeder \[[@B6]\], YMF \[[@B7]\], MultiProfiler \[[@B8]\], and Projection \[[@B9]\]. In general, PWM offers a more accurate description of motifs than consensus sequences, but the score of PWM is more difficult to optimize. On the other hand, consensus-based algorithms often rely on enumerating short subsequences, which may be impossible for longer motifs. For an excellent survey of the existing methods and an assessment of their relative performance, see \[[@B1],[@B8]\]. Recently several consensus-based motif finding algorithms have been developed using evolutionary algorithms, because of their efficiency in searching over multidimensional solution spaces. For example, GAME \[[@B10]\] and GALFP \[[@B11]\] are based on genetic algorithms, and have been shown to outperform many PWM-based algorithms. In a previous work, we proposed a motif finding algorithm based on the classical Particle Swarm Optimization (PSO) strategy \[[@B12]\], where we used the set of positions on each sequence together as a solution, and searched the solution space by PSO algorithm. To keep the solution space continuous, we restructured the original sequences using a sequence mapping. Although the algorithm shows a good performance on small input size (for example 20 sequences and 1000 bases for each sequence), the algorithm becomes slow for larger data sets, as the number of possible motif positions grows exponentially as the number of sequences increases. Several other motif finding methods have also been developed based on PSO, for example, Hybrid-PSO \[[@B13]\] and PSO-EM \[[@B14]\]. Hybrid-PSO uses a similar basic idea as our previous work \[[@B12]\], and therefore has the same problem we mentioned above. PSO-EM simply uses PSO to find candidate motifs, which are then used as seeds by other expectation-maximization based motif finding algorithms, such as MEME \[[@B2]\]. In this paper, we develop a novel algorithm, called PSO+, for finding motifs. This new method has the following contributions. First and most importantly, PSO+ differs from other motif finding algorithms by explicitly modeling gaps, which provides an easy solution to find gapped motifs. Many real motifs contain positions of low information (gaps), but the existing algorithms usually do not allow gaps, or require a user to specify the exact location and length of gaps, which is often impractical for real applications. Second, we use both consensus and PWM representations in our algorithm, taking advantage of the efficiency of consensus and the accuracy of PWMs. Our method also allows some input sequences to contain zero or multiple binding sites, which is common in real biology data set, but ignored by some of the algorithms. Finally, we propose a novel modification to the PSO update rule to accommodate discrete values, such as characters in DNA sequences, which may also be useful in other applications. Methods ======= Introduction to Particle Swarm Optimization ------------------------------------------- Particle Swarm Optimization (PSO), which has been shown to be effective in optimizing difficult multidimensional problems in many fields, is a population-based stochastic optimization technique for problem solving that is inspired by the social behaviors of organisms such as bird flocking \[[@B15]\]. The system is initialized with a population of random solutions and searches for the optimal solution by updating iteratively. Each potential solution, called particle (or agent), is represented by a point in the multiple-dimensional solution space. When searching for the optimum solution, particles fly around the solution space with a certain velocity (speed and direction). During flight, each particle adjusts its position and velocity according to its own experience and the experience of its neighbors. Specifically, each particle keeps track of the best solution it has encountered so far. This solution is called *pbest*, which stands for personal best. The system also keeps track of the global optimum of all the particles, hence called *gbest*. The fundamental concept of PSO consists of changing the velocity of each particle at each time step toward its *pbest*and *gbest*locations \[[@B15]\]. Method overview --------------- Figure [1](#F1){ref-type="fig"} shows the main structure of the algorithm, which contains three loops. The most inside loop, loop3, evaluates the fitness value of each agent and updates information for the whole system. Using its *current*solution, an agent first finds out a best match from each sequence, calculates the fitness value (see below), and updates *pbest*, *gbest*if necessary. Loop2 is the main part of the PSO+ algorithm. From a random initial solution, each agent continuously searches for better solutions in the neighborhood, taking information from its own experience (*pbest*), and the experience of all agents (*gbest*). The actual movement of each agent is determined by the update rule (see below). Finally, as a stochastic algorithm, the final solution of PSO+ depends on its starting solutions; the purpose of loop1 is therefore to restart the system several times, from independent random solutions, to ensure a high overall success rate. At the final step, we use post-processing to remove and/or add some binding sites, therefore allowing zero or multiple binding sites on each sequence. A flow chart is shown in Additional file [1](#S1){ref-type="supplementary-material"} Figure S1. {#F1} Solution space and fitness function ----------------------------------- To utilize the PSO+ algorithm, we need to represent a solution as a vector, and determine a fitness function appropriate for the problem. As discussed in Background, a motif of length *l*can be represented either as a position-specific weight matrix (PWM), which is a 4×l matrix of real numbers specifying the probability of each base at each position, or a consensus describing the most dominate base at each position. The matrix can be converted into a vector of length 4*l*, although some care needs to be exercised to ensure proper normalization. The consensus representation is more efficient in searching for new instances, and may lead to faster convergence, while the PWM is more powerful in representing weaker motifs and is more accurate in evaluating the motif quality. We decided to use both representations in our algorithm, to take advantage of both forms. The solution is initialized as a consensus. During the scanning stage, we use the consensus representation. After the best matches to the consensus are found from all the sequences, we compute a PWM based on the set of matches, and compute the fitness of the solution based on the PWM. Given a solution, i.e., a consensus, it is first used to scan the input sequences to find a best match on each sequence. Let *X*= (*x*~1~*x*~2~\... *x~l~*) be the consensus sequence and *Y*= (*y*~1~*y*~2~\... *y~l~*) be a putative binding site. The matching score between *X*and *Y*is computed using the following equations: $$\begin{matrix} {M(X,Y) = {\sum\limits_{i}{\sigma(x_{i},y_{i})}}\ \text{and}} \\ {\sigma(a,b) = \left\{ \begin{matrix} {1 + {\log}_{4}(0.25/p_{a})} & {\text{if}\,\, a = b} \\ {{\log}_{4}(0.25/\sqrt{p_{a}p_{b}})} & \text{otherwise} \\ \end{matrix} \right.} \\ \end{matrix}$$ where *p~a~*is the background frequency for base *a*in the input sequences or in the whole genome. As most genomes contain more AT\'s than CG\'s, this formula gives unequal weight to different types of matches/mismatches. A match between two bases with lower background frequency would have higher score than that between two bases with higher background frequency. For uniform base frequency *p~A~*= *p~C~*= *p~G~*= *p~T~*= 1/4, *σ*(*a*, *b*) = 1 if *a*= *b*and 0 otherwise, corresponding to an intuitive match/mismatch score. Given a set of matches of a consensus, *W*= *w*~1~, *w*~2~, \..., *w~n~*where each *w~i~*is a subsequence with length *l*, we compute the fitness of the consensus using the information content (IC) score: $$IC\, = {\sum\limits_{j = 1}^{l}{\sum\limits_{b}f_{b}}}(j){\log}_{2}(f_{b}(j)/p_{b}),$$ where *f~b~*(*j*) is the normalized frequency of nucleotide *b*on the column *j*of all instances in *W*and *p~b~*is the background frequency of *b*. Initial solutions and number of agents -------------------------------------- Similar to all stochastic algorithms, the performance of PSO+ partially depends on the initial solutions. The convergence of the algorithm can be significantly improved if at least one of the agents has an initial solution near the optimal solution. In this work we consider two strategies. The first strategy is to simply generate a set of random consensus. The second strategy is to randomly choose a subsequence from the input sequence as an initial solution. Although the first strategy would allow the maximum coverage, the probability that any randomly generated consensus is near the optimal solution is very low, as there are 4*^l^*possible solutions for a motif of length *l*. For the second strategy, we assume that the actual binding site is closer to the consensus than to random sequences. Since there is usually about one binding site per sequence, it is very likely that some agents may select a binding site as an initial solution. More precisely, assuming that the average sequence length is *L*and the motif length is *l*, the probability that a randomly selected sequence is a binding site is 1/(*L*- *l*+ 1). Also because of the Check Shift step in the algorithm, a random solution that contains a large suffix or prefix of the binding site can often lead to the recovery of the real motif quickly. We usually allow a binding site to be shifted by two bases to its left and right, respectively. Therefore, each true binding site can provide up to 5 initial solutions that are similar to the real motif. For this reason, we suggest the minimum number of agents to be (*L*- *l*+ 1)/5 to ensure a high convergence probability. In our experiments on both synthetic and real sequences, we have found that the second strategy usually leads to much faster convergence and therefore is implemented as the default option. Modified PSO+ update rule for discrete problems ----------------------------------------------- After each iteration, each agent needs to update its *current*solution based on the old *current*, its own *pbest*, and *gbest*of the system, each of which is a vector. The standard PSO algorithm is designed for optimization problems in the continuous domain; therefore, a new solution can be easily obtained by a sum of the three solution vectors multiplied by some random weights. In our case, however, each solution is a vector of ACGT\'s and they cannot be manipulated by multiplication and summation. In order to generate a new solution, we use the following rule, which is applied independently to each position of the motif: $$i* = \arg{\max}_{i}(c_{i}r_{i}\,\text{weight}(x_{i}))\,\text{and}\, x^{'} = x_{i*},$$ where *x\'*is the new character being generated, *x*~1~, *x*~2~, *x*~3~are the characters in *current*, *pbest*, and *gbest*, respectively, *x*~4~is a random character from ACGT, *c~i~*is a scaling factor to determine the relative importance of the four terms, *r~i~*is a uniform random number, and the function weight gives a higher weight to characters having lower background frequency. This strategy effectively suppress characters with lower occurrence, when compared to an alternative strategy that randomly picks a character proportional to its number of occurrences in *x*~1~, *x*~2~, *x*~3~and *x*~4~. For example, assuming *c~i~*= 1 and the weight function returns a constant, if *x*~1~= *x*~2~= *x*~3~= A and *x*~4~= C, the alternative strategy would select A with 3/4 probability and C with 1/4 probability; with our strategy, we would select A with 7/8 probability and C with 1/8 probability. This reduces the probability of drifting away when a solution is near the optimal solution. All agents stop their movements if the number of iterations exceeds an upper limit, or if there has been no update for a certain number of iterations and a minimum number of iterations has passed. Check shift ----------- Similar to many motif finding algorithms, the output of PSO+ algorithm may have a shift issue: the start positions of the binding sites may be one or two positions away from the real position, and it is difficult for the algorithm to escape from such local optima. To circumvent this problem, we periodically check whether shifting the binding sites by a small number (up to 3 bases) can improve the quality of the solution. This is done in the CHECK SHIFT step. Gapped motifs ------------- Many real motifs contain *do-not-care*positions in their consensus. Furthermore, these *do-not-care*positions are often consecutive, forming a motif with two short conserved regions with some fixed distance in between. We call these *do-not-care*positions gaps and the motifs gapped motifs. Most motif finding algorithms do not consider gaps explicitly. For consensus-based algorithms, ignoring gaps can lead to serious mistakes when a consensus is forced to be selected for a gap position, which has no dominant characters. ### Gap representation We solve this problem by asking the users to provide two parameters, motif length *l*, and gap length *k*. Importantly, when we search for gapped motifs, we allow gaps to appear as the suffix or prefix of a motif. This is very useful if the actual gap is shorter than *k*, or if the non-gap region of the motif is shorter than *l*- *k*. Furthermore, if a real motif contains no gaps, our algorithm will automatically put all gaps in the flank regions. While our algorithm may increase the guesswork from the user by asking for gap length as an additional parameter, it actually gives the user more flexibility in determining the appropriate motif and gap lengths. To find gapped motifs, we introduce a bit vector of length *l*. The bit vector contains exactly *k*0\'s and *l*- *k*1\'s, where a 0 means the position falls in a gap. This vector is initialized randomly and latter updated by masking out the columns in the PWM with low IC values. Given this vector, when calculating the match score between a consensus and a potential binding site, we only consider the positions with 1\'s. When we compute the fitness (IC score) of a motif given all the binding sites, we consider all columns, because in this case we have all the information to derive a PWM. ### Gap insert strategy We consider two types of gaps. The first type of gaps can be anywhere in the motif and do not need to be consecutive. To find this type of gapped motifs, we simply mark the columns with the lowest IC value as gaps. The second type of gaps are consecutive and are located in the center of a motif. For this type of gapped motifs, we require a motif to contain at most two consecutive non-gapped regions, while gaps can appear either as a prefix, a suffix, or in the center of the motif. An algorithm using the second type of gaps is less efficient, but can often result in better results for real motifs. This is the default option of our algorithm. Experimental results to support this default option are shown in Additional file [1](#S1){ref-type="supplementary-material"} Table S1 and Table S2. ### Flexible gap length If the actual gap is shorter than the given number, or if the motif is shorter than expected, gaps can appear as suffix and/or prefix of the motif. Furthermore, the algorithm will automatically put all gaps in the flank regions if the real motif contains no gaps. By default, we use *8-base motif with 0 gap*for short motifs, *12-base motif with 2 gaps*for medium-length motifs, and *16-base motif with 4 gaps*for long motifs. This strategy works well for the unknown length motif finding problems in this paper. The program allows a user to change the parameters by themselves. Post-processing --------------- The basic algorithm described above assumes that there is one and only one binding site on each sequence, which is certainly not always true. To address this problem, we use a statistically-inspired strategy to refine the binding sites. We assume that at least a good fraction of the sequences contain at least one binding site. We calculate the match score for each putative binding site returned from the basic algorithm. Let Q1, Q2, and Q3 represent the lower quartile, median, and upper quartile of the match scores. The inter-quartile range (IQR) of the match scores is then computed by Q3 - Q1. All binding sites with a match score below Q1 - IQR are dropped as false binding sites. We also rescan the input sequences using the consensus for additional putative binding sites. A binding site with match score higher than Q2 is considered a true binding site. A PWM is constructed using the final set of binding sites. Results and discussion ====================== To evaluate the performance of our algorithm, we tested it on two types of sequences. The first type of test data consists of synthetic DNA sequences, also known as the (l, d)-motif challenging problem \[[@B16]\]. The second type of data contains real promoter sequences. The algorithm is implemented in C. The test datasets are included in Additional file [2](#S2){ref-type="supplementary-material"}. Evaluation using synthetic data sets ------------------------------------ We tested our algorithm on the (*l*, *d*)-motif challenge problem. Each challenge problem includes *n*sequences of length *L*, each of which contains a variant of a pre-defined consensus of length *l*. The variants were generated by choosing *d*positions randomly from the consensus and changing them to random bases. In our first experiment, we focused on the (15, 4)-motif challenge problem, which is one of the most popular benchmarks for motif finding programs. We chose *n*= 20, and varied *L*from 400 to 1000. Table [1](#T1){ref-type="table"} (left half) shows the running time of our current algorithm (PSO+), another PSO-based algorithm we developed recently (PSO) \[[@B12]\], two evolutionary algorithms (GAME \[[@B10]\] and GALFP \[[@B11]\]), and three well-known combinatorial-search algorithms (Projection \[[@B9]\], Weeder \[[@B6]\], and MotifEnumerator \[[@B17]\]). We were not able to test Weeder by ourselves because the currently available implementation of the algorithm can only handle motifs of even lengths up to 12. Its running time was taken from the original publication and was based on an 89% success probability \[[@B6]\]. The results of the other algorithms were obtained by downloading the programs from the original authors\' websites and running with parameters that can recover the embedded motifs with 100% accuracy. Running time was based on the average of 10 runs on 5 sets of sequences. ###### Running time (seconds) on (l, d)-motif challenge problems Sequence length 400 500 600 800 1000 600 600 600 600 ----------------- ------------ ------------ ------------ ------------ ------------ ------------ ------------ ------------ ------------ **(*l, d*)** **(15,4)** **(15,4)** **(15,4)** **(15,4)** **(15,4)** **(11,2)** **(13,3)** **(17,5)** **(19,6)** Weeder 60 125 200 450 900 \- \- \- \- Projection 9 23 42 162 418 4 13 94 174 MotifEnumerator \- \- \- \- \- 5 119 \- \- PSO 18 34 57 137 288 72 58 61 54 GALFP 100 123 161 212 286 127 137 162 172 GAME 27 30 32 36 41 23 28 34 42 PSO+ 1.1 3.1 3.4 7.3 19.4 4.9 10.6 2.3 3.8 As shown in Table [1](#T1){ref-type="table"} PSO+ is significantly more efficient than the other algorithms. Our previous PSO algorithm was based on a very different problem formulation and is considerably slower than PSO+. It is worth noting that the running time of the evolutionary algorithms, GAME and GALFP, are much slower than PSO+ in general, but their running time only increases slightly with sequence lengths. This might be because these two algorithms spend a significant amount of time in initialization, independent of sequence lengths. Next, we compared these algorithms on their performance on challenge problems with varying motif lengths and number of error, while sequence lengths were fixed at 600 bp. The current algorithm outperforms the existing algorithms again on these test sequences (Table [1](#T1){ref-type="table"} right half). Similar to our previous PSO-based algorithm, the running time of PSO+ is relatively independent of motif lengths. Third, we ran PSO+ 100 times on independent sequences, and plotted the running time in Figure [2](#F2){ref-type="fig"}. As shown, the running time has a long-tail distribution, meaning in some rare cases the algorithm may need an extremely long time to converge. Because of this, the algorithm runs faster than the average running time in most cases. A better strategy for detecting local optima may eliminate some of these rare cases and improve the overall efficiency. The running time results for GALFP/GAME are shown in Additional file [1](#S1){ref-type="supplementary-material"} Figure S2 and Figure S3. {#F2} Finally, we tested our algorithm using synthetic sequences containing gapped motifs, and compared with the existing algorithms. We first generated sequences with embedded gapped motifs, much like the synthetic challenge problems, except that each embedded motif contains some gaps in the middle positions. Only two algorithms, namely, PSO+ and GALFP, can correctly identify these synthesized gapped motifs. It is not surprising to see that the other algorithms, which are not designed to find gapped motifs, failed at these synthetic test cases. Remarkably, GALFP had basically the same accuracy as PSO+. With further investigation, we found that using synthetic data sets with purely random background noise is insufficient to demonstrate the limitation of existing algorithms on finding gapped motifs. This is because gapped motifs can be approximately represented by a PWM with uniformly distributed base frequencies in the gapped positions, and that a motif generated by the synthetic model with a purely random background is either strong enough so that it can be found by both PSO+ and GALFP, or it is so weak that neither algorithm can find it. Therefore, we further generate two additional types of test cases with gapped motifs. Each test case of the first type consists of 20 sequences of length 600, and each sequence contains a gapped motif of length 15 and gap length 5 and a non-gapped motif of length 15. For this type of test cases, GALFP (as well as the other algorithms listed in Table [1](#T1){ref-type="table"}) always report the non-gapped motif; while the PSO+ algorithm report the gapped motif as first motif, and the non-gapped motif as the second motif. For the second type of test cases, we generate 30 sequences of length 600. We then generate a gapped motif of length 15 and gap length 5, and embed it on 10 of the sequences, while leave the remaining 20 sequences without any embedded motifs. For this type of test cases, only GALFP and PSO+ can report the correct motif. GALFP only reports a single motif and it has a 91 percent accuracy. PSO+ has an accuracy of 96 percent when only reporting a single motif for each test case, while its accuracy improved to 98 percent with two motifs each test case and 100 percent with eight motifs each test case. Experiments on real biological sequences ---------------------------------------- To test the performance of our algorithm on real biological sequences, we used the same eight representative test cases used in \[[@B11]\], which covered different lengths of motifs and both gapped and non-gapped motifs. In these data sets, some sequences may contain zero or more binding sites. Details of the data sets are available in \[[@B11]\]. Based on these data sets, it has been reported that two genetic algorithms, GALFP \[[@B11]\] and GAME \[[@B10]\], are significantly more accurate than five popular algorithms: MEME \[[@B2]\], Bioprospector (BP) \[[@B5]\], BioOptimizers based on MEME (BOM) and BioOptimizers based on Bioprospector (BOB) \[[@B18]\]. Therefore, we only compared PSO+ with GALFP and GAME directly. As in \[[@B11]\], we measure accuracy by *precision*, *recall*, and *F-score*. Precision is defined as *c/p*and recall is defined as *c*/*t*, where *c*, *p*and *t*are the number of correctly predicted binding sites, the number of predicted binding sites and the number of true binding sites in the sequences, respectively. The *F-score*combining both *precision*and *recall*is defined as *F*= 2 \* Precision \* Recall = (Precision + Recall). Table [2](#T2){ref-type="table"} shows the average results of PSO+, GALFP \[[@B11]\] and GAME \[[@B10]\] on 20 runs (the bold entries are the winners). As shown, PSO+ and GALFP have about the same accuracy. PSO+ has the best *F-scores*on 4 of 8 test cases while GALFP has the best *F-scores*on 3 test cases. More in-depth investigation reveals that PSO+ generally has the highest precision (5 out of 8), while GALFP has the highest recall (3 out of 8). This indicates that PSO+ reported less but more accurate binding sites than GALFP. ###### Comparisons of average performance on the 8 real datasets, the bold numbers are the best results among three algorithms GAME GALFP PSO+ ------ ----------------- -------------- --------------- ----------------- -------------- --------------- ----------------- -------------- --------------- ***Precision*** ***Recall*** ***F-score*** ***Precision*** ***Recall*** ***F-score*** ***Precision*** ***Recall*** ***F-score*** CREB 0.43 0.42 0.42 0.70 **0.84** **0.76** **0.76** 0.68 0.72 CRP 0.79 **0.78** 0.78 0.99 0.73 0.84 **1** **0.78** **0.88** ERE 0.52 0.78 0.62 0.82 0.76 0.79 **0.92** **0.92** **0.92** E2F **0.79** **0.87** **0.83** 0.77 0.85 0.81 0.68 0.7 0.69 MEF2 0.52 0.55 0.53 0.91 0.98 0.95 **1** **1** **1** MYOD 0.14 0.14 0.14 **0.57** **1** **0.72** 0.20 0.43 0.27 SRF 0.71 0.86 0.78 0.75 **0.89** **0.82** **0.80** 0.56 0.66 TBP 0.81 0.74 0.77 **0.87** 0.87 0.87 0.86 **0.91** **0.88** Interestingly, two of the eight cases (CRP, ERE) are clearly gapped motifs, as shown in Figure [3](#F3){ref-type="fig"}. PSO+ won in both cases, especially in ERE (average F-score: 0.92, 0.79, and 0.62, for PSO+, GALFP and GAME, respectively), indicating that the gap model in our algorithm is effective. On the other hand, even though our algorithm has a much larger search space by allowing gaps, its performance on the other six motifs that do not have gaps is still among the best. Figure [4](#F4){ref-type="fig"} shows the results of PSO+ on the ERE motif, with or without the gap option. Without the gap option, our algorithm attempts to maximize the total information content of all positions, which results in a motif with relatively uniform information content across all positions. In contrast, with the gap option, our algorithm automatically determines the low-information positions and treats them as gaps, and therefore improves the accuracy of the final result. With the gap option turned on, the Pearson correlation coefficient between the predicted and true motif PWMs is improved from 0.92 (*p*= 10^-22^) to 0.97 (*p*= 10^-33^). {#F3} {#F4} Finally, it takes our algorithm 20 to 60 seconds for each of the 8 test cases. In comparison, the average running time is about 62 seconds for GALFP, and 291 seconds for GAME. Conclusions =========== In this work, we have proposed a novel algorithm for finding DNA motifs based on Particle Swarm Optimization (PSO). Our contributions include a novel modification of the PSO update rule to allow discrete variables, a model to allow gapped motifs, and a simple method to ne-tune the motif when some sequences contain zero or multiple binding sites. Experimental results on synthetic challenge problems as well as real biological sequences show that our method is both more efficient and more accurate than several existing algorithms, especially when gaps are present in the motifs. We are working to finalize our program, which will be freely available to the research community soon. Competing interests =================== The authors declare that they have no competing interests. Authors\' contributions ======================= CL and JR conceived of the research and designed the study. CL implemented the algorithm and conducted the experiment. CL wrote the paper and JR helped with the manuscript preparation. Both authors read and approved the final manuscript. Supplementary Material ====================== ###### Additional file 1 **Supplemental Materials**. Additional figures, tables and discussion. ###### Click here for file ###### Additional file 2 **Experimental Dataset**. The synthetic data sets and real biological sequences used in the paper. ###### Click here for file Acknowledgements ================ This work was supported in part by a NIH grant SC3GM086305 and a grant from the San Antonio Life Sciences Institute. The authors also wish to thank the Computational Biology Initiative (UTHSCSA/UTSA) for providing access to their high-performance computing facility.

Introduction {#Sec1} ============ The Internet of Things (IoT) is a complex domain of application that allows objects to exist on the Internet. Creating systems in this domain is a challenge because it involves both software and hardware, sensing and actuating devices, a communication infrastructure, in addition to storage constraints. For this, a variety of IoT design patterns have been proposed in various categories to address variety of issues \[[@CR7]\]. They propose solutions for common and recurring problems to architects and designers in the IoT domain. Most of these patterns are presented visually and informally, there is no formal semantics associated with them. Hence, their meanings may be imprecise. They can lead to their misunderstanding and misuse. To remedy this problem, we propose an approach that allows to model and specify these patterns with a formal notation that allows to reuse them correctly. Our objective is to prove the relevance of these patterns. We illustrate our approach with different pattern examples. We propose a graphical modeling of these patterns in order to describe both their structural and behavioral features. Then, we propose a generic formal specification of these patterns using the formal Event-B method. Finally, we develop a graphical editor describing our approach using the Eclipse modeling platform. The rest of this paper is organized as follows. Section [2](#Sec2){ref-type="sec"} focuses on the structural modeling of IoT design patterns and Sect. [3](#Sec5){ref-type="sec"} focuses on the behavioral modeling. In Sect. [4](#Sec6){ref-type="sec"}, we present an application to a case study of our approach. Section [5](#Sec7){ref-type="sec"} describes how to formally specify IoT design patterns with the Event-B method. In Sect. [6](#Sec8){ref-type="sec"}, we present our tool, which implements the proposed approach. Section [7](#Sec9){ref-type="sec"} discusses related work. Section [8](#Sec10){ref-type="sec"} concludes and gives future work directions. Structural Patterns Modeling {#Sec2} ============================ We provide a modeling solution for describing IoT design patterns using a visual notation based on the graphical UML language in order to give readable models. We first describe a meta-model, then we present a model instance of the design pattern. The metamodel extends the component diagram of UML 2.0 (Unified Modeling Language). The use of UML is motivated by four distinct rationales: (i) It is a standard modeling language defined by OMG. (ii) It is used to describe software architectures. (iii) Component diagrams of UML allow us to represent structural features of patterns. (iv) Sequence diagrams of UML allow us to represent behavioral features of patterns. Structural features of patterns are generally specified by the types of entities. The configuration of the entities is also described in terms of static relationships between them \[[@CR16]\]. We model structural features of design patterns with the extended *Component* diagram. In the following, we present the proposed metamodel. An example of a corresponding model is presented and illustrated with case studies as follows. Metamodel {#Sec3} --------- The extended *Component* diagram describes, by a set of concepts, the structure of an IoT architecture. We use it to describe the architecture of IoT design patterns. More specifically, it is to define the entities that can be involved in the pattern, their types and their dependencies (connections). The metamodel presented in Fig. [1](#Fig1){ref-type="fig"} extends the metamodel of the component diagram of UML 2.0. In this metamodel, we concentrate on two categories of design patterns; "Bootstrapping Design Patterns" and "Registration Design Patterns". "Bootstrapping Design Patterns" allow configuring new devices. They are composed of "Medium Based Bootstrap Pattern" and "Remote Bootstrap Pattern". "Medium Based Bootstrap Pattern" allows to configure a new device on-site through a removable storage medium inserted in the device. This support contains the necessary information for configuration. "Remote Bootstrap Pattern" is a configuration pattern used in case that a device is placed far away and is difficult to reach. The configuration in this case is done by downloading configuration information from a bootstrap server. "Registration Design Patterns\" allow to register the attributes and the features of a new device on the Back-end server. The registration is used to facilitate the communication and the interrogation with other connected objects. There are many registration patterns. In this work, we present two patterns. So "Registration Design Patterns\" are composed of "Automatic Client Driven Registration Pattern" and "Server Driven Model Pattern". The "Automatic Client Driven Registration Pattern" allows the device to register on the Back-end server via an API call. The "Server Driven Model Pattern" is used to create a device model that includes its description and functionality.Fig. 1.Metamodel of IoT design patterns The basic elements of the metamodel are:  **Component and Object:**Entities, that make up the architecture of an IoT design pattern, can be either *Components* or *Objects*. All objects are components, but not all components are necessarily objects. An object can be connected to the internet, it can receive and send data.**Port:**Entities can have *Ports* that constitute interaction points with their environment. These *Ports* are related to one or more *provided* or *required* *Interfaces*.**Interface:**The interfaces are the points of communication that allow interaction with the environment. For an entity, there are two types of interfaces. The *Provided Interfaces* describe the services provided by the component. The *Required Interfaces* describe the required services that other components must provide for the good functioning of component. These interfaces are specified via the ports.**Connector:**The communication path between Entities within an architecture is called a *Connector*. It ensures the link between a *Provided* port and a *Required* port to form a complete and coherent system.**Device:**A device is an Object. It is the entry point of the physical environment, it is used to process sensor data and to control actuators.   General Pattern Model {#Sec4} --------------------- In this section, we present two general pattern models as instances of the proposed metamodel. We have used different notations that can be used as a graphical description of the entities presented in the model. We are based on the work of Reinfurt et al. \[[@CR8]\]. There are two possible general models depending on the location of the device. If the device is placed locally, we use the "Medium Based Bootstrap Pattern" as a solution to configure the new device. If it is placed at a distance, we use the "Remote Bootstrap Pattern" as a solution. In Fig. [2](#Fig2){ref-type="fig"}, we represent the general pattern model of the "Medium Based Bootstrap Pattern". The solution proposed by this pattern to configure a new local device is to use an object of type *Storage Medium* containing information configuration. In Fig. [3](#Fig3){ref-type="fig"}, we represent the general pattern model of the "Remote Bootstrap Pattern". The solution proposed by this pattern to configure a new remote device is to use a component of type *BootstrapServer* allowing the upload of the configuration information using the PushBD connector.Fig. 2.General pattern model of the Medium Based Bootstrap Pattern Fig. 3.General pattern model of the Remote Bootstrap Pattern In Fig. [4](#Fig4){ref-type="fig"}, we represent the general pattern model of the "Registration Design Pattern". The solution proposed by this pattern to register a new device. The device is related to the *BackEndServer* with a connector named PushDA in order to be registered on it via an *API* call. Meta-data entered by the device are recorded in the *RegistryDevice* through the PushAR connector. The *RegistryDevice* component has a connector named PushRDm to store a device template in a database component named *Device Model DataBase* of the "Server Driven Model pattern". A device can integrate an object of type *Sensor* or an *Actuator*. All objects of the patterns have ports to communicate with others.Fig. 4.General pattern model of the Registration Design Pattern Behavioral Patterns Modeling {#Sec5} ============================ To model behavioral features of the design patterns, we use the UML 2.3 sequence diagram. We describe through this diagram successive interactions between the different entities of the IoT application in order to represent the two categories of the design patterns. Figure [5](#Fig5){ref-type="fig"} represent the sequence diagram that illustrates this behavior. We grouped the interactions into two phases.Fig. 5.Sequence diagram of the used IoT patterns **Configuration phase:Local configuration:** In the configuration phase and at the "Medium Based Bootstrap Pattern" level, the configuration is done by cutting the storage medium configuration information (*Storage Medium*) to the device.**Remote configuration:** The configuration at the "Remote Bootstrap Pattern" is done by downloading information from a *Bootstrap Server* to the device.**Registration phase:** In the registration phase, the device triggers a registration process on the *Back-end Server*. After the registration, the metadata provided by the device are registered in the *RegistryDevice*. Subsequently, an instance of the model of this device is stored in a *Device Model Database*. If the device, go through the configuration and the registration phases, it becomes able to create communication links with other connected objects. The exchange of messages between them is done through a communication intermediary (*Device Gateway*). A connected object (smart thing) can interact directly with the *RegistryDevice* to retrieve information about a device if it is offline. Case Study: Smart Home {#Sec6} ====================== To validate our approach, we chose to apply a case study in the IoT application domain called "Smart Home". A smart home is usually made up of remote-controlled automated components which can be doors, windows, lamps, etc. It can include several other components that can be monitored and controlled remotely. Most of these components can be controlled by a mobile device or a computer. In our case study, we add a new device (a camera) to a smart home. This device makes it possible to control the various rooms of the house. For example, if a door or a window left open, the camera informs the user immediately through a notification sent to their smartphone. It can also send alerts when it detects unknown faces. First, the camera is added without any information to initiate its first connection. We then apply the Meduim Based Bootstrap design pattern to have its configuration information. This information is inserted into the device through a memory card. Second, we go through the registration procedure on the main server (BackEndServer). This procedure is done through the use of the two patterns of the Registration Design Patterns category which are the automatic client driven registration pattern and the server driven model pattern which allow registering the device on the main server. Finally, the camera became able to communicate and create connection links with its communication partners. The camera communicates with the user's smartphone to notify him of what is happening in real-time.Fig. 6.Smart home case study We model this application through the use of the model shown in Fig. [6](#Fig6){ref-type="fig"}. The camera is associated with an object of type "Device", the memory card is defined as an object of type "Storage Medium" and the Smart phone is associated as an object of type "SmartThing". The propagation of events between objects is done through a *DeviceGateway*. Patterns Specification {#Sec7} ====================== UML, as semi-formal language offers several benefits to the definition of IoT design patterns, such as visual and standard notation. This graphical aspect is certainly interesting and useful to an architect, in the sense that graphic design is easy. However, the fact that UML lack a precise semantics is a serious drawback because this language did not allow checks which we must carry. So, pattern models generated at the modeling approach can be ambiguous and imprecise. In addition, during the modeling phase, the architect can easily fall into the error. This is due to the absence of a precise formal semantics of UML that do not provide rigorous tools for verification and proof. However, any error or any bad modeling of a design pattern can cause serious problems that generate bad consequences. Thus, ensuring the reliability and the correctness of IoT design patterns is a goal that we have fixed. For this, we propose an approach to formally specify design patterns by using the formal method Event-B that is well suited to our needs and goals. Thus, each diagram graphically modeled will be accompanied by a formal semantics. This approach allows the validation of the modeling part and ensure the verification of the relevant properties of design patterns. Event-B method is well-suited for specifying IoT design patterns: (i) The primary concept in doing formal developments in Event-B is that of a model. It is made of several components of two kinds: machines and contexts. Machines contain the dynamic parts of a model, whereas contexts contain the static parts of a model \[[@CR1]\]. Thanks to this classification, Event-B allows the specification of structural and behavioral features of design patterns. (ii) Refinement techniques proposed by this method allow us to build patterns gradually and at different abstraction levels. (iii) Mathematical proofs allow verifying model consistency and consistency between refinement levels. (iv) The most important reason to use Event-B method is the availability of a supporting tool called the Rodin platform \[[@CR2]\]. It is an Eclipse-based tool set that provides effective support for modeling and automated proof. The platform is open source and is further extendable with plug-ins. A range of plug-ins have already been developed including ones that support animation and model checking like the Prob plug-in \[[@CR5]\] that we used. Extended *Component* diagram that model structural features of design patterns are transformed to a context in the Event-B method in which we specify entities of the architecture and their relations. The *Sequence* diagram is transformed into a machine in Event-B in which we specify events made between entities of the patterns. This transformation is proposed in order to attribute formal notations to IoT design patterns for the purpose of checking their design correctness in a second step. We explicitly defined a refinement strategy to follow. This strategy is interesting because it defines the pattern development process and improves the quality of the obtained models, and therefore the success of the formal development process. We defined specification levels by using a step-wise development approach. Tool Support {#Sec8} ============ We developed a graphical modeling tool that implements our approach; it ensures an easy and efficient modeling way for users. With our tool, we aim to make concrete the aforementioned concepts. The architect can model the solution of the IoT design patterns using an Eclipse plug-in that we propose. The tool, in its development, is based on EMF[1](#Fn1){ref-type="fn"} (*Eclipse* *Modeling* *Framework*) \[[@CR10]\]. This was chosen since we use models, which are basic building units, to develop our approach (Fig. [7](#Fig7){ref-type="fig"}).Fig. 7.The tool editor Related Work {#Sec9} ============ Research connected to design patterns in the field of software architecture, are mainly classified into four branches of work according to their architectural style. The first is about design patterns for Object-Oriented Architectures, the second is about design patterns for Enterprise Application Integration (EAI), the third is for Service Oriented Architectures (SOA) and the fourth one is for connected object architectures. Most of the proposed design patterns are described with a combination of a text description and a graphical representation sometimes using a proprietary notation in the aim of making them easy to understand. However, these descriptions make patterns ambiguous and may lack details. Some work so have proposed the semi-formal representations of these patterns using modeling languages \[[@CR4]\]. Some other works use or provide formal languages based on mathematical notation for a precise pattern specification \[[@CR16]\]. However, these approaches require knowledge of mathematics and first order logic to use them. Some research has chosen to combine the semi-formal and formal representations of patterns. This representation ensures a better understanding and precision of patterns. Generally speaking, there is a consensus on the elements that make up and define a design pattern. However, there is no consensus on the specification of the patterns. In past work \[[@CR11], [@CR13]\] we focused on both the modeling, the formal specification and the composition of SOA design patterns \[[@CR12], [@CR14]\] and established the link between them with an automatic transformation \[[@CR15]\]. We used the SoaML language for the pattern modeling that ease the understanding of pattern models. For the pattern specification, we used the Event-B formal method in order to attribute formal notations to SOA design patterns for the purpose of checking their design correctness. In this work, we are interested with the IoT design patterns. In this context we find several researchers who proposed a set of IoT design patterns in various categories. Eloranta et al. \[[@CR3]\] proposed patterns for the construction of distributed control systems. Qanbari et al. \[[@CR6]\] presented four patterns for the supply, deployment, orchestration and monitoring shipboard applications. Reinfurt et al. \[[@CR7], [@CR9]\] have published patterns for device power supply, operation and communication modes and a number of IoT design models. All these patterns are described with a visual and informal notation. There is no formal semantics associated. There is no research work that deals with the modeling of IoT patterns. In this paper, we present the modeling of IoT design patterns proposed by Reinfurt et al. \[[@CR7]\]. Conclusions {#Sec10} =========== In this paper, we presented an approach that allows to model and specify connected object architecture design patterns. In particular modeling the "Bootstrapping Design Patterns" category and the "Registration Design Patterns" category. The modeling phase consists of presenting models of the design patterns in order to present a meta-model that presents an abstract view of a model of the patterns. Subsequently, we described the structural and behavioral features of the pattern. Then, we formally specified these design patterns using the formal Event-B method. Finally, we developed a plug-in under the Eclipse Modeling platform that offers a graphical editor for modeling IoT design patterns. Currently, the transition from the SoaML modeling to the formal specification is achieved manually, we are working on automating this phase by implementing transformation rules. <https://wiki.eclipse.org/Eclipse_Modeling_Framework>.

There is an error in affiliation 1 for authors Gabriella Lakatos, Márta Gácsi, and Boróka Bereczky. Affiliation 1 should be: MTA-ELTE Comparative Ethology Research Group, Pázmány Péter sétány 1/C, Budapest 1117, Hungary.

1. Introduction {#sec1} =============== In recent years, a control system based on related concepts such as human mind, thought, and consciousness has developed rapidly. The difference between this system and traditional computerized control systems is that it is based directly on the human brain, using electroencephalogram (EEG) as the basis for data analysis. This class is also known as brain-computer interface (BCI), in which the human brain controls the corresponding system \[[@B1]\]. EEG was first discovered by Richard in 1875 while studying the brains of exposed rabbits. In 1924, Hans published the first paper on scalp signal which can trace to EEG. Although electroencephalography has been found for more than a hundred years, in earlier studies, because the brain\'s thoughts, minds, and other activities typically produced signals with much lower amplitude than the EEG, most of the signals were submerged in spontaneous potentials \[[@B2]\]. Based on the original EEG alone, the researchers were unlikely to get "specific sensory and complex cognitive information" about the subject \[[@B3]\]. This is a major difficulty in EEG analysis. 2. Technical Background {#sec2} ======================= Recently, the breakthrough of biotechnology and the rapid development of computer technology make the EEG and brain-computer interface, which is in the interdisciplinary field, develop rapidly. With the discovery of key characteristics of EEG such as event-related potential (ERP) \[[@B4], [@B5]\], P300 potential \[[@B6], [@B7]\] and contingent negative variation (CNV) \[[@B8]\], the technology gradually moved from ideal to reality. In 1999, *Nature* published the first paper on brain-computer interface. As a landmark achievement at that time, this paper showed that slow cortical potential (CSP) could realize the construction of brain-computer interface for spelling function \[[@B9]\]. In the 21st century, EEG technology has achieved the evolution from multinode invasive to multinode noninvasive and then to single-node noninvasive. EEG devices are moving from the laboratory to the mass market. Single-node portable EEG equipment is a prevailing research field and has great potential in the future. At the core of many of these kinds of devices is the TGAM (ThinkGear Asic Module) and its encapsulate algorithm, with the difference being the Bluetooth version, the type of API, and how data is received by terminals. However, even though the single-node EEG device of the mind has been adopted by many research teams and faculty members around the world, due to the limited cost of technology development, the algorithm effect of calculating data is not satisfactory. There has been a lot of research using the TGAM, such as EEG control of wheelchairs to develop disability services \[[@B10]\], robotic hand controlled \[[@B11], [@B12]\], and brain training \[[@B13]\]. This device has a wide range of application fields. In addition to the traditional service equipment for the disabled, it has profound application potential and use value in the fields of game entertainment, VR environment, education \[[@B14]\], and human-computer interaction \[[@B15]\]. As early as 2014, there was a precedent for using the TGAM to test users\' brain waves in games \[[@B16]\] which only detect the change of beta wave. However, the beta wave value given by the paper is quite different from the output value of the TGAM, so it can be seen that the modification module did not play a role in this experiment. In particular, the application of EEG in VR is a hot topic in recent years. As early as 2012, the idea of building a hybrid system based on the advantages of EEG and VR appeared \[[@B17]\], and it can be used to optimize the sense of immersion in VR environment \[[@B18]\]. EEG must be an important part of the development of VR technology in the future. The original TGAM was available in 2012, and despite some algorithm updates over the next eight years, the output data was still disappointing. The stability and accuracy of this module are far less than those of research-level equipment, and the existing open-source optimization method is quite feeble; whether official or private, there is no optimization algorithm for this widely used module, which is a pity for all the research projects using it. 3. The Design of Data Transmission via the TGAM {#sec3} =============================================== 3.1. Data Transmission Method {#sec3.1} ----------------------------- The TGAM locks the wired transmission protocol from the inside and can only use Bluetooth for data transmission. The transmitted data is divided into large packets and small packets, which are all represented in a hexadecimal system in this paper. A packet, whether large or small, consists of three components: the packet header, the payload, and the checksum. The small packets contain 8 bytes, one for every 2 ms. The big packets contain 36 bytes, with one big packet transferred after every 512 small packets. The data in the big packet (except for the data used for synchronization and checksum detection) is the result of calculating the raw wave value (raw data in short) in the previous 512 small packets and exporting it. In the small packets, the value of the first two bytes are 0xAA for data synchronization, and the value of the third byte is 0x04, which means that the package is a small package with only 4 bytes of payload. The first three bytes are called packet headers, while data starts with the fourth byte (after which all data except the checksum is called Payload). The fourth (0x80) and fifth (0x02) bytes represent the code and the length (2 bytes) of raw data, respectively. Sometimes, data like 0x80 and 0x02 are easily considered as packet header. However, it is clearly pointed out in the official manual that the checksum verifies all payloads, so 0x80 and 0x02 which are included in the checksum calculation formula are not packet header data, and the same is true for large packet data. The two-byte length of the raw data is in the sixth (high eight bits) and seventh (low eight bits) bytes and is artificially set to transmit eight bits high and then eight bits low. Therefore, two bytes need to be spliced when the data is extracted, but the data needs to be further processed for the convenience of analysis. To obtain the new raw data with the absolute value within 32678, the data processing logic can be expressed as follows. The last byte is the checksum, the module\'s own data detection mechanism. Add up all the payloads, invert, and then take the lower eight bits, so that the value and the eight-bit checksum are equal that means the packet is extracted correctly. That is Equation ([1](#EEq1){ref-type="disp-formula"}):, $$\begin{matrix} {\text{Checksum} = \text{Low}\_\text{eight}\_\text{bits }\left( {\text{Sum}\_\text{of}\_\text{payload XOR }0\text{xFFFFFFFF}} \right).} \\ \end{matrix}$$ The ratio of the number of packets transferred to the number of packets with checksum errors is the packet loss rate. The official manual states that the packet loss rate below 10% will not affect the final result \[[@B19]\]. The big packet also starts with two 0xAA for data synchronization. The third byte is 0x20 for 32 payloads. The first three bytes are packet headers. The payload starts from the fourth byte. The specific meaning of the data is not described here. It is important to note that the eight types of brainwave data transmitted in the packet are all divided into three bytes (the lowest bit is transmitted first) and need to be manually spliced. The transmission characteristics of the wave value can be found in Yin et al.\'s research \[[@B20]\]; the corresponding description will not be elaborated in this paper. 3.2. Limitations of TGAM Encapsulate Algorithm {#sec3.2} ---------------------------------------------- The encapsulate algorithm mainly has three limitations. First of all, it may be the problem of development cost that TGAM, as a single-node EEG device, does not have good enough antinoise and data analysis step. No matter, in the experiment of this paper or the experimental data in some articles \[[@B21]\], the attention data calculated by the module\'s own algorithm is very unsatisfactory in practical application, with large fluctuations, high delay, and low accuracy ([Figure 1](#fig1){ref-type="fig"}). Some research projects have used large amounts of data to get baselines for attention and meditation to improve accuracy \[[@B22]\], but the results have been extremely limited. Tests have shown that the aperiodical dips of attention value are caused by skin movements in the prefrontal, which is usually caused by the subjects blinking their eyes. In addition, subtle changes in the subjects\' facial expressions can lead to irregular changes in EEG signals and sometimes even long periods of constant values, as well as spikes or drops in the difference of values above 50 (which has a range from 0 to 100). Additionally, although there are 32 bytes of payloads in the big package (24 bytes of eight kinds of wave value data), it is difficult to make regular analysis because the hardware is a single-node sampling. In that case, the focus of the data analysis is mainly attention and meditation \[[@B23]\]. Both of these values are integers from 0 to 100, but the criterion of meditation is not as practical as attention in most cases, which is the reason why most research work using the TGAM will only analyze attention. When we tried to find the relationship between attention and meditation, we found that these two values which are an output by the encapsulate algorithm of the TGAM were actually divided into many parts artificially ([Figure 2](#fig2){ref-type="fig"}), and this conclusion derives from a data set which contains over ten thousand samples from the experiment. Lots of meaningless empty value areas will increase the analysis cost. Specifically, the empty value areas are null region data points that do not appear in [Figure 2](#fig2){ref-type="fig"}. In other words, theoretically, there are 100^∗^100 = 10,000 kinds of data, because attention and meditation are all integers from 0 to 100. However, a large part of the data will not appear in experiments, and these areas that will not appear are called null value areas. The lack of stability in the data is partly due to the fact that 100 numbers divide the level of attention too much. Finding the attention-meditation relationship is not the first of its kind; a research had analyzed the relationship and used the wave value to make data correction \[[@B24]\]. But perhaps because of the small amount of data, the study did not find that the data had been artificially divided. Finally, the encapsulate algorithm does not have error detection steps, even if there are a checksum and poor signal, which is used to verify the accuracy of the payload. In particular, the checksum is a standard for developers to check whether the data they receive from a module is correct. In other words, as long as the data is extracted correctly, the checksum will never report an error (our experiment can support this point). Poor signal is an integer between 0 and 200; 200 means that the developers cannot receive any signals by the wrong manner of wearing. This parameter is used to determine the quality of the device. Although the poor signal is 0, wrong data will still be output. And there are three obvious problems with these two data (poor signal and checksum) that can be used for detection: The output of a module does not consider checksum and poor signal. In other words, the data, such as attention, output of the encapsulate algorithm at each moment does not consider whether the checksum or poor signal at that moment is normal or notAfter many tests, it can be seen that as long as the correct collection method is used, the checksum is almost always correct, which is of no essential help to data optimization. A total of 9698 groups of big packets were collected in one test, and none of them had incorrect checksumPoor signal does not change continuously. In fact, there are only a few values such as 26, 51, 55, and 200, with a large number of null value areas. In addition, although the signal noise is zero in the case of data receiving, there are still unstable situations in [Figure 1](#fig1){ref-type="fig"}. It is obvious that this parameter is extremely limited 4. Attention Optimization Method {#sec4} ================================ 4.1. The Pipeline of the Proposed Optimization Method {#sec4.1} ----------------------------------------------------- Our algorithm optimizes two kinds of packages (big package and small package) to improve the performance of the data. The optimization process is summarized as follows (refer to [Figure 3](#fig3){ref-type="fig"} for the overall optimization process): Add checksum verification; that is, check whether the data value of the packet is consistent with the checksum and improves the accuracy of data collectionCross-check the data with the previous big package. A big package with 36 bytes of data detects, but only attention and meditation will be detectedBecause the eight kinds of wave value are a big integer with no practical significance which are defined personally, the result will be bad if dealt with directly. Consequently, use the Isolation Forest method to detect an outlier (abnormal data points) from a large number of wave value which is collected from experiment, and then find out the upper bound of eight wave values after the regularization. If at the same time there are three or more values beyond the upper bound, the packet at this moment will be judged as a wrong package. In addition, Isolation Forest is also used to find the relationship between the attention value and the meditation value and draw the decision boundary of the value of concentration on the meditation valueBecause the blink signal will cause the instability of the raw data, which is manifested as a sudden drop in the attention value, the blink can be detected by the shake change of the small packets\' data. When the user blinks, the algorithm will stabilize the value of attention from abnormal range to normal range. Additionally, there is a link between how often people blink and how focused they are \[[@B25], [@B26]\]. Therefore, a step was set to detect the blink frequency and to further optimize the data to improve the return value of attentionThe output of attention is step by step, which means two constantly updated attention values are used to make the data with strong shake much smoother There are three most important steps of this optimization algorithm, and the reasons and principle will be introduced in the next sections: Time interval setting in big packet data extractionBlink feedback mechanismOutput classification (removal of meaningless null) 4.2. Big Package Optimization via Isolation Forest {#sec4.2} -------------------------------------------------- Before optimization, big packets\' data extraction is the basis of the whole optimization algorithm system. Through experiments and investigations, it is found that the use of various versions of the official development kit or open-source code often encountered difficulties in extracting big packets\' data. This is because when extracting, a code needs to wait a certain amount of time between each byte to be extracted. It is really important. Big packets lose data in a continuous read because of the defects of the original encapsulate algorithm. A large number of experiments have proved that if the pause time is not set, a large packet of data is successfully extracted every 3-4 seconds in general, but in theory, the large packet data is one for every 1026 ms. In order not to affect the accuracy of reading data, the pause time is smaller than the byte transmission time of large packets, and our algorithm takes 10 *μ*s. After the checksum check is completed, cross-check with the previous packet is the second step of optimization. In a data set with more than 10,000 samples (under normal wearing conditions), no group of data has the same attention and meditation values as the previous group, and the two data are the same only when there is a problem in data collection. Therefore, cross-check can effectively eliminate incorrect data due to hardware contact or noise problems. After that, the poor signal at that moment will be extracted, which is located in the fifth byte of big packets. Poor signal can directly reflect the wearing quality. After that, there are two boundary detection mechanisms. Our algorithm uses Isolation Forest to detect abnormal points and outline the decision boundary. Isolation Forest is an unsupervised anomaly detection method, which is suitable for continuous data. Isolation Forest uses a binary search tree structure (also known as "isolated tree") to isolate samples and detect outliers through the isolation of sample points, which is different from the anomaly detection algorithm that expresses the degree of alienation between samples by quantitative indexes such as distance and density \[[@B27]\]. Because most of the attention values in the TGAM are in the correct range, the number of outliers is small and there is a significant difference from most samples. In that case, the isolation of sample points will cause the outliers to be isolated earlier; that is, the outliers will be closer to the root node, while the normal values will be further away from the root node. This is an important reason why this optimization algorithm uses Isolation Forest for upper- and lower-bound analysis. Finding the relationship between meditation and attention using the Isolation Forest method ([Figure 4](#fig4){ref-type="fig"}) is the first boundary detection mechanism. In the figure, there are four colored lines representing the boundary drawn by data from four different situations, and the final average of the four was taken. The boundary in the figure is the same as the trend of boundary which is drawn artificially, which is enough to prove the rigor and accuracy of this machine learning method. Another boundary detection, specifically, is the detection of eight wave value data (respectively, Delta, Theta, LowAlpha, HighAlpha, LowBeta, HighBeta, LowGamma, mid-gamma); if more than three wave values fall outside the normal upper bound, it is judged as wrong data. Before the decision, the wave value will be regularized, because it is a meaningless huge integer, which is too difficult to analyze. After trying standardization, normalization, regularization, and other methods and comparing with the untreated values, it is found that regularization is a method with obvious differentiation. Moreover, regularization will make the data fall between 0 and 1, and the upper and lower bounds of analysis are more convenient and effective. The results are as follows ([Table 1](#tab1){ref-type="table"}). The big packet optimization logic is shown below. 4.3. Small Package Optimization with Region Segmentation of Blink Frequency {#sec4.3} --------------------------------------------------------------------------- The calculation method of the raw data in the small packets is mentioned in [Section 3](#sec3){ref-type="sec"}, and the range after parsing is between -32767 and 32768. Through the data set analysis, when the EEG signal is transmitted regularly, the raw data will be limited to ±550. The highest value is no more than 528, the lowest value is no less than -427, and most (over 95%) are within ±350. The raw data output from the device in eight seconds were randomly extracted ([Figure 5](#fig5){ref-type="fig"}) in a total of 4143 groups, and the obvious wave peaks were caused by the blink moment (7 times) of the tester. It can be seen that the raw data is quite stable when a blink does not happen, and the blink activity can be monitored only by detecting the fluctuations of the raw data. In this algorithm, 1000 is set as the difference of the raw data to detect the blink signal, and the huge fluctuation of the raw data represents the blink movement of the user. The blink frequency can be calculated by calculating the time difference between each two blinks, and the attention value can be improved by calculating the specific frequency value, so as to further optimize the attention value. The method of judging the blink of the raw data can be expressed as follows. The Blink_type above refers to the region where the blink frequency falls, which will be explained in more detail below. Regarding the relationship between blinking and attention, Chen demonstrated that blinking not only affects the voltage of the cortex (that is, affects the EEG signal), but also negatively correlates the concentration with the blink frequency by comparing the blink frequency between the two groups of subjects during meditation and under normal conditions \[[@B26]\]. In the experiment, Chen concluded that relaxation and concentration resulted in a 3.8-6.0 times/min blink rate difference between the same subjects in different environments and at different times. In addition, Maffei and Angrilli have designed experiments that measure how quickly people blink when they complete tasks of different difficulty \[[@B25]\]. The harder and stranger the task, the less likely participants were to blink. This study also supports the interesting idea that when people focus on one thing, their concentration increases, but their reaction time to other things decreases \[[@B28]\]. The same conclusion emerged in a study of computer users who blinked significantly less when they looked at a screen \[[@B29]\]. To sum up, it is feasible to optimize the attention data by blinking. The blink frequency was divided into five parts for analysis. The frequency selection was derived from the experimental results of Chen. The specific distinction is in [Table 2](#tab2){ref-type="table"}. The reason for setting too small and too large frequency as an invalid zone is to effectively avoid collection errors caused by hardware or wearing problems. The fluctuations in raw data caused by the blink itself can induce attention to plumb. In most cases, the blink will bring an 8-20 drop in the next attention value and after that, the next value will become regular. Therefore, even if the frequency is not in the focus zone, "blink compensation" needs to be set to keep the data close to the previous packet. According to the data analysis and experimental feedback of blink moment, a relatively lower feedback value of 10 is used as the cardinal number of blink compensation. Additionally, frequency and maximum values are used to constitute the entire blink optimization system. It should be noted that after the blink signal is detected, the system does not immediately modify the attention value but waits for the next big packet to arrive before making corrections. Doing this, relative to the direct update, fundamentally avoids the occurrence of the next abnormal attention value. 4.4. Tuning on the Attention Level {#sec4.4} ---------------------------------- As mentioned in [Section 4.1](#sec4.1){ref-type="sec"}, there is no specific difference between each individual attention-meditation interval, and meaningless null value area will increase the analysis cost. The instability of the data is also partly due to the fact that 100 numbers divide the attention level too much. In the end, attention classification is a necessary step to optimize data, enhance accuracy, and reduce analysis costs. It is different from the numerical classification in similar articles \[[@B30]\]. The algorithm classification system is essentially reflecting the TGAM itself way of classification. As shown in [Figure 3](#fig3){ref-type="fig"}, if not done classification in advance, cannot explain why the null value area so coincidence slices the image into many areas. Therefore, we remove the null value areas and divide the rectangular regions which are composed of blue data points in [Figure 3](#fig3){ref-type="fig"}. In this way, it can be divided into seven levels ([Table 3](#tab3){ref-type="table"}) according to the abscissa (attention). The reliability of this attention level is proved in [Section 5](#sec5){ref-type="sec"}; it is crucial. Among them, 0-4 and 98-100 actually have the attention value, but the algorithm still abandons them. For the former, this is because only 147 (1.52%) of the 9698 data sets are between 0 and 4, and more than half of them have nonzero poor signal. In addition, there are only two forms of these data: one is the continuous occurrence of multiple times with a huge difference with the data before and after. The other is the occurrence of only once with no connection with the data before and after as well. For the latter, only 100 will appear in the interval of 98-100, accounting for only 1.15% of the total number. The data will appear for several times in a row, and the difference with the previous data is more than 20 in most of times. Combined with the fact that attention data rarely appear continuously and do not plunge or surge under regular circumstances, two areas can be abandoned. 5. Experimental Results {#sec5} ======================= In order to prove the reliability of this optimization algorithm, the following experiment is designed. Record the change of the attention value within a certain period of time in two specific application scenarios, and select game and study as test scenarios to correspond to the situations of intermittent concentration and high concentration, respectively. The time is about 20 minutes. Without any processing, about 1170 groups of attention value (one for every 1026 ms) will be collected, and 50 groups will be randomly extracted for data analysis. Subjects should not have a history of mental illness or have taken psychiatric drugs and no history of epilepsy and other diseases that have a greater impact on brain fluctuations. On the night before the experiment, they were told to have a good rest and not to drink stimulants the day or the day before the experiment. For the experimental environment, ensure that it is bright and quiet and try to choose a comfortable environment for the tested personnel. The volunteers were required to close their eyes and rest before starting the test. The EEG data reached a low noise and stable state, and then, the test of the corresponding scene was started at the command of the staff. The data in [Figure 6](#fig6){ref-type="fig"} is all after big packet optimization, where Attention_Cal is the result of small packet optimization (blink feedback). Obviously, even without small packet optimization, the attention value is still more stable and accurate than the data in [Figure 1](#fig1){ref-type="fig"}. Big packet optimization has eliminated most of the wrong data caused by hardware, microexpressions, and environment, while small packet optimization makes the change of attention more gradual and precise, with almost no abnormal sharp rise or plunge. [Figure 7](#fig7){ref-type="fig"} shows the attention levels of situations. After grading, it can be seen clearly that the overall attention in study is much higher than that in game. And the grade distribution itself also conforms to the principle of normal distribution, so it can be seen that the results of this algorithm conform to the expected and subjective expectations of the volunteers. No change across two or more levels, or other abnormal changes, can be a stable reflection of the trend of attention value. 6. Conclusion {#sec6} ============= There are three most important steps of this algorithm, which are as follows: the time interval of the big packet data extraction setting, blink feedback mechanism, and classified output based on the removal of meaningless nulls. The beneficial effect of this algorithm is to improve the performance of the attention value output from the TGAM without improving the real-time data transmission delay. The adverse effect of blink on EEG data is optimized and in turn feedbacks the degree of attention through the blink frequency, which greatly improves the stability and accuracy. Although the effect is significant, there are still two important problems. First, EEG data is not as objective as traditional data. For example, in the magnitude of the velocity, we only need to calculate the ratio of displacement to time. However, as the objective reflection of thinking, the strong subjectivity and difficulty in describing thinking doomed the researchers to set subjective consciousness as the benchmark and create the analysis method of objective EEG data. Since the discovery of EEG in the 20th century, there has been a contradiction between objective data and subjective consciousness. The purpose of analyzing objective data is to reflect subjective consciousness, but is the existing objective data reliable enough to reflect real subjective consciousness? In the process of compiling and improving this algorithm, although the data optimization model has been fitted to the subjective consciousness of the testers as much as possible, the effect is not 100% satisfactory. This is not only a problem of this optimization algorithm but also a common problem of all EEG devices. Second, because the encapsulate algorithm of the module is unknown, wave value analysis is almost impossible to achieve. Although we have tried to find the relationship between the eight wave value data and the attention, meditation, and raw data, we have failed in the case that the raw data analysis method is sealed. The TGAM itself is a product for developers, with more open-source data and more flexibility than a purely commercial EEG device. We believe that if the raw data analysis algorithm can also be open source, our optimization can be further improved to get better results. Finally, the optimization algorithm is the improvement of software. In the field of electronics, the software and hardware should be combined to play the most important role. However, in the case of limited hardware conditions, this optimization algorithm also has certain practicability and promotion value. 7. Future Work {#sec7} ============== No matter how good the algorithm is, it needs to be verified by practical application \[[@B31]\], and it should be improved and perfected continuously in this process. So our next step is to explore the possibilities of our algorithm in application scenarios that require EEG analysis. Especially in VR applications, we believe that EEG analysis will make a difference. EEG and BCI technologies are beginning to take off, and it is believed that such research will yield more brilliant results. Furthermore, we would like to apply EEG into a more realistic practical research framework with crossmodal \[[@B32], [@B33]\] and multitask \[[@B34]\] in action recognition \[[@B31]\] and HCI in VR \[[@B35]\]. The study is funded by the General Program of Sichuan provincial science and Technology Department (2018JY0528). This work is part of the research supported by the National Nature Science Foundation of China under Grant No. 61602088 and No. 61976156 and the Fundamental Research Funds for the Central Universities No. Y03019023601008011. Additionally, the authors thank the Glasgow College, UESTC for the assistance and support. Data Availability ================= Please contact us by email if any data used in the manuscript is demanded. Conflicts of Interest ===================== The authors declare that there is no conflict of interest regarding the publication of this paper. {#fig1} {#fig2} {#fig3} {#fig4} {#fig5} {#fig6} {#fig7} {#alg1} {#alg2} {#alg3} ###### Threshold of wave value after regularizing. Wave name Delta Theta LowAlpha HighAlpha LowBeta HighBeta LowGamma MidGamma ---------------------------------- ------- ------- ---------- ----------- --------- ---------- ---------- ---------- Threshold value (regularization) 0.635 0.610 0.640 0.600 0.615 0.605 0.620 0.630 ###### Region segmentation of blink frequency. Frequency (second/once) Region Attention management ------------------------- ------------------- ----------------------------- \<2 Invalid area None 2-4.81 Normal area Plus 10 4.81-6.51 Focus area Plus (10 + frequency × 0.8) 6.51-8.33 Highly focus area Plus (10 + frequency × 1.5) \>8.33 Invalid area None ###### Attention classification method. Attention level Null value region Boundary Attention range Meditation upper boundary Meditation lower boundary ----------------- ------------------- ---------- ----------------- --------------------------- --------------------------- 0 0-4 5-6 0-4 0 1 9, 12, 15 18-19 7-8 74 28 10-11 77 26 13-14 83 23 16-17 87 19 2 22, 25, 28 31-33 20-21 90 16 23-24 93 14 26-27 97 11 29-30 97 10 3 36, 39, 42 45-46 34-35 97 10 37-38 97 13 40-41 94 10 43-44 94 10 4 49, 52, 55 58-59 47-48 91 10 50-51 88 10 53-54 88 10 56-57 87 10 5 62, 65, 68 71-73 60-61 84 10 63-64 84 10 66-67 81 10 69-70 81 16 6 76, 79, 82 85-86 74-75 78 20 77-78 78 20 80-81 75 24 83-84 69 27 7 89, 92, 95 98-100 87-88 69 23 90-91 67 27 93-94 64 29 96-97 58 33 [^1]: Guest Editor: Chenxi Huang

Introduction {#Sec1} ============ Type 2 diabetes mellitus (T2DM) is a long-term metabolic disorder with high morbidity in humans around the world. The prevalence of diabetes is increasing rapidly worldwide, including in China^[@CR1]^. In China, diabetes was estimated to affect 144.4 million people aged 20--79 according to the report of the international diabetes federation in 2017^[@CR2]^. The prevalence of diabetes in a rural population of Henan province is high which can be seen in the Henan Rural Cohort Study^[@CR3]^. Although diabetes is an irreversible disease, it is largely preventable^[@CR4]^. The risk of developing diabetes will be reduced through early detection and lifestyle interventions. For individual patient care, physicians are well prepared to identify those at risk for T2DM. However, when trying to screen thousands of patients with high-risk conditions, the challenges faced by physicians become apparent. There is a need for analytics techniques to assist in T2DM mass screening. Many risk scores based on statistical knowledge have been developed for predicting individual's risk of developing T2DM, such as risk evaluation formula^[@CR5]^, Archimedes trial-validated diabetes model^[@CR6]^, the diabetes risk score^[@CR7]^, genetic risk score^[@CR8]^, the New Chinese Diabetes Risk Score^[@CR42]^ and the American Academy of Family Physicians risk model^[@CR9]^. These methods made the implicit assumption that each risk factor was linear to the outcome. The complex relationships between nonlinear interaction factors might be oversimplified, leading to the potential loss of related information^[@CR10],[@CR11]^. Moreover, when the number of variables increased, the hypothesis testing method became complicated^[@CR12]^. In contrast to traditional methods, machine learning can learn the nonlinear interactions iteratively from large amounts of data using computer algorithms^[@CR13]^, which have been applied in various fields, such as disease risk assessment and prediction^[@CR14],[@CR15]^. Recent research shows that machine learning methods can describe patients' characteristics and identify patients at risk of developing T2DM^[@CR16],[@CR17]^. A study illustrated the performance of support vector machine for detecting persons with diabetes and pre-diabetes^[@CR18]^. To assess the ability to estimate the risk of developing T2DM, a study evaluated the performance of different machine learning and statistical techniques, and the experimental results showed the comprehensive performance the ensembles of ANN was better than other models^[@CR19]^. A data mining pipeline based on classification algorithm was built to predict T2DM complications based on electronic health record data from nearly one thousand patients, which showed the validity of machine learning method^[@CR20]^. An ensemble approach with the use of the vote method with three Decision Trees was developed to predict incident diabetes using 13 attributes^[@CR21]^, and improved the value of AUC to 0.922. A novel joint clustering and classification (JCC) method which could discover hidden clusters features in the patient samples was developed to predict diabetes, and the method performed best among the methods that were applicable to the interpretation of prediction^[@CR22]^. A study used neural network, decision tree, and random forest to predict diabetes mellitus with 14 attributes, and the results showed that the highest accuracy method was random forest^[@CR23]^. Another study compared the performance of several machine learning techniques to predict the risk of developing T2DM in short, medium, and long term, and the results showed that logistic regression outperformed in short, medium term while support vector machines presented better performance in long term^[@CR24]^. A machine learning-based framework for identifying subjects with T2DM from EHR was constructed via feature engineering, and the results revealed that the framework performed higher identification compared with the expert algorithm^[@CR25]^. However, the current methods just focused on performance comparison of prediction techniques with fixed number of variables, and they were also done on a small population sample. To date, there has been no large-scale investigation applying machine-learning for risk assessment in the general rural population. Therefore, the purpose of this study was to (1) evaluate an array of machine learning algorithms for predicting the risk of T2DM in a rural Chinese population; (2) identify the important variables, and (3) reveal the model performance of each model on a varying number of variables. Method {#Sec2} ====== Study participants {#Sec3} ------------------ The participants of this study came from the Henan Rural Cohort Study (Registration number: ChiCTR-OOC-15006699). A total of 39259 participants aged between 18 to 79 years were recruited from five rural areas in Henan province of China over the period between July 2015 and September 2017. The design and population characteristics of the study have been described in the previous articles^[@CR26]--[@CR28]^. Data on socio-demographic characteristics, information on physical examination, and laboratory test data were collected. Participants were excluded if they: (1) were diagnosed with kidney failure (N = 18) or cancer (N = 332); (2) had type 1 diabetes mellitus (N = 4); (3) had gestational diabetes mellitus (N = 634); (4) had incomplete information on diagnoses of T2DM (N = 63); and (5) had incomplete information of potential covariates (n = 2127). Finally, 36,652 participants were included for the present study. Definition of T2DM {#Sec4} ------------------ After excluding participants with type 1 diabetes mellitus, gestational diabetes mellitus, and other special type diabetes, T2DM was a self-reported previous diagnosis of diabetes by a physician or fasting plasma glucose level ≥7.0 mmol/L according to the American Diabetes Association (ADA) diagnostic criteria^[@CR29]^. Machine learning methods {#Sec5} ------------------------ We used logistic regression, artificial neural networks, classification and regression tree, support vector machine, and ensemble learning (random forest and gradient boosting machine) to build the risk assessment model. From the description of basic characteristics of non-T2DMs and T2DMs, the data is imbalanced. The model is likely to be biased towards the dominant class, with poor accuracy in classifying negative cases. In view of this problem, the Synthetic Minority Over-Sampling Technique (SMOTE)^[@CR24],[@CR30],[@CR31]^ algorithm was used to address the data. All models were constructed using the package sklearn (0.21.3) of Python 3.7 programing language. Artificial neural networks {#Sec6} -------------------------- Artificial neural networks^[@CR32]^ are computing systems that are based on the neurons of the human brain. ANN can learn all complex and non-linear interactions between variables to look for patterns in the data. ANN is divided into multi-hidden layer neural network and single hidden layer neural network. Each layer contains a number of neurons connected by directed arcs with variable weights. In our study, the neural network consists of three layers: an input layer to accept all risk factors, a hidden layer to process information and an output layer to calculate responses. Classification and regression tree {#Sec7} ---------------------------------- A decision tree is a tree structure in which each internal node represents a test on an attribute, each branch represents a test output, and each leaf node represents a category^[@CR33]^. Typical algorithms of decision tree include ID3, C4.5, CART, and so on. Considering the extensive application of CART in clinical and basic research, we used CART in this study^[@CR34]^. CART is a non-parametric decision tree learning technology, which generates a classification tree or regression tree according to whether the dependent variable is classified or numerical^[@CR35]^. Logistic regression {#Sec8} ------------------- Logistic regression (LR) is a generalized linear regression analysis model, which works to find the best fitting model that can describe the relationship between dependent variables and independent predictors^[@CR36]^. LR model is most widely used when people are interested in predicting disease or health status^[@CR37]^. The LR model can compute the probability of an individual developing T2DM based on the risk factors input. If a subject suffers from T2DM, the value of Y is 1; otherwise, Y is 0. We defined the probability of an individual developing T2DM is $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\,p(Y=1|X)=p(X)$$\end{document}$. Then, the formula of the LR model is defined as follows.$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$logit(p)=ln\,\left[\frac{p(X)}{1-p(X)}\right]={\beta }_{0}+{\beta }_{1}{X}_{1}+{\beta }_{2}{X}_{2}+\cdots +{\beta }_{k}{X}_{k}$$\end{document}$$ and equivalently, after exponentiating both sides:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{p(X)}{1-p(X)}={e}^{{\beta }_{0}+{\beta }_{1}{X}_{1}+{\beta }_{2}{X}_{2}+\cdots +{\beta }_{k}{X}_{k}}$$\end{document}$$ The probability of an individual developing T2DM is$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p(X)=\frac{{e}^{{\beta }_{0}+{\beta }_{1}{X}_{1}+{\beta }_{2}{X}_{2}+\cdots +{\beta }_{k}{X}_{k}}}{1+{e}^{{\beta }_{0}+{\beta }_{1}{X}_{1}+{\beta }_{2}{X}_{2}+\cdots +{\beta }_{k}{X}_{k}}}$$\end{document}$$Where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$X=({X}_{1},{X}_{2}\cdots {X}_{k})$$\end{document}$ represents the risk factors, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta =({\beta }_{1},{\beta }_{2}\cdots {\beta }_{k})$$\end{document}$ are the coefficients estimated by using the method of maximum likelihood. Support vector machine {#Sec9} ---------------------- Support vector machine (SVM) is a kind of generalized linear classifier that classifies binary data according to supervised learning. Its decision boundary is the maximum margin hyper plane for the positive and negative classes^[@CR38]^. In our study, each data sample is made of 60 features. The value of each feature is a vector of a particular dimension. Then, we used SVM to construct a hyperplane in a high-dimensional space, which can distinguish the two classes nicely. Ensemble learning {#Sec10} ----------------- Ensemble learning is an algorithm that combines basic learners such as decision trees and linear classifiers. The main idea of ensemble learning is to use multiple learning algorithms to achieve better performance than any constituent learning algorithm alone. Common types of ensembles are boosting, bagging, random subspace. Random forest (RF) is an algorithm combines bagging ensemble learning theory with random subspace approach. RF generates many decision trees for splitting data randomly at training time. For each node of the base decision tree, a subset containing K attributes is randomly selected from the attribute set of that node, and then an optimal attribute is selected from the sub-set for partitioning. Each tree provides a classification as a vote for each tree, and the RF ultimately chooses the classification with the most votes^[@CR39]^. Gradient boosting machine (GBM) is an iterative algorithm whose core idea is to train different classifiers (weak classifiers) for the same training set, and then combine these weak classifiers to form a stronger final classifier (strong classifier). Through a series of iterations to optimize the classification results, each iteration is introduced into a weak classifier, to overcome the existing shortcomings of weak classifier combination. GBM is based on the residual of training data fitted by the previous weak classifier to enhance the model when training each weak classifier. Compared with most learning algorithms, it is less prone to over fitting. Figure [1](#Fig1){ref-type="fig"} showed the methodology of this study. In this study, risk assessment models for T2DM were developed using 6 ML algorithms on all variables. Next, algorithms were iteratively introduced to a growing number of ranked variables (5/10/15/...) selected by the algorithm itself. All models were trained and tested by 10-fold cross-validation during each iteration process, which was repeated 100 times. Performance of all models was calculated on the test samples. All models' parameters were determined using 10-fold cross-validation and grid search on the training data (Supplementary Table [2](#MOESM1){ref-type="media"}).Figure 1Methodology. Abbreviation: LR, logistic regression; CART, classification and regression tree; GBM, gradient boosting machine; ANN, artificial neural network; RF, Random forest; SVM, Support vector machine; PPV, positive predictive value; NPV, negative predictive value; AUC, area under curve; AUPR, area under precision recall curve. Statistical analysis and evaluation on the model {#Sec11} ------------------------------------------------ Model performance: Discrimination refers to the model's ability to identify who is at risk of developing T2D and who is not. We used sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), area under precision recall curve (AUPR) and area under curve (AUC) to evaluate discrimination. Sensitivity is a synonym for recall rate, true positive rate, and represents the proportion of real positive samples that are identified correctly. For instance, in our study, the subject who diagnosed with T2DM was defined as 1, namely the positive sample. Otherwise, it was a negative sample (0). Specificity indicates the rate of real negative samples can be detected correctly. PPV stands for the proportion of positive results in diagnostic tests that are true positive results. NPV is the proportion of negatives in diagnostic tests that are true negative results. For binary classification models, AUC and AUPR were also used to evaluate the performance. Variable importance: For the diabetes study, we also listed the importance of variables. For LR and SVM models, variable importance was determined by the coefficient effect size. CART model estimated the variable importance by summing changes in the mean squared error due to splits on every variable and dividing the sum by the number of branch nodes. Variable importance of RF was estimated by permutation of out-of-bag variable observations. GBM computed the variable importance by summing these estimates over all weak learners in the classification ensemble method. ANN used overall connection weights of variable to filter variables^[@CR40]^. In order to combine the variable importance of each method, the variable importance was also estimated using shapley additive explanations approach, which is a unified approach to explain the output of any machine learning model^[@CR41]^. Categorical variables were described as percentages (%), and continuous variables were shown as mean ± standard deviation (SD). Differences in the characteristics of T2DM and Non-T2DM groups were determined with the chi-square test for categorical variables and t-test for continuous variables. All statistical analyses were performed using SPSS (v.21, IBM) and a two-tailed P value \< 0.05 was considered statistically significant. Ethics approval {#Sec12} --------------- Ethics approval was obtained from the "Zhengzhou University Life Science Ethics Committee", and written informed consent was obtained for all participants. Ethic approval code: \[2015\] MEC (S128). The present study was conducted in accordance with the guidelines of the Declaration of Helsinki. Results {#Sec13} ======= Basic characteristics {#Sec14} --------------------- The general characteristics of the study population were presented in Table [1](#Tab1){ref-type="table"}. The study population consisted of 14,375 men and 22,277 women. Compared with participants without T2DM, individuals with T2DM tended to be with higher age, BMI, waist circumference, heart rate, waist to height ratio, urine glucose, low-density lipoprotein cholesterol, and were more likely to have a family history of T2DM, hypertension, coronary heart disease. In contrast, among participants without T2DM, higher creatinine, higher high-density lipoprotein cholesterol, and higher uric acid were more common. Further details were presented in Supplementary Table [1](#MOESM1){ref-type="media"}.Table 1General characteristics of the study population.VariableTotal (n = 36652)Non-T2DM (n~1~ = 33296)T2DM (n~2~ = 3356)*P-Value*Age (years)55.60 ± 12.1755.11 ± 12.3260.51 ± 9.20\<0.001Men, n (%)14375(39.22)13114(39.39)1261(37.54)0.040Education, n (%)\<0.001   ≤Primary school16432(44.83)14567(43.75)1865(55.57)   Middle school14614(39.87)13507(40.57)1107(32.99)   ≥High school5606(15.30)5222(15.68)384(11.44)Marry, n (%)0.027   Married/cohabitating32927(89.84)29949(89.95)29877(88.74)   Divorced/widowed/unmarried3725(10.16)3347(10.05)378(11.26)Average monthly individual income, n (%)\<0.001   \<100025111(68.51)22709(68.20)2402(71.57)   1000\~8833(24.10)8083(24.28)750(22.35)   ≥20002708(7.39)2504(7.52)204(6.08)High fat diet, (≥75 g/day)7088(19.34)6544(19.65)544(16.21)\<0.001Sweet flavor, n (%)\<0.001   No15872(43.30)13495(40.53)2377(70.83)   Mild14217(38.79)13500(40.55)717(21.36)   Middle5720(15.61)5494(16.50)226(6.73)   Heavy843(2.30)807(2.42)36(1.07)Waist circumference (cm)84.13 ± 10.3383.62 ± 10.2289.32 ± 10.01\<0.001Body mass index (kg/m^2^)24.85 ± 3.5324.72 ± 3.4926.20 ± 3.62\<0.001Waist to hip ratio0.89 ± 0.070.88 ± 0.070.93 ± 0.07\<0.001Pulse pressure (mm Hg)48.25 ± 13.0847.72 ± 12.8553.45 ± 14.22\<0.001Heart rate (beats/min)75.72 ± 11.1275.34 ± 10.9479.54 ± 12.13\<0.001Total cholesterol (mmol/l)4.75 ± 0.974.72 ± 0.955.01 ± 1.11\<0.001Triglyceride (mmol/l)1.68 ± 1.121.64 ± 1.072.13 ± 1.44\<0.001HDL-C (mmol/l)1.32 ± 0.331.33 ± 0.331.23 ± 0.32\<0.001LDL-C (mmol/l)2.87 ± 0.812.85 ± 0.803.06 ± 0.93\<0.001Insulin (ug/l)10.85 ± 5.3010.69 ± 5.0412.51 ± 7.19\<0.001Creatinine (umol/L)62.07 ± 14.0062.31 ± 13.7559.61 ± 16.08\<0.001Uric acid(umol/L)286.50 ± 79.29287.77 ± 79.19273.87 ± 79.22\<0.001Urinary protein, n (%)1087(2.97)797(2.39)290(8.64)\<0.001Urine glucose, n (%)915(2.50)125(0.38)790(23.54)\<0.001Hypertension, n (%)11943(32.58)10225(30.71)1718(51.19)\<0.001Coronary heart disease, n (%)1620(4.42)1368(4.11)252(7.51)\<0.001T2DM history of mother, n (%)1070(2.92)813(2.44)257(7.66)\<0.001T2DM history of father, n (%)532(1.45)432(1.30)100(1.45)\<0.001Abbreviations: SD, standard deviation; HDL-C, high-density lipoprotein cholesterol; LDL-C, low-density lipoprotein cholesterol; T2DM, type 2 diabetes mellitus. Variable importance analysis {#Sec15} ---------------------------- The top 10 variables according to the variable importance of each algorithm were presented in Table [2](#Tab2){ref-type="table"} (Supplementary Table [3](#MOESM1){ref-type="media"}). Elevated urine glucose level was presented as top-ranked variables by all algorithms. Indicators of obesity repeatedly were appeared at the top of the list, such as waist to hip ratio, and waist to height ratio. This phenomenon confirmed that obesity is a risk factor of T2DM. Hypertension was ranked as an important factor of T2DM by most models, perhaps reflecting the relationship between hypertension and the development of T2DM. The risk factors in the New Chinese Diabetes Risk Score included sex, age, family history of diabetes, waist circumference, BMI, SBP. Several risk factors of the New Chinese Diabetes Risk Score (age, family history of diabetes, sex, and SBP) were shown in the list of top-ranked variables in our study. Common variables for diabetes were also identified by machine learning methods, such as genetic factors, hypertension, insulin, and so on. Also, new important variables (urinary parameters) were not found in previous risk prediction methods but determined by machine learning. Furthermore, the LR, SVM and ANN models prioritized genetic factor and urinary parameters, such as T2DM history of mother/father, urine glucose, urine protein, and so on.Table 2The top-10 ranked variables by the variable importance for each algorithm.RankMachine-learning algorithmsLRCARTGBMANNRFSVM1Urine glucoseUrine glucoseUrine glucoseUrine glucoseUrine glucoseUrine glucose2Diabetes history of motherSweet flavorSweet flavorDiabetes history of motherSweet flavorUrinary protein3Urinary proteinSour flavorWaist to hip ratioUrinary proteinWaist to hip ratioDiabetes history of mother4Diabetes history of fatherWaist to hip ratioHypertensionUrine latent bloodAgeDiabetes history of father5Urine ketone bodiesAgeMore vegetables and fruitsSweet flavorCreatinineUrine ketone bodies6HypertensionDiabetes history of motherAgeDiabetes history of fatherUric acidHypertension7Coronary heart diseaseWaist to height ratioUrinary vitamin CUrine ketone bodiesHeart rateCoronary heart disease8Low-density lipoprotein cholesterolInsulinUrine PHGenderInsulinLow-density lipoprotein cholesterol9Urine PHPulse pressureSour flavorSystolic blood pressureTriglycerideUrine PH10Urine nitriteHeart rateDiabetes history of motherHypertensionWaist to height ratioUrine nitriteAbbreviation: LR, logistic regression; CART, classification and regression tree; GBM, gradient boosting machine; ANN, artificial neural network; RF, Random forest; SVM, Support vector machine. We analyzed the importance of variables based on all models using the shapley additive explanations approach (Supplementary Table [4](#MOESM1){ref-type="media"}). As shown in Table [3](#Tab3){ref-type="table"}. Among the top-10 variables across all methods were sweet flavor, urine glucose, age, heart rate, creatinine, waist circumference, uric acid, pulse pressure, insulin, and hypertension.Table 3Variable ranking based on the mean rank of all models based on shapley additive explanations approach.ModelLRCARTGBMANNRFSVMMean rankFeature importance rankSweet flavor3214132.33Urine glucose5136213Age2425423.17Heart rate810410687.67Creatinine71369968.33Waist circumference4201171149.5Uric acid101971412711.5Pulse pressure1671011102012.33Insulin1281415181313.33Hypertension153291851115LR indicates logistic regression; CART, classification and regression tree; GBM, gradient boosting machine; ANN, artificial neural network; RF, Random forest; SVM, Support vector machine. Comparison of model performance {#Sec16} ------------------------------- Table [4](#Tab4){ref-type="table"} presented the comparison results of machine learning algorithms. Using all available variables, all models for predicting risk of T2DM demonstrated strong predictive performance, with AUCs ranging between 0.811 and 0.872. The GBM model performed best (AUC = 0.872 with laboratory variables), and also presented better specificity (81.71%), positive predictive value (28.83%), and AUPR (0.546). In terms of accuracy and negative predictive value, the data showed that RF model remained strong predictive performance (85.90%, and 97.52% respectively). ANN model's sensitivity was best among all models'. Using only non-laboratory data, such as BMI, age resulted in large declines in model performance. Moreover, the models with only non-laboratory data were also significantly better than the New Chinese Diabetes Risk Score^[@CR42]^ based on statistical knowledge only using non-laboratory data (AUC = 0.728, p \< 0.05) (Supplementary Figure [1](#MOESM1){ref-type="media"}).Table 4Performance of the machine-learning algorithms.LabModelAUCAccuracy(%)Sensitivity(%)Specificity(%)PPV(%)NPV(%)AUPR**With lab**LR0.841(0.825--0.858)75.2378.4974.9123.3797.280.493CART0.811(0.793--0.829)80.0666.9781.3325.9196.190.433GBM0.872(0.858--0.886)81.2076.0481.7128.8397.220.546ANN0.858(0.842--0.873)74.0180.9573.3422.8397.530.520RF0.868(0.854--0.883)85.9079.5778.1426.1997.520.538SVM0.835(0.818--0.851)76.4274.6576.5923.7196.880.490**No lab**LR0.804(0.787--0.821)75.0672.3575.3322.2396.550.313CART0.767(0.749--0.784)62.7979.2661.1816.6096.800.235GBM0.817(0.801--0.833)70.2878.9669.4320.1197.130.345ANN0.808(0.791--0.825)70.5278.0369.7920.1197.020.328RF0.803(0.786--0.820)70.7775.5870.3019.8796.730.327SVM0.800(0.783--0.818)76.4670.5177.0423.0396.400.316Abbreviation: LR, logistic regression; CART, classification and regression tree; GBM, gradient boosting machine; ANN, artificial neural network; RF, Random forest; SVM, Support vector machine. Figure [2](#Fig2){ref-type="fig"} displayed the receiver operator characteristic curves of each model with all variables. This visualization revealed that the GBM performed similarly to RF model, and the two models exhibited greater superiority than the ANN model (*p* \< *0.05*), with 0.872, 0.868 and 0.858 respectively. The above three models (GBM, RF and ANN) performed significantly better than CART (AUC = 0.11), LR (AUC = 0.841), and SVM (AUC = 0.835) (*p* \< *0.05*). The area under precision recall curve also showed the same results (Figure [3](#Fig3){ref-type="fig"}).Figure 2Receiver operating characteristic curve of different machine learning models. Abbreviation: LR, logistic regression; CART, classification and regression tree; GBM, gradient boosting machine; ANN, artificial neural network; RF, Random forest; SVM, Support vector machine.Figure 3Precision recall curve of different machine learning models. Abbreviation: LR, logistic regression; CART, classification and regression tree; GBM, gradient boosting machine; ANN, artificial neural network; RF, Random forest; SVM, Support vector machine. Model performance with a varying number of variables {#Sec17} ---------------------------------------------------- In order to compare the performance of different models on a varying number of variables, the highest 5/10/15... ranked variables of each model were consecutively incorporated into each model. As shown in Figure [4](#Fig4){ref-type="fig"}. Overall, as the number of variables increased, the graph showed the increase of AUC values except the CART model. Before introducing with 30 variables, the LR, GBM, ANN, SVM, and RF models presented dramatic rise trends on the values of AUC. Performance of the five models plateaued when introduced 30 variables to each model. After that, all trends showed slight fluctuation, but the changes were modest. The CART model maintained a constant trend of AUC value.Figure 4Performance variation of different models on a varying number of variables. LR indicates logistic regression; CART, classification and regression tree; GBM, gradient boosting machine; ANN, artificial neural network; RF, Random forest; SVM, Support vector machine. Discussion {#Sec18} ========== Using machine learning methods, this study developed several risk assessment models for characterizing the risk of developing T2DM. High predictive performance was achieved by all models, with AUCs ranging from 0.811to 0.872. Compared to other models, the GBM model performed the best, with an AUC value of 0.872 (95% 0.858--0.886) and the models' performance significantly better than the traditional risk score. In addition to common factors for diabetes, new important factors (urinary parameters) were not found in previous risk assessment methods, but determined by machine learning in our study. Our study demonstrated that machine learning technologies are uniquely positioned to identify significant risk factors in large-scale epidemiological studies. To our knowledge, this is the first study to assess the importance of variables and characterize the risk of developing T2DM with use of different machine learning methods in a Chinese rural population. Our results were consistent with the previous findings. The New Chinese Diabetes Risk Score showed that sex, age, family history of diabetes, waist circumference, BMI, SBP were important risk factors^[@CR42]^. Our results also revealed their prominent presence on the top-10 key factors for T2DM. Our data also indicated that obesity was a major risk factor for the development of T2DM^[@CR43]^. The previous studies have demonstrated the significant role of boosting method in other medical fields, such as urinary tract infections^[@CR44]^, hepatocellular carcinoma diagnosis^[@CR45]^, prediction of hip fracture^[@CR46]^. Our results confirmed the outstanding performance of the boosting method in the risk assessment of T2DM. Identifying the key factors is of great clinical significance in the risk assessment of T2DM. The severity of T2DM is often estimated through a lot of factors in different aspects, including socio-demographic characteristics, anthropometric measures and laboratory test data. Given that the diversity and massive of factors in the development phase of T2DM, it is difficult to choose a specific number of variables for risk assessment. Compared to models with no laboratory data, the inclusion of laboratory data resulted in significant increase in the identification capabilities of models. This phenomenon shows that adding effective laboratory data can help identify the risk of T2DM patients. Our study also showed that the importance of different factors depended on the modeling technique. For LR, SVM and ANN models, the genetic factors and urinary indicators, such as diabetes history of mother/father, urine glucose, occupied center stage in the risk assessment of T2DM. Moreover, our results showed 30--35 variables were needed when the model performance reached a plateau, and the model performance would not be improved with too many variables. Over the past decade, the ability to collect data has become faster and cheaper, but we need to pay more attention to the model with too many features. Data analysis techniques or model fitting ability are important in disease risk assessment and prediction. With the use of traditional statistical approaches, many risk scores and prediction models have been developed based on logistic regression. If the relationship among the data is linearly separable, traditional methods will fit better^[@CR47],[@CR48]^. If not, such models may oversimplify complex relationships among factors with nonlinear interactions, leading to the potential loss of important relevant information. This suggests the important of choosing an appropriate model according to the characteristics of the data set. According to the data characteristics of the Henan Rural Cohort Study, our results showed the boosting model fit data best. The performance of diagnosis model based on machine learning will be better if the number of training samples is large^[@CR49]^. Compared to previous studies, the major strength of our study was the relatively large sample size including 36652 subjects from the rural population in China. Also, we compared the model performance from two aspects: the fixed number of variables and the dynamic number of variables, which confirmed models with several variables could perform no worse than the model with all variables^[@CR50]^. Furthermore, the superiority and feasibility of nonparametric algorithms were proved compared with the model based on logistic regression. However, several limitations should be worth mentioning. Firstly, the research findings were derived from a cross-sectional study without follow-up data; therefore, we may not be able to determine the causal and temporal associations. Secondly, we need to do future research with external validation and other machine learning methods to assess the model performance. In addition, it's difficult to explain the inherent complexity of variable interactions and their impacts on outcomes due to the "black box" nature of machine learning methods. In conclusion, using a series of machine learning models, we developed a data mining approach to characterize risk ofT2DM and compared the model performance from the fixed number of variables and the dynamic number of variables. Our results showed the advantage ability of machine learning to identify risk factors and predict outcomes across a wide range of data and an increasing number of variables, which leading to greater insights on disease risk factors with no prior assumption of causality. Data sharing statement {#Sec19} ---------------------- All relevant data are within the paper and its Supporting Information files. Contact to Dr. Chongjian Wang (tjwcj2005\@126.com) for additional information regarding data access. Supplementary information ========================= {#Sec20} Supplementary information. **Publisher's note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Supplementary information ========================= is available for this paper at 10.1038/s41598-020-61123-x. The authors thank all of the participants, coordinators, and administrators for their support and help during the research. This research was supported by the National Key Research and Development Program Precision Medicine Initiative of China (Grant NO: 2016YFC0900803), National Natural Science Foundation of China (Grant NO: 81573243, 81602925, 21806146), Henan Natural Science Foundation of China (Grant NO: 182300410293), Science and Technology Foundation for Innovation Talent of Henan Province (Grant NO: 164100510021), Science and Technology Innovation Talents Support Plan of Henan Province Colleges and Universities (Grant NO: 14HASTIT035). The funders had no role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. Z.F.W. and C.J.W. conceived and designed the study. L.Y.Z., M.M.N. and Y.K.W. coordinated data collection. L.Y.Z. and Y.K.W. conducted the analyses. L.Y.Z. wrote the manuscript. All co-authors critically revised the manuscript. The authors declare no competing interests.

Background {#Sec1} ========== Mood disorders are chronic mental health conditions that cause a range of disabilities for patients, generating a negative impact on the individual, health systems and society \[[@CR1]\]. Due to their multifactorial etiology, mood disorders are known to be influenced by genetic, personal and/or environmental factors \[[@CR2], [@CR3]\]. While mood disorders are prevalent, a significant proportion of individuals with mood disorders go undiagnosed due to the spectrum of severity and prognosis \[[@CR4], [@CR5]\]. In this context, it is important to find ways to improve the identification of its risk factors, leading to appropriate treatment management and consequently preventing unfavourable outcomes. The etiology of mood disorders has been extensively studied, and some chronobiological factors have been found to play an essential role in the pathophysiology of mood disorders \[[@CR6]\]. For instance, previous studies have revealed that alterations in circadian rhythms are highly associated with major depression and bipolar disorder \[[@CR7]--[@CR9]\]. Specifically, altered variations of the clock genes, those involved with rhythmicity and timing of biological rhythms at a molecular level, have been found in individuals with bipolar disorder, as well as major depression \[[@CR10]\]. Recent meta-analytic studies have concluded that the abnormal sleep rhythms are consistently observed in patients with these two major mood disorders \[[@CR11], [@CR12]\]. These disturbances can be a potential predictor of declining mental health, as they have been shown to contribute to escalated mood levels and the triggering of manic episodes in patients \[[@CR13]\]. For instance, studies have shown that sleep deprivation and jet lag can trigger or aggravate depressive, hypomanic or manic episodes \[[@CR14]\]. In addition, studies have shown that discrete patterns of daily activity rhythms can distinguish specific mood disorder subgroups, such as bipolar depression and mania, or non-melancholic and melancholic depression \[[@CR15], [@CR16]\]. Notably, lower stability and weakened amplitude in rest-activity rhythms have been associated with greater symptom severity (e.g. impulsivity and mood instability) in individuals with borderline personality disorder \[[@CR17]\]. Therapies that target circadian rhythms synchronization might be useful in the management of mood disorders, such as bright light therapy and interpersonal and social rhythm therapy \[[@CR18]--[@CR20]\]. A better understanding of the rhythmicity of mood symptoms can help to identify individuals whose severity of mood symptoms follow an altered circadian rhythm. However, despite the increasing evidence linking mood disorders and circadian rhythms disruption, little is known regarding the rhythmicity of mood symptoms due to the lack of validated clinical questionnaires. In order to fill this gap, we have developed the Mood Rhythm Instrument (MRhI), a clinical tool aiming at assessing the self-perceived rhythmicity of mood symptoms. The MRhI is a self-reported questionnaire developed to evaluate the presence and timing of daily patterns for mood-related symptoms over the last 15 days. Each of the 15 items comprises a categorical and a continuous question. The original version of this instrument was created in Brazilian Portuguese \[[@CR21]\], which was then translated and validated in Spanish \[[@CR22], [@CR23]\]. In a large study with 708 participants that completed the MRhI, we found that the rhythmicity of specific mood-related symptoms and behaviors, such as pessimism and motivation to exercise were associated with higher risk for psychiatric disorders \[[@CR24]\]. Notably, we also found specific cultural differences in comparing Spanish and Brazilian samples in terms of the daily patterns of mood-related symptoms \[[@CR25]\]. These results are consistent with previous studies suggesting that cultural differences, as seen in different populations' sleep/wake habits \[[@CR26]--[@CR28]\], as well as ethnic differences \[[@CR29]\], and as racial differences in tau and circadian phase shifting, are relevant factors in circadian rhythm research. Thus, this study aims to validate the MRhI English version in an English-speaking Canadian sample. Methods {#Sec2} ======= Step 1. Translation of the Mood Rhythm Instrument (MRhI) {#Sec3} -------------------------------------------------------- The translation process is detailed in Fig. [1](#Fig1){ref-type="fig"} and was composed by five steps, including forward translation, correctness, back translation, back translation review and harmonization. Fig. 1Translation process of MRhI into English. The main steps for translation and cultural adaptation of Brazilian Portuguese into English are detailed The instructions on how to answer the questionnaire was updated with the removal of one sentence. In the translated version, the following sentence was written: "Answer the following questions according to the previous 15 days, taking into account how you have felt during most of the time, on the majority of the days and in the absence of any events that have caused you distress". In order to improve clarity, this sentence was changed to: "Answer the following questions according to the previous 15 days, taking into account how you have felt during most of the time". All sentences had their conjugation changed to the Present Perfect Tense, as the MRhI intends to assess the self-perceived rhythmicity of mood-related symptoms in the last 15 days. For question 11, the word "prone" was changed to "motivated". For question 12, the sentence "you memorize" was changed to "your memory". For question 15, the sentence "when you feel your best" was changed to "when you have had more energy and motivation to do things". Furthermore, instead of a 24-h format scale as it stands in the Portuguese version, the English version has an am/pm format scale. The final English version of the MRhI can be found as [Supplementary Methods](#MOESM1){ref-type="media"}. Step 2. Validation of the Mood Rhythm Instrument {#Sec4} ------------------------------------------------ ### Participants and procedures {#Sec5} Data collection was conducted between January 2016 and September 2018. We recruited the study sample through poster advertisements at McMaster University and St. Joseph's Healthcare Hamilton campuses, and online research recruitment within the Department of Psychology, Neuroscience & Behaviour at McMaster University. The final study sample comprised 401 individuals (age: 18--60; mean age: 22. ±7), predominantly women (72%), with a mean of 15 ± 3 years of schooling. The study was approved by the Hamilton Integrated Research Ethics Board and was conducted in accordance with the Declaration of Helsinki. All study participants provided written informed consent before study entry. ### Instruments {#Sec6} #### Mood Rhythm Instrument (MRhI) {#FPar1} The MRhI is composed of 15 items referring to physical, psychological and behavioural aspects related to mood, and each item provides a categorical (presence or absence of a daily peak) and a continuous (peak time in a 24 h period) variable. Subjects answered if, in the last 15 days, there was a specific time of the day when they experienced a peak in mood-related symptoms. We have recently completed a study where we tested the agreement rates between the MRhI and a daily version of the MRhI, and we found high agreement rates between the two instruments, thus suggesting that the MRhI may not be significantly influenced by memory bias \[[@CR25]\]. The MRhI displayed a satisfactory internal consistency (i.e. Cronbach's alpha) in Brazilian (0.73) and Spanish (0.70) populations \[[@CR21], [@CR23]\]. #### Reduced Morningness-Eveningness questionnaire (rMEQ) {#FPar2} The rMEQ provides a self-evaluation of chronotype, which is a unidimensional construct and offers a classification that varies between evening and morning types. This questionnaire was developed by Adan and Almirall \[[@CR30]\] and includes items 1, 7, 10, 18, and 19 of the original MEQ. These 5 questions comprise the smallest possible number of items that provides the maximum amount of information relative to the time when each individual feel to be more prone to perform daily activities and to sleep. Higher numbers indicate morning tendencies and lower numbers indicate evening tendencies (scores 4--11: evening type; 12--17: intermediate type; 18--25: morning type). The rMEQ has been widely used due to its practicality, allowing parallel recording of other variables, especially in large sample studies. The psychometric properties of the rMEQ has been evaluated in many countries of Europe, America, as well as in Kingdom of Saudi Arabia, China, India, Iraq and Iran. In most of these previous studies rMEQ showed similar values for internal consistency \[[@CR31]\]. ### Reliability and validity process {#Sec7} Internal consistency was measured with Cronbach's alpha. A Cronbach's alpha value between 0.7 and 0.9 was considered acceptable \[[@CR32]\]. Psychometric properties of MRhI-English were assessed through the exploratory factor analysis (EFA). The EFA was carried out using a tetrachoric correlation matrix since our data has a binary feature \[[@CR33]\]. Maximum Likelihood and Varimax were the extraction and rotation methods, respectively. Compared to other commonly used extraction methods, Maximum Likelihood uses the full information solution of the 2p contingency table \[[@CR34]\]. For practical implications, Maximum Likelihood is considered preferable for tests with few factors (stated as 1 to 3 factors), which is the main reason we opted for this specific extraction method \[[@CR35]\]. Satorra-Bentler corrected model estimation algorithm was used to surmount biased estimates. Varimax is a widely used rotation technique, being suitable for the present data as it shows excellent results by differentiating groups in several simulation scenarios \[[@CR36]\]. Factors extraction was initially performed through Velicer's minimum average partial (MAP) \[[@CR37]\], Horn's Parallel Analysis (PA) \[[@CR38]\], and Comparison Data (CD) \[[@CR39]\], obtaining 2, 5, and 3 factors respectively. As supported by Ruscio and Roche \[[@CR39]\], CD method performs better than MAP and PA in terms of accuracy and precision, with nearly unbiased results. Thus, the three-factor model was considered for the analysis. A confirmatory factor analysis (CFA) using Comparative Fit Index (CFI) \> 0.95, Tucker-Lewis Index (TLI) \> 0.95, Root mean square error of approximation (RMSEA) \< 0.06, and Standardized Root Mean Square Residual (SRMR) \< 0.08 as model fit indices were conducted \[[@CR40]\]. The CFA model fit indices showed suitable or slightly less than the good fit values (χ^2^ = 178.8, df = 87, CFI = 0.875, TLI = 0.850, RMSEA = 0.05, SRMR = 0.06). ### Statistical analysis {#Sec8} Variables were tested for normality by the Shapiro-Wilk test. Comparisons of the frequency of the dichotomous MRhI-English according to sex were analyzed by Chi-square test (χ^2^). Linear-circular correlations between time peaks of MRhI items and MEQ scores were performed \[[@CR41]\]. The distribution of MRhI-English items peaks were shown as a circular mean and compared between sexes according to Mardia-Watson-Wheeler test, considering that the data do not follow a normal distribution for circular data \[[@CR42]\]. R version 3.4.1 (package "Directional v3.3") and NCSS 12.0.9 were used for circular analysis. R version 3.4.1 (packages "psych", "lavaan" and "RGenData") and PASW Statistics Version 18 (SPSS Inc., Chicago, IL) were used for statistical analyses. Statistical significance was accepted at *p* \< 0.05. Results {#Sec9} ======= Reliability and validity of the English version of the mood rhythm instrument {#Sec10} ----------------------------------------------------------------------------- The MRhI-English presented a Cronbach's alpha of 0.75 in this sample, which suggests good internal consistency. Table [1](#Tab1){ref-type="table"} presents the three factors obtained with the factorial analysis of the categorical MRhI items. The first factor was predominantly composed by cognitive items such as *problem-solving, concentration, memory, talking to friends* and *energy.* Items related to affective aspects were in the second factor, e.g. *self-esteem, anxiety, sadness* and *pessimism.* Finally, the third factor grouped *alertness, sleepiness, irritability* and somatic items like *appetite*, *sexual arousal,* and *motivation to exercise*. Table 1Exploratory Factor Analysis of the English version of the Mood Rhythm Instrument items based on a three-factor solutionItemsFactor 1Factor 2Factor 31. Alertness0.430.16**0.45**2. Sleepiness0.260.2**0.72**3. Problem-Solving**0.76**0.090.284. Self-esteem0.15**0.45**0.445. Concentration**0.64**0.10.316. Appetite0.120.19**0.54**7. Sexual Arousal0.110.26**0.45**8. Irritability0.150.37**0.49**9. Anxiety0.1**0.66**0.2110. Sadness0.1**0.85**0.0111. Motivation to Exercise0.2−0.12**0.45**12. Memory**0.72**0.24013. Pessimism0.18**0.63**0.1414. Talking to Friends**0.32**0.310.1715. Energy**0.59**0.110.46Eigenvalues2.382.262.32% of variance0.160.150.15 The frequency of self-reported rhythmicity for each MRhI item and the comparison between sexes is shown on Table [2](#Tab2){ref-type="table"}. Items with the highest reported occurrence (\> 70%) of a daily peak were *alertness*, *sleepiness*, *concentration*, *appetite* and *energy*. On the other hand, less than 40% of subjects reported a daily peak in *self-esteem*, *sexual arousal*, *sadness*, *memory* and *pessimism*. The comparison between sexes showed that women reported a higher frequency of daily patterns in *irritability*, *anxiety*, *sadness* and *talking to friends*, while men reported in *problem-solving*, *sexual arousal* and *motivation to exercise*. Table 2Frequency of self-reported rhythmicity of the Mood Rhythm Instrument (MRhI) items -- English versionMRhI itemsTotal (*n* = 401)Men (*n* = 112)Women (*n* = 289)χ^2^, *p* valuen (%)n (%)n (%)Alertness313 (78)85 (76)228 (79)0.42, *p* = 0.52Sleepiness375 (94)109 (97)266 (92)3.71, *p* = 0.05Problem-solving273 (68)85 (76)188 (65)**4.36,*p*** **\< 0.05\***Self-esteem144 (36)40 (36)104 (36)0.00, *p* = 0.96Concentration318 (79)86 (77)232 (80)0.60, *p* = 0.44Appetite292 (73)82 (73)210 (73)0.01, *p* = 0.91Sexual Arousal125 (31)47 (42)78 (27)**8.44,*p*** **\< 0.01\*\***Irritability243 (61)58 (52)185 (64)**5.06,*p*** **\< 0.05\***Anxiety172 (43)34 (30)138 (48)**9.97,*p*** **\< 0.01\*\***Sadness157 (39)30 (27)127 (44)**9.98,*p*** **= 0.01\*\***Motivation to exercise248 (62)79 (70)169 (58)**4.97,*p*** **= 0.05\***Memory127 (32)37 (33)90 (31)0.13, *p* = 0.72Pessimism122 (30)30 (27)92 (32)0.97, *p* = 0.32Talking to Friends205 (51)43 (38)162 (56)**10.08,*p*** **\< 0.01\*\***Energy310 (77)86 (77)224 (78)0.02, *p* = 0.88Chi-square test; \**p* \< 0.05; \*\**p* ≤ 0.01 The comparison of MRhI time variables distribution according to sex is displayed in Rose plots (Fig. [2](#Fig2){ref-type="fig"}). The items did not vary between men and women (all *p* \> 0.05). Notably, *sleepiness*, *appetite*, *anxiety*, *motivation to exercise*, *pessimism* and *talking to friends* seemed to have a multimodal (i.e. more than one peak time in a 24 h period) pattern of occurrence. Fig. 2Rose plots for the Mood Rhythm Instrument (MRhI) time variables divided by groups of sex. The circle represents 24 h and each bin represents a 3-h period. The petals magnitudes are based on the proportion of participants of each group presenting a daily peak in the corresponding 3-h period. The bin width is divided equally by the two groups and the petals are laid out sequentially in the bin. The lines represent the circular mean for each group. Men and women do not differ in any MRhI item considering a level of significance of α = 0.05 (Mardia-Watson-Wheeler test for equal distributions) According to linear-circular correlation, all MRhI timing items were significantly correlated with rMEQ scores (Table [3](#Tab3){ref-type="table"}). However, we cannot establish if items are positively or negatively correlated to rMEQ scores due to the nature of a circular measure. Figure [3](#Fig3){ref-type="fig"} presents the frequency of which subjects responded with having a peak for MRhI items and the circular means of the reported peaks according to chronotype. The later the circular means appeared for cognitive and somatic items, the more eveningness the chronotype became (e.g. *alertness, problem-solving, concentration, appetite, motivation to exercise, memory, talking to friends* and *energy*). The opposite occurred with *irritability*, *sleepiness*, *anxiety* and *pessimism*. *Sexual arousal* and *sadness* did not seem to vary among the different chronotypes and showed little rhythmicity. Table 3Linear-circular correlations between time of peak of MRhI items and rMEQ scoresMRhI items (n)R-squared*p*-valueAlertness (309)0.290≤0.001Sleepiness (370)0.106≤0.001Problem-Solving (270)0.147≤0.001Self-esteem (142)0.206≤0.001Concentration (315)0.247≤0.001Appetite (289)0.028≤0.001Sexual Arousal (123)0.066≤0.001Irritability (241)0.203≤0.001Anxiety (170)0.089≤0.001Sadness (157)0.046≤0.001Motivation to Exercise (246)0.162≤0.001Memory (126)0.231≤0.001Pessimism (120)0.107≤0.001Talking to friends (203)0.0200.018Energy (305)0.221≤0.001*Abbreviations*: *rMEQ* Reduced Morningness-Eveningness QuestionnaireFig. 3Frequency and peak of each MRhI item. The circular mean of each 24-h peak for mood symptoms is depicted on the x-axis and frequency (%) is depicted on the y-axis Discussion {#Sec11} ========== The translation of the MRhI to English was adjusted to language and clarity. Importantly, the time scale was modified into the 12-h clock format for a cultural adaptation for the Canadian and most English-speaking countries. The Cronbach's alpha was 0.75, meaning that the items had an acceptable internal consistency and were adequate, similar to previous validation studies of the MRhI \[[@CR21], [@CR23]\]. The three factors solution grouped the items based on the nature of their features. Considering that the first factor explains the greatest percentage of the variance, items in this factor are considered to have an important contribution to the explained variance \[[@CR43]\]. It seems that in this Canadian sample, cognitive items are more important in assessing the profile of mood rhythmicity than affective and somatic ones. Yet, in the recent Spanish validation of MRhI, psychometric analysis showed that the first factor grouped somatic items except for *problem-solving* \[[@CR23]\]. In line with previous MRhI studies, cognitive and somatic items had more reported peaks than affective items \[[@CR21], [@CR23]\]. With the exception of *anxiety*, the affective items exhibited perceived peaks on less than 40% of subjects from the whole sample. When considering sex to compare the frequency of daily peaks, we found significant differences in *irritability*, *anxiety* and *sadness*, which were more frequently reported by women. Notably, these results are consistent with epidemiological data pointing to a higher prevalence of mood and anxiety disorders in women, which is corroborated by biological \[[@CR44]\] and socioeconomic \[[@CR45]\] contributors. Another factor that may be related to the sex differences observed are alexithymic behaviors, which are more prevalent in men, resulting in a lack of report of negative emotions in this population \[[@CR46]\]. Women also reported a higher frequency of daily peaks for *talking to friends*, which is in accordance to a previous British study that evaluated social interaction components of circadian rhythms through phone calls monitoring. Results showed that, comparatively to men, women spent more time on calls with friends in the evening and at night \[[@CR47]\]. Interestingly, in terms of sexual arousal, men reported more frequency of sexual arousal peaks than women, which is exactly the opposite from what we observed in Spanish population \[[@CR23]\] and also distinct from Brazilians that did not display differences in sexual arousal peak between women and men \[[@CR21]\]. Moreover, *problem-solving* and *motivation to exercise* were more frequently pointed as having a daily peak in men. This result can be in part supported by the fact that women are more prone to be extrinsically motivated (with expectations to gain rewards or outcomes) to exercise than intrinsically motivated (aiming personal satisfaction and/or enjoyment), resulting in less motivation for regular physical activity in comparison to men \[[@CR48]\]. In contrast to the Brazilian sample, where *alertness* was the only item that differed between sexes, with women reporting more frequently to have a peak than men \[[@CR21]\], no differences with regards to *alertness* were found in the Canadian sample. Overall, we observed that the Canadians reported more sex differences with regards to frequency of perceived peaks than the Brazilian sample. Regarding negative mood and somatic symptomatology, women reported more frequent peaks than men (irritability, anxiety and sadness), while for positive cognitive and somatic activity behaviors men reported more frequent peaks than women (sexual arousal, problem solving and motivation to exercise). As aforementioned, higher prevalence of mood symptoms in women possibly contrasts observations related to the affective items in a sample mostly composed by them. Thus, future research exploring these factors which controls for the relationship between sex and psychiatric symptoms shall bring valuable insights related to the sex differences observed. The time when items peak did not differ between men and women in any of the MRhI items. Considering that participants could only choose one time peak, even though men reported to have a *sexual arousal* peak in the morning, pointed morning peaks were much more frequent as among women and, therefore, circular means were similar between women and men. A Polish study reported that women had evening peaks of "greatest need for sex", whereas men had both morning and evening peaks \[[@CR49]\]. This multimodal pattern of occurrence was also identified in *appetite* and usually varies among breakfast, lunch and dinner \[[@CR50]\]. Subjects also reported peaks of *sleepiness* in the morning, right after midday and at night which is in line with previous analyses of sleepiness expression \[[@CR51]\]. As we expected, the circadian typology, measured by means of chronotype, is significantly correlated to all MRhI items time peaks \[[@CR52]\]. Participants classified as morning types reported earlier peak times for *concentration*, *alertness*, *problem-solving*, *energy*, *memory*, *motivation to exercise* and *self-esteem*. In contrast, individuals classified as evening types reported that these items peaked later in the day. These results are consistent with previous studies from Europe and United States showing that individuals with morning chronotype performed better in terms of attention, alertness and working memory in the morning and afternoon when compared to individuals with an evening chronotype \[[@CR53], [@CR54]\]. Also, depending on the type of problem to solve (e.g. insight or analytic), individuals classified as having a later circadian arousal perform better during later afternoon sessions (between 4 pm and 5:30 pm) \[[@CR55]\]. Cognitive performance has been shown to be correlated with individual's body temperature rhythm. Wright et al. \[[@CR56]\] found that cognitive tasks were performed better when body temperature is high and near its circadian peak. Alongside these findings, another study found that individuals with a morning chronotype have earlier temperature rhythms, thus their peaks in cognitive performance such as memory and alertness would also occur earlier than those with evening chronotypes \[[@CR57]\]. Our data revealed that individuals with evening chronotype reported earlier peaks of *sleepiness* compared to individuals with morning and intermediate chronotypes. However, it is possible that this finding does not reflect spontaneous behaviors, but rather the consequences of sleep disruption related to diurnal social demands, a hypothesis that is endorsed by the same finding regarding *irritability, pessimism* and *anxiety* (earlier in evening chronotypes). Due to the mean age of 22.2 ± 7 years of our sample, which in general is related to later chronotypes \[[@CR52]\], the opposite maladaptation for nocturnal activities could be observed in *talking to friends*. This item peaked at a similar time for intermediate and evening types, and earlier in morning types. Finally, in our sample *sexual arousal, appetite* and *sadness* displayed little variation between chronotypes. This study has some limitations. We are aware that the MRhI does not reflect mood rhythms independently of external or social factors, as responsibilities and schedules of participants probably influence their responses. However, it is of our interest to evaluate individuals' self-perception of rhythmicity of mood-related symptoms when inserted in a real-life setting, rather than assessing internal rhythm alone. External factors that exist in a person's environment are intricate experiences that can also influence how symptoms of mood disorders present themselves. Another limitation is that only one external validation measure was used, albeit it is a well-established questionnaire to evaluate chronotype \[[@CR30]\]. Longitudinal monitoring of cognitive, affective, and somatic symptoms using Ecological Momentary Assessment methods should also be considered when validating instruments like the MRhI. Conclusions {#Sec12} =========== In conclusion, the results obtained with the English version of the MRhI are consistent with previous chronobiology studies, suggesting that this instrument might be useful to enhance the knowledge of self-perceived daily patterns of mood-related symptoms. The Cronbach's alpha analysis suggests good internal consistency of this instrument. Cognitive, affective and somatic items presented different frequency of reported peaks and regarding its timing, they seemed to behave accordingly to chronotype. The future directions will be the use of the MRhI instrument in a large sample of individuals with mood disorders, aiming to provide a better understanding of the relationship between daily patterns of mood variability and mental disorders. Supplementary information ========================= {#Sec13} **Additional file 1.** MRhI : Mood Rhythm Instrument rMEQ : Reduced Morningness-Eveningness Questionnaire MRhI-English : English version of the Mood Rhythm Instrument MAP : Minimum average partial CD : Comparison Data CFA : Confirmatory factor analysis TLI : Tucker-Lewis Index RMSEA : Root mean square error of approximation **Publisher's Note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Melissa A. B. Oliveira and Kristina Epifano contributed equally as first authors Supplementary information ========================= **Supplementary information** accompanies this paper at 10.1186/s40359-020-00397-2. M.A.B.O., A.C., A.P.F., A.A., M.P.H. and B.N.F. designed the study. M.A.B.O., K.E., M.S., S.M., F.G.C., A.C., A.P.F. and L.L.S.G. collected and organized the data. M.A.B.O, K.E., M.S., F.G.C., A.C. and M.P.H. analyzed the data. M.A.B.O., K.E., M.S., A.C., L.L.S.G., M.P.H and B.N.F. wrote the first draft of the manuscript. All authors have read, revised and approved the final manuscript. Financial support was provided by CAPES (MO, FC), Propesq-UFRGS (MS), PQ-CNPq (MH), PVE-CNPq and PRINT-CAPES (BF), the Spanish Ministry of Economy, Industry, and Competitiveness (grant PSI2015--65026; MINECO / FEDER / UE) (AA). The funding bodies had no role in the study design, data collection, analysis, or interpretation, or in the writing of the manuscript. All data generated or analyzed during this study are included in this published article. All individuals who agreed to participate provided written informed consent. The study was approved by the Hamilton Integrated Research Ethics Board and was conducted in accordance with the Declaration of Helsinki. Not applicable. The authors declare that they have no competing interests.

INTRODUCTION ============ Core build-up materials are often required to reconstruct and provide an ideal morphology to severely damaged teeth prior to their preparation for indirect foundation restorations. Despite substantial documented evidence of the long-term success of large amalgam restorations^[@r26],[@r40]^, resin composites, since the early days of self-cured materials, have also been used for this purpose. More recently, light-cured core build-up materials that are more convenient to use than chemically cured composites have been widely used^[@r33]^. Both these materials, however, have their disadvantages^[@r12],[@r29]^. While chemically cured materials do not allow clinicians to adjust the setting time individually, light-cured resin composites do not ensure adequate polymerization in areas with limited access to the curing light. However, resin-based composites are associated with polymerization shrinkage, causing stress development under confined conditions^[@r36]^. Several strategies have been used to overcome the limitations of polymerization shrinkage. Various layering techniques have been suggested to minimize shrinkage stress^[@r20],[@r23]^. Nevertheless, time limitations when placing core build-up materials restrict clinicians from using elaborate multi-layering techniques. Therefore, dual-cured resin composites, in which polymerization is chemically initiated in the deeper portions of the canal or preparations, have been developed for use as core build-up materials; this has allowed clinicians to build extended foundation restorations quickly, and in bulk. At the coronal region, dual-cured core build-up resin composites are mainly polymerized through photo-initiated reactions, whereas, in the apical region, polymerization is chemically initiated. However, the incorporation of self- and light-curing modes in the same material does not ensure maximal curing of the material. Due to incomplete compensation for deficient light activation, lower hardness values of dual-cured core build-up resin composites have been observed with increased depth of cavity^[@r01]^. It has been speculated that a delay in light activation would be beneficial in enhancing the degree of conversion of dual-cured resin cements, since immediate exposure to light could interfere with the chemical-curing mechanism^[@r24]^. On the other hand, it has been reported that time delay and duration of light exposure does not increase microhardness at different depths of a dual-cured core build-up resin composite 2 weeks after light irradiation^[@r39]^. Moreover, light-activation delayed by 5 minutes after seating the fiber-reinforced post did not affect the microhardness of dual-cured resin cements at 3 regions of the root after 3 months of storage in water^[@r28]^. Although dual-cured core build-up resin composites have been recently used to prepare prefabricated posts and core or coronal-radicular build-ups, their hardness behavior at greater depths of cavity is unknown. Moreover, no information is available in the literature regarding the effect of applying a bonding agent on the cavity wall on polymerization of the dual-cured core build-up resin composite. Based on these considerations, the purpose of the present study was to investigate the influence of light-exposure durations (20, 40, and 120 seconds) and application of a bonding agent on the extent of polymerization by measuring the microhardness of 2 dual-cured core build-up resin composites at different depths of cavity without prefabricated posts and various post-irradiation times. The following null hypotheses were tested: (1) an increase in light-exposure time results in no difference in hardness, regardless of the depth of cavity and type of dual-cured resin composite; (2) hardness is not affected by depth of cavity and post-irradiation time; and (3) application of the bonding agent does not improve hardness. MATERIAL AND METHODS ==================== Specimen preparation -------------------- Forty semi-cylindrical cavities with a diameter of 3 mm and a depth of 11 mm were prepared in 5x10x16 mm acrylic resin blocks ([Figure 1](#f01){ref-type="fig"}). Two acrylic resin blocks, with or without a semi-cylindrical cavity, were placed in a silicon impression material mold (15x15x20 mm). Either 1 of 2 dual-cured core build-up resin composite pastes \[Clearfil DC Core Plus (DCP) or Unifil Core EM (UCE) ([Figure 2](#f02){ref-type="table"})\] was filled directly in the cavities using auto-mixing tips, being sure to avoid entrapment of air, according to the manufacturer\'s instructions. The upper surface in the resin composite material was covered with a plastic strip and pressed with a thin cover glass to remove any excess resin. Light-irradiation was provided by placing the tip of the LED light unit (power density: 1,000 mW/cm^2^; Pencure, J. Morita MFG Corp., Kyoto, Japan) on the plastic strip. Power output was verified with a curing radiometer (Cure Lite; Dentsply Caulk, Milford, CT, USA) immediately before every light-activation throughout the study. The core build-up resin composites were light-cured for different durations using 1 of the 4 following light-exposure methods: (1) light-irradiation for 20 seconds on the plastic strip after filling the core build-up resin composite (20 s method); (2) light-irradiation for 40 seconds (40 s); (3) application of the bonding agent on the cavity wall with a brush, followed by air-drying, and light-irradiation for 10 seconds on the top of the cavity before the 20 s method (B+20 s method); (4) light-irradiation for 40 seconds on the plastic strip, followed by removal of the acrylic blocks from the silicon impression material mold, and irradiation of both sides of each acrylic resin block for 40 seconds each (120 s method). After irradiation of all the specimens, the acrylic resin blocks were removed from the silicon mold and separated. Five specimens of each dual-cured core build-up resin composite were irradiated by each of the 4 methods. {#f01} ###### Dual-cured core build-up resin composites used in this study **Materials** **Manufacturer** **Composition** ----------------------------------------------- ----------------------------------------------- -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Core material     Clearfil DC Core Plus (DCP) (Batch \# 0002AA) Kuraray Noritake Products Corp., Tokyo, Japan A paste: Bis-GMA, hydrophilic aliphatic dimethacrylate, hydrophobic aliphatic dimethacrylate,hydrophobic aromatic dimethacrylate, silanized barium glass filler, silanized colloidal silica, colloidal silica, chemical-initiator, photo-initiator, pigments     B paste: TEGDMA, hydrophilic aliphatic dimethacrylate, hydrophobic aromatic dimethacrylate, silanized barium glass filler, silanized colloidal silica, aluminum oxide filler, photo-accelerator , chemical-accelerator Bonding material     Clearfil S^3^ Bond Plus (Batch \# 00024B) Kuraray Noritake Products Corp., Tokyo, Japan Bis-GMA,HEMA, MDP, hydrophilic dimethacrylate, hydrophobic dimethacrylate, colloidal silica, ethanol, water, photo-initiators, photo- and chemical-accelerator, sodium fluoride Core material     Unifil Core EM (UCE) (Batch \# 1107011) GC Corp., Tokyo, Japan Base: UDMA, dimethacrylates, fluoro-alumino-silicate glass, silicon dioxide, photo-initiator, accelerator     Catalyst: UDMA, dimethacrylates, fluoro-alumino-silicate glass, silicon dioxide, chemical-initiator, pigment Bonding material     Self-etching bond A (Batch \# 1205181) GC Corp., Tokyo, Japan dimethacrylates, 4-META, silicon dioxide, ethanol, water, photo-initiator Self-etching bond B (Batch \# 1206011) GC Corp., Tokyo, Japan ethanol, accelerator Bis-GMA: bis-phenol-A-glycidyldimethacrylate; TEGDMA: triethyleneglycol dimethacrylate; UDMA: urethane dimethacrylate; HEMA: 2-hydoxyethyl methacrylate; MDP: 10-methacryloyloxydecyl dihydrogen phosphate; 4-META: 4-methacryloyloxyethyl trimellitate anhydride Hardness measurements --------------------- Hardness was measured at the following depths from the light-irradiated surface of the cavity: 0.5, 2.0, 4.0, 6.0, 8.0, and 10.0 mm. For each specimen, the Knoop hardness number (KHN) was measured 5 times at each depth using a microhardness tester (FM-700: Future-Tech Corp., Kawasaki, Japan) at 0.5 hours, 24 hours, and 7 days post-irradiation. A Knoop diamond indenter was applied under a load of 25 g for a dwell time of 15 seconds; the load was then removed, and the long diagonal of the indentation was measured under 400x magnification. KHN, which is inversely proportional to the square of the long diagonal, was thus calculated. All specimens were stored under dry and dark conditions in a box, which was placed in a biochemical incubator at 37ºC to avoid exposure to light, and was accessed only to obtain measurements. Statistical analysis -------------------- The KHN data were statistically analyzed by the repeated measures three-way ANOVA test. The independent variables analyzed were depths of cavity and post-irradiation times for within-subject analysis, and light-exposure methods and type of dual-cured core build-up resin composite for between-subject analysis. A one-way ANOVA with the *post-hoc* Tukey\'s compromise test was used to establish specific differences in KHN values between the groups (α=0.05). RESULTS ======= The results of this study are summarized in [Tables 1](#t01){ref-type="table"} and [2](#t02){ref-type="table"}, which show the mean KHN and standard deviation of all the experimental groups in DCP and UCE, respectively, at 6 depths of cavity after 3 different post-irradiation durations by 4 different light-exposure methods. ###### Mean KHN±SD for Clearfil DC Core Plus (DCP) at 6 depths of cavity after 3 post-irradiation times by 4 light-exposure methods **Exposure method** **Post-irradiation time** **Depth of cavity (mm)** ------------------------------------------ --------------------------- -------------------------- ------------------ ------------------ ------------------ ------------------ ----------------- Exposure time of 20 s (20 s) 0.5 hours 43.48±0.56^αaC^ 39.89±0.92^αbC^ 37.56±0.40^αcC^ 35.89±0.73^αcdC^ 34.88±1.27^αdC^ 33.08±1.71^αeC^ 24 hours 45.76±0.74^αaB^ 41.13±0.68^αbB^ 39.39±0.32^αcB^ 39.19±0.31^αcAB^ 38.67±0.21^αcAB^ 38.59±0.12^αcAB^ 7 days 46.79±0.59^αaA^ 42.94±0.59^αbA^ 41.62±0.28^αbA^ 40.03±1.14^αcA^ 39.62±1.28^αcA^ 39.04±1.07^αcA^ Bonding + exposure time of 20 s (B+20 s) 0.5 hours 41.71±0.54^βaC^ 38.63±0.60^αbC^ 37.67±0.58^αbcC^ 37.07±0.58^αcC^ 35.49±0.98^αdB^ 35.01±0.90^αdB^ 24 hours 46.11±0.59^αaB^ 43.18±0.75^βbB^ 41.84±0.62^βcB^ 41.70±0.49^βcB^ 41.34±0.82^βcA^ 41.17±0.84^βcA^ 7 days 47.50±0.25^αaA^ 45.51±0.95^βbA^ 43.34±0.83^βcA^ 43.35±0.88^βcA^ 42.54±0.70^βcA^ 42.24±0.90^βcA^ Exposure time of 40 s (40 s) 0.5 hours 48.83±0.36^γaB^ 44.42±0.33^βbC^ 42.83±0.50^βcC^ 40.83±0.87^βdC^ 38.60±0.58^βeC^ 37.57±0.29^βfC^ 24 hours 51.89±0.61^βaA^ 49.04±0.66^γbAB^ 47.57±0.57^γcAB^ 46.73±0.46^γcAB^ 44.76±0.59^γdAB^ 43.98±0.58^γdAB^ 7 days 52.30±0.53^βaA^ 49.47±0.52^γbA^ 47.92±0.76^γcA^ 47.03±0.51^γcA^ 45.36±0.68^γdA^ 44.76±0.77^γdA^ Exposure time of 120 s (120 s) 0.5 hours 48.57±0.51^γaC^ 48.74±0.40^γaC^ 48.84±0.34^γaC^ 48.95±0.26^γaC^ 48.98±0.15^γaC^ 48.90±0.26^γaC^ 24 hours 52.40±0.25^βaB^ 52.35±0.27^δaB^ 52.20±0.24^δaB^ 52.40±0.28^δaB^ 52.67±0.26^δaB^ 52.26±0.50^δaB^ 7 days 53.06±0.19^βaA^ 53.03±0.21^δaA^ 52.98±0.09^δaA^ 53.00±0.28^δaA^ 53.27±0.33^δaA^ 52.92±0.26^δaA^ The same lower-case superscript Greek characters indicate no statistically significant differences between exposure methods at same post-irradiation time in the same depth of cavity; the same lower-case superscript letters indicate no statistically significant differences between depths of cavity at same post-irradiation time in the same exposure method; the same upper-case superscript letters indicate no statistically significant differences between post-irradiation times at same depth of cavity in the same exposure method (p\>0.05) KHN=Knoop hardness number ###### Mean KHN±SD for KNH for Unifil Core EM (UCE) at 6 depths of cavity after 3 post-irradiation times by 4 light-exposure methods **Exposure method** **Post-irradiation time** **Depth of cavity (mm)** ------------------------------------------ --------------------------- -------------------------- ------------------ ------------------ ------------------ ----------------- ----------------- Exposure time of 20 s (20 s) 0.5 hours 36.26±1.31^αaC^ 34.18±0.97^αbC^ 32.99±0.64^αbC^ 31.69±0.64^αcC^ 29.10±0.49^αdC^ 27.31±0.39^αeC^ 24 hours 41.22±0.81^αaB^ 38.42±1.14^αbB^ 38.35±1.07^αbB^ 38.42±1.19^αbB^ 37.42±1.33^αbB^ 36.87±1.50^αbB^ 7 days 42.86±0.80^αaA^ 40.52±1.04^αbA^ 39.79±1.20^αbcA^ 39.56±1.15^αbcA^ 38.37±1.01^αcA^ 38.08±1.23^αcA^ Bonding + exposure time of 20 s (B+20 s) 0.5 hours 32.23±0.97^βaC^ 30.78±0.79^βbC^ 30.58±0.65^βbC^ 30.54±0.47^βbC^ 30.10±0.93^αbC^ 29.49±0.79^βcC^ 24 hours 40.12±0.76^αaB^ 37.97±1.08^αbB^ 37.56±0.25^αbcB^ 37.50±0.11^αbcB^ 37.39±0.21^αbcB^ 36.79±0.42^αcB^ 7 days 42.06±0.40^αaA^ 40.21±0.28^αbA^ 39.08±0.24^αcA^ 39.19±0.26^αcA^ 38.82±0.26^αcA^ 38.09±0.43^αdA^ Exposure time of 40 s (40 s) 0.5 hours 37.19±0.24^γaC^ 36.50±0.37^γaC^ 35.28±1.15^γbB^ 34.14±0.61^γcC^ 32.22±0.17^βdB^ 30.36±0.13^γeC^ 24 hours 42.31±0.38^βaB^ 40.93±0.53^βbB^ 40.90±0.61^βbA^ 40.62±0.49^βbcB^ 39.80±1.10^βbcA^ 39.51±0.87^βcB^ 7 days 43.07±0.15^βaA^ 42.12±0.45^βbA^ 41.87±0.45^βbA^ 41.97±0.47^βbA^ 40.76±0.61^βcA^ 40.77±0.69^βcA^ Exposure time of 120 s (120 s) 0.5 hours 37.35±0.32^γaC^ 37.60±0.35^δaC^ 37.46±0.25^δaC^ 37.59±0.18^δaB^ 37.36±0.21^γaC^ 37.32±0.25^δaB^ 24 hours 42.58±0.30^βaB^ 42.67±0.62^γaAB^ 42.60±0.46^γaAB^ 42.59±0.46^γaA^ 42.62±0.55^γaAB^ 42.7±0.44^γaA^ 7 days 42.78±0.40^βaA^ 42.98±0.77^γaA^ 42.97±0.86^γaA^ 42.81±0.45^γaA^ 43.00±0.85^γaA^ 43.06±0.60^γaA^ The Same lower-case superscript Greek characters indicate no statistically significant differences between exposure methods at same post-irradiation time in the same depth of cavity; the same lower-case superscript letters indicate no statistically significant differences between depths of cavity at same post-irradiation time in the same exposure method; the same upper-case superscript letters indicate no statistically significant differences between post-irradiation times at same depth of cavity in the same exposure method (p\>0.05) KHN=Knoop hardness number For the DCP, the KHN was effected by the light-exposure method (p=0.0001; F=1652.29), post-irradiation time (p=0.0001; F=606.09), depth of cavity (p=0.0001; F=1184.99), and all the interactions between all the aforementioned factors (p=0.0001). For UCE, significant differences were found between the light-exposure methods (p=0.0001; F=201.27), post-irradiation times (p=0.0001; F=857.15), and depth of cavity (p=0.0001; F=488.76); in addition, significant interactions were also found between all the aforementioned factors (p=0.0001). For both resin composites, at 0.5 hours post-irradiation, the 20 s and 40 s methods resulted in the highest KHN values at the depth of 0.5 mm; these values gradually decreased with increasing depths of cavity (p\<0.05). On the other hand, at 24 hours or 7 days post-irradiation, the KHN values of DCP or UCE were not significantly different between the depths of 2.0 mm, 4.0 mm, and/or 6.0 mm, or 8.0 mm and 10.0 mm, respectively (p\>0.05). The 40 s method resulted in significantly higher KHN values than the 20 s method at all depths of cavity and post-irradiation times for both resin composites, with the exception of UCE at the depth of 0.5 mm (p\<0.05). With the 120 s method, the KHN values of both resin composites were not significantly different among the 6 depths of cavity at all post-irradiation times (p\>0.05), but they were significantly higher than those of the 40 s method, with the exception of that at the depth of 0.5 mm at all post-irradiation times (p\<0.05). For DCP, with the B+20 s method, no significant differences in the KHN values were observed at 0.5 hours post-irradiation at all depths of cavity, except 0.5 mm (p\>0.05). However, at 24 hours and 7 days post-irradiation, the B+20 s method resulted in significantly higher KHN values than the 20 s method at all depths of cavity, except 0.5 mm (p\<0.05). On the other hand, for UCE, at 0.5 hours post-irradiation, the B+20 s method showed significantly lower KHN values than the 20 s method up to the depth of 6.0 mm; however, the KHN values were significantly higher at the depths of 8.0 mm and 10.0 mm (p\<0.05). At 24 hours and 7 days post-irradiation, no significant differences in KHN values were observed between the B+20 s and 20 s method at all depths of cavity (p\>0.05). For all light-exposure methods, except the 120 s method, the UCE exhibited significantly higher KHN values at all depths of cavity, except at 4.0 mm in the 40 s method, at 7 days post-irradiation than at 0.5 hours and 24 hours post-irradiation (p\<0.05). It was found that the KHN values of DCP were significantly higher than those of UCE at all depths of cavity and post-irradiation times, regardless of the exposure method (p\<0.05). DISCUSSION ========== The results of this study indicate that the KHNs of dual-cured core build-up resin composites depend on the light-exposure method, including irradiation duration, use of bonding agent, depth of cavity, post-irradiation time, and material brand. Therefore, the research hypotheses formulated for this study must be rejected. KHN has been shown to be a good indicator of the degree of conversion/polymerization based on its good correlation with infrared spectroscopy^[@r09],[@r10],[@r31]^. However, the prediction of an absolute value of degree of conversion by means of an absolute hardness value is not achievable, since other factors such as type and size of filler, filler load, monomer composition, quantity of initiators, and the ratio of chemical- and light-cured components strongly influence the final quantity of reacted monomers^[@r02],[@r05],[@r14]^. Microhardness data from the same resin cement only should be compared according to the depth of the root canal or time elapsed since luting^[@r03]^. KHNs could be used to reflect the degree of conversion at different depths of a resin composite^[@r32]^. Therefore, KHNs were measured in the present study to reflect monomer conversion at different depths of cavity and post-irradiation times in the dual-cured core build-up resin composites. The KHNs of both core build-up resin composites irradiated for 20 or 40 seconds were the highest at the depth of 0.5 mm, and gradually decreased with increasing depth. This phenomenon could be simply attributed to the direction of photo-initiation. Light irradiation was focused on the top surface of the cavity. Therefore, polymerization of the resin composites, by means of photo-activated free radicals, may occur immediately at the shallow depths of cavity. This finding, that is, KHN of light- and dual-cured resin composites are affected by depth of cavity, has been reported previously^[@r04],[@r06]-[@r08]^. The chemical-curing mechanism of dual-cured resin composites is usually based on a redox reaction of benzoyl peroxide with aromatic tertiary amines, which generates free radicals that break the aliphatic carbon double bonds to initiate the polymerization process. It is supposed that immediate photo-activation after light irradiation, despite causing a rapid increase in the viscosity of the polymer matrix, does not hinder migration of the activated free radicals responsible for further chemically induced polymerization. Although the photo-activated free radicals at shallow depths of cavity could induce chain propagation of the resin polymer in the downward direction, the exact polymerization mechanism of dual-cured core build-up resin composite at greater depths of cavity remains unknown. It is difficult to distinguish clearly between the depths of cavity at which polymerization of the resin composite occurs through photo-initiation and those at which polymerization occurs by means of chemical initiation alone. Evaluation of the duration of photo-activation revealed that a longer exposure time of 40 seconds resulted in significantly higher KHN values than did a shorter exposure time of 20 seconds at all depths of cavity and post-irradiation times for both resin composites; however, this trend was not observed in UCE at the depth of 0.5 mm. Thus, our study showed that prolonged irradiation durations resulted in increased hardness values. Light-curing units with lower power density needed longer light-exposure times to produce a similar microhardness value of resin composite as that of light sources with high power density^[@r11]^. Therefore, with longer light-exposure durations, which result in higher energy densities at a given irradiance, more photo-sensitizer molecules are activated, which in turn increase the free radical concentration and consequently the conversion of double bonds. However, at the depth of 0.5 mm, no significant differences were observed in the KHN values between the light-irradiation durations of 40 seconds and 120 seconds (40 seconds each on the top and either side of the cavity) for both resin composites. This finding might be attributed to the fact that the polymer network developed during light exposure for 40 seconds does not allow any additional mobility of the polymer chains in order to increase monomer conversion, indicating that the light-irradiation duration of 40 seconds alone is sufficient. Dentin hardness ranges from KHN values of 50 to 70, depending on the distance from the amelodentinal junction^[@r21]^. The mean microhardness (KHN) value of 52.9 for DCP with 120 s method at the depth of 10.0 mm at 7 days post-irradiation therefore predicts similar mechanical properties to that of dentin. The equal degree of polymerization within the core material may support a uniform distribution of stress along the tooth-material interface under load. Longer light-exposure durations result in superior microhardness, but, at the same time, contribute to an increase of shrinkage and contraction stress of the dual-cured core build-up resin composite^[@r38]^. In cases of significant coronal destruction, it is necessary to replace the lost tooth structure with a core build-up material to attain full-coverage restoration. The cast post and core, prefabricated post and core materials, and coronal-radicular build-ups are the available options for this purpose. The fracture resistance and survival probability of post and core restorations depend on several factors such as the post material, luting agent, amount and condition of residual tooth structure, core material, preparation of the tooth for restorative procedures, and, finally, the fixed restoration^[@r16],[@r22],[@r25],[@r34]^. Therefore, when using dual-cured core build-up resin composites, it is preferable to prepare the composites by using the effective exposure method. Before core build-up resin composites are used, a bonding agent must be applied on the cavity wall. The effectiveness of the bonding agent on the KHN as an indirect method of monomer conversion differed between the 2 resin composites. For DCP, at 24 hours and 7 days post-irradiation, application of the bonding agent resulted in significantly higher KHNs than the 20 s method at all depths of cavity; however, this was not observed with UCE. The initiator and/or accelerator present in the bonding agent would promote monomer conversion of the dual-cured core build-up resin composites. On the other hand, the acidic monomer in the bonding agent inhibits the amine co-initiator in the dual-cured materials^[@r16]^, which in turn adversely affects polymerization of dual-cured core build-up resin composites. To prevent this inhibitory effect, aromatic sulfinic acid sodium salts are sometimes added to the bonding agent^[@r15],[@r37]^. The differences in compositions between the 2 core build-up resin composites might be responsible for the difference in their KHN behaviors. In the present study, both resin composites at all depths of cavity had a post-curing effect 7 days after irradiation, showing statistically higher KHN values than those at 0.5 hours post-irradiation for all light-exposure methods. These results are consistent with those of a previous study^[@r03]^, but not with those of another study^[@r41]^, which did not find changes in microhardness values 24 hours after irradiation. However, the polymerization reaction of the dual-cured materials might be specific^[@r19]^, and these previous studies used resin cements, whereas dual-cured core build-up resin composites were used in the present study. In fact, in the former study, neither did the authors evaluate dual-cured core build-up resin composites nor did they evaluate hardness behavior of dual-cured core build-up resin composite in simulated depths of cavity. Dual-cured resin composites have been introduced for use as both luting and core build-up materials. Superior physical properties are important for a successful restoration. In this study, it was found that the KHNs of DCP were superior to those of UCE, regardless of exposure methods. Various factors can influence the microhardness of a resin composite, such as filler load, type, or size, or resin matrix type^[@r17],[@r18]^. Increasing the monomer viscosity decreased the hardness^[@r27]^. In this study, the association between KHN values of the resin composite and composition of filler and matrix resin was not evaluated. On the other hand, the differences between the KHNs at the depths of 0.5 mm and 10.0 mm in UCE were lower than those in DCP, regardless of the light-exposure method. Dual-cured materials differ markedly in terms of the relative content of light- and self-activated catalysts^[@r13]^. Differences in the degree of conversion among materials when subjected to various curing protocols may consequently be attributed to variations in catalyst systems. In the present study, since similar KHNs were observed irrespective of the light-exposure method at greater depths of cavity in UCE rather than DCP, it might be inferred that the former exhibits high levels of chemically curing activators compared with the latter, compensating for attenuation of light energy at greater depths of cavity. Indeed, the polymerization behavior of dual-cured resin composites is strongly related to the material and can vary as a function of composition^[@r35]^. The speed of the polymerization reaction is strongly influenced by inhibitor concentration in the unfilled light-cured methacrylate-based systems^[@r30]^. Therefore, the curing mechanism of a specific composite material may not be applicable to other materials. CONCLUSIONS =========== Within the limitations of the present study, the following conclusions can be drawn: 1\. The microhardness of the dual-cured core build-up resin composites vary depending on the light-exposure method, including irradiation duration, use of bonding agent, depth of cavity, post-irradiation time, and material brand. 2\. For both resin composites, irradiation for 120 seconds does not result in significant differences in KHNs among all 6 depths of cavity at all post-irradiation times. 3\. Based on the physical property of microhardness behavior, dual-cured core build-up resin composites are preferably prepared by the effective exposure method.

Background ========== The peroxidase from horseradish roots (*Armoracia rusticana*; HRP, isoform C) is the most widely studied peroxidase, due mainly to its many diverse uses in biotechnology \[[@B1]\]. Several reports of HRP immobilisation exist, commonly employing adsorption as the immobilisation technique \[[@B2]-[@B4]\]. Although simple to implement, this approach suffers from several drawbacks, including the removal of adsorbed protein by stringent washing, high contamination (due to non-specific protein binding) and denaturation of proteins adsorbed onto hydrophobic surfaces \[[@B5]\]. Covalent immobilisation of proteins offers a more robust approach. Traditionally, immobilisation of plant HRP relied on periodate oxidation of the enzyme\'s carbohydrates. In recent years, however, immobilisation technology has developed rapidly, primarily due to the increase in number and range of activated solid supports and the ability to engineer or chemically modify biomolecules to enable easier covalent attachment. Several protein-engineering strategies have introduced reactive amino acids to promote directed protein immobilisation, with lysine being the most common. Abian and co-workers \[[@B6]\] selected asparagine, glutamic acid or arginine residues of Pencillin G Acylase (PGA) for replacement by lysine; these substitutions resulted in improved thermal stability of the immobilised PGA, due to multipoint attachment to a glyoxyl agarose-activated solid phase. HRPC is a highly cationic protein that contains twenty-one arginine residues \[[@B7]\]; however, only Arg residues on the opposite side to the active site entrance were considered for substitution in the present study, as attachment to the solid phase via the \"*back*\" of the molecule should allow an orientated rHRP immobilisation. This primary condition was supplemented with additional structural criteria (e.g. avoidance of helices, position in relation to the active site entrance, exposure to solvents) and led to the substitution of Arg118, Arg159 and Arg283 (Figure [1](#F1){ref-type="fig"}) by lysines, thus permitting the investigation of orientated covalent attachment to two activated solid phases, polyethersulfone membrane and CNBr-activated Sepharose™. {#F1} Results ======= Immobilisation on PES Membrane ------------------------------ Use of a modified PES membrane permitted successful orientated rHRP immobilisation. Introduction of Arg to Lys substitutions (Mutant 1; Table [1](#T1){ref-type="table"}) on the opposite face of the molecule to the active site, together with the removal of wildtype reactive lysine residues (Mutant 3, Table [1](#T1){ref-type="table"}; see also Figure [1](#F1){ref-type="fig"}), forces the protein to orientate its active site away from the membrane and towards the bulk solution phase during covalent immobilisation. Figure [2](#F2){ref-type="fig"} demonstrates faster and more intense colour development for Mutant 3 over an eighteen-hour period compared with wildtype. Densitometrical analysis (see Table [2](#T2){ref-type="table"}) using Image J software confirms and underscores this point. Equal concentrations of Mutant 1, Mutant 2 (Mutant 1 with reversion at position 238), Mutant 3, Mutant 4 (see Table [1](#T1){ref-type="table"}), and wildtype rHRP were applied (see Methods). Increased catalytic activity towards DAB substrate is attributed to optimal orientation of the rHRP molecule (Mutant 3), in contrast to the random orientation of wildtype rHRP or reduced binding capabilities of K283R revertants (Mutants 2 and 4). Furthermore, this experimental set-up also demonstrated reusability. Immobilised Mutant 3 was stained repeatedly (up to three times, data not shown) with TMB (non-precipitating substrate) over a period of several hours, with appreciable catalytic activity noted each time. ###### HRP mutants described in this paper **Name** **Mutations** ---------- ------------------------------------ Mutant 1 R118K, R159K, R283K. Mutant 2 R118K, R159K. Mutant 3 R118K, R159K, K232F, K241N, R283K. Mutant 4 R118K, R159K, K232F, K241N. R, Arg; K, Lys. The numbering system is based on the HRP structure with the PDB accession number 1ATJ. ###### \% Relative densitometrical analysis of modified PES immobilisation **Name** **% Relative Densitometry Values** ---------- ------------------------------------ Plant 94 WT 100 Mutant 1 127 Mutant 2 140 Mutant 3 163 Mutant 4 138 Plant, WT, Wildtype; Mutant 1, R118K/R159K/R283K; Mutant 2, R118K/R159K; Mutant 3, R118K/R159K/K232N/K241F/R283K; Mutant 4, R118K/R159K/K232N/K241F. % values were calculated using the \'Analyse\' function of ImageJ software package \[35\]. {#F2} Stability Properties of Arg → Lys Mutants ----------------------------------------- Stabilities in free solution of Mutants 1--4 were compared with wildtype. Mutant 1 displayed almost identical thermal and solvent stabilities (except in MeOH) to wildtype rHRP. Mutant 3 was notably less stable (50% decrease in thermal and 20% decrease in MeOH stabilities). This was unexpected, as the chemically modifiable proximal lysine residues (232 and 241, \[[@B8]\]) had been mutated to the stabilising Phe and Asn respectively (Ryan and O\'Fágáin, in preparation). Mutant 4 (Mutant 3 with reversion at position 283) displayed a 2.6-fold increase in t~1/2~at 50°C over Mutant 3 and a 1.3-fold increase versus wildtype. However, Mutant 1 (Arg at position 283) and Mutant 2 (reversion at position 283) displayed very similar thermal stabilities. Solvent stability was also affected by reversion to Arg at position 283: Mutant 2 exhibited a 1/3 increase in C~50~in DMF, but decreases in DMSO (1/6) and MeOH (1/5) tolerances compared with Mutant 1. Mutant 4 displayed decreased stabilities in DMSO (1/7) and MeOH (1/5). Interestingly, both reversion mutants displayed greater H~2~O~2~stability in free solution than their unreverted counterparts (16% for Mutant 2 versus Mutant 1 and 94% for Mutant 4 versus Mutant 3; see Tables [1](#T1){ref-type="table"} and [3](#T3){ref-type="table"}). ###### Stability Characteristics of HRP Immobilisation Mutants in free solution and immobilised on CNBr-activated Sepharose **Thermal Stability** **Solvent and H~2~O~2~Stability** ------------------- ----------------------- ----------------------------------- ----------------- ----------------- --------------- --------------- -------------- ------------------- **Mutant** **T~50~**(°C) ***k*-value**(min^-1^) **SD**(min^-1^) **t~1/2~**(min) **DMSO**C~50~ **MeOH**C~50~ **DMF**C~50~ **H~2~O~2~**C~50~ **Free Solution** WT 50 0.0559 ± 0.0034 12 35 53 14 17 Mutant 1 51 0.0535 ± 0.0005 12 34 44 12 38 Mutant 2 50 0.0568 ± 0.0057 12 30 35 16 44 Mutant 3 48 0.0109 ± 0.0007 6 35 45 25 17 Mutant 4 50 0.0445 ± 0.0003 17 30 35 16 33 **Immobilised** WT \- 0.0255 ± 0.0023 27 32 40 14 6 Mutant 1 \- 0.0223 ± 0.0020 3 24 32 17 5 Mutant 2 \- 0.0544 ± 0.0044 13 28 38 10 5 Mutant 3 \- 0.0152 ± 0.0017 5 42 43 29 5 Mutant 4 \- 0.0491 ± 0.0049 14 28 38 15 5 WT, Wildtype; Mutant 1, R118K/R159K/R283K; Mutant 2, R118K/R159K; Mutant 3, R118K/R159K/K232N/K241F/R283K; Mutant 4, R118K/R159K/K232N/K241F. Modelled *k*-values were calculated using the *Enzfitter*™ software package (Version 1.05. Cambridge UK: Biosoft Ltd.; 1987). Values are the mean of three independent determinations in each case. Standard deviations were \< 5% for solvent and \< 10% for H~2~O~2~studies. Units of C~50~are % v/v for solvents and mM for H~2~O~2~. Immobilisation on CNBr-activated Sepharose ------------------------------------------ Increased lysine content correlated with increased immobilisation on CNBr-activated Sepharose: Mutant 4 (3 sites: Lys 118, 159, 174; 42% immobilisation) \< wildtype (3 sites: Lys 174, 232, 241; 50% immobilisation) \< Mutant 3 (4 sites: Lys 118, 159, 174, 283; 57% immobilisation) \< Mutant 2 (5 sites: Lys 118, 159, 174, 232, 241; 69% immobilisation) \< Mutant 1 (6 sites: Lys 118, 159, 174, 232, 241, 283; 86% immobilisation). These calculations were based on measured units of activity before and after immobilisation, assuming that no inactivation occurred during the mild immobilisation process. Immobilised wildtype demonstrated more than two-fold thermal stabilisation, whilst immobilised Mutant 2 displayed a marginal 5% increase. Mutant 1 (4-fold), Mutant 3 (1.25-fold) and Mutant 4(1.25-fold) all displayed decreased thermal stability following immobilisation. Solvent stabilities of wildtype and mutant (both free and immobilised) are set out in Table [2](#T2){ref-type="table"}. Small variations (\< 2-fold) in solvent stabilities are noted. All four mutants and wildtype show a dramatically decreased resistance to inactivation by H~2~O~2~following immobilisation onto CNBr-activated Sepharose, approximating to a basal C~50~value of 5 mM H~2~O~2~(Table [3](#T3){ref-type="table"}). This contrasts sharply with variations in H~2~O~2~resistance noted in free solution (Mutants 2, 1 and 4 display a 2.6, 2.2 and 2-fold increase in stability respectively compared to wildtype and Mutant 3; Table [3](#T3){ref-type="table"}). Discussion ========== Lopez-Gallego and co-workers \[[@B9]\] recently proposed chemical modification to enrich the protein surface with reactive groups that would facilitate multipoint covalent attachment during protein immobilisation. The present work extends this concept by substituting additional lysines in a specific region of rHRP so as to facilitate directed, multipoint immobilisation onto a commercially available, modified polyethersulfone (PES) membrane. Lysines are often exploited for covalent immobilisation, as they occur frequently in proteins and are active nucleophiles when unprotonated, i.e. at alkaline pH. They are usually located on surface-exposed regions of proteins, due to their charged nature. This characteristic also allows for simple immobilisation reactions, as lysine residues do not require activation prior to immobilisation \[[@B10]\]. To allow directed immobilisation of rHRP to an activated matrix, arginine residues were selected for conservative substitution by lysine. Of the twenty-one arginine residues in rHRP, only three were deemed suitable for the proposed replacements: Arg118, Arg159 and Arg283 (see Methods). These positions are located on the opposite side of the molecule to the active site; hence, immobilisation via these \"new\" lysines forces the protein to orientate its active site away from the membrane and towards the bulk solution. Maximal colour development was noted during DAB catalysis for Mutant 3 (Lys 118, 159, 174 and 283 available), which can bind only in a uni-directional manner: the reactive wildtype lysines 232 and 241 \[[@B8]\], on a different face of the molecule, have been removed. This directional immobilisation should allow optimal substrate access and product egress. Although introduction of additional lysines permits directional immobilisation, a stability penalty is incurred. Despite its conservative nature, an Arg-to-Lys substitution can alter hydrogen bonding, as lysine can form only a single hydrogen bond compared with Arg \[[@B11]\]. Reduced hydrogen bonding resulting from the loss of three Arg, coupled with substitution by a bulky aromatic Phe (K232F) residue, may explain the instability of Mutant 3 in free solution. Arg283 is located on the last turn of Helix J \[[@B12]\]. Simple reversion of this residue rescued (or even enhanced) poor thermal and H~2~O~2~stabilities of the R283K rHRP mutant. Whilst this reversion reduced the number of potential immobilisation sites, the stabilising effect of the K232F and K241N mutations (Ryan and O\'Fágáin, in preparation) was evidenced (Table [2](#T2){ref-type="table"}; Mutant 4 versus Mutant 2 in free solution). Further protein engineering is required to re-introduce stability to the immobilisation mutants. The work of Strausberg and co-workers \[[@B13]\] has parallels with the present study. They removed a Ca^2+^binding loop from subtilisin so as to enhance its stability under metal-chelating conditions. Their initial \"loop-less\" mutant was much less stable than the wildtype, but it proved possible to \"retro-engineer\" stability into the mutant, such that it out-performed wildtype under chelating conditions. Currently, there are very few reports of either direct or random immobilisation of recombinant HRP in the literature. Rojas-Melgarejo and colleagues \[[@B2]\] noted the difficulty in immobilising recombinant HRP expressed in *E. coli*to traditional adsorbent solid phases, since it lacks the carbohydrate residues commonly used for this type of immobilisation. Instead, cinnamic esters were successfully employed (63% immobilisation rate) as an immobilisation platform using physical adsorption in conjunction with glycosylation-independent hydrophobic interaction. Wildtype recombinant HRP immobilised in such a fashion yielded an 8% increase in thermal stability. Abad and co-workers \[[@B14]\] recently described directed recombinant HRP immobilisation via an N-terminal His tag-cobalt (II)-terminated gold nanoparticle interaction. Although no stability data were reported for this method, its potential application in the convenient attachment of rHRP onto gold electrodes for biosensing is obvious. Although our rHRP possesses a C-terminal His Tag, we chose not to exploit this feature for immobilisation and instead adopted the strategy outlined here. The immobilisation of enzymes can increase protein stability \[[@B15]\]; however, by implementing orientation-dependent protein immobilisation, researchers may not only stabilise a protein, but also promote superior performance in applications such as biosensors. Recent developments in the biosensing area focus on the type of bio-recognition involved in immobilisation. Traditionally, lysines (as in this study) or cysteines can be engineered into a protein backbone to allow for immobilisation; however, research has progressed towards controlled deposition and orientation of immobilised recombinant oxidoreductases for optimal Direct Electron Transfer (DET). The shorter the electron transfer distance, the greater the chance of DET occurring in a biosensor configuration; hence, controlled directional immobilisation of enzymes is important for the progression of third generation biosensors. Typically, physical adsorption is utilised for peroxidase immobilisation onto a metal electrode; however, other methods such as entrapment \[[@B16]\] and direct covalent attachment \[[@B17]\] are employed. Often, these methods lead to the formation of a randomly orientated layer, either on the surface of the electrode (metal) or within cavities (carbon). Physical adsorption may result in enzyme denaturation, due to multiple contacts with the solid phase. Binding of ligands, or substrate/product access and egress, may also be hindered, all resulting in poor bioelectrocatalytic activity \[[@B18]\]. Self-assembled monolayers (SAM) are an attractive alternative for controlled, directed enzyme immobilisation. These highly organised monomolecular layers form spontaneously upon submersion of a solid phase into a solution containing amphifunctional molecules. The lengthy hydrophobic chain of the SAM can be altered and terminated with a functional group that will interact only with a specific residue on an enzyme backbone \[[@B19]\]. In this instance, a site directed mutation could be utilised to dictate the location of the specific residue and, hence, control the orientation of the molecule (e.g additional Cys residue for thiol immobilisation \[[@B20]\]). This concept has recently been adapted for use with chemically-modified plant-derived HRP by Suarez and co-workers \[[@B21]\]. Other methods of immobilisation have developed in recent years, from His-tag chelation to biotin-avidin interaction. Polypeptide scaffolds comprise a topical generic immobilisation method, employing a hydrophobic anchor attached to a leucine-zipper protein capture domain utilised to directionally immobilise several different proteins to a hydrophobic solid phase \[[@B22]\]. Fuentes and colleagues \[[@B23]\] recently noted that the rate of plant HRP immobilisation depended exponentially on the concentration of reactive groups both on the protein and the support. Hence, a protein is mainly immobilised by an area of its surface having the highest density of reactive groups. Lysine is the key amino acid responsible for protein immobilisation onto CNBr-activated Sepharose \[[@B24]\], as reflected in the present study. Only bound Mutant 4 displayed improved stability over free-solution figures, suggesting that multipoint attachment, most evident in the case of Mutant 1, increases strain on the rHRP molecule. This, combined with restricted flexibility, leads to a notable decrease in thermal stability for the mutants. Previous reports of destabilising multipoint immobilisation of glucoamylase also cite increased protein strain as reason for decreased thermal stability \[[@B25]\], while the type of immobilisation resin can also destabilise the immobilised protein \[[@B26]\]. Our results contrast with previous reports of increased thermal stability following multipoint attachment \[[@B6]\]. However, Abian and co-workers \[[@B6]\] focused their mutations on a lysine-rich area of Pencillin G Acylase, which previous chemical modification studies had proven not to be critical for protein stability, and so minimised the chance of a deleterious mutation in this area. This group also utilised a glyoxyl agarose-activated solid phase for protein immobilisation. Plant-derived glycosylated HRP has previously been immobilised onto a variety of solid phases, by a range of different techniques, to increase thermal stability. For example, Rojas-Melgarejo and co-workers \[[@B27]\] noted increased thermal stability (30%) for plant HRP absorbed onto cinnamic carbohydrate esters, similar to that noted in this study for wildtype rHRP (27%), covalently immobilised onto CNBr-activated Sepharose. The reduced solvent stabilities of the mutants can similarly be ascribed to increased strain and lack of protein flexibility. Due to the confined nature of the immobilised protein, organic solvent molecules that have stripped and replaced water molecules are maintained in close proximity to the protein surface, leading to accelerated protein denaturation \[[@B27]\]. It is well known that many enzymes can function in organic solvents \[[@B28]\]; however, protein denaturation often occurs in water-miscible solvents, by a process referred to as *\"water-stripping\"*. Gorman and Dordick \[[@B29]\] described how T~2~O (tritiated water) is desorbed from HRP by organic solvents such as MeOH and DMF. DMF, despite its higher dielectric constant, desorbed significantly less T~2~O than did MeOH. The difference is possibly due to structural similarities between MeOH and H~2~O, which allow MeOH to strip and replace water molecules close to the protein surface. There is a major requirement for water for HRP activity \[[@B29]\], so any water removal will have a detrimental effect on enzyme activity. Small losses of solvent stability (e.g. Mutant 3) can be attributed to the reduced number of potential attachments for covalent immobilisation. Hence, the immobilisation of Mutant 3 is not as rigid as that of Mutant 1, permitting some protein flexibility. Immobilisation on CNBr-activated Sepharose notably decreased H~2~O~2~tolerance of rHRP wildtype and mutants: the basal C~50~value was 5 mM H~2~O~2~, much lower than any free-solution value. Previous immobilised HRP H~2~O~2~tolerance results depend on the type and reactivity of the activated solid phase. Rojas-Melgarejo and co-workers \[[@B27]\] report increased H~2~O~2~tolerance for HRP adsorbed onto cinnamic carbohydrate esters when the stability analysis was carried out at pH 7.0; at pH 4.5, however, there was a dramatic loss of H~2~O~2~stability. Those authors believe that reactive oxygen species such as superoxide, generated during H~2~O~2~catalysis, may have been absorbed by the cinnamic support at pH 7.0 instead of attacking the essential enzyme residues. At lower pH, however, carbonyl groups are oxidised and do not provide a sink for oxidising reactive oxygen species, leading to dramatic enzyme destabilisation \[[@B27]\]. Conclusion ========== Directional orientated immobilisation of rHRP mutants onto an activated, modified polyethersulfone (PES) membrane has been achieved with excellent retention of catalytic activity. Despite some loss of thermal, solvent and H~2~O~2~stabilities both in free solution and following immobilisation onto CNBr-activated Sepharose, orientated rHRP immobilisation has many potential applications including diagnostics and biosensor development. Ultimately, use in these applications will require re-engineering acceptable stability characteristics into the \"immobilisation mutants\". Methods ======= Materials --------- The HRP gene was a generous gift from Prof. Frances H. Arnold (Caltech, CA, USA). The pQE60 vector was purchased from Qiagen (Valencia, CA); XL 10 Gold cells and QuickChange™ Mutagenesis Kit were purchased from Stratagene (La Jolla, CA). Pall *UltraBind*™ PES modified membranes were obtained from AGB Scientific (Dublin, Ireland). All other reagents (including CNBr-activated Sepharose) were purchased from Sigma Aldrich and were of analytical grade or higher. Cloning ------- The HRP gene was directionally cloned into the pQE60 vector as a fusion with the N-terminal pectate lyase (PelB) leader sequence \[[@B30]\] and a C-terminal hexa-histidine purification tag, to generate the plasmid pBR_I. Initially, the PelB leader was cloned via a *Nco*I -- *BamH*I double restriction. This introduced a novel *Not*I site 5\' to the existing *BamH*I site in the modified pQE60 vector. The HRP gene was then *Not*I -- *Bgl*II directionally cloned into a *Not*I -- *BamH*I restricted PelB-modified pQE60 vector. This cloning strategy incorporated the poly-His tag, present in the pQE60 vector, at the C-terminus of HRP. Bacterial Strains and Plasmids ------------------------------ *E.coli*XL 10 Gold was used as host strain to express the HRP protein. The plasmid (pBR_I), carrying the HRP gene and coding for the HRP fusion protein, was used for expression and site directed mutagenesis. Recombinant DNA Techniques -------------------------- All DNA manipulations were carried out by standard techniques \[[@B31]\]. Site directed mutagenesis was carried out as described by Wang and Malcom \[[@B32]\] utilising the QuickChange™ method. Mutant primers were supplied by MWG-Biotech (Germany). Mutations were confirmed by commercial di-deoxy sequencing (Fusion Antibodies, Belfast, Northern Ireland). Expression and Purification --------------------------- A single cell transformed with pBR_I (or mutant derivative) was grown in LB medium containing 100 μg/mL ampicillin and 2% w/v glucose until the OD~600\ nm~reached 0.4; the cells were removed via centrifugation at 2,000 × *g*for 5 min and resuspended in fresh LB supplemented with 100 μg/mL ampicillin, 1 mM δ-ALA and 2 mM CaCl~2~. The cells were then allowed to grow at 30°C, 220 rpm for 16 h. Following overnight expression, the cells were centrifuged at 2,000 × *g*for 5 min and the supernatant was treated with 50% w/v (with respect to the original supernatant volume) ammonium sulphate for 2 h at room temperature. The cells were periplasmically lysed \[[@B33]\] and the periplasmic contents were similarly treated with ammonium sulphate. Proteins precipitated by 50% w/v ammonium sulphate were collected via centrifugation, resuspended in 50 mM phosphate buffer pH 8.0 and dialysed versus the same buffer overnight at 4°C. Sodium chloride (1 M) and GnCl (200 mM) were added to the dialysed fractions, and these latter were subjected to nickel affinity chromatography at room temperature. Sodium acetate (25 mM, pH 4.5) was utilised to elute the bound HRP. The eluted HRP was again dialysed versus 50 mM phosphate buffer pH 7.5 overnight at 4°C, after which the protein was concentrated (Amicon-Plus 20 concentrator tubes), filter sterilised and stored at 4°C. Substitution residue selection ------------------------------ Arginine residues in the wildtype were pinpointed for conservative replacement by the reactive side chain lysine, since both have similar size and charge; this should result in minimal secondary structure rearrangement \[[@B11]\]. The twenty-one arginines present in wildtype HRP were viewed in *DeepView*\[[@B34]\] and assessed for suitability based on secondary structure and location. Beta sheet-forming residues were preferred over alpha-helical ones, and surface-exposed residues were selected in preference to buried ones. Only residues on the plane opposite to the active site entrance were considered. These criteria resulted in the selection of R118, R159 and R283 as targets for replacement by lysines (see Figure [1](#F1){ref-type="fig"}). Of the six wildtype lysines, only 174, 232 and 241 are available for immobilisation under mild conditions \[[@B8]\]. Lys 174, modified to only 20% compared with 100 and 85% for 232 and 241 \[[@B8]\], was left intact throughout this study, but positions 232 and 241 were altered in some of the mutants (see Results section and Figure [1](#F1){ref-type="fig"}). ### rHRP immobilisation #### (a) Pall UltraBind™ Modified PES Membrane possesses activated aldehyde functional groups tailored for amine-based covalent immobilisation. HRP (plant or recombinant) was resuspended in 50 mM sodium phosphate buffer, pH 7.5 (enzyme concentration 30 pM), and directly spotted onto the activated membrane; the latter was allowed to air-dry completely at room temperature for 10 min. The remaining binding sites were then blocked with 1% w/v solution of food-grade non-fat dry milk in 50 mM sodium phosphate buffer, pH 7.5, for 1 hour at room temperature. The membrane was then washed with 50 mM sodium phosphate buffer, pH 7.5, and allowed to air-dry completely at room temperature for 10 min. DAB (precipitating chromogen) and TMB (non-precipitating chromogen) were utilised to locate immobilised HRP prior to imaging. #### PES Immobilised HRP-Imaging Developed membranes were imaged with a Hewlett Packard Scanjet 5590, connected to a Dell Optiplex Computer with the following parameters: output type, true colour (24 bit); output scale, 100%; output resolution, 2400 dpi; sharpen level, extreme; scan form, scanner glass; highlights, -100; shadows, -100; and midtones, -100. True colour images were converted to greyscale using Hewlett Packard Scanjet 5590 image software. ### Image J Analysis Saved greyscale images were opened in Image J software \[[@B35]\]. The background was subtracted from the image using the Process\>subtract background function. A rolling ball radius of 50 and a white background were selected. The image threshold was also automatically adjusted to black and white. The image was cropped and desitometrical analysis carried using the \'Analyse\' function. The results were expressed as a %age of the wildtype value. #### (b) CNBr activated Sepharose^®^4B HRP was suspended in coupling buffer (0.1 M NaHCO~3~/Na~2~CO~3~containing 0.5 M NaCl, pH 8.5; enzyme concentration 30 pM). CNBr-activated Sepharose^®^4B was aliquoted into a clean purification column (2 × 8 cm), then washed and swollen in cold 1 mM HCl for at least 30 min. The resin was then washed with 10 column volumes of distilled water and, finally, with coupling buffer. Immediately, the resin was transferred to the solution containing HRP and mixed on an end-over-end rotator for 2 hours at room temperature. After coupling, any unbound protein was removed using several washes of coupling buffer. Any remaining unreacted groups on the Sepharose particles were blocked with 0.2 M glycine, pH 8, for 2 hours at room temperature. Extensive washing with high (A: 0.1 M NaHCO~3~/Na~2~CO~3~buffer containing 0.5 M NaCl, pH 8.5) and low (B: 0.1 M acetic acid-sodium acetate buffer, pH 4) pH buffers, A and B respectively, removed the glycine and consolidated the HRP-to-resin covalent bonds. #### H~2~O~2~and Thermal Tolerance Analysis H~2~O~2~stability of recombinant HRP, and mutant variants, was determined as described in ref. \[[@B36]\]. In brief, rHRP (360 nM in 50 mM phosphate buffer, pH 7.0) was incubated with increasing concentrations of H~2~O~2~(0--100 mM). H~2~O~2~concentrations were determined spectrophotometrically at 240 nm using 43.6 M^-1^cm^-1^as the extinction coefficient \[[@B37]\]. Samples were exposed to the relevant H~2~O~2~concentration for 30 min at 25°C in a temperature-controlled waterbath. Residual catalytic activities were then measured (below) (note that the H~2~O~2~concentration used with a reducing co-substrate in the activity assay below was significantly lower than the concentration of H~2~O~2~as sole substrate that led to 50% inactivation (C~50~) \[[@B38]\]). The thermal stability parameters of recombinant HRP and mutant variants were determined as follows. A single peroxidase stock solution was placed for 10 min at progressively increasing temperatures, at which time aliquots were withdrawn, chilled on ice and eventually assayed under optimal conditions. This procedure gives T~50~, or the temperature of 50% inactivation. For thermal inactivation at a constant 50°C, a single peroxidase stock solution (room temperature) was plunged into a waterbath held at 50°C and aliquots were withdrawn every minute for ten minutes onto ice and later assayed under optimal conditions. The solvent parameters of recombinant HRP and mutant variants were determined as described for plant HRP \[[@B39]\] except that the solvents employed were methanol, dimethylsulfoxide and dimethylformamide. Following temperature, solvent or H~2~O~2~incubation, aliquots (50 μL) were withdrawn and the remaining catalytic activities were assayed using a standard microtitre-based TMB assay. This comprised 150 μL of 32 mM TMB substrate (in 100 mM citric acid buffer, pH 5.5, containing 3 mM H~2~O~2~) and 50 μL of rHRP in each well. The microplate was shaken as the initiating enzyme was added and the absorbance at 620 nm was recorded after 6.5 min reaction time. The C~50~value (mM H~2~O~2~or % v/v solvent where 50% of maximal HRP activity still remains) was utilised to compare H~2~O~2~/solvent stabilities across the mutant matrix, whilst the *t*~1/2~(half-life) was employed to compare mutant thermal stabilities (see Table [3](#T3){ref-type="table"}). Abbreviations ============= ABTS, 2,2\'-azino-bis(3-ethylbenzthiazoline-6-sulfonic acid; δ-ALA, delta aminolevulinic acid; C~50~, concentration of H~2~O~2~leading to 50% inactivation after 30 min at 25°C; DAB, diaminobenzidine; DMF, dimethylformamide; DMSO, dimethylsulfoxide; dpi, dots per inch; HRP, horseradish peroxidase isoenzyme C; GnCl, guanidine hydrochloride; LB, Luria-Bertani medium; MeOH, methanol; rHRP, recombinant horseradish peroxidase isoenzyme C; PES, polyethersulfone; t~1/2*app*~, apparent half-life; T~50~, temperature leading to 50% inactivation after 10 min; *t*~1/2~, half-life; TMB, 3,3\',5\'5-tetramethyl benzidine; v/v, volume per volume; w/v, weight per volume. Authors\' contributions ======================= BJR designed this study and participated in its conception. BJR also carried out all experimental work, data collection, analysis and worked jointly on manuscript drafting. CF conceived the study, participated in its design and co-ordination, and worked jointly on manuscript drafting. Both authors read and approved the final manuscript. Acknowledgements ================ We thank the Irish Research Council for Science, Engineering and Technology (Embark Initiative postgraduate scholarship to BJR) and Dublin City University (Postgraduate Accommodation Award to BJR and Albert College Award to CÓ\'F) for financial support. We are also grateful to Prof FH Arnold and California Institute of Technology for the generous gift of a recombinant horseradish peroxidase plasmid. Dr Donal O\'Shea and Ms Lynne Dobson are also thanked for the use of the Hewlett Packard Scanjet 5590. The National Centre for Sensors Research was established under the Higher Education Authority\'s Programme for Research in Third Level Institutions.

Background {#Sec1} ========== Periodic menstrual bleeding is an integral component of the human female reproductive cycle, \[[@CR1]\] but the way this phenomenon is viewed and the meanings attached to it vary considerably in different societies around the world \[[@CR2]--[@CR5]\]. Most cultures require that menstrual flow be regulated or contained in some fashion. This cultural requirement creates a recurrent challenge to all menstruating girls and women. These menstrual management challenges are often most acute at menarche, because newly-menstruating adolescent girls often lack the knowledge, experience, and self-confidence necessary to deal with the problems of menstrual hygiene \[[@CR5]--[@CR15]\]. Factors that influence menstrual hygiene management include knowledge of reproductive biology; the background beliefs about menstruation prevalent in society; the limitations (if any) that beliefs about menstruation place on female activities; and the ease of access that girls and women have to menstrual management supplies such as commercially-manufactured sanitary pads (either disposable or reusable), tampons, or traditional menstrual management materials obtained from the home environment such as strips of cloth torn from old bedding, skirts, shawls, or other articles of clothing. Particularly in low-income countries, menstrual hygiene management may present a formidable obstacle to the education and social advancement of adolescent girls, who are often confused, embarrassed, and unprepared for the sudden and unexpected appearance of menstruation \[[@CR12]--[@CR20]\]. The Dignity Period Project is a collaborative community development project between Mekelle University (a large Federal university in northern Ethiopia), the Mariam Seba Sanitary Products Factory (a woman-owned social enterprise), and Dignity Period (a not-for-profit public charity based in the United States, [www.dignityperiod.org](http://www.dignityperiod.org)) which seeks to improve knowledge about menstruation in northern Ethiopia and to provide free, locally-produced, environmentally friendly, reusable menstrual hygiene products to schoolgirls in this region. The project hopes to improve the psychosocial environment of Ethiopian girls with respect to menstruation, to reduce school absences from menstruation-related causes, and to stimulate the growth of a viable commercial industry that someday can meet the need for menstrual hygiene supplies in an efficient, locally-sourced, and cost-effective manner. The project provides educational materials about menstruation to girls and boys in Ethiopian schools through a bilingual (English/Tigrigna) educational booklet \[[@CR21]\] and on-site, in-school training on menstrual hygiene management. The project also provides schoolgirls with free menstrual hygiene kits containing 2 pairs of underwear and 4 washable, reusable menstrual pads \[Fig. [1](#Fig1){ref-type="fig"}\]. By the end of 2018, the project will have reached over 250,000 students in Tigray and in the contiguous Afar Region of Ethiopia. The present study is part of ongoing efforts to understand menstruation and menstrual hygiene in this part of the country.Fig. 1Menstrual hygiene kit distributed by the Dignity Period Project, consisting of 2 pairs of underwear and 4 re-usable menstrual pads. The pads have a waterproof backing; a soft, cotton absorbent lining; and they fasten over the crotch of a pair of underwear using a button. The pads are cleaned by soaking in water, hand-washing with soap, rinsing, and then air-drying in the sun. Ultraviolet radiation from direct sunlight has a bacteriocidal/bacteriostatic effect. Photo by the authors Methods {#Sec2} ======= An ethnographic survey of menstrual beliefs in Tigray, northern Ethiopia, was carried out through focus groups, in-depth interviews, and individual case studies in conjunction with a previously-published community-based survey of this topic \[[@CR22]\]. There were 40 focus group discussions, 64 in-depth interviews with key informants, and 16 individual case histories. A total of 322 individuals participated, including 240 people in six different types of focus groups: pre-menarchal girls (*n* = 36), menstruating adolescents (*n* = 48), mature married women of reproductive age (*n* = 48), post-menopausal women (*n* = 48), adolescent males (*n* = 30), and mature married men (*n* = 30). In-depth interviews were also carried out with 80 individuals, including Orthodox Christian priests (*n* = 12), imams from the Muslim community (*n* = 8), secondary school principals (*n* = 12), teachers (*n* = 16; 6 male, 10 female) and nurses (*n* = 16; 5 male, 11 female), as well as menstruating schoolgirls (*n* = 16). Focus groups were run by native speakers of the local language (Tigrigna), except in the Tahtay Adyabo subdistrict, where indigenous speakers of Kunamigna were used as translators for this ethnic group. The sessions were recorded, transcribed, and then translated into English. Audio data were transcribed and translated, then broken down into discrete codes using Atlas Ti software (version 7.5.4, Atlas.ti Scientific Software Development Mnbh, Berlin) and grouped into related families and sub-families based on their content. The results were synthesized by themes to construct a narrative concerning menstruation in Tigray. The study protocol was reviewed and approved by the institutional review boards at Ayder Comprehensive Specialist Hospital, College of Health Sciences, Mekelle University, Mekelle, Ethiopia, and by the Washington University School of Medicine in St. Louis, MO, USA. Informed consent was obtained prior to each interview or focus group discussion. Study subjects were compensated for their time and participation with 50 Ethiopian Birr (approximately US\$2.50 at the time of the study). Results {#Sec3} ======= A number of recurrent themes emerged in the course of these interviews and focus-groups. We have organized these themes into four broad categories for the purpose of discussion: the nature and biological function of menstruation; menarche; menstrual restrictions; and menstrual needs, hygiene accidents, and associated stigma. The nature and biological function of menstruation {#Sec4} -------------------------------------------------- There is universal agreement that menstruation is a normal physiological process linked to reproduction. Most people refer to it as a "gift from God" which permits women to become pregnant and bear children, to give the gift of new life, to become mothers. As one woman said, *"It means the body is ready to reproduce."* Only the most educated members of our participant population (teachers, nurses, school directors, and a few older students) could provide a reasonably accurate description of the biology of conception and its relationship to menstruation. A typical approximation of these responses was given by one male who noted *"menstrual blood is part of life and the process of making life. It becomes blood if it is not fertilized"---*a recognition that menstruation regularly follows when fertilization does not occur. It is generally understood that menstruation stops during pregnancy. Menstrual periods are described as occurring monthly (roughly every 21--35 days) with some variability, and bleeding is said to last for roughly 3--7 days. Periods of longer duration or more unpredictable occurrence are regarded as potential reasons to seek medical advice. There is good awareness that certain contraceptive technologies (oral contraceptive pills, medroxyprogesterone acetate injections, etc.) can cause changes in the menstrual cycle. There is also a recognition that menstruation ceases when menopause occurs (biologically around age 50, but commonly described by participants as occurring somewhere between ages 40 and 60). At menopause the potential for becoming pregnant also ceases. The phenomenon of menstruation has different names in different Ethiopian languages. Menstruation is commonly called the "monthly flower" (in Amharic, *yewor abeba;* in Tigrigna, *werhawi tsigya*). As one Orthodox Christian priest explained it: *"Menstruation is like flowers. Flowering plants finally produce fruit. The same is true of girls after menstruation."* Another common expression for menstruation in Tigrigna is *nay adetatna* ("\[the thing\] belonging to our mothers"), which also links it clearly to reproduction. In the Kunamigna language menstruation is called *mara* ("menstruation") or *ginayaka* ("monthly bleeding from our mothers"). Among the Raya people and others (both Muslim and Christian) it is often called *adef* ("dirty matter"). As discussed below, menstruation is commonly regarded as ritually polluting. Many women also believe that it is a cleansing process. One menopausal woman remarked *"the blood that comes during menstruation is bad blood and after bleeding the woman makes good blood."* Another woman said *"It releases bad blood out of our body; it cleanses the system."* Still another woman remarked "*It voids useless fluids from your body."* The notion that loss of menstrual blood cleanses or purges the body is common in many other cultures \[[@CR2], [@CR3]\]. Menarche {#Sec5} -------- Males and females both agree that menstruation occurs for the first time in adolescence (roughly between ages 14--18, but in some cases as early as age 7 or 8) and that it heralds the onset of female reproductive capacity like the first "blossoming of a flower." There is a widely held consensus that the timing of the onset of menstruation varies depending on the girl's constitution and her living conditions: in the view of many people, good health and good nutrition are linked to an earlier occurrence of menarche. There is also a common, but not universal, belief (more commonly held by males) that menstruation does not start until a girl has sexual intercourse for the first time. The existence of this false belief presents a potential hazard for girls approaching menarche if they live in families where this belief is held. We know of some girls who have been beaten by their fathers at menarche because they were assumed to be sexually active and misbehaving, when, in fact, they were merely bewildered adolescents having their first menstrual periods. The emotional scars of such experiences are likely to be deep and long-lasting. This belief also may result in some schoolgirls being disparaged by boys as having lost their virginity if they suffer a hygiene accident at school that results in their menstrual status becoming known. Many Tigrignan girls enter adolescence with little or no knowledge of menstruation. Their first periods are often unexpected, traumatic experiences, as several participants described. *"It \[menarche\] happened to me when I was going back to my house from a test I had taken in the eighth grade. I had not been feeling well during the day. In the middle of the road I started to bleed and I was scared. I used a piece of cloth and put it in my underwear. Nobody saw me; but I was very worried and frightened. I didn't tell my parents."* Another girl said "*I went to school and while I was learning suddenly the blood was visible through my uniform. I was shocked and embarrassed. I covered up my uniform with my school bag and went home. I didn't tell my mother what happened and I immediately went to my room so as not to be seen."* An older woman remarked: "*I was in school when I first had my period. When I felt it, I thought that it was urine and I had to go to the bathroom. When I went to the bathroom, I felt the blood. I had no underwear. I had to use the inside of my skirt and wrap it around my body. Then I went back to class and sat on my exercise book. I managed to avoid getting my dress soaked in blood by keeping it away from the blood. I had to wait until all of the other students were gone before I stood up and left the classroom. Then I went to my house and didn't talk to my parents or siblings. In the meantime, I had to steal my older sister's underwear and use it. I used dirty, worn-out clothes that were not useful for other purposes. Finally, I told my friend about it. She was my schoolmate. She spoke with our teacher and they bought me underwear and a pad."* Stories of this kind are common. Menstrual restrictions {#Sec6} ---------------------- Numerous cultural restrictions are imposed during menstruation in Tigray. Although girls and women are expected to continue their normal household chores while menstruating, there is a general belief that menstruants are vulnerable and should not do heavy physical labor while on their periods. Hard physical work is thought to increase the bleeding. One woman explained *"A woman's womb is open during menstruation and if you load yourself with a heavy load, it can cause a problem and too much bleeding."* The underlying belief appears to be that the uterus is like a sack full of blood and if you strain, more blood comes out. Women probably do notice an increase in bleeding with straining or lifting, particularly if they lack underwear or effective sanitary pads to contain the menstrual flow. On the other hand, there is also the pragmatic realization that the daily work of the household must get done irrespective of what else is happening. For poor women in particular, there is no alternative but to do the work themselves. Economic necessity overrides personal preferences. Many informants expressed the belief that menstruating women should avoid exposure to direct sunlight. There is a prevalent belief that if a menstruating woman is out in direct sunlight she will develop a debilitating and potentially serious medical condition called *michi*. Thus, to the ritually polluting nature of menstruation is added the belief that being "out" risks serious health consequences to the woman herself---a belief that also keeps the subject of menstruation "in the shadows," both figuratively and literally. A more common concern than the development of *michi*, however, is the presence of menstrual cramps and how to manage them. Dysmenorrhea (which is fairly common among adolescents, especially in the first few years after menarche) is often cited as a reason that menstruating girls do not attend school. School absences are also increased if they have difficulty accessing adequate menstrual hygiene supplies. Acetominophen (paracetamol) is frequently used as a treatment for cramps, but it is not always readily available in schools and access to more effective non-steroidal anti-inflammatory drugs such as ibuprofen or naproxen is almost non-existent. Many girls report trying to manage menstrual cramping by taking hot drinks, applying hot water bottles, or resorting to traditional folk-remedies such as rubbing the abdomen with butter to soothe the cramping. Lack of access to effective pain medications for menstrual cramps while at school was cited as a common problem that adversely impacted school attendance and learning. The intake of certain foods and fluids is thought by many to influence menstrual flow, but these beliefs are inconsistent and often contradictory, reflecting an uncertain understanding how they might work. (In fact, there is no physiological reason to believe that food or fluid intake would affect the quantity or duration of menstrual flow). Many people say menstruating girls should take hot drinks while menstruating, perhaps for their soothing qualities; but others deny this and say that hot drinks only serve to increase the amount of menstrual flow and that menstruants should take only cooling drinks instead. In many cultures there is a belief that "cooling" the blood leads to clot formation and sluggish menstrual flow, whereas heat is thought to increase menstrual circulation \[[@CR2]\]. Something conceptually similar to this may be in play here. Most respondents believe that women should take "good" (i.e., nourishing) food (many also mentioned iron supplements, perhaps a reflection of public health education pertaining to anemia) and stated that girls and women should avoid overwork while menstruating in order to conserve their strength. Unlike some parts of the world, there do not appear to be restrictions in Tigray on the ability of menstruating women to prepare food (assuming they keep themselves clean and wash regularly). Only one or two men said they would prefer *not* to eat food cooked by a menstruating woman. A more typical response was offered by one Orthodox priest (a married man), who, when asked whether or not he would eat food cooked by his wife if she was menstruating, said somewhat incredulously: *"Well, who else would prepare it for me?"* Cooking is a responsibility that women are expected to assume in the household and this is not abrogated by menstruation. Nonetheless, menstruation is foreign to male physiological experience and is not usually discussed openly. There is a general male wariness about menstrual blood. Many informants reported that diseases could be transmitted through menstrual blood (including genital infections, HIV, hepatitis, and even tetanus). One said: *"If women don't dispose of hygienic materials properly and if it touches body parts that may be sensitive to disease, it may cause disease."* Such beliefs about the potential infectiousness of blood have likely been increased by HIV/AIDS awareness campaigns in Ethiopia. When asked if it was permissible to have sexual relations with a woman while she was menstruating, there was generally strong disapproval of this practice by both males and females. Although some people believed that such intercourse would result in pregnancy, almost all respondents voiced cultural, religious, and aesthetic disapproval of having intercourse with a menstruating woman. One woman explicitly likened this to incest. Referring to menstruation as *"\[the thing\] of our mothers,*" she said *"If a man sleeps with his wife when she is on her menstrual period it is like he was sleeping with his mother,"*\--a very strong aversion indeed! Orthodox priests and Muslim imams both said that sex was not permitted during menses. *"I won't have sex with a menstruating girl,"* one young man reported, *"because her blood may contaminate me."* Another young man said *"It is forbidden to have sex with a menstruating girl in our tradition. Besides this,"* he added, *"girls smell bad during their menstrual periods."* A specific aesthetic objection was thereby added to the general cultural disapproval of this practice. There is a strong cultural tradition among both Orthodox Christians and Muslims that menstruating girls and women are ritually impure and should be excluded from religious activities. Menstruants are prohibited from entering the mosque or the church; rather, they should remain outside and pray by themselves. This largely appears to be a self-imposed restriction, however; we are not aware of any attempts to check the menstrual status of women at sites of religious worship. As such, a girl's menstrual status is a matter between her and God, but many people seem to feel that God Himself would be personally offended by a menstruating girl or woman who comes into His holy presence. One pre-menarchal girl declared with certainty: *"She \[a menstruating girl\] can't fast or pray in a church for it is useless and not accepted by God."* Many people also expressed the belief that menstruants should not touch sacred objects such as the Bible, a cross, holy water, or the Koran. Such beliefs about menstruation and ritual pollution put women at a serious disadvantage in a society where religious influences remain very strong and permeate many aspects of the culture. Hygiene needs, menstrual pads, and menstrual stigma {#Sec7} --------------------------------------------------- How to manage menstrual hygiene is an omnipresent concern for females, particularly for poor girls and women in rural areas where living standards may be often marginal. The embarrassment and shame that result from a public menstrual hygiene accident can be overwhelming for the girl who experiences it, particularly during her vulnerable early adolescent years. We heard many reports of schoolgirls who were taunted by boys (and sometimes girls) who ran after them, shouting and making fun of them if they had visible menstrual staining on their clothes at school. The mortification that results from such public shaming can be severe. One girl reported an incident in her school in which a very bright female student was called to the front of the classroom to write on the blackboard (an everyday experience in Ethiopian schools) only to realize that the entire class was staring at her backside. To her horror she discovered that the seat of her dress (which was turned directly to the class as she wrote on the blackboard) was soaked in menstrual blood. She was so ashamed that she left the classroom immediately and decided to change schools rather than return to the scene of her humiliation. Other girls who have experienced such accidents stop coming to school while on their periods; some simply drop out altogether. One older woman reported concerning her own girlhood: *It was difficult back in the day* \[when you were on your period.\] *You sat down and stayed in one place. If you were making coffee* \[a common and important Ethiopian social ritual for guests\]*, you would prepare everything but not move at all. You would be careful not to get blood on your dress. You just covered yourself and were self-conscious. You would be especially careful if there was a guest or a boy who might see you.* Another woman said: *"Back then there were no things like underwear and sanitary pads. I lived in a village and we didn't know about it. It was very difficult to socialize without having underwear. It was common to miss two or three days \[of school each month\]. I used to miss class. I menstruated at 15 or 16 but there was no underwear or sanitary pads. We wore traditional \[white\] dresses and it was not easy to control menstruation. We were farmers. If I was seen bleeding in class the students would make fun of me."* The rest of her focus group chimed in with enthusiastic support, shouting *"That is it! She has said it!"* Her comments touched a "social nerve" that was felt by all the women present. Many teachers reported that poor girls routinely skipped class during their menstrual periods for fear of being publicly exposed to such shaming. One adolescent boy remarked, *"I don't think girls ever stand up without looking at their seats because they know what will happen if students see blood."* One adolescent schoolgirl said *"At school there is not any kind of support* \[for menstruation\]. *If we bleed, we have to go home because we have financial problems* \[and cannot afford to buy menstrual pads\]*."* The fear of exposure of their menstruating state due to an accident of hygiene is constantly on the minds of girls and keeps them from concentrating on their lessons. One informant said, *"A menstruating girl is afraid to go to school and when in class she cannot pay attention due to fear of menstrual staining."* Another girl said, poignantly, *"My menstrual pad fell out while I was playing football \[soccer\] at school. All the students laughed at me and shouted at me and I missed school for three days."* The stress of participating in required vigorous physical activity while trying to wear a menstrual pad with no underwear to support it was extremely stressful and several women reported similar incidents in their lives. One of the most common obstacles to school attendance is lack of adequate menstrual hygiene supplies. Commercially produced, disposable pads are often not accessible, but even when they are available, they are too expensive for poor rural woman to buy. Proper disposal of such pads is also problematic. The price of a package of disposable menstrual pads was stated as 18--25 Birr (roughly \$US 0.80 to \$US 1.10) per month, but as several women stated, *"It is very expensive. We are poor"* and *"For a poor woman, it's difficult even to afford one pair of underwear."* One woman remarked, *"It's expensive, let alone 18 Birr! Even 8 Birr \[\$US 0.35\] is expensive!"* Another said, *"Some can't even afford three or four Birr"* \[\~\$US 0.18\]. A Muslim iman with a large family lamented, *"They cost 18 Birr---if you have three daughters it will be especially difficult."* For him, menstruation was a familial financial catastrophe. The alternative to using commercially-produced sanitary pads is to make do with strips of cloth torn from old dresses, pieces of old mattresses, or whatever absorbent materials may be at hand, but such remedies are often not effective. As one teacher summarized the problem, *"Most of the girls do not attend class* \[when they are menstruating\] *since they don't have menstrual pads or even clean pieces of cloth."* There is a general consensus that good hygiene is important. The most common view is that girls and women should wash themselves several times per day when they are menstruating and that this is particularly important if they don't have menstrual pads; but scarcity of water is a constant, pervasive problem throughout Tigray, particularly in schools. Women in rural areas may have to carry water several miles from the nearest water-source in order to have a supply at home, which makes water a precious household commodity. Lack of water for washing (let alone drinking) is very common, particularly in rural schools. Many schools have no source of running water on site and the only water available is whatever students can carry from home in containers for their daily use. There is also a belief (more notable in rural areas and among older women) that washing or bathing during menstruation increases the flow of blood. In the past, many girls were actually forbidden to wash while menstruating, which undoubtedly only increased their discomfort. These beliefs may be intertwined with one another. There was general agreement that schools should furnish menstrual pads to girls and that schools should also have medications for menstrual cramps available for student use. Both strategies were felt to be beneficial for improving school attendance. There was also general agreement that toilet facilities at schools were inadequate---especially for girls\-\--and needed much improvement. This is readily demonstrable in almost any school in the region \[Fig. [2](#Fig2){ref-type="fig"}\]. There was agreement that girls' toilets should be well separated from those used by males to ensure better privacy, especially when dealing with intimate issues such as washing and changing menstrual pads during the day. When asked in a focus group discussion what should be provided in school latrines, one woman stated forcefully: *"I would build a latrine with a shower and toilet inside. I would have well-made menstrual pads and soap! And it would have a continuous supply of water and soap!" \[much laughter and cheering from the group\].* Achieving this ideal seemed hopelessly remote, given the local economic circumstances, but nonetheless such improvements were greatly desired, and it was thought that such facilities would contribute substantially to improved gender equity in the Tigray Region.Fig. 2Typically inadequate toilet facilities at a secondary school in rural Tigray. Photo by the authors Discussion: Menstrual hygiene Management in Context {#Sec8} =================================================== In recent years menarche and menstruation increasingly have become recognized as important public health issues with particular salience for the health and psychosocial well-being of adolescent girls \[[@CR13]--[@CR16], [@CR20], [@CR22]--[@CR24]\]. The increasing attention paid to these issues has produced several qualitative studies from East Africa (Kenya, Tanzania, Uganda, and Ethiopia) which document the impact that lack of adequate resources for coping with menstrual hygiene has on adolescent girls \[[@CR6], [@CR12], [@CR19], [@CR22]--[@CR34]\]. Menstrual hygiene management is difficult and inadequate throughout Ethiopia (<http://www.pma2020.org/sites/default/files/ETR5-MHM%20brief-v1-2017-08-14_0.pdf>). Our study confirms and enlarges the qualitative research on menstruation, with specific reference to northern Ethiopia. The challenges Tigrayan schoolgirls face in managing their menses is rooted in a general cultural reluctance to discuss the subject of menstruation, a problem which is made worse by inadequate scientific education concerning sexuality and human reproductive biology. The failure to discuss menstruation at home (which is often grounded in parental discomfort with the topic, personal embarrassment, and a denial of their daughters' impending sexual awakening) means that many (perhaps most) girls arrive at menarche unprepared for the experience of menstruation. This phenomenon is certainly not limited to Tigray \[[@CR4]--[@CR20], [@CR30], [@CR31]\]. When their menstrual bleeding first begins---often unexpectedly---girls may be frightened, embarrassed, and suffer unnecessary psycho-social trauma. Some people in this region believe that menstruation does not begin until a girl has sexual relations \[[@CR22]\], which may precipitate unfounded, scurrilous rumors about sexual misbehavior and provoke baseless family conflict. Not knowing what is really happening during their menses, many girls are worried, fearful, and go to great lengths to hide their bleeding from others, especially from family members. For these girls, containing the menstrual flow is a major, anxiety-ridden preoccupation while they are on their periods. This task is made more difficult by lack of access to satisfactory menstrual management materials, which are either unavailable, too expensive, or have inadequate absorbent capacity. Male family members who control financial resources often do not understand the importance of this aspect of their daughters' lives because of ignorance or benign neglect \[[@CR35]\]. The resultant anxiety over possible menstrual hygiene accidents at school or elsewhere in public is often so high that girls stay home while menstruating and fall behind in their studies. The stigma associated with menstruation is increased by negative religious attitudes that categorize menstruating women as ritually unclean, as well as by thoughtless, mean, and sometimes cruel taunting and teasing behavior by fellow students (especially adolescent boys) when menstrual hygiene accidents occur. The school environment is perceived as difficult for menstruating girls due to the social pressures surrounding this normal physiological process, and this is worsened by pitifully inadequate toileting facilities, poorly constructed latrines, lack of access to soap and water for cleaning, lack of menstrual pads when menstruation begins unexpectedly, and lack of basic medications to help with menstrual cramps when these occur. Our findings from Tigray confirm the presence in that region of the problems so commonly reported elsewhere \[[@CR36]\]. The Kenyan comment that "the girl with her period is the one to hang her head" \[[@CR6]\] applies equally to the Tigray Region of northern Ethiopia. We have not, however, found evidence thus far that Tigrayan girls trade sexual favors for access to menstrual hygiene supplies as has been reported in Kenya \[[@CR29]\]. This should be a topic for further investigation. Meeting the challenges surrounding menstrual hygiene requires a multilateral approach. There is a great need for better instruction in the science of human reproductive biology, including the female hormonal cycle and how it relates to fertility and menstruation. A strong understanding of basic human reproductive biology would go far towards creating a more compassionate attitude towards menstruation. Cultural attitudes towards menstruation in Tigray are strongly influenced by religious perspectives on menstruation that developed centuries before the science underlying the female reproductive cycle was understood. A broader understanding of the biology of menstruation throughout the general population could open the way for more nuanced theological discussions of this and related topics. In addition, there is a great need for improving the physical environment of Tigrayan schools, especially as it relates to hygiene. In many places latrines are substandard, even unsafe (sometimes located well away from the school grounds and unprotected by fencing that would keep interlopers away). Boys' latrines should be segregated from girls' latrines and there should be enough room and enough privacy within girls' latrines to allow easy changing of menstrual pads and attention to basic washing and cleaning \[[@CR36]\]. The provision of fresh water is a particular challenge in Tigray, where drought is common and on-site water sources often do not exist. Each school should make sure that it has an acceptable, functioning gender office which can supply sanitary pads on an emergency basis, basic medication for the relief of menstrual cramps, and a place for girls to rest if necessary. The Tigray Regional Educational Bureau is taking steps to make this happen. By itself, improved education about menstruation appears to have a positive impact on menstrually-related absences by adolescent schoolgirls \[[@CR37]\]. It seems likely that provision of menstrual hygiene supplies to students also has a positive impact on school attendance by girls, although further research in this area is needed to solidify this perception \[[@CR37]--[@CR39]\]. In light of the fact that menstruation is an unavoidable experience for girls in low-resource countries, effective programs to lower the barrier this presents to school attendance and their further education is highly desirable. Strong evidence demonstrates that further education improves health and family outcomes, lowers child mortality, and produces positive economic benefits \[[@CR40], [@CR41]\]. These are all valuable public policy goals. A recalibration of school and community values to make menstrual hygiene a higher priority than it is at present will be necessary to bring this about. Conclusion {#Sec9} ========== Changes in the educational system are required to improve the understanding of menstruation by all students, male and female, to foster gender equity and to create a society more understanding and accepting of menstruation. Not applicable. Funding {#FPar1} ======= The study was made possible by a project grant to the College of Health Sciences, Mekelle University, Mekelle, Ethiopia, by Dignity Period, a not-for-profit public charity based in St. Louis, MO, in the United States. The study was jointly designed by Dignity Period and Mekelle University. Data analysis was carried out independently by the School of Public Health at Mekelle University and Makaio Consultants. Availability of data and materials {#FPar2} ================================== The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request. LLW: Protocol/project development; data management; data analysis; manuscript writing/editing. KT: Data management; data analysis; manuscript writing/editing. AD: Data management; data analysis; manuscript writing/editing. SB: Protocol/project development; data collection and management; data analysis; manuscript writing/editing. All authors have read and approved the final manuscript. Ethics approval and consent to participate {#FPar3} ========================================== All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards. Informed consent was obtained from all individual participants for whom identifying information is included in this article. The study protocol was reviewed and approved by the institutional review boards at Ayder Comprehensive Specialist Hospital, College of Health Sciences, Mekelle University, Mekelle, Ethiopia (ERC 0570/2015), and at Washington University in St. Louis, MO, USA (IRB \#201503071). Informed consent was obtained prior to each interview or focus group discussion, both oral (audio recording) and written consent was obtained from all adults. Assent to participate was obtained from minors, along with the written consent of the parent/guardian. The youngest study participant was 10 years old. Study subjects were compensated for their time and participation with 50 Ethiopian Birr (approximately US\$2.50 at the time of the study). Consent for publication {#FPar4} ======================= Not applicable. Competing interests {#FPar5} =================== L. Lewis Wall serves as a non-compensated member of the board of directors of the charity, Dignity Period. The other authors have nothing to declare. Publisher's Note {#FPar6} ================ Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Introduction {#Sec1} ============ Hemophagocytic lymphohistiocytosis (HLH) is a hyperinflammatory syndrome caused by excessive cytokine release, triggered by genetic or acquired overactivation of macrophages, T and natural killer (NK) cells \[[@CR1]\]. Clinical presentation may include fever, cytopenias, organomegaly, and hyperferritinemia, none of which are specific for this rare though life-threatening condition \[[@CR2]--[@CR4]\]. As HLH shares similarities with other inflammatory states, e.g., sepsis, its diagnosis is challenged by clinical overlap particularly in the intensive care unit (ICU) \[[@CR2]\]. Consequently, HLH is likely to be under-recognized in critically ill patients where evidence for clear definition and correct diagnostic workup is lacking \[[@CR5]\]. So far, diagnosis largely relies on data derived from studies conducted in pediatric patients \[[@CR6]\]. Henter et al. developed the HLH-2004 criteria whereby a diagnosis of HLH is confirmed if five out of eight criteria are fulfilled \[[@CR7]\]. However, these guidelines lack prospective validation in adult HLH patients. Moreover, the specificity of some criteria has been questioned. According to HLH-2004 guidelines, a ferritin ≥ 500 μg/L meets the criterion of hyperferritinemia \[[@CR7]\]. However, markedly higher ferritin levels have been seen in adult HLH patients \[[@CR8]\]. In fact, we detected best prediction accuracy at a ferritin cutoff level of 9083 μg/L with 92.5% sensitivity and 91.9% specificity for HLH in critically ill patients, thereby providing satisfying discrimination of HLH patients \[[@CR8]\]. The HScore published by Fardet et al. \[[@CR9]\] calculates a sum score of nine variables allowing to assess the probability of HLH. Each variable was assigned a maximum number varying between 18 and 64 points. The authors found the best discriminatory performance at an HScore of 169 with a sensitivity of 93.0% and specificity of 86.0%. Unlike the HLH-2004 criteria which are composed of parameters derived from a pediatric population, the HScore was developed in an adult cohort including patients aged ≥ 18 years. Yet, only non-ICU patients were included, possibly limiting the tool's generalizability to critically ill patients. Hence, it is unclear whether HLH-2004 criteria and HScore reliably detect and discriminate HLH in adult critically ill patients. We therefore calculated the sensitivity and specificity of HLH-2004 criteria and HScore, respectively, in a cohort of adult patients admitted to ICUs at an academic medical center. Methods {#Sec2} ======= Patients {#Sec3} -------- This further analysis of a retrospective observational study \[[@CR8]\] was conducted at the university hospital Charité -- Universitätsmedizin Berlin. Data of patients who were admitted to at least one adult surgical, anesthesiological, or medical ICU between January 2006 and August 2018 were reviewed and extracted from two electronic patient data management systems operated at the Charité -- Universitätsmedizin Berlin (COPRA, Sasbachwalden, Germany and SAP, Walldorf, Germany). We included all patients aged ≥ 18 years who had at least one ferritin value measured during ICU stay and hyperferritinemia of at least 500 μg/L according to HLH-2004 criteria \[[@CR7]\]. Of all patients included, we extracted data for body temperature, ferritin, blood counts, triglycerides, fibrinogen, soluble interleukin-2 receptor (sIL-2R), and aspartate aminotransferase (AST). Ultrasound, computed tomography (CT) scans and autopsy findings were reviewed to determine the presence of hepatomegaly and/or splenomegaly. Medical reports were screened for evidence of preexisting immunosuppression, while bone marrow findings were reviewed for hemophagocytosis. All variables were recorded at day of maximum ferritin assessment. If no assessment was documented that day, we extended the period to a plausible time range for each parameter according to our protocol (Table [1](#Tab1){ref-type="table"}). Using the obtained data, HLH-2004 criteria and HScore (Supplement Table S[1](#MOESM1){ref-type="media"}) were determined in all non-HLH patients. To avoid bias by pending parameters at the day of ferritin maximum, we used the highest number of fulfilled HLH-2004 criteria and maximum HScore in all HLH patients. The study period was defined from ICU admission until hospital discharge, transfer, or death. Table 1Data collection of variables for HLH-2004 criteria and HScoreVariablesTime range with regard to maximum ferritin (when not assessed at day of ferritin maximum)Hemoglobin, platelets, white blood cell count \[min\]± 3 daysFibrinogen \[min\]± 3 daysTriglycerides \[max\]± 5 daysBody temperature \[max\]± 5 daysAST \[max\]± 3 daysNK cell activity, CD107a \[max\]± 10 dayssIL-2RStudy periodHepatomegaly, splenomegalyStudy periodHemophagocytosisStudy periodPreexisting immunosuppressionObtained from medical records before study period*AST* aspartate aminotransferase, *Max* maximum, *Min* minimum, *NK* natural killer cell, *sIL-2R* soluble interleukin-2 receptor Diagnosis of HLH {#Sec4} ---------------- The charts of all included patients were reviewed for clinically diagnosed or suspected HLH. In parallel, we searched for all adult ICU patients diagnosed with ICD-10 codes for HLH (D76.1, D76.2, and D76.3). Only cases with previously suspected or diagnosed HLH by clinicians were reviewed by two HLH experts who confirmed or rejected HLH diagnosis based on HLH-2004 criteria and HScore (Supplement Table S[1](#MOESM1){ref-type="media"}) while considering patient's history and clinical presentation, according to current recommendations \[[@CR6]\]. Importantly, the diagnosis of HLH was confirmed before HLH-2004 criteria and HScore were determined in all non-HLH patients, i.e., patients who were not previously diagnosed or suspected for HLH by clinicians. Though HLH-2004 criteria and HScore were determined in latter patients, these were not reviewed for HLH by the experts. Of note, HLH patients comprise 7 cases of previously undiagnosed HLH who have been retrospectively detected and described by our research group \[[@CR5]\]. Statistical analysis {#Sec5} -------------------- Results are reported as median (percentiles) or as counts (relative frequencies) according to their scaling. Comparison of HLH and non-HLH patients was conducted using the non-parametric Mann-Whitney *U* test for continuous variables and the chi-square test for categorical variables. Receiver operating characteristics (ROC) analysis was performed to determine the best prediction accuracy of each diagnostic variable for HLH diagnosis and for fixed combinations. As a post hoc analysis, we rerun ROC analyses for HLH-2004 criteria while raising hyperferritinemia cutoff from 500 to 3000 μg/L based on the lowest ferritin maximum in HLH patients (3102 μg/L). To analyze the best fever cutoff, we reiterated ROC analyses for HLH-2004 criteria using fever cutoffs from 38.0 to 38.5 °C (38.3 °C was used for main analyses as shown in Supplement Table S[1](#MOESM1){ref-type="media"}). For analysis of hepatomegaly, we extended splenomegaly to spleno- and/or hepatomegaly in another post hoc ROC analyses for HLH-2004 criteria. As a sensitivity analysis, we rerun ROC analyses for HLH-2004 criteria and HScore with restriction to patients with at least 5 assessed HLH-2004 criteria. Multivariable logistic regression analysis was performed to assess associations between HLH-2004 criteria and HScore, respectively, with in-hospital mortality while adjusting for age, sex, body mass index (BMI), and maximum sequential organ failure assessment (SOFA) score. All tests should be understood as constituting exploratory data analysis. No adjustments for multiple testing were made. A two-tailed *P* value \< 0.05 was considered statistically significant. All numerical calculations were performed with IBM© SPSS© Statistics, Version 26, © Copyright 1989, 2010 SPSS Inc. Results {#Sec6} ======= Study population and characteristics {#Sec7} ------------------------------------ Between January 2006 and August 2018, 6340 of 116,310 ICU patients had at least one ferritin measurement during ICU stay and were ≥ 18 years old. Of these, 2623 patients with hyperferritinemia (≥ 500 μg/L) were included into the final analyses. Among those, 50 patients had initially been diagnosed or suspected with HLH by clinicians of whom 40 cases were confirmed by the experts (Fig. [1](#Fig1){ref-type="fig"}). The remaining 10 patients (Supplement Table S[2](#MOESM1){ref-type="media"}) had either low number of fulfilled HLH-2004 criteria (\< 4) or low HScore (\< 190); in three of the 50 patients, clinical judgment was decisive. Basic patient characteristics, values of HLH-2004 criteria and HScore, and outcome parameters are shown in Table [2](#Tab2){ref-type="table"}. Distribution of fulfilled HLH-2004 criteria and HScore over all patients is shown in Fig. [2](#Fig2){ref-type="fig"}. The group of HLH patients has been described in detail previously \[[@CR4]\]. The overall cohort of 2623 patients has already been published \[[@CR8]\] to analyze hyperferritinemia between patients with HLH, sepsis, septic shock, and other diagnoses. Fig. 1Consort diagramTable 2Basic patient characteristics, biomarkers, and outcome parametersParametersHLH patients (***n*** = 40)Non-HLH patients (***n*** = 2583)***P*** valueAge \[years\]47 (33--62)62 (49--73)\< 0.001^‡^Male sex \[*n*\] (%)26 (65.0%)1588 (61.5%)0.650^†^Body mass index \[kg/m^2^\]23.0 (21.0--26.5)25.0 (22.0--29.0)0.094^‡^Sepsis without shock \[*n*\] (%)12 (30.0%)1003 (38.8%)0.255^†^Septic shock \[*n*\] (%)23 (57.5%)626 (24.2%)\< 0.001^†^Hemodialysis \[*n*\] (%)29 (72.5%)1357 (52.5%)0.012^†^ECLA/ECMO \[*n*\] (%)6 (15.0%)188 (7.3%)0.064^†^ICU admission SOFA score9 (6--13)6 (3--9)\< 0.001^‡^Maximum SOFA score17 (12--19)11 (7--15)\< 0.001^‡^HLH-2004 criteria Measured7 (6--7)4 (4--5)\< 0.001^‡^ Fulfilled5 (4--6)2 (1--2)\< 0.001^‡^HScore258 (225--280)62 (33--101)\< 0.001^‡^Bi-/pancytopenia \[*n*\] (%)\*, *n* = 40\|258237 (92.5%)471 (18.2%)\< 0.001^†^Hemoglobin \[g/dL\]6.9 (6.4--7.6)8.8 (7.9--9.9)\< 0.001^‡^Platelet count \[/nL\]18 (5--34)170 (88--270)\< 0.001^‡^Leukocyte count \[/nL\]0.9 (0.2--2.7)9.0 (6.1--13.3)\< 0.001^‡^Hypofibrinogenemia or hypertriglyceridemia \[*n*\] (%), *n* = 40\|151730 (75.0%)288 (19.0%)\< 0.001^†^Fibrinogen \[mg/dL\], *n* = 39\|11442.0 (1.0--3.0)3.8 (2.5--5.3)\< 0.001^‡^Triglycerides \[mg/dL\], *n* = 39\|855376 (245--563)158 (104--247)\< 0.001^‡^Max. core body temperature \[°C\], *n* = 40\|245239.1 (38.5--39.8)38.2 (37.5--38.9)\< 0.001^†^Splenomegaly \[*n*\] (%), *n* = 40\|200826 (65.0%)401 (20.0%)\< 0.001^†^Hepatomegaly \[*n*\] (%), *n* = 40\|203723 (57.5%)328 (16.1%)\< 0.001^†^Hemophagocytosis \[*n*\] (%), *n* = 31\|22116 (51.6%)15 (6.8%)\< 0.001^†^AST \[U/L\], *n* = 40\|2327171 (119--498)47 (26--108)\< 0.001^‡^Pre-existing immunosuppression \[*n*\] (%)30 (75.0%)728 (28.2%)\< 0.001^†^ICU duration \[d\]20.0 (11.3--37.3)19.0 (6.0--47.1)0.522^‡^In-patient duration \[d\]27.7 (18.6--77.4)38.3 (18.1--76.1)0.682^‡^Deceased \[*n*\]24 (60.0%)741 (28.7%)\< 0.001^†^Diagnostic parameters with *n* representing the number of patients with available data in each group, if not available in all patients; continuous quantities in median with quartiles. CD107a testing as a functional marker for identification of NK cell activity was performed in 4 patients only; however, three showed values within the normal range and one could not be analyzed due to low NK cell count. Values of ferritin and sIL-2R of the cohort were already described in Lachmann et al. \[[@CR8]\]*ECLA* extracorporeal lung assist, *ECMO* extracorporeal membrane oxygenation, *ICU* intensive care unit, *SOFA* sequential organ failure assessment^‡^*P* values calculated using the Mann-Whitney *U* test^†^*P* values calculated using the *χ*^2^ test\*Leukopenia was assumed by white blood cell count \< 1.67/nLFig. 2Distribution of fulfilled HLH-2004 criteria and HScore over all patients Sensitivity and specificity of HLH-2004 criteria and HScore {#Sec8} ----------------------------------------------------------- Twenty-nine HLH patients (72.5%) and 33 non-HLH patients (1.3%) fulfilled at least 5 HLH-2004 criteria. ROC curves of fulfilled HLH-2004 criteria and HScore are shown in Fig. [3](#Fig3){ref-type="fig"}. The best prediction accuracy for HLH was seen for a cutoff of 4 fulfilled HLH-2004 criteria and an HScore cutoff of 168. The sensitivity and specificity of each criterion are shown in Table [3](#Tab3){ref-type="table"}, while the analyses of ferritin and sIL-2R were already published in Lachmann et al. \[[@CR8]\]. When analyses were restricted to patients with at least 5 measured HLH-2004 criteria (*n* = 1303), we found 95.0% sensitivity and 88.0% specificity for the cutoff of 4 fulfilled HLH-2004 criteria (AUC 0.968 (95% CI 0.950--0.986)) as well as 100% sensitivity and 88.6% specificity for the HScore cutoff of 168 (AUC 0.984 (95% CI 0.975--0.993)). Fig. 3Receiver operating characteristic curves of fulfilled HLH-2004 criteria and HScoreTable 3Sensitivity and specificity of fulfilled HLH-2004 criteria and HScoreParametersAUC (95% CI)CutoffSensitivity/specificity**HLH-2004 criteria**0.982 (0.971--0.993)2100%/38.4%3100%/77.8%495.0%/93.6%572.5%/98.7%645.0%/99.9%710.0%/100%**HScore**0.992 (0.987--0.996)140100%/88.2%150100%/90.7%160100%/93.1%168100%/94.1%17097.5%/94.2%18095.0%/95.3%19095.0%/96.3%20090.0%/97.3%21085.0%/98.3%Bi-/pancytopenia0.871 (0.822--0.920)Yes92.5%/81.8%Hypofibrinogenemia or hypertriglyceridemia0.780 (0.702--0.859)Yes75.0%/81.0% Fibrinogen \[g/L\]0.760 (0.677--0.843)1.543.6%/91.1%3.179.5%/64.5% Triglycerides \[mg/dL\]0.830 (0.776--0.883)22984.6%/71.9%26566.7%/77.4%Max. core body temperature \[°C\]0.737 (0.656--0.818)38.580.0%/59.8%Splenomegaly0.725 (0.638--0.812)Yes65.0%/80.0%Hemophagocytosis0.724 (0.611--0.837)Yes51.6%/93.2%Receiver operating characteristics (ROC) analysis to determine the best prediction accuracy of each diagnostic variable for HLH diagnosis. CD107a testing as a functional marker for identification of NK cell activity not shown as performed in 4 patients only. The predictive values of ferritin and sIL-2R of the cohort were already analyzed in Lachmann et al. \[[@CR8]\]*AUC* Area under the curve, *CI* confidence interval Post hoc analyses of HLH-2004 criteria cutoffs {#Sec9} ---------------------------------------------- When hyperferritinemia cutoff was raised from 500 to 3000 μg/L, the specificity of 4 fulfilled HLH-2004 criteria increased to 96.1%, while sensitivity remained 95.0% (AUC 0.989 (95% CI 0.983--0.996)). By analyses of different fever cutoffs, best prediction accuracy was found for 38.2 °C (97.5% sensitivity and 93.5% specificity for 4 fulfilled HLH-2004 criteria (AUC 0.984 (95% CI 0.975--0.994))). By adjusting cutoffs of both hyperferritinemia to 3000 μg/L and fever to 38.2 °C, the sensitivity and specificity of 4 fulfilled HLH-2004 criteria were 97.5% and 96.1%, respectively (AUC 0.991 (95% CI 0.985--0.996)). Extension of splenomegaly to spleno- and/or hepatomegaly reduced specificity of 4 fulfilled HLH-2004 criteria from 93.6 to 92.2% while the sensitivity of 95.0% was unchanged (AUC 0.981 (95% CI 0.968--0.993)). Analyses of fixed combinations of fulfilled HLH-2004 criteria showed less prediction accuracy compared to independent combinations of at least 4 fulfilled HLH-2004 criteria (Supplement Table S[3](#MOESM1){ref-type="media"}). HLH-2004 criteria and HScore for prediction of mortality {#Sec10} -------------------------------------------------------- Multivariable logistic regression analysis including age, sex, BMI, and maximum SOFA score as confounders revealed statistically significant associations between in-hospital mortality and fulfilled HLH-2004 criteria or HScore, respectively (Table [4](#Tab4){ref-type="table"}). In-hospital mortality of fulfilled HLH-2004 criteria and HScore strata is shown in Supplement Table S[4](#MOESM1){ref-type="media"}. Table 4Multivariable logistic regression analyses for in-hospital mortalityCovariatesHLH-2004 criteriaHScoreOR95% CI***P*** valueOR95% CI***P*** valueAge, years1.0251.018--1.031\< 0.0011.0311.024--1.038\< 0.001Sex (male)0.9430.778--1.1430.5500.8970.737--1.0910.276BMI, kg/m^2^0.9790.965--0.9940.0050.9800.965--0.9950.009SOFA score max1.1631.140--1.185\< 0.0011.1481.126--1.171\< 0.001Fulfilled HLH-2004 criteria1.5131.372--1.667\< 0.001--HScore--1.0111.009--1.013\< 0.001Multivariable logistic regression analyses were performed with in-hospital death as dependent variable*BMI* body mass index, *CI* confidence interval, *OR* odds ratio, *SOFA* sequential organ failure assessment Discussion {#Sec11} ========== This is the largest study investigating the diagnostic performance of HLH-2004 criteria and HScore in an adult intensive care population. We found the best prediction accuracy of HLH diagnosis for a cutoff of 4 fulfilled HLH-2004 criteria (95.0% sensitivity and 93.6% specificity) and an HScore cutoff of 168 (100% sensitivity and 94.1% specificity). Analyses of each single HLH-2004 criterion revealed good sensitivity and specificity for a ferritin cutoff of 9083 μg/L described previously \[[@CR8]\] while all other HLH-2004 criteria had unsatisfying predictive ability in our study, including sIL-2R, which was also described previously \[[@CR8]\]. The combination of 4 fulfilled HLH-2004 criteria provided better diagnostic accuracy compared to each single criterion. By adjusting HLH-2004 criteria cutoffs of both hyperferritinemia to 3000 μg/L and fever to 38.2 °C, sensitivity and specificity increased to 97.5 and 96.1%, respectively. Our approach to analyze fixed combinations of fulfilled HLH-2004 criteria showed less prediction accuracy compared to independent combinations of at least 4 fulfilled HLH-2004 criteria. Both HLH-2004 criteria and HScore were independently associated with in-hospital mortality. The HLH-2004 criteria, currently the standard in HLH diagnosis, have been developed in pediatric populations but so far have not been validated in adult patients. According to current recommendations for HLH in adults, HLH diagnosis requires ≥ 5 fulfilled HLH-2004 criteria which should be considered along with patient's history and clinical presentation \[[@CR6]\]. In daily practice, clinical presentation might be suggestive of HLH, while less than 5 out of 8 HLH-2004 criteria are present. Moreover, the diagnostic value of some criteria, e.g., fever, is limited, particularly in critically ill patients where the use of antipyretic agents and devices such as extracorporeal membrane oxygenation (ECMO) and hemodialysis are frequently seen rendering body temperature an unreliable or even invalid parameter. Of note, the cutoff of 4 fulfilled HLH-2004 criteria had the best sensitivity and specificity possibly allowing faster HLH diagnosis, prompt treatment, and thus improved survival. Yet, these findings need further confirmation in prospective studies to validate safe HLH diagnosis in adults with only 4 fulfilled HLH-2004 criteria. Ongoing studies could contribute to improve safe HLH diagnosis in adult critically ill patients \[[@CR10]\]. Our analysis of sensitivity and specificity of single HLH-2004 criteria is broadly in line with data reported in pediatric HLH patients. Hypofibrinogenemia is known to have high specificity but rather low sensitivity as only 53% of children with HLH had fibrinogen levels \< 1.5 g/L \[[@CR11]\]. Also, we found sensitivity for fibrinogen of 1.5 g/L to be at 43.6% while specificity was at 91.1%. In contrast to studies in pediatric populations, sIL-2R proves to be of insufficient diagnostic value in adult patients \[[@CR8]\] whereas levels ≥ 2400 U/L in children provided good sensitivity and excellent specificity of 93.0% and 100%, respectively \[[@CR11]\]. The diagnostic value of ferritin in the present cohort has been described previously by our research group and appeared as a good screening marker \[[@CR8]\]. Importantly, the presence of 4 fulfilled HLH-2004 criteria provides higher sensitivity and specificity for HLH diagnosis than ferritin alone. Of note, sensitivity and specificity of 4 fulfilled HLH-2004 criteria increased to 97.5% and 96.1%, respectively, when cutoffs of both hyperferritinemia and fever were adjusted to 3000 μg/L and 38.2 °C, respectively. In this context it is noteworthy that 5 out of 8 HLH-2004 criteria can be fulfilled in critically ill non-HLH patients. NK cell activity was assessed in four patients only in whom HLH was likely considered as a differential diagnosis. For practical guidance, we recommend assessment of body temperature, cytopenias, ferritin, triglycerides, fibrinogen, splenomegaly, and wherever available sIL-2R. Hemophagocytosis, even though the eponymous feature of HLH with high specificity, is an unreliable diagnostic marker with only poor sensitivity, again particularly in critically ill patients with sepsis \[[@CR12]\]. However, the latest recommendations for HLH in adult patients advise bone marrow investigation as it helps to detect occult hemato-oncological malignancies and to differentiate between cytopenias caused by chemotherapy from patients who have actually underlying HLH \[[@CR6]\]. The HScore developed by Fardet et al. \[[@CR9]\] provides a tool to predict the probability of HLH diagnosis in adults. The authors found the best cutoff at an HScore of 169 yielding 93.0% sensitivity and 86.0% specificity in a cohort of non-ICU patients. Our present study included ICU patients only and revealed an HScore of 168 to have the best sensitivity and specificity of 100% and 94.1%, respectively, thereby providing slightly superior prediction accuracy compared to the HLH-2004 criteria. Importantly, the similar found cutoff underlines the value of the HScore for HLH diagnosis and its reliability in critically ill patients. One previous study by Meena et al. \[[@CR13]\] also analyzed the diagnostic performance of HLH-2004 criteria and HScore in critically ill patients. The authors included 445 patients with ferritin assessment among whom ten were diagnosed with HLH. They reported an HScore of 143.5 for best possible classification and found 5 out of 6 criteria to be the cutoff for HLH-2004 criteria with 70% and sensitivity and 97.2% specificity. However, we present a larger cohort of 2623 patients including 40 HLH cases. Yet, the work by Meena et al. and our study are currently the only data available investigating the diagnostic standard for HLH diagnosis in the adult ICU population. Both HLH-2004 criteria and HScore were associated with in-hospital mortality suggesting that both indicate disease severity. This relationship has been reported previously by Gualdoni et al. who found increased 30-day mortality correlating with HLH-2004- or HScore-positive patients \[[@CR14]\]. Our study has several limitations. As this is a retrospective study, data availability had to rely on patients who had a ferritin assessment during their ICU stay. This might constitute an important selection bias as patients with ferritin assessment might have been more severely ill. For instance, suspicion of inflammation or diagnostic of anemia was likely when ferritin assessment was considered. Thus, our findings might not be generalizable to ICU patients without hyperferritinemia. In addition, not all variables of HLH-2004 criteria and HScore were available in all patients which reflects clinical practice where rather rare diagnostic tests such as NK cell activity might be unavailable. Our study bears a considerable risk that HLH cases could have remained undiagnosed depending on physicians' expertises, particularly in patients with ≥ 5 fulfilled HLH-2004 criteria. Conclusions {#Sec12} =========== This is currently the largest study investigating the diagnostic performance of HLH-2004 criteria and HScore in an adult ICU cohort. Four fulfilled HLH-2004 criteria as cutoff for a diagnosis of HLH had a sensitivity of 95.0% and a specificity of 93.6%. By adjusting cutoffs of both hyperferritinemia to 3000 μg/L and fever to 38.2 °C, sensitivity and specificity increased to 97.5% and 96.1%, respectively. An HScore cutoff of 168 revealed a sensitivity of 100% and a specificity of 94.1%, thereby providing slightly superior diagnostic accuracy compared to HLH-2004 criteria. With regard to single criteria, ferritin demonstrated the best diagnostic performance of all 8 HLH-2004 criteria warranting its use as a reliable screening parameter for HLH diagnosis. Both HLH-2004 criteria and HScore proved to be of good diagnostic accuracy and consequently might be used for HLH diagnosis in critically ill patients. Supplementary information ========================= {#Sec13} **Additional file 1: Supplemental Table S1.** HLH-2004 criteria and HScore ([@CR7], [@CR9]). AST*, aspartate aminotransferase;* hb, *hemoglobin;* mM, *mmoles/liter;* plt, *platelets;* U, *Units.***Supplemental Table S2.** Fulfilled HLH-2004 criteria and HScore of suspected HLH patients where HLH was not confirmed. **Supplemental Table S3.** Sensitivity and specificity of fixed combinations of fulfilled HLH-2004 criteria. *\*Patients with complete obtained data in each category. AUC, Area under the curve; CI, confidence interval. Receiver operating characteristics (ROC) analysis to determine best prediction accuracy of each category (dichotomous variable) for HLH diagnosis.***Supplemental Table S4.** In-hospital mortality of fulfilled HLH-2004 criteria and HScore strata. AST : Aspartate aminotransferase AUC : Area under the curve BMI : Body mass index CI : Confidence interval CT : Computed tomography ECLA : Extracorporeal lung assist ECMO : Extracorporeal membrane oxygenation Hb : Hemoglobin HLH : Hemophagocytic lymphohistiocytosis HPS : Hemophagocytic syndrome ICD : International classification of diseases ICU : Intensive care unit MAS : Macrophage activation syndrome NK : Natural killer cell OR : Odds ratio Plt : Platelets ROC : Receiver operating characteristics sIL-2R : Soluble interleukin-2 receptor SOFA : Sequential organ failure assessment U : Units **Publisher's Note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Supplementary information ========================= **Supplementary information** accompanies this paper at 10.1186/s13054-020-02941-3. We thank the Department of Cardiovascular Surgery, the Department of Surgery CCM/CVK, the Medical Department, Division of Nephrology and Internal Intensive Care Medicine (CVK/CCM), the Medical Department, Division of Infectiology and Pneumonology, the Medical Department, Division of Cardiology (CVK), the Department of Cardiology (CBF), the Department of Neurology with Experimental Neurology, and the Department of Anesthesiology and Operative Intensive Care Medicine (CBF) for being part of our study, providing the data and excellent collaboration. We are grateful to Oguzhan Mizrak for his help with data acquisition. We acknowledge support from the German Research Foundation (DFG) and the Open Access Publication Funds of Charité -- Universitätsmedizin Berlin. Conceived and designed the study: CK, GL. Obtained the data: CK, PN, FSS, PH, FB, GL. Analyzed the data: CK, GJ, GL. Wrote the manuscript: CK, PL, GJ, GL. Commented on the manuscript: all authors. The authors read and approved the final manuscript. Gunnar Lachmann is a participant of the Berlin Institute of Health (BIH) Charité Clinician Scientist Program funded by Charité -- Universitätsmedizin Berlin and BIH. Due to legal restrictions imposed by the data protection commissioner of the Charité -- Universitätsmedizin Berlin, public sharing of study data with other researchers or entities is restricted to anonymized data. Requests may be sent to dai-researchdata\@charite.de. Ethics approval was obtained from the institutional review board (Ethikkommission der Charité -- Universitätsmedizin Berlin, EA1/176/16). Not applicable The authors declare no competing interests.

During intraoperative manipulation of colorectal carcinomas, levels of circulating tumor cells (CTC) can be demonstrated in peripheral and portal blood \[[@CR1]\]. Recently, it has been demonstrated that the cumulative percentage of CTC in peripheral and portal blood was significantly lower after laparoscopic resection in which no-touch technique was used \[[@CR2]\]. This term refers to the principal of early lymphovascular ligation before manipulation of the tumor. In 1966, Turnbull et al. showed that early ligation increased survival rates \[[@CR3]\]. A randomized trial suggested favorable patient-free survival and overall survival rate, however without significant difference \[[@CR4]\]. Today, the no-touch technique is not considered the standard surgical approach in open surgery. However in laparoscopic colorectal surgery, no-touch principles can optimally be represented with the preferable medial to lateral approach where the supplying vessels are ligated before the tumor is manipulated. If fewer CTC are detected in blood during laparoscopic surgery, it can be hypothesized that the same holds true for lymphatic flow with occult tumor cells (OTC) passing the lymphatic sinus. Based on the sentinel lymph node (SN) concept in which the lymphatic drainage from the primary tumor follows a specific order, we showed that OTC are preferentially found in the SN of patients with colorectal cancer \[[@CR5]\]. The objective of this study is to assess the effect of the surgical approach (i.e., open lateral to medial versus laparoscopic no-touch medial to lateral approach) on levels of occult tumor cells in sentinel lymph nodes of patients with stage I and II colorectal cancer. Patients and methods {#Sec1} ==================== Study population {#Sec2} ---------------- A prospective consecutive series of patients operated on between November 2006 and June 2009 were analyzed. Only patients undergoing potentially curative resection for biopsy-proven colorectal cancer were eligible. Patients with solid organ metastases detectable by preoperative radiological staging or intraoperative visualization were excluded, as were patients with macrometastases on conventional H&E slides. In total, 107 patients with stage I and II colorectal cancer were included (Fig. [1](#Fig1){ref-type="fig"}).Fig. 1Patient stream by application of inclusion criteria In 87 patients the choice of open or laparoscopic approach depended on the surgeon to whom the patient was referred. Twenty patients also participated in a randomized trial comparing open and laparoscopic surgery (ISRCTN: 79588422) \[[@CR6]\]. All procedures were performed by experienced colorectal surgeons performing more than 20 procedures each year. All laparoscopic procedures were performed by one surgeon expert at laparoscopic surgery. All these resections were performed according to the no-touch isolation technique (i.e., medial to lateral approach with early vessel ligation). Lateral to medial approach was performed during open resection. Patients who had early conversion, e.g., because of dense adhesions, were analyzed in the open group if the vascular trunk was not ligated before mobilization of the bowel and the lateral to medial approach was used. The study was done in accordance with the guidelines of the local ethics committee. Sentinel node procedure {#Sec3} ----------------------- A SN procedure was performed in all patients. Ex vivo sentinel lymph node mapping was used, as described previously in detail \[[@CR7]\]. After resection, 0.5--2 ml patent blue (depending on the volume of the tumor) was injected below the subserosa, around the tumor, with the colonic specimen left intact. The first one to four blue lymph nodes were identified as sentinel nodes and either dissected or marked with a suture. Pathological examination {#Sec4} ------------------------ The surgical resection specimens were analyzed immediately at the department of pathology using a standardized protocol. Tumor stage and grading were classified according to the sixth edition of the AJCC tumor--node--metastasis (TNM) classification \[[@CR8]\]. All lymph nodes (SNs and non-SNs) were stained with hematoxylin and eosin and evaluated for tumor involvement. For histologically proven N0 patients, three serial sections at 500-μm intervals were immunohistochemically stained with three different monoclonal antibodies to reveal OTC. The anti-epithelial cell antibody Ber-EP4, directed against membrane glycoproteins (DAKO, The Netherlands), was combined with two anticytokeratin antibodies: CK20 with its expression limited to gastrointestinal epithelial cells (Euro Diagnostica, Arnhem, The Netherlands) and Cam5.2, directed against cytokeratin 7 and 8 expressed in all epithelial cells (Becton and Dickinson, Alphen aan den Rijn, The Netherlands). Immunohistochemically detected cells with any of the three antibodies were considered as OTC only when they showed unequivocal morphological features of cancer cells. These tumor cells within lymph nodes were subdivided into two categories according to the AJCC revised guidelines: tumor cell deposits between 0.2 mm and 2.0 mm were referred to as micrometastases, and those smaller than 0.2 mm as isolated tumor cells (ITC) \[[@CR9]\]. Statistical analysis {#Sec5} -------------------- Statistical calculations were performed using the Statistical Software Package version 14.0 (SPSS Inc., Chicago, IL, USA). Categorical variables were compared using the chi-square test, while the *t*-test was used for continuous data with normal distribution. Univariate logistic regression analysis was used to identify factors predictive for detection of OTC in histologically proven N0 patients. Method of resection, T-stage, differentiation grade, and tumor diameter were selected a priori as important cofactors in identification of CTC. Since the limited number of events (from a statistical point of view) meant that only a restricted number of possible predictors could be included \[[@CR10]\], variables with multiple categories were recoded into dichotomous variables by combining categories with comparable prognosis (T-stage: I and II versus III and IV; differentiation grade: well and moderate versus poor). For tumor diameter, a cutoff of 35 mm was decided upon, as suggested in a study of a specifically constructed receiver-operating curve \[[@CR11]\]. All tests were performed two sided, and *p* \< 0.05 was considered statistically significant. To determine the clinicopathological characteristics predictive for presence of OTC, logistic regression was performed. Results {#Sec6} ======= Characteristics of the study population {#Sec7} --------------------------------------- During the study period, 107 patients with stage I and II colorectal cancer were included (Table [1](#Tab1){ref-type="table"}). Sixty-two patients were analyzed in the open group, whereas 45 patients were analyzed in the laparoscopic group (Fig. [1](#Fig1){ref-type="fig"}). There were 48 right-sided colectomies, 5 left-sided colectomies, 53 (recto)sigmoid resections, and 1 subtotal colectomy.Table 1Clinicopathological characteristics of the included patientsAll patients (*n* = 107)Open group (*n* = 62)Laparoscopic group (*n* = 45)*p* ValueAge (years)^a^70 ± 1170 ± 1069 ± 120.8Gender Male^b^4828200.9 Female ^b^593425Body mass index (kg/m^2^)^c^26 (18--31)24 (18--28)26 (22--31)0.7Preoperative CEA level ^c^2.2 (0.5--10)2.4 (0.9--10)2.2 (0.5--8.4)0.2Tumor location ^b^0.2 Colon875334 Rectum20911Tumor diameter (cm)^a^5.1 ± 24.2 ± 2.00.04Number of resected lymph nodes^a^15.5 ± 715.5 ± 715.6 ± 71.0Number of identified SN ^a^2.0 ± 21.9 ± 2.12.1 ± 2.40.7Depth of invasion^b^0.7 pT1835 pT2291712 pT3684127 pT4211Differentiation grade^b^0.2 Well954 Moderate804337 Poor18144Lymphovascular invasion^b^0.1 Absent975443 Present1082^a^Mean (SD)^b^Absolute numbers^c^ Median (min.--max.) The patient and tumor characteristics were not different between the two patient groups, except for tumor size, which was slightly larger in the open group (Table [1](#Tab1){ref-type="table"}). Immunohistochemical detection of occult tumor cells {#Sec8} --------------------------------------------------- A total of 208 SNs were analyzed. Ten patients had true micrometastases of more than 0.2 mm in one or more SNs (9.3%, pN1mi+), whereas ITC were found in 23 patients (21.5%, pN0itc+). Presence of micrometastases was equally distributed between the open and laparoscopic groups (*n* = 5 in both groups) (Table [2](#Tab2){ref-type="table"}).Table 2Presence of occult tumor cells in SN in patients with colorectal cancer: comparison between open and laparoscopic resectionAll patients (*n* = 107)Open group (*n* = 62)Laparoscopic group (*n* = 45)*p* Value Difference (%)Occult tumor cells332310Micrometastases10550.60.2--2.0 mm8%11%--3% (--15 to 8%)Isolated tumor cells231850.03\<0.2 mm29%11%18% (3--33%) In contrast, isolated tumor cells were more frequently found in the conventional open resection group when compared with the no-touch laparoscopic group. Eighteen patients (29%) in the open group had ITC, compared with five patients (11%) in the laparoscopic group. This results in an absolute risk reduction for incidence of ITC of 18% and a number needed to treat of six in favor of the laparoscopic group (Table [2](#Tab2){ref-type="table"}). One of the five patients in the laparoscopic group with ITC had a larger cluster of 0.18 mm with evident signs of proliferation and stromal reaction, which could therefore be considered as a micrometastasis, although we adhered to the quantitative characteristics of the AJCC. ITC were most frequently found in the sinuses of the lymph nodes, as would be expected when these cells are considered as transiently shed cells with limited lifespan. Predictive factors for presence of occult tumor cells {#Sec9} ----------------------------------------------------- A relation between presence of OTC and larger tumors (both infiltration depth and tumor diameter) was found. The occurrence of true micrometastases in the SN was strongly related to lymphovascular invasion, irrespective of surgical approach. However, this could not be demonstrated for occurrence of isolated tumor cells. Presence of these cells was only related to method of resection, with odds ratio of 3.3 (1.1--9.6) for open lateral to medial resection (Table [3](#Tab3){ref-type="table"}).Table 3Logistic regression analyses of predictors for presence of occult tumor cells, micrometastases (0.2--2.0 mm), and isolated tumor cells (\<0.2 mm) in histologically N0 patients with colorectal cancerVariableOccult tumor cellsIsolated tumor cellsOR^a^ (CI^b^)*p* ValueOR^a^ (CI^b^)*p* ValueMethod of resectionOpen versus laparoscopic0.5 (0.20--1.16)0.13.3 (1.11--9.63)0.03Depth of invasionpT3/4 versus pT1/22.5 (0.99--6.6)0.062.2 (0.75--6.6)0.2Differentiation gradePoor versus well and moderate2.1 (0.75--5.78)0.22.1 (0.70--6.45)0.2Tumor diameter\>3.5 cm versus \< 3.5 cm2.7 (1.05--7.08)0.041.7 (0.61--4.86)0.3MicrometastasesIsolated tumor cellsLymphovascular invasionPresent versus absent18.4 (4.0--85.1)\<0.0010.38 (0.05--3.16)0.4^a^Odds ratio^b^95% CI Discussion {#Sec10} ========== This study demonstrates that the incidence of isolated tumor cells in lymph nodes of patients with stage I and II colorectal cancer is lower after no-touch laparoscopic resection when compared with lateral to medial open resection. These results are in line with other studies demonstrating reduced levels of CTC in peripheral and portal blood during laparoscopic resection \[[@CR2], [@CR12]\]. This is the first study analyzing tumor cells in lymph nodes using immunohistochemical techniques, thereby overcoming the historically controversial issues regarding reliability and reproducibility of CTC measurement in blood \[[@CR13]\]. An advantage of analyzing OTC in lymph nodes is that histology can be preserved, discriminating between OTC and falsely positive stained cells. Most studies sampling peripheral or portal blood for CTC during surgery use reverse-transcriptase polymerase chain reaction (RT-PCR) or quantitative real-time RT-PCR. However, histological confirmation is lost in these techniques, which harbors the danger of identifying shed debris or hematopoietic cells as tumor cells. The clinical relevance of our finding is hard to assess. Although the prognostic role of OTC seems to be established in metastasized breast, colon, and prostate cancer patients \[[@CR14]--[@CR16]\], the clinical relevance of peroperatively detected cells is still debated. So far, these cells are generally considered to be transiently shed cells without prognostic significance. The metastatic process is extremely inefficient, involving survival in circulation or lymphatics, arresting at a distant target organ, extravasation into surrounding tissue, and survival in the foreign microenvironment, followed by proliferation and induction of angiogenesis while evading apoptotic death or immunological response \[[@CR17]\]. Tumors can shed millions of cells into the bloodstream daily, but it has been demonstrated that only a small percentage of tumor cells (0.05%) can survive and initiate a metastatic focus \[[@CR17]\]. In contrast, data on patients with breast cancer show that OTC in sentinel lymph nodes are related to reduced 5-year disease-free survival \[[@CR18]\]. Cultured single cells from rib marrow of patients with esophagogastric cancer were viable when inoculated subcutaneously in athymic nude mice \[[@CR19]\]. In addition, in patients with colon cancer, improved 5-year survival rates were reported combined with reduced frequency of cancer cells in portal blood using the open no-touch isolation technique \[[@CR3]\]. The only randomized trial on no-touch technique in patients with colon cancer did not show a significant increase in disease-free survival, but data demonstrated better results in all analyses for the no-touch patient group (time to recurrence, incidence of liver metastases, and disease-free survival) \[[@CR4]\]. So far, several randomized trials comparing laparoscopic with open surgery with long-term follow-up did not show survival differences \[[@CR20], [@CR21]\]. This would be in line with our hypothesis that the decreased number of detected ITC is merely a mechanical process depending on the surgical approach, rather than being viewed as a prognostic factor. Better disease-free survival was demonstrated in one early trial comparing laparoscopic with open resection, but this improvement was only seen in stage III patients, and therefore these results cannot be compared with our findings in histologically N0 patients \[[@CR22]\]. Previously, it has been suggested that the decreased detection rate of CTC could be attributed to early ligation of the lymphovascular trunk (medial to lateral approach) \[[@CR2]\]. We have demonstrated in a previous study that OTC are preferentially found in peritumoral SNs \[[@CR5]\]. Since lymphatic drainage to peritumoral lymph nodes is not disturbed by early ligation, it is unlikely that solely a medial to lateral approach would result in decreased numbers of ITC. Therefore, our results suggest that predominantly the no-touch technique of the laparoscopic approach explains the lower incidence of ITC when compared with conventional open resection. Even without demonstrated prognostic relevance, this would make laparoscopic no-touch resection the treatment of choice for patients with stage I and II colorectal cancer. The development of all metastases will start with a single cell, and so far there are no techniques available to predict the possible survival of ITC. Although most cells will not survive circulation and immunological responses, activation of blood coagulation and relative immune suppression due to surgical stress might enhance the metastatic potential of these intraoperatively spilled cells \[[@CR23], [@CR24]\]. A major drawback of our study is the nonrandomized comparison. However, the decrease in ITC is not likely to be attributed to patient selection since patient and tumor characteristics, carcinoembryonic antigen (CEA) levels, and number of resected lymph nodes were comparable in the two treatment groups. There was a difference in tumor diameter, with smaller tumors in the laparoscopic group. However, regression analysis showed no relation between presence of ITC and tumor diameter. The only parameter predictive for presence of ITC was method of resection, with OR of 3.3 for open lateral to medial surgery. Conclusions {#Sec11} =========== Laparoscopic no-touch surgery results in fewer isolated tumor cells in lymph nodes compared with open lateral to medial surgery in patients with stage I and II colorectal cancer. Although prospective survival analyses in these patients with ITC and randomized trials should be awaited, every attempt should be made to prevent worsening of the prognosis of patients during surgery. In that respect, laparoscopic resection with the no-touch technique may have benefits over open surgery in patients with stage I and II colorectal cancer. Authors van der Zaag, Buskens, Vlug, Peters, Bouma, and Bemelman have no conflicts of interest or financial ties to disclose. **Open Access** This article is distributed under the terms of the Creative Commons Attribution Noncommercial License which permits any noncommercial use, distribution, and reproduction in any medium, provided the original author(s) and source are credited.

**Session:** 151. Viruses and Bacteria in Immunocompromised Patients *Friday, October 5, 2018: 12:30 PM*

Estimating the genetic relatedness of modern individuals is routinely achieved by employing the use of microsatellites (synonymous with short tandem repeats (STRs)) or other genomic markers that estimate kinship coefficients based on probabilities of identity-by-descent (IBD)[@b1][@b2]. These methods, however, cannot be applied to cases where the DNA presents high levels of fragmentation and damage, as is common in ancient DNA (aDNA). Upon an organism's death, its genetic material starts to degrade and accumulate damage as repair enzymes no longer maintain the integrity of the molecular structure of DNA[@b3][@b4]. Among the many factors that contribute to the rate and severity of this phenomenon are temperature, the acidity of the surrounding depositional environment, ambient level of humidity, and the eventual invasion of environmental microbes into the organism's cells. As a result, DNA fragments extracted from preserved tissue (in most cases bone and teeth) that is recovered from either ancient or semi-ancient (e.g. many forensic cases) human remains are short in length, ranging from 30 to 70 base pairs (bp). The degradation process has a major impact on the success rates and authenticity of many PCR-based aDNA identification techniques[@b3][@b4][@b5][@b6]; however, analysis of these short and damaged DNA molecules was revolutionised with the onset of Next Generation Sequencing (NGS) just over one decade ago. Next-generation shotgun sequencing has enabled aDNA studies to progress at a much faster rate than before, and when applied in conjunction with optimised bone tissue isolation, DNA extraction, and sequencing technologies, large amounts of genetic information can be obtained even from samples with poor molecular preservation. Relatedness estimation is a topic of relevance and interest in both anthropological and forensic studies. Before NGS, PCR-based studies were affected by a limited capacity to authenticate aDNA results and an inability to retrieve the required quantity of data from most aDNA samples[@b7][@b8][@b9][@b10]. However, there are some methods that have been adapted to work specifically with this type of NGS or ancient DNA data present in software such as PLINK2[@b11] and NGSrelate[@b12]. Both software packages utilise Single Nucleotide Polymorphism (SNP) data, shown to work well with maximum likelihood approaches, and rely on genotypes, genotype likelihoods and minor allele frequencies. However, these packages require the input of relatively high amounts of genetic data (i.e. large numbers of loci), which is oftentimes challenging and expensive to retrieve from ancient skeletal material[@b1][@b2][@b12][@b13]. Our method overcomes these challenges by substantially reducing the amount of input data required without sacrificing the confidence of the relatedness estimation. Here, we apply this novel method to identify the century-old skeletal remains of a famous Irish Rebel, Thomas Kent. Thomas Kent (1865--1916), an Irish rebel native to Castlelyons, grew up in Bawnard House located just outside the town of Fermoy in County Cork, Ireland. A week after the Easter Rising insurrection in April of 1916, the Royal Irish Constabulary (RIC) raided the family home on 1^st^ of May. An RIC officer was shot dead during the raid. Brothers Thomas and William Kent were arrested. Following court martial, William was acquitted, but Thomas received a death sentence and was subsequently one of 16 men executed by British forces following the Easter Rising. He was executed in the early hours of the 9^th^ of May, 1916 at Cork Detention Barracks and then buried adjacent to where he fell[@b14]. The remains of Thomas Kent lay in these barracks, which subsequently became Cork Prison, until June 2015, when they were exhumed by a team led by the National Monuments Service of the Department of Arts, Heritage and the Gaeltacht. Poorly kept records from the era of Thomas Kent's execution and throughout the intervening 99 years resulted in confusion surrounding his final resting place and uncertainty in the identification of his remains. The presumed identity of the remains was solely based on circumstantial evidence, and though attempted, it was soon determined that traditional DNA analysis utilizing STRs was not an option because of the expected DNA degradation resulting from the remains' ancient/archaeological origin. The National Forensic Coordination Office at the Garda Technical Bureau and Forensic Science Ireland contacted University College Dublin (UCD) to lead the development of a new DNA identification method based on optimisation techniques involving the use of the osseous inner ear part of the petrous part of the temporal bone[@b15] which has been applied successfully for over \~1000 archaeological samples from temperate regions spanning between 40,000--500 years before present (average endogenous yields range of 50--70% and with an overall success rate of \~80%[@b16]). Using low-coverage shotgun sequencing data obtained from a single sequencing run on the Illumina MiSeq platform, we compared modern genetic data from two of Thomas Kent's living relatives to his century-old genetic material in order to identify his remains. Based on the success of our analytic approach in a low-coverage data scenario, we propose a NGS SNP-based method for relatedness estimation that is based upon a symmetrical Rxy estimator algorithm developed by Queller and Goodnight[@b17] and uses "forced homozygote" allele data to estimate relationship coefficients. Similar to other available software, the approach reported in this study relies on SNP data but is unique in its requirement for a substantially lower amount of input data than the methods mentioned previously without sacrificing any accuracy. This makes it widely applicable budget-efficient forensic applications, as well as to the rapidly-expanding field of ancient DNA studies, where other methods are not an option because low coverage homozygous data is the norm[@b10][@b16]. Here we detail the success of our approach in the identification of the historical remains of the Irish revolutionary Thomas Kent. Results and Discussion ====================== Authentication of Sequencing Data --------------------------------- As expected, DNA preservation differed noticeably between the modern individuals and the supposed archaeological remains of Thomas Kent (hereafter, TK). Because of that, we followed the methodologies used for ancient DNA analysis. For standardization purposes, after separate DNA extractions, which required different protocols due to the use of different biological tissues, we prepared the modern samples for sequencing in exactly the same way as TK. The average sequence read length from TK was predicted to be shorter than his modern relatives (E81 and E82) due to the historic nature of this sample; average fragment length was determined to be 54.01 bp, with a wide standard deviation of ± 11.57 bp ([Table 1](#t1){ref-type="table"}). In contrast, the modern relatives' DNA size averaged 64.48 bp, with a standard deviation of ± 1.52 bp, which is extremely close to the sequencing length used (65 bp). During the analysis of the raw sequencing data, the presence of adapters was detected in very few reads for the modern individuals as compared to the ancient sample (38% for E81 and E82, against 72% for TK), further supporting the notion that these endogenous modern DNA fragments were longer than 65 bp. This was the expected outcome for modern DNA samples, indicating that these non-damaged sequences were possibly of lengths greater than or close to 65 bp. Due to the archaeological nature of Thomas Kent's genetic material and the possibility of modern DNA contamination, raw data for this sample was first analysed to confirm the authenticity of the retrieved DNA as endogenous and ancient. To authenticate the DNA of TK as ancient, we utilised a widely-used approach developed for ancient DNA that quantifies deamination frequencies at the terminal ends of the DNA molecule, looking in particular for C \> T substitutions at 5′ overhangs that characterize the deamination of cytosines. Using the mapDamage v 2.0 software[@b18][@b19], the deamination frequencies present in TK's DNA, 0.14 C \> T at the 5′ end and 0.10 G \> A at the 3′ end ([Fig. 1](#f1){ref-type="fig"}) appear consistent with the expectation of molecular degradation for century-old bones interred in a shallow grave in the presence of quicklime[@b14]. In contrast, the modern DNA from TK's living relatives did not show significant damage patterns at the ends of the sequences. However, because fragments with shorter size than that of the sequencing length (65 bp) are not expected to be overwhelmingly present in modern DNA, these deamination frequencies are not informative as it is probable that the ends of the molecules were not read. No rescaling of base quality scores was applied. Mitochondrial haplogroups were also estimated for the three individuals ([Table 1](#t1){ref-type="table"}) using Phy-Mer[@b20]. For ethical reasons, the determined haplogroups are not reported in this paper; however, the two modern relatives, as expected, shared the same haplogroup, with a Phy-Mer confidence score of 0.61. This score estimates how well the given data matches the assigned haplogroup in the 0--1 interval. Thomas Kent's haplogroup was different from that of the relatives, with a confidence score of 0.77. All three haplogroups were consistent with expectations for historic or modern individuals native to Ireland. Relatedness Estimations ----------------------- We estimated relatedness among the putative TK remains and two surviving members of the Kent family using very low coverage shotgun data (ranging from 0.04X to 0.1X) obtained from one MiSeq sequencing run using a 50 cycles v2 kit sequenced to 65 bp, which currently generates a maximum of 25 million reads. Because we did not use a targeted enrichment or hybridization capture method to selectively identify and obtain common loci within the human genome, the output data for each individual was a random pool of overlapping reads. Along with the negative controls, these three samples were the only samples placed on the sequencing run. A total of 25% of TK's total reads aligned to the human genome, representing a genomic coverage of 0.04X. This amount of endogenous DNA is considered relatively high in an ancient DNA context and was made possible to retrieve because of improved DNA extraction methodologies[@b15][@b21]. A total of 4817884 total reads were retrieved from E81, with 3855705 aligning to the human genome (80% endogenous contents and 0.08X coverage) and a slightly higher total of 6081215 total reads were retrieved from E82, from which 4758208 were of human origin (78% endogenous contents and 0.1X coverage) ([Table 1](#t1){ref-type="table"}). None of the negative controls prepared along with the samples rendered human sequences. Using the dataset of SNPs developed for population and evolutionary genetic studies employed in ref. [@b10], we called genotypes for 354,212 positions for each individual, obtaining 17403 SNPs called for TK, 34195 for E81, and 42066 for E82. Out of these, we extracted only the shared SNPs between each dyad: TK:E81 (1328 SNPs), TK:E82 (1592 SNPs), and E81:E82 (3480 SNPs). As the total genome coverages were very low ([Table 1](#t1){ref-type="table"}), virtually all SNPs called had only one 1X read depth. Because we did not have more than one read per SNP position, we forced each SNP to be homozygous by repeating the called base to generate a diploid locus; this is referred to as the "forced homozygote" approach. For SNPs with more than 1X coverage, one call with phred quality above 30 was randomly selected and then "forced" homozygous by repeating the base as explained above. We estimated relationship coefficients for each of the three dyads using the Queller and Goodnight (1989) algorithm incorporated in the software SPAGeDi1-5a (build04-03-2015)[@b22], using the correspondent European allele frequencies downloaded from the 1000 Genomes Phase 3 website. As anticipated, the use of the forced homozygote approach resulted in relatedness coefficients (Rxy) of half the expected values. For the pair E81:E82, we observed an Rxy of 0.2794, consistent with second order relatedness, but with first order relatedness in the forced homozygote approach (i.e. equivalent to a Rxy of 0.50 if heterozygous data had been used). For TK:E81, the Rxy was estimated at 0.1336, and for TK:E82, it was estimated at 0.1236. These values are consistent with a second order of relatedness (25% if heterozygous data had been used, or 12.5% under our forced homozygote approach) between TK and the two living relatives, supporting the positive identification of his remains. The expected hypothesis that Thomas Kent was related to the two living relatives by a second order relationship and the two living relatives are related to each other by a first order relationship (Hypothesis \#4, [Table 2](#t2){ref-type="table"}) is unambiguously supported by the data, comprising nearly the entire posterior probability of the set of hypotheses. Using the posterior probabilities, the odds that this hypothesis is incorrect given the observed data is less than one in one million (8.15 E-07). Indeed, the Odds Ratio of the summed posterior probabilities for the four hypotheses proposing that the remains of Thomas Kent are related, in any manner, to both relatives versus the odds that he is unrelated to at least one of the two is in excess of 5 trillion, indicating conclusively that the TK remains are related to the two living members of the Kent family. *In silico* Simulations of Relatedness -------------------------------------- In order to assess the accuracy of the relatedness estimations using forced homozygote data, we computed relatedness coefficients using the forced homozygote approach on three relatedness classes -- unrelated individuals, first order, and second order, on two different sets of data. Each of the three possible dyads - TK:E81, TK:E82, E81:E82 shared a unique set of SNP loci. Therefore, three simulated data sets, based on 1000 Genomes Project frequencies of shared SNPs for each specific dyad, were simulated. Each simulated data set consisted of 2000 first (e.g. full siblings), second (half siblings) and unrelated individuals respectively. All simulated individuals were heterozygous but were forced to exhibit 100% homozygosity by random removal of one of the alleles at each locus and replication of the remaining allele. Pairwise relatedness coefficients were calculated in SPAGeDI, and the distribution visualised, as shown in [Fig. 2](#f2){ref-type="fig"} (details in the Methods section). The peaks of the distributions are at the expected half-values of the relationship coefficients and it is clear that the results obtained for the three relative pairs fall within the expected ranges of variation. We then applied the same approach for three pairs of samples of known relatedness for each order from the 1000 Genomes Project Phase 1 ([Table 3](#t3){ref-type="table"}), as Phase 3 does not include related individuals. We randomly downsized each sample to approximately 50.000 SNPs and then ran the simulations for the shared SNPs between each dyad. The number of common SNPs varied from 2040 to 2307, which is in between the values shared by TK and relatives, and one relative and another. The relatedness coefficients for each pair were calculated using exactly the same "forced homozygote" approach and then six hundred estimations per order or relatedness were simulated. These were plotted using the correspondent frequencies of the common SNPs, showing that the coefficients for each pair match their known order of relatedness ([Fig. 3](#f3){ref-type="fig"}). An R script with two sets of functions (TKrelated and CybRSex) was developed to automate the two processes: 1) simulations down to a true coefficient of relationship of 25%, and 2) actual data tests. By using data in PLINK format as input, our package can runs SPAGeDI for the desired pairs of individuals, and generate X number of homozygous individuals for the given set of SNPs and their allele frequencies. A function allows to plot the three coefficients of relatedness used for simulations (0%, 25%, 50%), making it possible to visualise the distribution of simulated relatedness estimates for any given relatedness class with the expected ranges of variation from the specific input SNP data. The tests on pairs of individuals are performed by a function that requires two input files and an allele frequency file. This package is freely available under the GNU General Public License v3 at <https://github.com/danimag/tkrelated> and includes a detailed walkthrough manual. Conclusions =========== A unique interdisciplinary research opportunity on this historical matter has allowed us to develop an efficient and accurate method for relatedness estimations using small amounts of genetic data. We were able to identify the skeletal remains of Thomas Kent, whose state funeral took place on the 18^th^ of September of 2015, shortly after the identification of his remains. Applicable to both forensic and ancient DNA research, our method for relatedness estimation has important added additional benefits in comparison to existing methods. When compared to the software packages PLINK2[@b11], NGSrelate[@b12] and KING[@b13], the approach we present here requires substantially lower genomic coverage. This will prove helpful when large amounts of genomic data are unavailable, such as in the case of most ancient DNA studies. We assessed the existing algorithms NGSrelate and KING with our data, both in forced homozygote and unmodified states. However, the results from all three algorithms were ambiguous, either because of our data having too few loci or absence of heterozygotes due to the very low coverage, or the inability to work with homozygote data only. Korneliussen and Moltke[@b12] showed that using NGSrelate with 1X coverage resulted in large variance and ambiguous relatedness estimates, yet it still outperformed PLINK2 with these low read depths. With our data, NGSrelate was not able to correctly estimate any of the relationships, including the known one between E81 and E82 (data not shown). Manichaikul and colleagues[@b13] successfully implemented KING on data sets with 5000 SNPs to estimate 1^st^ and 2^nd^ degree relationships. However, the algorithms in KING failed to produce results when implemented on the forced homozygosity data with relatedness estimations for our data ranging from −34 to −21 between dyads, very far from the theoretical \[0, 1\] interval (data not shown). However, the method we present was effective with very low coverage ranging from 0.04X to 0.1X. We also designed an R script to simulate virtual groups of (un)related individuals and their relatedness coefficients, based on Queller and Goodnight's Rxy, from a given set of SNPs and corresponding allele frequencies. This should prove useful in ancient DNA, where low endogenous DNA contents are often the norm and where target enrichment approaches for SNP capture are becoming more common. With the implementation of the "forced homozygote" method, estimating the relatedness between individuals in contexts such as multiple or mass burials may become a more routine task in future studies. This will benefit research in archaeology and anthropology, where the relationships of individuals found interred in multiple burials are often only hypothesized. Materials and Methods ===================== Archaeological Bone Sampling ---------------------------- To obtain genetic material from the skeletal remains of Thomas Kent, fine bone powder was retrieved from the cochlea of the left petrous part of the temporal bone that was detached from the rest of the cranium. While the petrous part of the temporal bone is accepted as yielding systematically higher endogenous DNA compared to other skeletal elements[@b21], the cochlea in particular was chosen because of research that demonstrated that the otic capsule, and particularly the cochlea, provides the highest endogenous DNA yield from any part of the petrous[@b15]. The powder was obtained using a minimally-destructive direct drilling technique developed at University College Dublin aimed at reducing any possible damage to the bone. A Dremel 9100 Fortiflex rotary tool, fitted with a small-sized spherical grinding bit (1.5 mm) previously treated with bleach and ethanol, was set to medium speed and used to obtain approximately 100 mg of bone powder. The cochlea was accessed from the superior aspect of the petrous bone, limiting visible damage to a 2--3 mm hole on the superior surface of the petrous. The bone powder generated from drilling the cochlear cavern was collected in a clean weighing boat and transferred to a 1.5 mL sterile Eppendorf tube. This procedure was conducted in a clean sample preparation facility at UCD. Blood Sampling and DNA Extraction for Modern Relatives ------------------------------------------------------ Blood samples were collected from Thomas Kent's living relatives in accordance with the prescribed methods employed by Forensic Science Ireland in the investigation of any unidentified remains. DNA extracts were then sent to University College Dublin for further processing. Informed consent was obtained by the Gardaí for the genetic analysis of this biological material. DNA Extraction for Thomas Kent ------------------------------ DNA was extracted from Thomas Kent's bone powder following the protocol from ref. [@b23] which improves upon the optimized silica-based extraction technique described in ref. [@b6]. Extraction took place in a physically separated ancient DNA lab at UCD in adherence with stringent anti-contamination protocols. Approximately 50 mg of bone powder was combined with 1 mL of an extraction buffer solution containing 0.5 M EDTA and Proteinase K (Roche Diagnostics). The bone powder was suspended by vortexing and incubated at 37 °C with rotation for 18 hours in a ThermoMixer C (Eppendorf AG) and subsequently centrifuged for 2 minutes at 17.000 g in a Heraeus Pico 17 microcentrifuge (Thermo Scientific) to separate the undissolved bone from the supernatant solution. The supernatant solution was collected and added to 13 mL of binding buffer solution containing guanidine hydrochloride (MW 95.53, 5 M), isopropanol, Tween-20 (10%), and sodium acetate (3 M) in a custom-made binding apparatus. This binding apparatus was constructed by forcibly fitting a reservoir removed from a Zymo-Spin V column (Zymo Research) into a MinElute silica spin column (Qiagen). This apparatus was then placed into a 50 mL falcon tube[@b23]. The 14 mL solution of binding buffer and DNA extract was added to the extension reservoir in the falcon tube, the cap was secured, and the falcon tube was centrifuged for 4 minutes at 2500 rpm, rotated 90°, and centrifuged for another 2 minutes at 3,000 rpm. The extension reservoir was then disassembled and the MinElute column was placed into a 2 mL collection tube. The column was dry-spun for 1 minute at 13,300 rpm, and two wash steps were subsequently performed using 650 μL of PE wash buffer. Finally, the column was placed into a clean 1.5 mL Eppendorf tube and the DNA was eluted into 25 μL of TET buffer. DNA Library Preparation ----------------------- Libraries for next-generation sequencing were built for all three DNA extracts using a modified version of ref. [@b24] as outlined in ref. [@b21], where blunt end repair was performed using NEBNext End-Repair (New England Biolabs Inc.) and Bst was inactivated by heat (20 minutes at 80 °C). Thomas Kent's DNA library was prepared in a dedicated ancient DNA lab whereas the libraries for the DNA of two modern relatives were prepared in a modern DNA lab in UCD Earth Institute's Area 52. Indexing PCRs were performed with AccuPrime Pfx Supermix (Life Technology), with primer IS4 and an indexing primer. 3 μL of the indexed library was added to 21 μL of freshly prepared PCR mix, and combined with 1 μL of unique index, enabling the pooling of samples for multiplex sequencing. This resulted in a final volume of 25 μL. PCR amplification was performed using the following temperature cycling profile: 5 minutes at 95 °C, 12 cycles of 15 sec at 95 °C, 30 sec at 60 °C, and 30 sec at 68 °C, and a final period of 5 minutes at 68 °C. PCR reactions were then purified using MinElute PCR Purification Kit (Qiagen), following the manufacturer's instructions. Assessment of the PCR reactions were performed on the Agilent 2100 Bioanalyzer following the guidelines of the manufacturer. Based on the concentrations indicated by the Bioanalyzer, samples were pooled in equimolar ratios for sequencing. Next-Generation Sequencing -------------------------- Libraries were sequenced on an Illumina MiSeq platform at the UCD Conway Institute of Biomolecular and Biomedical Research using 65 base pair (bp) single-end sequencing. Bioinformatics Analysis ----------------------- A custom ancient DNA bioinformatics pipeline written by the Pinhasi Lab was applied for processing short length raw MiSeq data. The software cutadapt v1.5[@b25] was used to trim adapter sequences. Minimum overlap was set to 1 (-O 1) and minimum length to 17 bp (-m 17). Alignment to the human reference genome (hg19, GRCh37) was processed by the Burrows-Wheeler Aligner v.0.7.5a-r405[@b26] with disabled seed (-l 1000) and filtering for reads with a minimum phred quality score of 30. Duplicated sequences were removed using samtools v0.1.19--96b5f2294a[@b27]. To assess the authenticity of Thomas Kent's DNA as ancient, damage patterns were assessed using the mapDamage v.2.0.6 tool[@b19]. Single nucleotide polymorphisms were called using the Genome Analyzer Tool Kit's (GATK) v.3.3--0-g37228af Pileup tool for the 354,212 positions present in the Harvard's "Fully public genotype dataset" described in ref. [@b10]. Relatedness Analysis -------------------- Most loci were represented by 1X reads, and this low read depth prevented identification of heterozygote loci for the vast majority of SNP loci in all three analysed individuals, although some loci had greater coverage. These results are the norm in ancient DNA studies, and so we proceeded according to the established protocols. To be able to fully leverage the set of SNP loci, we modeled relatedness on a sample of single-read loci. For loci with greater read depth, we randomly selected one representative allele to reduce the bias that might have been introduced by allowing for some heterozygote loci. By ensuring that all loci contained only one allele we forced a "homozygote" structure on the data. This will necessarily impact the Queller & Goodnight coefficient, as only half the genome is being interrogated, reducing the anticipated relatedness between dyads by a factor of one-half (i.e, reducing first order relationships from 0.5 to 0.25 and second order relationships from 0.25 to 0.125). In theory, it would be possible to retain the heterozygous structure of the few common loci that were covered more than 1x for each pair in addition to the forced homozygous loci with only 1X read coverage. However, this would result in a mixture of heterozygote and, perhaps falsely, homozygote (i.e. 1X read) loci as the basis for relatedness estimations. This would result in estimation of full siblings at 0.5 using the heterozygous data and an estimation of full siblings at 0.25 using the forced homozygous data. This would be problematic because of the heterogyzote-to-homozygote ratio that would be significantly different from the one present in the actual individuals. Although the simulations could take heterozygotes into consideration as well when providing a relatedness coefficient estimate, the combination of heterozygous and forced homozygous data would create non-intuitive relatedness classes that would vary for each test and render our estimations inaccurate. ### Thomas Kent Simulations We reduced the list of genotyped loci to only those loci shared for each dyad (i.e, Thomas Kent and Relative1, Thomas Kent and Relative2, Relative1 and 2). European allele frequencies at the shared loci for each comparison were retrieved from the 1000 Genomes Project (Phase 3, release 20100804 <http://www.1000genomes.org/>) using tabix (<http://www.htslib.org/doc/tabix.html>), and these were used as the reference frequencies for estimating degree of relatedness (symmetrical Rxy estimator, Queller and Goodnight 1989) using SPAGeDi1--5a (build04-03-2015)[@b22]. This was done using the TKrelated set of functions in the R package developed for this project/approach (detailed walkthrough at <https://github.com/danimag/tkrelated>). This function reads sample and allele frequencies data in non-binary text PLINK format, \*.*ped/map*, and \*.*frq*, respectively. It then makes that data SPAGeDI-ready, and runs the estimations. It also exports some files that can be used for the virtual simulations. For the three dyads of relatedness comparison (Thomas Kent and Relative1, Thomas Kent and Relative2, Relative1 and 2), we simulated nine data sets, each with 2000 virtual pairs of full siblings (first order), half siblings (second order) or unrelated individuals, using the observed alleles held in common for the each of the comparisons and the correspondent European allele frequencies (Phase 3, release 20100804 <http://www.1000genomes.org/>). Within the R package, the set of functions CybRsex take these allele frequencies and generate a desired number of pairs of unrelated individuals, first order relatives, and second order relatives. Each function for each order of relatedness starts by generating random unrelated individuals based on the frequencies of the SNPs from the input file. For first order, it pairs these unrelated individuals and produces one offspring from their homozygous genotypes. The same approach is followed for second order simulations, pairing the common parent with a new unrelated individual. For each simulated data set, we forced the same homozygote condition, resulting in a comparable set of loci represented by one allele. We assessed the degree of relatedness for the simulated data sets with SPAGeDi1--5a. The output relatedness coefficient for each simulated data set was tabulated to create an empirical distribution for three degrees of relatedness (first order, second order, unrelated) for the particular set of loci observed to be held in common for the three subjects. The distribution of relatedness coefficients was nearly normal ([Fig. 2](#f2){ref-type="fig"}). Using mean and variance parameters fit to the empirical distributions, we calculated maximum likelihood (ML) fits of the observed degree of relatedness for each dyad to the three relatedness distributions[@b28][@b29]. Potential relatedness hypotheses constitute the set of potential combinations of full sibling/parental and half sibling/uncle between the three subjects, producing a set of eleven potential hypotheses ([Table 2](#t2){ref-type="table"}). The ML fit of each hypothesis is then the sum of the ML fits of the observed relatedness coefficient between the two individuals for the appropriate empirical distribution. ### 1000 Genomes Simulations To test the robustness of our approach, we applied it to three pairs of individuals with known relatedness from the 1000 Genomes Project. Since related individuals are excluded in Phase 3, we downloaded the variant calls from Phase 1 for all chromosomes (release 20101123 from <http://www.internationalgenome.org/data>, accessed on 27/09/2016). Using PLINK v.1.90b3.41[@b11] we converted and merged the data. We selected the individuals shown in [Table 3](#t3){ref-type="table"}. They were isolated from the dataset and randomly sub-sampled to approximately 50.000 SNPs. We then ran our script for estimating relatedness and simulating individuals. As our approach has been designed to be used with samples from very low-coverage scenarios such as in ancient DNA studies, we ran these tests with approximately 2000 common SNPs and 600 simulations. We retrieved allele frequencies from the populations from where each pair of individuals was originated, e.g. for the second order test on a pair of individuals originating from Great Britain we used the allele frequencies of that same population. The results of these simulations were consistent with known relatedness and therefore confirm the robustness of our approach when dealing with low-coverage data ([Fig. 3](#f3){ref-type="fig"}). Ethics Statement ================ The investigation into the authentication of Thomas Kent's remains was tasked to the Irish Police, An Garda Siochana, on behalf of the State, and therefore obliged to adhere to specific ethical and legal considerations. Informed consent was obtained when collecting the blood from Thomas Kent's living relatives in regard to analysis of the genetic data and dissemination of the results. Kent's remains, legally considered archaeological, were handled with the permission of the correspondent legal authorities The request for assistance from UCD by An Garda Siochana to identify the remains recovered from Cork Prison in early 2015 was made to progress that element of the overall investigation. An Garda Siochana are tasked with such investigations, on behalf of the State, and do not require an ethics committee to initiate enquiries. An Garda Siochana may enlist the expertise of any agency or academic entity to pursue lines of enquiry, and such was the case with UCD. In this case, the original request for help in identifying the remains came from the Department of An Taoiseach (Head of Government) to the National Forensic Coordination Office, who then managed the overall investigation. The integrity of all evidence, samples and results was managed by the Head of the National Forensic Coordination Office, who was also the investigating officer in this case. All ethical considerations and legal obligations under the Data Protection Acts were his responsibility as Investigating Officer, and he was the person who sought the assistance of UCD on behalf of An Garda Siochana. He reviewed all evidence or results before they were communicated to the relevant parties. In such investigations ethical considerations form part of the overall review, in addition to many layers of legal consideration and all requirements were met. Data Availability ================= The genetic data from the study has been stored in the repository of the National Forensic Coordination Office of An Garda Siochana, and although it is of public access, legal considerations require that it complies with the Data Protection Act, making it restricted. The data will be available from the Police repository by request, under the reference number NFCO-01-244103/15. The following email can be used <forensic.coordination@garda.ie>. Additional Information ====================== **How to cite this article**: Fernandes, D. *et al*. The Identification of a 1916 Irish Rebel: New Approach for Estimating Relatedness From Low Coverage Homozygous Genomes. *Sci. Rep.* **7**, 41529; doi: 10.1038/srep41529 (2017). **Publisher\'s note:** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. We would like to thank Dr. Sudipto Das for his comments on sequencing methods; Dr. Eppie Jones for comments on the manuscript; the Irish Government for their support throughout the Thomas Kent identification process; and Dr. Olivia Cheronet for helping with the R script. This research was supported by R.P.'s European Research Council Starting grant ERC- 2010-StG 263441 (<https://erc.europa.eu>). D.F. was supported by an Irish Research Council Post-Graduate grant GOIPG/2013/36 ([www.research.ie](http://www.research.ie)); K.S. was supported by NSF grant BCS-1613577. M.N. was supported by an Irish Research Council Post-Doctoral Fellowship (GOIPD/2013/1). The authors declare no competing financial interests. **Author Contributions** D.F., J.C., K.S., M.N. and R.P. designed the experiments. D.F., E.C., J.E.L.C., K.S. and M.N. carried the experimental work. D.F., E.F., J.C., J.F. and K.S. analysed the data. D.F., J.B., J.C., J.F. and K.S. wrote the manuscript. Tables and Figures were created by D.F. {#f1} {#f2} {#f3} ###### Sequencing data analysis and relatedness coefficient results. Individual TK E81 E82 ----------------------------------- ------------------------------------------ -------------------------------------------- -------------------------------------------- Total reads 9168617 4817884 6081215 Trimmed reads 6603060 1805684 2335337 Aligned reads 2359538 3855705 4758208 Endogenous (%) 25.73 80.03 78.24 GC content (%) 37 40 40 Average bp size (stdv) 54.01 (+/−11.57) 64.48 (+/−1.52) 64.48 (+/−1.45) MapDamage 5′ \| 3\' 0.14 \| 0.10 N/A \| N/A N/A \| N/A Molecular sex Male Female Female Coverage 0.04X 0.08X 0.1X SNPs[\*](#t1-fn1){ref-type="fn"}¹ 17403 34195 42066 SNPs with coverage \> 1 476 1796 2686 Common SNPs E81-1328/E82-1592 TK-1328/E82-3480 TK-1592/E81−3480 mt Haplogroup (score) α1[\*](#t1-fn1){ref-type="fn"}^2^ (0.77) β5γ2[\*](#t1-fn1){ref-type="fn"}^2^ (0.61) β5γ2[\*](#t1-fn1){ref-type="fn"}^2^ (0.61) Relatedness coefficients E81-0.1336/E82-0.1236 TK-0.1336/E82-0.2794 TK-0.11236/E81-0.2794 ^\*^¹From[@b10]. ^\*2^Real haplogroup information is not shown due to ethical constraints. ###### Set of potential relatedness hypotheses for the combinations of full sibling/parental and half sibling/uncle between the three subjects. Hypothesis \# TK - E81 TK - E82 E81 - E82 Model LnL Posterior Probability --------------- ----------- ----------- -------------- ----------- ----------------------- 1 Unrelated Unrelated Unrelated −99.3158 4.6096 E-48 2 Unrelated Unrelated Full Sibling −5.58593 2.3444 E-07 3 Unrelated Unrelated Half Sibling −32.6766 4.0244 E-19 4 Uncle Uncle Full Sibling 9.680128 0.9999 5 Parent Parent Full Sibling −4.6796 5.8030 E-07 6 Parent Uncle Half Sibling −23.3091 4.7092 E-15 7 Uncle Parent Half Sibling −25.8717 3.6312 E-16 8 Parent Unrelated Unrelated −97.7177 2.2789 E-47 9 Uncle Unrelated Unrelated −91.8191 8.3072 E-45 10 Unrelated Parent Unrelated −100.008 2.3079 E-48 11 Unrelated Uncle Unrelated −91.5465 1.0910 E-44 ###### Information about the samples from the 1000 Genomes Project used for testing of the forced homozygote approach. Population CHS CHS GBR GBR CHS CHS ---------------------- ------------- -------------- ----------- ----------- ----------- ----------- Coriell Sample ID HG00501 HG00524 HG00119 HG00124 HG00403 HG00446 SRA Accession Number SRS008629 SRS008634 SRS008504 SRS008508 SRS008598 SRS006919 Sex Female Male Male Female Male Female Known Relatedness First order Second order Unrelated

Background {#Sec1} ========== Kenya has witnessed a significant decrease in the prevalence of malaria over the past years. Parasitaemia declined from 11% in 2010 to 8% in 2015 \[[@CR1]\]. This progress has led to a profound optimism to achieve pre-elimination targets, defined as period of re-orientation of malaria programme between the sustained control and elimination stages \[[@CR2]\]. However, further progress in reducing or eliminating malaria will likely be challenging. This is related to a number of factors, including reduced funding for control activities as authorities lower priority on malaria following reducing prevalence and the emergence of resistance to insecticides used in indoor residual spraying (IRS), which has resulted in the discontinuation of indoor residual spraying (IRS) in the majority of Kenyan counties in the highest endemic region \[[@CR3]\]. Furthermore, there is considerable heterogeneity in malaria transmission, with the prevalence of the disease still high in the expansive Western region along and around the Lake Victoria where prevalence stands at 27% as per 2015 Malaria Indicator Survey (MIS). Such persistent high prevalence hotspots hamper a national transition to a pre-elimination phase and will likely delay progress towards malaria elimination. In order to implement targeted strategies to control malaria in this region, there is need to assess malaria risks at local levels to identify vulnerable populations and implement prevention strategies that are context-specific and tailored to address observed disparities in the prevalence of the disease. This requires understanding of the modifiable risk factors that increase the risk of disease in this region so that steps can be taken to mitigate them. Previous studies in settings similar to the Kenya has shown that factors such as age of child, socio-economic status of households, level of education of household head, housing conditions, land use, residence (rural *vs* urban), and sources of water for domestic use are risk factors associated with malaria parasitaemia \[[@CR4]--[@CR7]\]. At present, it remains unknown how these factors contribute to increased or decreased risks of malaria to individuals and households in different malaria endemicity regions in Kenya, especially in the declining phase of the disease. Such information, if available, would provide guidance to programme managers and policymakers in understanding the predictors of parasitaemia and in designing targeted control efforts to reduce malaria prevalence and accelerate progress towards pre-elimination, especially in the remaining endemic areas. Using data collected during the MIS 2015, analysis of malaria parasitaemia prevalence by county (administrative areas consisting of multiple districts) within the Lake region of Kenya was performed. In addition, the associations between malaria parasitaemia in children and selected risk factors in this high endemic region was explored. Methods {#Sec2} ======= Study area {#Sec3} ---------- The Lake endemic region is administratively made up of eight counties of Kisumu, Siaya, Migori, Homabay, Kakamega, Busia, Bungoma, and Vihiga (Fig. [1](#Fig1){ref-type="fig"}). With the exception of Kakamega, Vihiga and Bungoma, the counties in this region share a border with one of the largest freshwater lakes in Africa, Lake Victoria. Generally, the area has a tropical climate with average temperatures ranging from 19 to 27 °C, while altitude ranges from 1200 to 1700 m above sea level. With two rainfall patterns annually, mosquito vector population in this region are usually high and malaria transmission is intense due to suitable climate conditions. As a result, the region experiences stable malaria transmission throughout the year.Fig. 1Study areas showing the counties with the Lake endemic region Study design {#Sec4} ------------ Cross-sectional data collected during the 2015 Malaria Indicator Survey (MIS)---a household survey was analysed. Eligibility of households was determined through a representative probability sample designed to get independent estimates of the prevalence of malaria parasitaemia nationally, by endemicity zone and by urban/rural residence. Sampling involved a two-stage stratified cluster sampling. In the first stage, 246 clusters including 131 rural and 115 urban clusters were selected with equal probability from the country's national sample survey and sampling programme \[[@CR3]\]. The second stage involved systematically selecting 30 households from each cluster. In each household, all eligible women of childbearing age were invited for interview and only children of the women who were interviewed were included in the survey. Data were collected on characteristics of the household, including number of people in the house by age and gender, sanitation status, asset ownership, bed net ownership, and housing conditions. Thereafter, peripheral blood was collected from all children in the household aged 6 months to 14 years to test for malaria parasitaemia. Definitions {#Sec5} ----------- As a dependent variable in the logistic regression analysis, Parasitaemia was defined as the detection of any malaria parasite in a peripheral blood smear examined by microscopy. For most of the independent variables tested in our logistic regression, the definitions of the MIS 2015 without any alteration was used \[[@CR3]\]. However, housing characteristics was defined as shown in Table [1](#Tab1){ref-type="table"}.Table 1The definition of key variables on housing condition \[[@CR8]\]VariableDefinitionRoof materialTraditional = (thatch/grass/makuti (interwoven palm tree leaves), and dung/mud/sod)Modern = (iron sheets, asbestos, concrete, tiles)Wall materialTraditional = (cane/palms/trunks, dung/mud/sod, bamboo with mud, re-used wood)Modern = (iron sheets, cement, stone with cement, bricks, blocks, covered adobe, and wood planks)Floor materialTraditional = (Earth/sand, dung, wood planks, palm/bamboo).Modern = (Polished wood, Vinyl PVC, asphalt, ceramics, cement and carpet) Laboratory methods {#Sec6} ------------------ MIS used rapid diagnostic tests (RDT) and microscopy to test for malaria in the selected children. For RDT, SD BIOLINE Malaria Ag P.f/Pan (Catalogue Number: 05FK60) was used. Briefly, blood was obtained by a finger prick, collected with a sterile loop applicator and placed in the RDT device. A buffer was added to the cassette so that the blood sample flows through the reaction area of the kit by capillary action. After 15 min, the results were read and recorded as either positive, negative or indeterminate. Indeterminate results were repeated until the test met the validity criteria. Peripheral blood smears were collected and sent to a malaria diagnostic centre for staining and examination by an expert microscopist. At the malaria diagnostic centre, the thin smear was fixed with methanol and both the thick and thin blood smears were then stained in 20% Giemsa stained for 15 min. The smears were dried and examined microscopically with a high power objective. For quality assurance purposes, every slide was read by two trained microscopists at different times. A third reader was called to settle any discrepancy between the two readers. As microscopy is considered the gold standard technique for the diagnosis of malaria, all analysis presented in this report are based on malaria microscopy results alone and not on RDT. Ethical considerations {#Sec7} ---------------------- This study involved the secondary analysis of data collected in 2015 and archived for further explorative analysis. No ethical approvals were necessary. Kenyatta National Hospital and University of Nairobi institutional review boards approved the primary protocol for MIS Kenya. Data analysis {#Sec8} ------------- Analysis was done with Stata version 12.0 software (STATA Corp, USA). Prevalence rates were assessed by counties and for children of different age groups. Univariate and multivariate logistic regressions were done to test for the association between malaria parasitaemia and selected risk factors, taking into account survey design effects. Microscopy confirmed malaria parasitaemia was used as dependent variable. Selected independent variables included the child's age and gender, gender of the household head, socio-economic status of households defined by wealth index, house characteristics (modern *vs* traditional), sources of water for domestic use (pipe, wells, open sources), occupation of household head, and residence (urban *vs* rural). To build a predictive model, all potential risk factors showing at least borderline significant statistical association with malaria (p \< 0.1) were included in a multivariate logistic regression model. A stepwise backwards-elimination was used to identify final covariates in the model. Wald test was used to check for the effect of removing factors from the model. Factors maintained in the final model are presented as odds ratio plot with 95% confidence interval. Results {#Sec9} ======= Nationally, MIS selected 7380 households but included 6667 (90.3%) in the final survey. A total of 10,721 equal to 93% of eligible children in the households participated \[[@CR3]\]. In the Lake endemic region, 1290 households equivalent 20% of households that were sampled nationally and 2442 children equivalent to 22% of all the children selected nationally participated in the survey; the mean age was 89 months (7.4 years) and 50.2% were female; 89.3% of households reported ownership of at least one insecticide-treated net. Table [2](#Tab2){ref-type="table"} shows all the general characteristics of the children assessed in this report.Table 2General characteristics of children from Lake endemic region enrolled in MIS 2015CharacteristicsNumber%Total number of children244222.0Age distribution Below 5 years77131.5 5--10 years87736.0 10--14 years79432.5Gender Male122849.8 Female121450.2Malaria Number tested by microscopy225393 Children with malaria60427.3Place of residence Rural166178.6 Urban78121.4Counties of origin Siaya23211.6 Kisumu28613.6 Migori26112.2 Homabay32113.4 Kakamega33014.6 Vihiga23311.0 Bungoma27210.9 Busia29212.1 In general, of the 2253 children that were tested for malaria by microscopy, 604 (27.1%) tested positive compared to 8.0% positivity at national level. However, prevalence in the different counties was variable, ranging from highest 37% in Busia county to lowest 18% in Bungoma county (Fig. [2](#Fig2){ref-type="fig"}a). The counties of Busia and Migori, which border Lake Victoria, and Kakamega which is classified under the Highlands, had highest prevalence as shown in Fig. [2](#Fig2){ref-type="fig"}b.Fig. 2**a** Prevalence of parasitaemia in counties within the Lake endemic region (counties in random sequence). **b** Prevalence of parasitaemia in counties within the Lake endemic region Regarding prevalence in children of different age groups, parasitaemia was lowest in the youngest age group, which had a mean parasitaemia of 14%. Gradually parasitaemia increased with age and reached a peak level of 35% at the age group of 13--14 years. The prevalence of parasitaemia in children from the Lake endemic region was relatively high compared to that of the national level at any given age group as shown in Fig. [3](#Fig3){ref-type="fig"}.Fig. 3The prevalence of parasitaemia in different age categories in children from the Lake endemic region of Kenya compared to the national trend of malaria parasitaemia. Time points indicate 2-year age brackets In univariate analysis, the main risk factors for parasitaemia were age of a child, socio-economic status, main materials for roof, floor and wall, practicing agriculture, residence (urban vs rural), and county of residence. In the multivariable analysis, only age, socio-economic status, agricultural farming and county of residence remained significantly associated with parasitaemia (Table [3](#Tab3){ref-type="table"}).Table 3Univariate and multivariate analysis of the risk factors for malaria in children in the Lake endemic region in KenyaCovariablesMultivariate analysisUnadjustedAdjustedOR (95% CI)*P*-valueOR (95% CI)*P*-valueAge groups 0--5 years1 (Reference)1 5--10 years2.03 (1.56--2.65)*\<0.001*2.20 (1.69--2.87)*\< 0.001* 10--14 years2.63 (1.93--3.58)*\<0.001*3.04 (2.26--4.10)*\< 0.001*Child's gender Male1 Female0.95 (0.80--1.12)0.330Gender---household head Male11 Female0.73 (0.53--1.01)0.0580.70 (0.51--0.96)*0.030*Education---household head No11 Yes0.78 (0.58--1.06)0.1150.88 (0.69--1.13)0.320Wealth Index---SES Poorest3.54 (1.93--6.50)*\<0.001*2.06 (1.30--3.27)*0.003* Poorer2.52 (1.49--4.26)*0.001*1.50 (0.93--2.41)0.091 Middle2.24 (1.45--3.46)*0.001*1.50 (0.99--2.29)0.054 Rich11Bed--net ownership Owns a bed net0.67 (0.41--1.08)0.1010.74 (0.44--1.24)0.251 Slept under LLIN0.63 (0.47--0.85)*0.004*0.69 (0.55--0.87)*0.002*Floor material Modern11 Traditional3.12 (2.58--3.78)*\<0.001*1.62 (1.08--2.45)*0.039*Main wall material Modern11 Traditional2.66 (2.19--3.23)*\<0.001*1.64 (0.88--2.96)0.115Roofing material Modern11 Traditional2.52 (1.90--3.34)*\<0.001*1.54 (1.02--2.34)*0.042*Source of water Piped1 Wells/boreholes1.32 (0.70--2.47)0.378 River/canal1.33 (0.77--2.31)0.302 Does not practice agriculture11 Practices agriculture2.34 (1.64--3.34)*\< 0.0011.76 (1.23--2.52)0.003* Does not own a livestock11 Owns livestock1.33 (0.98--1.81)0.0641.01 (0.74--1.39)0.941Residence Urban11 Rural2.88 (1.40--5.92)*0.005*1.68 (0.96--2.95)0.059Counties Bungoma11 Siaya2.45 (1.25--4.78)*0.010*3.40 (1.70--6.83)*0.001* Kisumu1.64 (0.54--5.00)0.3782.58 (0.87--7.61)0.085 Migori2.54 (0.87--7.42)0.0864.64 (2.09--10.30)*\< 0.001* Homabay1.63 (0.84--3.16)0.1442.18 (1.13--4.24)*0.022* Kakamega2.61 (1.18--5.79)*0.019*2.65 (1.29--5.43)*0.009* Vihiga1.41 (0.45--3.25)0.6981.40 (0.48--4.07)0.532 Busia4.09 (1.91--8.75)*0.001*4.12 (2.07--8.19)*\< 0.001*SES: Socio-economic status, *P* values less than 0.05 are highlighted in italics Compared to children living in Bungoma County, which had the lowest observed prevalence, the adjusted odds of having malaria parasitaemia was significantly higher for children living in Migori (aOR = 4.64), Kakamega (aOR = 2.65) and Busia (aOR = 4.12). To identify the main risk factors that drive malaria prevalence in the Lake endemic region, a predictive logistics regression model was built. Factors predicting increased risk for parasitemia included age group, wealth index, the county of residence, materials used in making roofs and floors of houses, and farming (see Fig. [4](#Fig4){ref-type="fig"}). Sleeping under a long-lasting insecticide-treated net (LLIN) predicted a decreased likelihood. There was no evidence of interaction between the factors in the model.Fig. 4Plot of adjusted odds ratios and 95% confidence interval for risk factors in the final model in the Lake endemic malaria region of Kenya Discussion {#Sec10} ========== Using data from MIS 2015, prevalence and risk factors associated with parasitaemia in children from the Lake endemic region of Kenya was assessed. MIS is a highly standardized survey and is considered a gold standard surveillance tool to measure both the burden of malaria and the uptake of control strategies \[[@CR3]\]. Analysis of MIS data has brought forth a number of important insights into the epidemiology of malaria in the Lake endemic region of Kenya. Firstly, despite the average malaria prevalence in this region standing at 27%, significant variations were evident in the levels of parasitaemia in children from different counties in the same region. Expectedly, the counties of Migori and Busia that border Lake Victoria had the highest prevalence relative to other counties in this region. This may be because environmental conditions around the Lake create conducive environments to sustain the mosquito vector and increase parasite transmission. A previous study from Mbita, a district on the shores of Lake Victoria has shown that important malaria vectors, such as *Anopheles funestus* and *Anopheles gambiae* breed on vegetation and in lagoons along the shores of the Lake \[[@CR9]\]. The availability of water and breeding conditions throughout the year from this freshwater lake may provide a perennial source of transmission of malaria for these counties. Surprisingly, Kakamega County, which is geographically considered a highland area but classified under the Lake endemic regions for operational convenience by National Malaria Control Programme (NMCP), had a malaria parasitaemia prevalence of 33%, the second highest in the region. This observation was unexpected as highlands areas are generally considered to have unsuitable environmental conditions for mosquito breeding and survival. Despite this, malaria in the highlands is increasingly becoming a health problem not only in Kenya but also in other sub-Saharan Africa countries such as Ethiopia, Rwanda and Burundi \[[@CR10]\]. Such newly emerging risks need particular attention when moving towards pre-elimination phases, as they can jeopardize control efforts in the traditional key areas. It is therefore important to understand the factors fuelling transmission in this particular county so that preventive measures can be taken to interrupt ongoing transmission, avoid an outbreak and control the disease among the communities living within the region. The differences in the levels of parasitaemia between the counties in the same endemic region require practical and policy considerations. In counties such as Migori, Kakamega and Busia, more than 30% of children tested were parasitaemic indicating a high level of exposure. Identifying these high-risk counties is important step towards implementation of targeted interventions. Considering the halted application of IRS in this region due to reported resistance to pyrethroids, the provision of universal access to bed nets alone, even though beneficial, may not be sufficient \[[@CR3]\]. Strengthened environmental management and other integrated vector control strategies deployed simultaneously will likely have an impact in the control of vector population and reduce malaria transmission in these counties. In addition, it may be time to consider piloting the application of seasonal malaria chemoprophylaxis (SMC) for adults and older children with appropriate anti-malarials as a tool to fast track control towards pre-elimination, as feasibility and effectiveness of such programmes has been reported in Senegal \[[@CR11]\]. The shift in the burden of the disease to older age groups reported here agrees with similar findings in Tanzania, Uganda and The Gambia \[[@CR6], [@CR8], [@CR12]\]. It has been argued that as malaria declines, the dynamics of the disease changes, with older children and adults suffering most from the disease. Mathews et al. hypothesized that the numerous public health messages and specific interventions, such as LLINs, explicitly directed towards the under 5 years and pregnant women, which starts in the earliest term of the pregnancy and is continuous throughout the duration of pregnancy, may have contributed towards the protection of children under age five \[[@CR6]\]. Other studies have proposed behavioural changes in the mosquito vector where the female *Anopheles* mosquito bites occur during early and late hours of the day as the reason for higher levels of parasitaemia in the older children who are likely to be playing outside \[[@CR13]\]. The shift in the age burden to the older age groups requires strategies that address malaria in older children, as older children and adults may become the primary reservoirs of the parasite \[[@CR14]\]. Given that older children are more likely schoolgoers, strategies are needed that target this group of children, including providing LLINs and ensuring consistent use through health educational programmes in schools. These efforts should happen without neglecting or reducing the current attention towards the under-fives. Importantly, the high prevalence of parasitaemia in children in areas with reported 80% of household ownership of a mosquito net and 75% consistent use according to MIS report 2015 \[[@CR3]\] raises a fundamental question as to what extent malaria prevalence can be further reduced by LLIN use alone. As changes occur in mosquito-feeding behaviour, the shift in transmission time will require programmes that curtail transmission outdoors as a complement to those that prevent malaria indoors so that maximum reduction in transmission can be achieved. A potential strategy that could act as an adjunct to universal ITN use is the application of intermittent preventive therapy for older children (IPTc) with an effective long-acting anti-malarial drug to purposely curtail the effects of chronic parasitaemia in this important group. While the effectiveness of IPTc using combination of different anti-malarial drugs in younger children in this region has been reported by Odhiambo et al. the challenge may lie in timing its application as their effectiveness has been shown to wane \[[@CR15]\]. Wealth index and the nature of materials used for the roof, the wall and the floor of houses, which are both considered proxy measures of socio-economic status of households, have been shown to have an association with being parasitaemic. The richer households enjoy a purchasing power which allows them to not only dwell in better conditioned houses but also enables them to afford a LLIN so that exposure of their household members to mosquitoes is lowered and risk of getting malaria reduced. These findings collaborate with the widely held view that malaria is a disease of poverty at the household levels. Similar findings relating to the associations between household socio-economic status and malaria have been reported in other studies in sub-Saharan Africa \[[@CR7], [@CR8], [@CR16], [@CR17]\]. Contrary to this result, Pullan et al. and Snow et al. who did similar studies in Uganda and Kenya, respectively, did not find any association between socio-economic status and malaria \[[@CR18], [@CR19]\]. However, the difference between this report and those with similar findings compared to these two studies with different results are that the latter was conducted in single communities living in rural settings. As a result, asset index data collected in such a small and rural setting may not be comparable to MIS that encompasses both urban and rural populations as well as a mix of different communities. Children whose households reported that they slept under LLIN had 30% lower likelihood of having parasitaemia after adjustment for age, socio-economic status and other factors. Like other studies of similar findings, this result adds to the current evidence of the efficacy of LLINs which are the mainstay of malaria prevention. The modest protective effect of LLINs reported here against parasitaemia is similar to many other findings in different studies \[[@CR5], [@CR12], [@CR18], [@CR20]\]. Malaria parasitaemia was also associated with households that practised agricultural farming. People living in the Lake endemic region of Kenya generally practice large-scale farming of cane and rice and engage in small-scale farming of corn and other vegetations, which all create the right breeding environment for the mosquito vector. This increased occupational risk for farmers is in agreement with many other findings reported previously \[[@CR8], [@CR21], [@CR22]\]. To mitigate risk related to farming, which remains one of the major economic activities for people of the Lake endemic region, it is essential that Kenya's NMCP programme integrates environmental management programmes, including community managed larval control programmes with routine agricultural practices. Lessons on this can be drawn from the implementation of community-based larviciding activities implemented in Dar es Salaam, Tanzania, which has reported the successful outcome in the reduction of malaria cases in that region \[[@CR23]\]. Even though the design of MIS is generally suitable for the analysis carried out in this study, potential limitations within the data need to be noted. Firstly, like all studies based on a cross-sectional survey, the associations observed may be unsuitable to draw causal relationship as they may be confounded by unmeasured factors \[[@CR20]\]. Secondly, socio-economic status described in this report uses wealth index that was designed using principal component analysis (PCA) of assets in the households. It has been stated that expenditure data may be a better measure of how rich or poor households are in a survey, compared to wealth index measurement using PCA \[[@CR19]\]. Lastly, the result of the malaria microscopy test was used to measure parasitaemia, i.e., dependent variable, even though it is known that microscopy is a less sensitive method compared to molecular techniques \[[@CR24]\]. For MIS Kenya, examination of slides by two research microscopist with the quality assurance processes that included the confirmation of slide status by an expert microscopist improved the sensitivity of the technique. Conclusion {#Sec11} ========== Despite the positive progress made by NMCP in the control of malaria in Kenya in the last decades, there are still some regions of the country where prevalence of parasitaemia is very high and heterogenous. It is therefore imperative to ensure that the current malaria control tools are deployed to get a maximum impact by targeting the areas and individuals who are at greater risk of the disease. Designing control strategies aimed at protecting school-age children and adults and implementing targeted interventions in counties with higher burden may be a useful strategy to make an impact on the levels of transmission of the parasite, to try to achieve nationwide pre-elimination. IPTc : intermittent preventive therapy---children IRS : indoor residual spray ITN : Insecticide Treated Nets IVM : integrated vector management LLIN : Long-Lasting Insecticide Treated Nets MIS : Malaria Indicator Survey NMCPP : National Malaria Control and Prevention Programme OR : odds ratio PCA : principal component analysis RDT : rapid diagnostic tests SES : socio-economic status SSA : sub-Saharan Africa WHO : World Health Organization **Publisher\'s Note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. The authors express gratitude to Demographic and Health Survey, USAID for granting permission to use these data for analysis. We thank EPCO Hasker of ITM, Antwerp for important inputs in the statistical analysis. IMB acquired the data, did the analysis and drafted the manuscript. MVDS provided valuable edits and revised the manuscripts. NN revised the manuscripts. All authors read and approved the final manuscript. Not applicable. Data was provided by Demographic and Health Survey---USAID project. Not applicable. Not applicable. The authors declare that they have no competing interests.

Introduction ============ The pathogenesis of transient left ventricular contractile dysfunction (TLVCD) remains uncertain. Acute emotional or physical stress is suggested to cause TLVCD, with the possible involvement of sympathetic nerve activation on coronary microvessel spasms ([@B1]). TLVCD is a synonym of "ampulla (Takotsubo) cardiomyopathy", as the left ventricular (LV) shape mimics an octopus pod, showing abnormal left ventricular wall motion, typically at the apical and/or mid-ventricle walls, except for the basal segments. Recent reports have shown that a number of medical conditions are also associated with the development of TLVCD ([@B2],[@B3]). Furthermore, anorexia nervosa in eating disorders leads to TLVCD in patients with severe hypoglycemia ([@B4]-[@B7]). We herein report a malnourished man who developed TLVCD during the treatment of rhabdomyolysis. Our case raises awareness that a predisposed electrolyte imbalance and refeeding after a long period of malnutrition may be involved in the pathophysiology of TLVCD. Case Report =========== A 52-year-old Japanese man with a height of 172 cm and weight of 51.4 kg (body mass index 17.5 kg/m^2^) was admitted to our emergency room due to a disturbance of consciousness. He had abused alcohol and had frequent diarrhea for two weeks, and he had not eaten anything for at least three days before admission. He had never been prescribed any sort of medications. His blood pressure was 117/81 mmHg, pulse rate 100 beats/min and regular, respiratory rate 30 breaths/min, and body temperature 37.7°C. A physical examination revealed a Glasgow coma scale score of 14 (E4V4M6), but there were no apparent signs of heart failure or focal neurological deficits except for muscular grasping pain. As shown in ([Table 1](#t001){ref-type="table"}), blood testing revealed a glucose level of 168 mg/dL and increased aspartate transaminase and alanine transaminase levels of 247 and 134 U/L, respectively, but decreased ion concentrations of potassium (1.1 mmol/L; normal range 3.6-4.8), phosphate (2.0 mg/dL; normal range 2.7-4.6), magnesium (1.4 mg/dL; normal range 1.8-2.3), and albumin-corrected calcium (8.63 mg/dL; normal range 8.8-10.1). Blood testing also showed an elevated creatine kinase level (11,064 U/L; normal range 59-248) with 28 IU/L for the MB isoform (creatine isozyme BB, 0%; MB, 0%; MM 100%), and metabolic alkalosis (pH 7.62, carbon dioxide partial pressure (pCO~2~) 46.5 mmHg, HCO~3~^-^ 47.7 mmol/L). Chest X-ray showed a 47% cardio-thoracic ratio without pulmonary congestion ([Fig. 1A](#g001){ref-type="fig"}), and a 12-lead surface electrocardiogram demonstrated a sinus rhythm at 100 beats/min and prolongation of the corrected QT interval (654 ms) without ST elevation ([Fig. 2A](#g002){ref-type="fig"}). To prevent refeeding syndrome, electrolytes, minerals and vitamins were corrected intravenously with frequent monitoring, and he was started on 300 kcal/day of enteral alimentation ([Fig. 3](#g003){ref-type="fig"}). ###### Laboratory Data on Admission. -------------------- -------- ----------- White blood cell 12.5 ×10^3^/uL Red blood cell 3.3 ×10^6^/uL Hemoglobin 12.1 g/dL Hematocrit 31.9 \% Platelet 280 ×10^3^/uL Total protein 5.30 g/dL Albumin 2.87 g/dL Urea nitrogen 10.0 mg/dL Creatinine 0.68 mg/dL Total bilirubin 3.7 mg/dL Direct bilirubin 1.7 mg/dL Glucose 168 mg/dL Sodium 139 mmol/L Potassium 1.1 mmol/L Chloride 79 mmol/L Calcium 7.5 mg/dL Magnesium 1.4 mg/dL Phosphate 2.0 mg/dL AST 247 U/L ALT 134 U/L LD 770 U/L ALP 185 U/L ChE 165 U/L Creatine kinase 11,064 U/L CK-MB 28 IU/L Amylase 72 U/L Folic acid 1.6 ng/mL Aldosterone 6.1 ng/dL ACTH 62.8 pg/mL Cortisol 19.8 ug/dL BNP 47.3 pg/mL C-reactive protein 1.55 mg/dL -------------------- -------- ----------- AST: aspartate transaminase, ALT: alanine transaminase, LD: lactate dehydrogenase, ALP: alkaline phosphatase, ChE: cholinesterase, ACTH: adrenocorticotropic hormone, BNP: brain natriuretic peptide {#g001} {#g002} {#g003} On day 4 after admission, he complained of shortness of breath after drinking a large amount of water, and chest X-ray revealed bilateral pulmonary congestion ([Fig. 1B](#g001){ref-type="fig"}). A 12-lead electrocardiogram revealed a prolongation of the corrected QT interval (626 ms) and T-wave symmetrical inversion at the precordial leads V~2-4~ ([Fig. 2B](#g002){ref-type="fig"}). Blood testing showed re-elevation of the creatine kinase level (14,123 U/L) accompanied by increased MB isoform (60 IU/L) and troponin-T levels (0.1 ng/mL; normal range ＜0.02). In addition, the brain natriuretic peptide level had elevated from 47.3 pg/mL on the day of admission to 361 pg/mL, but the magnesium (5.1 mg/dL) and potassium (4.8 mmol/L) ions were corrected to the normal ranges. Transthoracic echocardiogram revealed severe hypokinesis of the apical and mid-ventricular segments, except for the basal segments of the LV wall (ejection fraction of 35%) ([Fig. 4A, B](#g004){ref-type="fig"}). Ten milligrams of furosemide was administered intravenously, leading to relief of his symptoms. The patient refused to undergo coronary angiography. Thallium-201 myocardial scintigraphy performed on day 7 suggested a preserved perfusion, but ^123^I-metaiodobenzylguanidine scintigraphy showed a 34% increased washout rate. Follow-up chest X-ray showed improved pulmonary congestion on day 8 ([Fig. 1C](#g001){ref-type="fig"}), and a trend toward improvement in the T-wave inversion at the precordial leads was noted on his electrocardiogram on day 12 ([Fig. 2C](#g002){ref-type="fig"}). Transthoracic echocardiogram performed on day 12 revealed a dramatic improvement in the LV systolic function (ejection fraction of 55%) ([Fig. 4C, D](#g004){ref-type="fig"}). He was transferred to the rehabilitation hospital 15 days after admission. {#g004} Discussion ========== In the present case, we posited three hypotheses regarding the cause of TLVCD: \[1\] rhabdomyolysis, \[2\] refeeding syndrome, and \[3\] coronary multivessel vasospasm. The cause of rhabdomyolysis was initially presumed to be hypokalemia and hypophosphatemia due to a shortage of food intake, chronic diarrhea, or alcohol abuse ([@B8]). The patient exhibited metabolic alkalosis, and his pCO~2~ level remained high despite tachypnea. pCO~2~ increases at 0.6-0.7 mmHg per 1-mmol/L increase in HCO~3~^-^, and the pCO~2~ level was expected to be 55-64 mmHg (normal range 42-50 mmHg) under 47.7 mmol/L of HCO~3~^-^ (normal range 23-27 mmol/L) in venous blood. We speculate that the tachypnea and pCO~2~ value of 46.5 mmHg were a result of compensation. Several reports have described the relationship between rhabdomyolysis/myopathy and TLVCD ([Table 2](#t002){ref-type="table"}) ([@B9]-[@B15]). We did not assess the echocardiogram on the admission day, but the patient did not show any physical signs, chest X-ray findings, or brain natriuretic peptide levels suggestive of heart failure, despite severe hypokalemia and marked elevation of creatine kinase. This implies that rhabdomyolysis is unlikely to have played a major role in the development of TLVCD in this case. ###### Takotsubo Cardiomyopathy Complicating Rhabodomyolysis/Myopathy. Reference Age Sex Symptom (s) Possible trigger (s) Manifestation of LV dysfunction CPK max (U/L) Creatinine (mg/dL) Phosphate (mg/dL) Potassium (mmol/L) Glucose (mg/dL) LV wall motion LVEF (%) at onset LVEF (%) at recovery ----------- ----- ----- --------------------------------------------------- --------------------------- ------------------------------------------- ------------------------- -------------------- ------------------- -------------------- ----------------- ---------------- ------------------- ---------------------- ---------- \[9\] 73 M general fatigue dysbasia rouvastatin on the day of admission 16,538 1.4 nd 3.4 226 apical ballooning 52 65 \[10\] 61 F retrosternal chest pain diaphoresis lighteaedness shortness of breath fever inflammatory myopathy on the day of admission 31,241 nd nd nd nd apical ballooning 28 63 \[11\] 58 F burn burn injury emotional and physical stress on the day of admission nd nd nd nd nd apical ballooning 5 to 10 55 to 70 \[12\] 39 M collapse heat stroke on the day of admission 4,517 2.7 nd nd nd apical ballooning 40 65 \[13\] 78 M fall anxiety on the day of admission 5,342 nd nd nd nd apical ballooning 15 45 \[14\] 55 F general fatigue weakness of extremities vomiting nd on the day of admission 7972 nd nd nd nd apical ballooning 41 nd \[15\] 67 F chest discomfort rouvastatin fenofibrate on the day of admission 19,000 3.6 nd nd nd apical ballooning 25 47 Our case 52 M shortness of breath malnutrition alchohol abuse 4th day of admission 14,123 0.68 2.0 1.1 168 apical ballooning 35 55 We searched the reports in PubMed using the following key words; \"takotsubo\", \"myopathy\", \"rhabdomyolysis\", and \"cardiomyopathy\". nd indicates \"not described\", LVEF, left ventricular ejection fraction assessed by transthoracic echocardiogram or magnetic resonance image. \"Apical ballooning\" indicates hypokinesis to akinesis (aneurymal apex) of the left ventricle, except in the basal region. The LV ejection fraction is usually maintained within normal limits in patients with anorexia nervosa ([@B16]); however, anorexic patients are susceptible to TLVCD if they have severe hypoglycemia on the day of admission ([@B4]-[@B6]). An excessive catecholamine release against hypoglycemia is postulated to result in coronary microvessel spasms. Artificial feeding (either enteral or parenteral) in malnourished/cachectic patients can lead to multisystem organ failure due to a shortage of phosphate, potassium, magnesium, calcium, and vitamins, known as "refeeding syndrome" ([@B17],[@B18]). This failure usually occurs 2-5 days after the beginning of nutritional repletion. Cardiovascular failure includes sudden death, arrhythmias, hypertension, and congestive heart failure. Consistent with these previous findings, heart failure occurred four days after starting alimentation in this case, accompanied by re-elevation of the creatinine kinase levels. This was comparable to TLVCD associated with rhabdomyolysis, which usually exhibits left ventricular dysfunction on the day of admission ([Table 2](#t002){ref-type="table"}). The LV morphology mimicked a Takotsubo-like configuration, which is rarely recognized as a form of acute heart failure in this setting ([@B19],[@B20]) ([Table 3](#t003){ref-type="table"}). Of additional note, hypoglycemia appears to be a trigger for the development of TLVCD in these cases. ###### Takotsubo Cardiomyopathy Complicating Refeeding Syndrome. Reference Age Sex Symptom (s) Possible trigger (s) Manifestation of LV dysfunction CPK max (U/L) Creatinine (mg/dL) Phosphate (mg/dL) Potassium (mmol/L) Glucose (mg/dL) LV wall motion LVEF (%) at onset LVEF (%) at recovery ----------- ----- ----- ------------------------------------ ---------------------- ----------------------------------------- ---------------------------- -------------------- ------------------- -------------------- ----------------- ---------------- ---------------------------- ---------------------- ---- \[19\] 54 F impaired consciousness palpitation malnutrition urge to ingest a coppius dinner 1 day after coppius dinner 80 nd 2.2 2.7 19 Inverted apical ballooning 25 nd \[20\] 18 F apetite loss anorexia nervosa enteral nutrition 2nd day of admission nd nd 6.2 4.7 21 apical ballooning nd nd \[20\] 58 F drowsy anemia administration of a vitamin with saline 4th day of admission nd nd 2.9 3.4 19 apical ballooning nd nd We searched the reports in PubMed using the following key words; \"takotsubo\", \"myopathy\", \"refeeding\", and \"cardiomyopathy\". nd indicates \"not described\", LVEF, left ventricular ejection fraction assessed by transthoracic echocardiogram. \"Apical ballooning\" indicates hypokinesis to akinesis (aneurymal apex) of the left ventricle, except in the basal region. Inverted apical balooning indicates dyskinesia of basal and mid-ventricular segment, with hyperkinesia of the left ventricular apex. The mechanisms underlying the development of TLVCD in the present patient remain unclear. We acknowledge that the patient did not show hypoglycemia as observed in previous reports ([@B21],[@B22]). Similarly debatable is whether or not hypophosphatemia itself causes left ventricular dysfunction ([@B8]). Our case supports the notion that exaggerated sympathetic nerve activity may be involved in accelerating the form of Takotsubo-like configuration. We speculate that the depletion of myocardial energy production by a series of electrolyte derangements as well as food intake after long-term malnutrition may have resulted in the reduced left ventricular contractile force, leading to overt heart failure following excessive water intake. We did not perform coronary angiography; therefore, we were unable to distinguish the multivessel coronary spasm from ampulla (Takotsubo) cardiomyopathy based solely on scintigraphy with ^123^I-metaiodobenzylguanidine ([@B23],[@B24]). Under the National Institute for Health and Care Excellence (NICE) guidelines, our patient was identified as being at high risk of developing refeeding syndrome given his chronic alcoholism, low nutritional intake for more than 10 days, and low levels of serum potassium, phosphate, and magnesium before feeding. Accordingly, the patient was administered vitamins and minerals, phosphate, potassium, and magnesium intravenously, followed by low-calorie enteral alimentation, and he eventually recovered. In conclusion, the present case reminds us that TLVCD can be associated with rhabdomyolysis or refeeding syndrome, and meticulous care may help prevent the development of life-threatening events in such patients. **The authors state that they have no Conflict of Interest (COI).** [^1]: Correspondence to Dr.　Toshihiro Tsuruda, <ttsuruda@med.miyazaki-u.ac.jp>

Introduction {#s1} ============ The largest collection of microbes resides in our gut, which harbours trillions of bacteria [@pone.0061608-Frank1]. Although these populations are highly stable, they are still prone to perturbations by environmental insults [@pone.0061608-Sullivan1], with important consequences for our physiology and, consequently, our health. Shaped by millennia of co-evolution, host and bacteria have developed beneficial relationships, creating an environment for mutualism, were the mutual survival of harboured microbiota and human host is interdependent [@pone.0061608-Ley1], [@pone.0061608-Chow1] **.** Alterations in the development or composition of the microbiota could affect the cross-talk between microbiota and host compromising human health. Microbiota ecosystem unbalance, named dysbiosis, has been found in different pathologies such as Inflammatory Bowel Diseases (IBD) [@pone.0061608-Willing1]--[@pone.0061608-Nell1], Celiac disease (CD) [@pone.0061608-Schippa1], [@pone.0061608-DiCagno1] and Cystic fibrosis (CF) [@pone.0061608-Cox1]. In IBD patients, for example, it has been showed a decrease in prevalence of members of the human commensal microbiota (i.e. *Clostridium* IXa and IV groups, *Bacteroides, Bifidobacteria*) and a concomitant increase in detrimental bacteria (i.e. sulphate-reducing bacteria, *Escherichia coli*) [@pone.0061608-Frank1], [@pone.0061608-Willing1], [@pone.0061608-Qin1]. In patients with susceptible genetics the gut microbiota could drive the expansion of 'pro-inflammatory' species and the restriction of the protective/regulating species. Although the factors driving the composition dynamics of resident microbial communities are not well defined, understanding the mechanisms and relationships that control the delicate balance among the microbial groups is of primary importance. Bacteria, in their natural environments, are subjected to predation from bacteriophages, protists and predatory prokaryotes. *Bdellovibrio bacteriovorus* is a predatory gram-negative bacterium belonging to the δ-proteobacteriaceae sub-family, which attacks other gram-negative bacteria and invade their periplasm, using them as substrates for growth and reproduction [@pone.0061608-Rendulic1]. This particular intracellular niche allows *B. bacteriovorus* to feed without competition and, through its high lytic capability, can rapidly reduce gram-negative populations. From its discovery in 1963 [@pone.0061608-Stolp1], *B. bacteriovorus* was under investigation for decades, but only in recent years it took the stage [@pone.0061608-Strauch1]. Its life cycle consists of two major steps: a 'free-swimming/gliding' one spent searching for prey in water or soil, and a double 'attack phase/growth' stage spent inside the periplasm of the prey bacterium [@pone.0061608-Seidler1], [@pone.0061608-Lambert1]. Once the resources of the prey have been consumed, the *B. bacteriovorus* divides into progeny which then lyses the residue cells and swim away to chase new hosts [@pone.0061608-Kessel1]. Depending on the prey concentration and environment, this life cycle takes roughly 3--4 h. Due to its selective preying capacity on gram-negative bacterial species [@pone.0061608-Rogosky1], [@pone.0061608-Jurkevitch1], *B. bacteriovorus* is probably involved in keeping the balance between the different bacterial species living together in a community, acting as an ecological balancer [@pone.0061608-Dwidar1]. For example, *Bdellovibrio* delivered orally to live birds caused an increase in some gram-positive bacteria, a reduction in gram-negative bacteria (like *Salmonella*) and a parallel reduction in mucosal inflammation, without affecting the bird health [@pone.0061608-Atterbury1]. Moreover, *B. bacteriovorus* population levels respond to the Lotka-Volterra prey-predator oscillation [@pone.0061608-Varon1], where great *B. bacteriovorus* expansions [@pone.0061608-Gray1] are balanced by a phenotypic resistance of the prey [@pone.0061608-Shemesh1]. Thus, also in the intestine, in which polimicrobial cohorts coexists [@pone.0061608-Kim1]--[@pone.0061608-Macfarlane2], it could be expected that a similar self-regulation for this predatory bacterium [@pone.0061608-Hobley1]. *B. bacteriovorus* is ubiquitous in terrestrial and aquatic environments, even if at low counts [@pone.0061608-Jurkevitch1], [@pone.0061608-Peng1]--[@pone.0061608-Richardson1], and it has also been isolated from intestines of vertebrates [@pone.0061608-Chu1]--[@pone.0061608-Westergaard1], indicating that it is probably present in the human gut [@pone.0061608-Schwudke1]. The hypothesis underlying this study was that a dysbiotic microbial community, as found in inflammatory diseases, could be linked to differential prevalence and abundance of bacterial species acting as ecologic equalizer, such as *B. bacteriovorus*. To this purpose, we evaluated by PCR the presence of the bacterial predator *B. bacteriovorus* in bioptic samples collected from IBD and Celiac patients, as well as in faecal samples taken from Cystic fibrosis patients. Sex- and age- matched controls were also enrolled to compare the prevalence and relative abundance of *B. bacteriovorus* in the healthy status. Materials and Methods {#s2} ===================== Ethics Statement {#s2a} ---------------- Intestinal biopsies were obtained from gastro-duodenoscopy (for 'Celiac group') and colonoscopy (for 'IBD group') procedures carried out at the 'Policlinico Umberto I' Hospital at Rome, Italy. Ethics approval for this study was granted by the Ethics Committees of the 'Sapienza' University and 'Policlinico Umberto I' Hospital, Italy. Written informed consent was obtained from parents of all subjects enrolled in this study who were under 18 years of age, while the written informed consent was given autonomously by subjects over (or equal to) 18 years. In any case, written informed consent was obtained upon instructions on ethics, aims, and methodologies employed in the study. Patients {#s2b} -------- ### 'Inflammatory bowel disease group' {#s2b1} Twenty-three paediatric patients referred to the Paediatric Gastroenterology and Liver Unit of the Sapienza University of Rome, Italy, for suspected IBD were studied: active CD was diagnosed in 9, and active UC was diagnosed in 6. The remaining 8 subjects with functional gastrointestinal disorders (lymphonodular hyperplasia), and normal colonoscopy and histology findings, served as controls. In [table 1](#pone-0061608-t001){ref-type="table"} were reported the patients' baseline demographics. The patients' groups did not differ significantly by age and disease duration. All children with CD had ileocolonic involvement, and all had disease activity in the moderate to severe range. All UC patients had endoscopic evidence of pancolitis, and the disease was classified severe. The diagnostic workup of UC and CD was done according to international protocols. Children with CD were assessed using the Paediatric Crohn's Disease Activity Index (PCDAI), which is a multi-item score based on recall of the preceding week's symptoms, laboratory parameters (erythrocyte sedimentation rate, haematocrit and albumin levels), and physical examination. A score of 10 implies inactive disease, one of 11 to 30 implies mild disease, and one of \>30 implies moderate to severe disease. It is noteworthy that our CD paediatric patients did not have any previous endoscopic assessment (see [table 1](#pone-0061608-t001){ref-type="table"}), nor any kind of treatment, so they had naïve mucosal inflammation. Patients with UC were evaluated using the Paediatric Ulcerative Colitis Activity Index (PUCAI), which is based only on clinical symptoms. A score of 10 indicates inactive disease, one of 11 to 34 implies mild disease, one of 35 to 64 implies moderate disease, and one greater than 65 implies severe disease. Patients within UC cohort had also a naïve mucosal inflammation. None of the children included in the study, either CD, UC and controls, was treated with antibiotics for at least 3 months before the sampling time. The study protocol was approved by the Committee on Ethical Practice of the 'Sapienza' University of Rome and 'Policlinico Umberto I' hospital, as reported in the 'Ethics statement'. All paediatric patients underwent ileocolonoscopy after parental informed written consent was provided. During ileocolonoscopy, for routine histological assessment and bacteriological study, two biopsy samples were taken from each region (ileum, descending colon, rectum), in a non-inflamed patch adjacent to an inflamed one. Specimens were collected in 2-ml screw-cap vials filled with 0.85 ml of brain heart infusion broth (Oxoid, Cambridge, United Kingdom) and 0.15 ml of glycerol (Sigma-Aldrich, St. Louis, MO) and immediately stored at −80°C. 10.1371/journal.pone.0061608.t001 ###### Demographics and baseline disease characteristics of the patients' groups. {#pone-0061608-t001-1} Crohn's Disease Ulcerative Colitis Controls Celiac Disease Controls Cystic Fibrosis Controls -------------------------------------------------------- ----------------- -------------------- ---------- --------------------------------------------------- ----------- ----------------- ----------- No. of cases 9 6 8 10 8 35 16 Age (years; mean ± SD) 14.0±2.5 12.0±4.1 12.7±1.8 8.3±3.7 11.7±3.5 18.6±2.2 16.3±1.7 M/F 4/5 3/3 4/4 5/5 4/4 18/17 8/8 Weight (Kg) (mean ± SD) 30.6±12.2 33.9±13.1 36.2±9.6 29.7±3.2 40.9±3.2 45.2±2.4 48.7±5.6 Height (cm) (mean ± SD) 133.8±19.1 138.3±22.1 140±25.3 128.9±5.5 147.9±3.7 150.5±2.9 159.2±6.2 BMI (Kg/m^2^) (mean ± SD) 16.4±2.0 17.0±2.2 20.0±2.1 16.7±0.5 18.5±0.8 19.4±0.5 19.8±0.4 Disease duration (weeks; median and range) 6 (2--12) 8 (3--12) -- 36 (32--56) -- -- -- Duodenum involvement[(a)](#nt101){ref-type="table-fn"} -- -- -- 10/10 (100%) -- -- -- Ileo-colon involvement 9/9 (100%) 0/6 (0%) -- -- -- -- -- Colon involvement 0/9 (0%) 6/6 (100%) -- -- -- -- -- Rectum involvement 0/9 (0%) 6/6 (100%) -- -- -- -- -- Previous endoscopic assessment 0/9 (0%) 0/6 (0%) 0/8 (0%) 0/10 (0%) 0/8 (0%) -- -- Current treatment [(b)](#nt102){ref-type="table-fn"} 0/9 (0%) 0/6 (0%) 0/8 (0%) 0/10 (0%) 0/8 (0%) 35/35 (100%) -- PCDAI (mean ± SD)[(c)](#nt103){ref-type="table-fn"} 42.5±11 -- -- -- -- -- -- PUCAI (mean ± SD) [(c)](#nt103){ref-type="table-fn"} -- 71.4±7.5 -- -- -- -- -- Marsh index -- -- -- IIIA 3/10 (30%); IIIB 3/10 (30%); IIIC 4/10 (40%) -- -- -- FEV1%[(c)](#nt103){ref-type="table-fn"} -- -- -- -- -- 91.8±2.6 -- Biopsies were taken in the second part of duodenum. IBD: Sulfasalazine, 5-aminosalicylic acid drug, Steroid; CF: ciproxin 500 mg (x2)/die (not provided for 2 months prior sampling). PCDAI, Paediatric Crohn's Disease Activity Index; PUCAI: Paediatric Ulcerative Colitis Activity Index; FEV1%, Forced Expiratory Volume (in percentage). ### 'Celiac disease group' {#s2b2} Two groups of children referred to the Paediatric Gastroenterology and Liver Unit of the "Sapienza" University of Rome were included in this study: 10 Celiac patients in active state and 8 controls undergoing upper gastrointestinal endoscopy for functional dyspepsia. Controls tested negative for antitransglutaminase and antiendomysial antibodies with normal IgA levels, while histology of duodenum did not reveal features of Celiac disease. Diagnosis of Celiac disease had been performed according to ESPGHAN criteria [@pone.0061608-Stenhammar1], and [table 1](#pone-0061608-t001){ref-type="table"} summarizes clinical features of the studied population. It is noteworthy that our CD paediatric patients did not have any previous endoscopic assessment (see [table 1](#pone-0061608-t001){ref-type="table"}), nor any kind of treatment, so they had naïve mucosal inflammation. Size-appropriated and well-oriented endoscopic biopsy specimens were obtained from the second part of the duodenum, in a non-inflamed patch adjacent to an inflamed one. The histopathological diagnosis (Marsh index) was based on typical mucosal lesions with crypt cell hyperplasia, villous atrophy, and increased number of intra-epithelial lymphocytes (IELs) [@pone.0061608-Marsh1]. All untreated Celiac patients were positive for antiendomysial and antitransglutaminase antibodies at the time of diagnosis. None of the children included in the study was treated with antibiotics for at least 3 months before the sampling time. The study protocol was approved by the Committee on Ethical Practice of the 'Sapienza' University of Rome and 'Policlinico Umberto I' hospital, as reported in the 'Ethics statement'. Children were enrolled in the study after written informed consent from their parents. Specimens were collected in 2-ml screw-cap vials filled with 0.85 ml of brain heart infusion broth (Oxoid, Cambridge, United Kingdom) and 0.15 ml of glycerol (Sigma-Aldrich, St. Louis, MO) and immediately stored at −80°C. ### 'Cystic fibrosis group' {#s2b3} Thirty-five CF patients, referred to the Cystic Fibrosis Centre of the Department of Paediatrics at Hospital 'Policlinico Umberto I' of Rome, were enrolled after written informed consent. Sixteen age- and sex-matched controls were also enrolled. The diagnostic work up of CF was according to international protocols. The study protocol was approved by the Committee on Ethical Practice of the 'Sapienza' University of Rome and 'Policlinico Umberto I' hospital, as reported in 'Ethics statement'. All patients did not received antibiotics during 2 months prior to the beginning of the study, and were fasting prior faecal sampling. Faecal samples were collected in a 50 mL tube, and immediately frozen at −80°C. Biopsy DNA Extraction {#s2c} --------------------- Biopsy specimens were taken from fasting patients (Crohn's disease, Ulcerative colitis, Celiac disease) and proper age−/sex-matched controls by the same gastro-endoscopist, and immediately stored in 2 mL tubes at −80°C, as described above. Within 2 hours from sampling, biopsies were first quickly washed three times in 500 µL of physiologic saline with 0.016% dithiothreitol (DTT) to remove luminal bacteria and the mucus layer, and then utilized for DNA extraction procedure by DNeasy tissue kit (Qiagen, Hilden, Germany) according to the manufacturer's instructions. In order to obtain maximum yield of Gram-negative bacteria, a special step in DNA purification protocol was added, following DNeasy tissue kit manual. Briefly, 180 µL of ATL buffer were added to sample followed by 180 µL volume of enzymatic lysis buffer (20 mM Tris·Cl, pH 8.0, 2 mM sodium EDTA, 1.2% Triton X-100, lysozyme to 20 mg/ml), and incubated for 30 minutes at 37°C. Next, 25 µL of proteinase K solution and 200 µL of buffer AL were added, followed by an incubation step at 56°C for 30 minutes. DNA concentration was determined using a NanoDrop spectrophotometer (Thermo Fisher Scientific, Wilmington, Delaware, U.S.A.) at 260 nm. The 260/280 ratio was also measured to ensure a proper DNA purity. We obtained similar DNA concentrations after kit extraction both from patients and controls biopsies: a Mann-Whitney U test was performed on total DNA concentration (*P* = 0.117), indicating a similar amount of extracted DNA in all the cohorts ('IBD group' and 'Celiac group'). Faecal DNA Extraction {#s2d} --------------------- Faecal samples were taken from fasting CF patients and age- and sex-matched controls in a sterile environment placed into the 'Policlinico Umberto I' hospital, collected in a 50 mL tube, and immediately frozen at −80°C. Total DNA was extracted within 1 hour from sampling by QIAmp Stool Mini Kit (QIAGEN, Hilden,Germany) following manufacturer's instructions. Starting faecal amount was set at 500 mg, picked up from different chunks within the sample itself, in order to minimize the sampling error. Upon extraction, DNA concentration was quantified by a NanoDrop spectrophotometer (Thermo Fisher Scientific, Wilmington, Delaware, U.S.A.) at 260 nm, along with the 260/280 ratio, and integrity checked through 1% agarose gel electrophoresis containing EtBr 0.5 µg/ml. We obtained similar DNA concentrations after kit extraction both from CF patients and controls biopsies. A Mann-Whitney U test was performed on total DNA concentration (*P* = 0.348), indicating a similar amount of extracted DNA in both cohorts within the 'CF group'. *B. bacteriovorus* Cultivation and DNA Extraction {#s2e} ------------------------------------------------- The predatory bacteria used was *B. bacteriovorus* strain HD 100 (DSM No.: 12732) taken from Deutsche Sammlung von Mikroorganismen und Zellkulturen (DSMZ, Braunschweig, Germany), maintained as plaques in double-layered diluted nutrient broth (DNB) agar, a 1∶10 dilution of nutrient broth amended with 3 mM of MgCl~2~·6H~2~O and 2 mM of CaCl2·2H2O \[pH 7.2\] and agar (0.6% agar in the top layer) [@pone.0061608-Starr1]. To initiate a lysate, co-cultures were obtained by adding a plug of agar containing *B. bacteriovorus* plaque to washed prey cells in DNB and incubated at 30°C on a rotary shaker set at 200 rpm until the co-culture became clear (stock lysate). To harvest the predators, co-cultures were prepared in which 2 ml of washed host cells (10^9^ CFU/ml) was incubated with 2 ml of stock lysate in 20 ml of DNB. The co-cultures were incubated for 18 h to reach a final concentration of 10^8^ plaque-forming units (PFU)/ml of predator. At this point, the lysate was passed three times through a 0.45-µm Millex pore-size filter (Millipore, Billerica, MA, USA) to remove residual prey and cell debris (filtered lysate). As a control, filtered sterilized lysate was prepared by sequentially passing the lysates through three 0.22-µm pore-size filters. After filtration, no predator, as judged by PFU, could be detected [@pone.0061608-Kadouri1]. DNA was extracted from the 0.45-µm-filtered suspension of *B. bacteriovorus* through thermal lysis, and used in subsequent PCR as positive control. PCR Procedure Work-flow {#s2f} ----------------------- The present study was aimed to evaluate, by PCR technique, the presence of *B. bacteriovorus* in intestinal and faecal biopsy specimens from paediatric healthy subjects and patients suffering from IBD, Celiac disease and CF. To this purpose, we carried out the following PCR reactions: i) 'β-globin gene amplification' on the total DNA extracted from all samples collected, in order to verify DNA suitability for amplification; ii) *Bdellovibrio* species-specific end-point PCR and qPCR on the total DNA extracted from all samples collected, in order to verify the *Bdellovibrio* presence; iii) 16S rDNA amplification on negative *Bdellovibrio* species-specific PCR samples, in order to asses if overall bacterial DNA was present in such samples, thus ensuring their negativity for *Bdellovibrio*. Detection of *B. bacteriovorus* in Intestinal Biopsies and Faecal Samples by Species-specific PCR {#s2g} ------------------------------------------------------------------------------------------------- A first-step assessment of DNA suitability for subsequent *B. bacteriovorus*-specific PCR was achieved through a human β-globin gene amplification for each sample, as described previously [@pone.0061608-Conte1]. Briefly, aliquots of each DNA sample (50 ng) were amplified with specific primers: forward primer, 5′-CAACTTCATCCACGTTCACC-3′; reverse primer, 5′-GAAGAGCCAAGGACAGGTAC-3′. Amplification reactions were carried out in a 50-µl volume containing 1× PCR buffer II (Applied Biosystems, Roche, California, USA), 3 mM magnesium chloride, 200 µM each deoxynucleoside triphosphate, 50 pmol each primer and 1.5 U of AmpliTaq Gold polymerase (Applied Biosystems). The PCR was carried out under the following conditions: 1 cycle of 95°C for 7 min, 35 cycles of 94°C for 1 min, 55°C for 1 min and 72°C for 1 min and 1 cycle of 72°C for 7 min. Twenty microliters of each PCR were run on a 1% agarose gel containing EtBr 0.5 µg/ml, and photographed with DigiDoc-It system (UVP, Cambridge, UK). *B. bacteriovorus* specific primers Bd529F (5′-GGTAAGACGAGGGATCCT-3′) and Bd1007R (5′-TCTTCCAGTACATGTCAAG-3′) were used to amplify a 481-bp trait of the 16S rDNA gene [@pone.0061608-Davidov1]. At the same time, primers targeting a 910-bp trait of the *hit* locus, specific for *B. bacteriovorus* [@pone.0061608-Schwudke1], [@pone.0061608-Cotter1], were employed: Hit_FW (5′-GACAGATGGGATTACTGTCTTCC-3′) and Hit_RW (5′-GTGTGATGACGACTGTGAACGG-3′). We used these two sets of primers because they are reported to be specific for *B. bacteriovorus* PCR amplification from total DNA, and we decided to use them both to strength the results. PCR was performed with Taq DNA-polymerase (Hot Star Taq Plus, QIAGEN, Hilden, Germany). PCR reaction (25 µl) contained 1X buffer for PCR, 300 µg/mL bovine serum albumin (Sigma-Aldrich, St. Louis, Missouri, United States), 2.5 mM MgCl2, 200 µM for each dNTP, 0.5 µM of each primer, 1.25 U of Taq polymerase and 100 ng of total DNA (from biopsies or faeces). For 16S rDNA trait, sample DNA was amplified under the following conditions: 95°C for 5 min, 20 cycles of 95°C for 1′, 53°C for 1′, 72°C for 1′, and a final step of 72°C for 10 minutes. For *hit* locus, sample DNA was amplified under the following conditions: 95°C for 5 minutes, 20 cycles of 95°C for 1′, 60°C for 1′, 72°C for 1′, and a final step of 72°C for 10 minutes. Appropriate positive (*B. bacteriovorus* strain HD100) and negative (water) controls were employed. In order to minimize the PCR bias, and to avoid the problem due to the expected low levels of *B. bacteriovorus* in samples, we performed three individual PCR reactions for each sample. The individual PCR reactions were unified, analysed by electrophoresis on 2% agarose gels containing ethidium bromide to determine their size (481 bp), and concentrated with SpeedVac (Savant, Holbrook, NY, USA) to reach a final volume approximately equal to 1/3 of the original. The unified PCR reactions were titrated using two different methods: firstly, twenty-five microliters of each concentrated PCR were loaded on a 1% agarose gel containing EtBr 0.5 µg/ml, run for 1 hour at 80 V, photographed with DigiDoc-It system (UVP, Cambridge, UK), and analysed for densitometry with Phoretix 1D software (TotalLab, Newcastle upon Tyne, United Kingdom); secondly, measure of DNA density was performed by NanoDrop spectrophotometer (Thermo Fisher Scientific, Wilmington, Delaware, U.S.A.) at 260 nm, using one microliter of unified PCR. The results obtained by such measures were in agreement one each other and, upon normalization, were used for subsequent analyses. All positive PCR products were sequenced using the BigDyeTM terminator chemistry (Applied Biosystems, Foster City, USA) and the sequencing mixture was analysed on a DNA sequence analyser ABI3730 (Applied Biosystems, Foster City, USA). Sequencing was performed in both directions and sequences were analysed with SEQ MATCH at the Ribosomal Database Project II website (<http://rdp.cme.msu.edu/seqmatch/seqmatch_intro.jsp>). The obtained sequences were also compared to gene sequences of known bacterial identities available in GenBank by means of the Basic Local Alignment Search Tool (BLAST) present at National Centre for Biotechnology Information (NCBI) website (<http://www.ncbi.nlm.nih.gov>). *In silico* PCR Amplification {#s2h} ----------------------------- To ensure that both primer pairs used, for 16S rDNA trait and *hit* locus, were targeted towards *B. bacteriovorus* genome, we performed an *in silico* PCR method through the online tool present at the website of the University of the Basque Country, employing the sequenced genome of the strain HD100: <http://insilico.ehu.es/PCR/index.php?mo=Bdellovibrio>. Such an approach allowed us to determine that *B. bacteriovorus* has two 16S rDNA operons (first target, from 820075 bp to 820555 bp, and second target, from 1688139 bp to 1688619 bp), while it harbours only one *hit* locus on the reverse strand (from 96861 bp to 97770 bp). The information so obtained was used to properly normalize the relative abundances of *B. bacteriovorus*. Bacterial 16S rDNA Amplification with Universal Primers on Negative PCR Samples {#s2i} ------------------------------------------------------------------------------- In order to confirm the significance of results, a further 16S rDNA PCR was performed on all *B. bacteriovorus* PCR-negative samples, detecting the presence of bacterial DNA. Universal primers U968 (5′-GAA CGC GAA GAA CCT TAC-3′) and L1401 (5′-GCG TGT GTA CAA GAC CC-3′) were used to amplify the V6--V8 region of bacterial gene coding for 16S rRNA. PCR was performed with Taq DNA-polymerase (Hot Star Taq Plus, QIAGEN). PCR reaction (25 µl) contained 1X buffer per PCR, 2.5 mM MgCl~2~, 200 µM for each dNTP, 0.5 µM of U968-GC and L1401 oligos, 1.25 U of Taq polymerase and 100 ng of total DNA. The samples were amplified under the following conditions: 95°C for 5 min, cycles at 94°C for 45″, 53°C for 45″, 72°C for one min and 10″, and a final step of 72°C for 30 min. To rule out non-specific products, a 'touchdown PCR' was performed with a starting annealing temperature of 58°C and decreasing it by 0.5°C each cycle to reach 53°C, followed by 30 cycles at 53°C. In order to minimize the PCR bias, three 'touchdown PCR' reactions were performed for each sample and subsequently pooled. To minimize hetero-duplex formation and single-stranded DNA (ssDNA) contamination during PCR amplification, 5 additional cycles of 'reconditioning PCR' were performed, taking 1/10 of the previous pooled PCR volume as template in a new reaction. In order to minimize the PCR bias, three 'reconditioning PCR' reaction were done for each sample and subsequently pooled. Successful reaction and DNA concentration was quantified by spectrophotometer measurements at 260 nm and DNA integrity checked through 1% agarose gel electrophoresis containing EtBr 0.5 µg/ml. Quantitative-PCR {#s2j} ---------------- We performed a quantitative PCR (qPCR) approach with standard curve in order to assay the relative abundances of *B. bacteriovorus* in our samples. Primers and hydrolysis probe used were taken from the literature [@pone.0061608-VanEssche1], amplifying a *B. bacteriovorus*-specific region of 16S rDNA. Briefly, primers used were Bd347F (3′-GGAGGCAGCAGTAGGGAATA-5′) and Bd549R (5′-GCTAGGATCCCTCGTCTTACC-3′), while the probe was Bb396P (5′FAM-TTCATCACTCACGCGGCGTC-TAMRA3′). The qPCR reaction mix was made up with 10 µl of '2X qPCR ProbesMaster' from Jena Bioscience GmbH (Jena, Germany) (composition: qPCR Pol, dATP, dCTP, dGTP, dUTP, reaction buffer with KCl, (NH~4~)~2~SO~4~ and MgCl~2~, ROX, stabilizers), 900 nM for both of the primers, 50 nM for the probe, and 5 µl of template (corresponding to 200 ng of total DNA). The final reaction volume was adjusted to 20 µl with PCR-grade distilled water. Thermal cycling conditions were: 2 min at 95°C (initial denaturation), followed by 50 repeats of 15 s at 95°C (denaturation) and 1 min at 60°C (annealing and extension). Data were collected during the annealing phase. Three different replicates were performed for each sample. In each run no template controls (NTC) were included, along with positive controls. To construct the standard curve, a plasmid (pUC57) containing one copy of the target was provided by Biofab Research (Rome, Italy), quantified by spectroscopy at 260 nm, and 10-fold serially diluted in the range of 10^9^-10^0^ copies [@pone.0061608-Vandecasteele1]. Results of qPCR were normalized for the number of 16S rDNA operons within *B. bacteriovorus* genome [@pone.0061608-VanEssche1], and expressed as 'number of genome copies/mg of sample' (biopsies or faeces), following the Minimum Information for publication of Quantitative real-time PCR Experiments (MIQE) guidelines [@pone.0061608-Bustin1]. Statistical Methods {#s2k} ------------------- To assess putative differences among and within the patients' groups (IBD, Celiac, CF) in terms of prevalence of *B. bacteriovorus* in subjects and samples, we used the χ^2^ test. To assess significant differences among the aforementioned groups in terms of relative abundance of *B. bacteriovorus* in samples, a Mann-Whitney U test was employed. For both tests, a *P* value less than or equal to 0.05 was considered significant. Results {#s3} ======= *B. bacteriovorus* Prevalence in Subjects and Samples {#s3a} ----------------------------------------------------- We analysed overall 87 intestinal biopsies from healthy controls (n = 8) and patients affected by Crohn's (n = 9), Ulcerative colitis (n = 6) and Celiac disease (n = 10), and 51 faecal samples from healthy controls (n = 16) and Cystic fibrosis patients (n = 35). Species-specific end-point PCRs and qPCR were employed to detect the presence of *B. bacteriovorus* in the aforementioned patients. All end-point PCR-positive reactions were sequenced and submitted to SEQ MATCH and BLAST, finding a percentage similarity of 100% against *B. bacteriovorus* strain HD100 for all the sequences. Specifically, the percentage similarity was 100% across 481 bp for 16S rDNA locus, while it was 100% across 910 bp for *hit* locus. We found reliable results among the two sets of primers used. We had reliable results among the two approaches used to detect *B. bacteriovorus* prevalence. Within the 'IBD group', *B. bacteriovorus* was significantly found in 8/8 (100.0%) of control subjects, in 6/6 (100%) of UC patients, and in 1/9 (11.1%) of CD patients. Within the 'Celiac disease group', *B. bacteriovorus* was found in 8/8 (100%) of control subjects, and in 8/10 (80.0%) of Celiac patients. Within the 'CF group', it was found in 2/16 (12.5%) of controls, and in 4/35 (11.4%) of CF patients, with no significant difference. These results, together with *P* values, are summarized in [figure 1](#pone-0061608-g001){ref-type="fig"}. {#pone-0061608-g001} Results of *B. bacteriovorus* prevalence in biopsy samples of the 'IBD group', taken within a single subject from different districts, were reported in [table 2](#pone-0061608-t002){ref-type="table"}. Prevalence of *B. bacteriovorus* was assessed taking into account both end-point and qPCR data. We observed a preponderant prevalence of *B. bacteriovorus* in ileum, colon, and rectum of control biopsies (20/24, 83.3%) in respect to CD (1/27, 3.7%, *P*\<0.0001). No difference was found among overall UC (15/18, 83.3%) and control samples *P* = 0.6760). It is noteworthy that all 'IBD group' biopsy samples (ileum, colon, rectum) were treated with the reducing agent dithiothreitol (DTT), in order to remove the mucus layer and the luminal flora, as reported in [Materials and Methods](#s2){ref-type="sec"} section. Thus, the mucosa-associated flora was retained, along with mucosa-associated *B. bacteriovorus*. 10.1371/journal.pone.0061608.t002 ###### Prevalence of *B. bacteriovorus* (% of positive samples) in the 'IBD group'. {#pone-0061608-t002-2} CD (n = 27) UC (n = 18) Controls (n = 24) -------- ------------- ------------- ------------------- Ileum 0/9 (0) 5/6 (83.3) 7/8 (87.5) Colon 1/9 (11.1) 5/6 (83.3) 7/8 (87.5) Rectum 0/9 (0) 5/6 (83.3) 6/8 (75) Numbers within brackets are percentage values. *B. bacteriovorus* Relative Abundance in Subjects and Samples Assessed by qPCR {#s3b} ------------------------------------------------------------------------------ Data about *B. bacteriovorus* abundance were collected by qPCR. Within-cohort mean values, expressed as number of copies/mg of sample (biopsies or faeces), were compared after a proper correction for the number of 16S operons, namely two in *B. bacteriovorus* genome, as assessed by *in silico* PCR. Standard curves gave a mean slope of −3.3392±0.0405, a mean intercept of 39.4547±0.3976, and a mean R^2^ equal to 0.9970±0.0007 ([Figure S1](#pone.0061608.s001){ref-type="supplementary-material"}). The mean reaction efficiency (E), calculated from the mean value of the slope by means of the equation E = 10^(−1/slope)^, was 99.4% ±1.6%. The lowest reproducible detection level of the qPCR was 10 plasmids per reaction, each containing one target sequence. Results of qPCR were: 549±156 copies/mg and 35±20 copies/mg, for controls *vs* Celiac; 23±6 copies/mg, 2±2 copies/mg, and 9±2 copies/mg for controls, CD and UC. In faecal samples of 'CF group', *B. bacteriovorus* abundance was always under the limit of detection in both controls and CF patients. We found that relative abundance of *B. bacteriovorus* was significantly higher in control bioptic samples of both 'IBD group' and 'Celiac group' ([Figure 2](#pone-0061608-g002){ref-type="fig"}). In particular, significant differences in relative abundances were found among controls *vs* CD (*P*\<0.0001), UC *vs* CD (*P*\<0.0001), and controls *vs* Celiac (*P* = 0.00053). In UC, even though *B. bacteriovorus* is always present ([Figure 1](#pone-0061608-g001){ref-type="fig"}), we observed a drop in its relative abundance of 59.1% in respect to controls ([Figure 2](#pone-0061608-g002){ref-type="fig"}). Also in Celiac patients we found an important presence of *B. bacteriovorus* (80%) similar to controls (100%), but, contrarily, the 93.5% drop in its relative abundance was highly significant (*P*\<0.0005) ([Figure 2](#pone-0061608-g002){ref-type="fig"}). {#pone-0061608-g002} *B. bacteriovorus* Relative Abundance is District-dependent {#s3c} ----------------------------------------------------------- In order to analyse the relative abundance of *B. bacteriovorus* in different intestinal districts, we compared qPCR results obtained from ileum, colon, and rectum biopsy samples, and from faeces encompassing luminal flora. Results from controls clearly showed a decreasing trend of *B. bacteriovorus* relative abundance along the intestine ([Figure 3](#pone-0061608-g003){ref-type="fig"}, grey bars), in a cephalic-caudal direction. Interestingly, we observed the highest *B. bacteriovorus* levels in duodenum (*P* = 0.0012), and the lowest in rectum, while it was under the lower limit of qPCR detection in faeces. Furthermore, results in [figure 3](#pone-0061608-g003){ref-type="fig"} evidenced a kind of 'discrete distribution' of B. *bacteriovorus* abundances, recognizing three different levels: starting from 10^2^ copies/mg in duodenum, 10^1^ in ileum/colon, we arrived at an order of 10^0^ in rectum. {#pone-0061608-g003} In order to compare biopsy samples from controls and diseased subjects, mostly UC samples were taken into account, because CD samples resulted all *B. bacteriovorus*-negative (except one, a colonic biopsy). Comparing control and disease statuses (Celiac, duodenum; UC, ileum, colon, rectum), we observed a significant difference only in duodenum (*P* = 0.00053) and ileum districts (*P* = 0.0434) ([Figure 3](#pone-0061608-g003){ref-type="fig"}). We didn't observe such a difference among control and disease status in colon (*P* = 0.7721) or rectum (*P* = 0.2432) districts. Discussion {#s4} ========== This is the first extensive survey of *B. bacteriovorus* presence in humans, both in healthy and disease status. A precedent study [@pone.0061608-Schwudke1] reported *B. bacteriovorus* presence in one human faecal sample. The occurrence of these predatory bacteria in the intestinal tract raises questions about their role in this habitat. Predation is an important mechanism in nature to keep bacterial populations under control and plays a major role in the cycling of nutrients through the microbial loop [@pone.0061608-Chen1]. Predatory bacteria have been found, always at low population levels, in almost all prey-containing niches, ranging from marine sediments to fresh water and animal intestines. Our study hypothesis was that an imbalanced microbial community, such as observed in different diseases like IBD, Celiac disease and Cystic fibrosis, goes along with a loss/reduction of bacterial species acting as ecologic equalizer, such as *B. bacteriovorus*. We assessed *B. bacteriovorus* presence by end-point and qPCR. Our results showed that *B. bacteriovorus* is significantly present in paediatric healthy subjects and almost completely absent in CD and Celiac patients. Moreover, we found a significant difference among CD and UC for *B. bacteriovorus* prevalence in subjects ([Figure 1](#pone-0061608-g001){ref-type="fig"}), while UC and controls were similar. However, UC samples harboured a diminished *B. bacteriovorus* relative abundance (59.1%) in respect to controls. These last results could underscore the difference between CD and UC patients in terms of microbiota composition, and it could support the idea that UC patients are closer to controls in terms of intestinal habitat, in agreement with already reported studies [@pone.0061608-Qin1], [@pone.0061608-Iebba2], [@pone.0061608-Schippa2]. As reported in [figure 3](#pone-0061608-g003){ref-type="fig"}, we found a differential cephalic-caudal distribution of *B. bacteriovorus* levels along the intestine in controls. The preferential duodenal colonization could reflect its predatory nature, or the existence of particular habitat conditions (pH, redox state, pO~2~, shear-stress). Duodenal district is usually colonized by a bacterial load in the order of 10^2^ CFU/ml, and our finding of a triple-level abundance of *B. bacteriovorus* along the gut, with a 10-fold drop in ileo-colonic districts, should reflect the suitable conditions for prey/predator equilibrium, as stated by Lotka-Volterra equation [@pone.0061608-Varon1], [@pone.0061608-Hobley1], principally in duodenal district. However, we found that *B. bacteriovorus* could strive also in ileo-colonic region at mucosal level, a low-pO~2~ district. It was found how *Bdellovibrio* could survive in anaerobic conditions [@pone.0061608-Atterbury1], thus enabling the colonization of gut districts with low pO~2~. Interestingly, disease status affected *B. bacteriovorus* relative abundance mainly in upper GI districts, such as duodenum and ileum. Future studies should describe the *B. bacteriovorus* distribution in the human gut in a fine-tune fashion, to better understand its role within the bacterial community. In this study, for the first time, we showed the presence of *B. bacteriovorus* at the intestinal mucosal level, indicating that this predatory bacterium could be a permanent inhabitant of the gut ecosystem. We didn't find it in faecal samples at significant prevalence or abundance. We cannot exclude that treatment with DTT, a reducing agent that loosen the disulphide bonds among mucin proteins, ultimately eliminating the mucus layer, should have removed a portion of the *B. bacteriovorus* population from biopsies. Using DTT, we left only the mucosa-attached microbiota, along with the predators inside their gram-negative preys, and, possibly, with a little percentage of free-swimming *B. bacteriovorus*. The intestinal mucus layer is 40--240 µm thick [@pone.0061608-Swidsinski1] and it should be a good 'hunting field' for *B. bacteriovorus* to chase prey: its long polar flagellum [@pone.0061608-Seidler2], [@pone.0061608-Lambert2] drives a speed of around 160 µm/s, (in certain cases up to 400 µm/s), and this speed would be even higher under certain viscosity conditions, as found for other bacterial species [@pone.0061608-Swidsinski2]. Thus, mucus removal could have excluded from DNA extraction, and subsequent PCR detection, a major part of free-swimming *B. bacteriovorus*. In this view, future experiments should address the free-swimming/intra-periplasmic ratio of *B. bacteriovorus* population within a specific sample, both with and without DTT treatment, to deepen the knowledge about its relative abundances along the intestinal tract. The mucosa-associated microbiota, due to its close proximity to the intestinal epithelium and the underlying mucosal immune system, is believed to play a main role in maintaining the homeostasis in healthy gut and in the inflammatory response in diseases. The key role of *B. bacteriovorus* in the gut microbial community is exerted through its selective gram-negative predatory activity, as found in terrestrial and aquatic ecosystems [@pone.0061608-Jurkevitch1], [@pone.0061608-Peng1]--[@pone.0061608-Richardson1]. The low abundance of *B. bacteriovorus* in the intestinal mucosa of IBD and Celiac patients could bring to an uncontrolled bacterial overgrowth, as already reported in these subjects, where an increased number of mucosa-associated bacteria was found [@pone.0061608-Schippa1], [@pone.0061608-DiCagno1], [@pone.0061608-Conte1], [@pone.0061608-Walker1]. In conclusion, the low prevalence of *B. bacteriovorus* at mucosal level, in both IBD and Celiac paediatric patients, could support the idea that loss in microbiota biodiversity could also involve species, such as predatory *Bdellovibrio*, that are normally underrepresented, and that exert a role in regulating the bacterial population levels. A recent study found that all BALOs species accounts for 0.3%--0.7% of the bacterial soil community, with *Bdellovibrio* genus representing around 80% of BALOs [@pone.0061608-Fulthorpe1]. Further studies will be necessary to improve our results, along with an increase in number of subjects to be enrolled. The present study, performed on naïve paediatric patients and controls, enabled us to investigate an untouched gut ecosystem, with no previous endoscopic assessments or pharmacological treatments, thus supporting the idea of an early gut colonization by *B. bacteriovorus* from the environment. In the future, it will be interesting to check the *B. bacteriovorus* presence also in adult controls and patients affected by dysbiotic diseases. Thus, *B. bacteriovorus* could be a good candidate for new therapy strategies, such as shown in birds [@pone.0061608-Atterbury1], aimed to restore the ecosystem balance, and, ultimately, to control the dysbiotic events. Supporting Information {#s5} ====================== ###### **Standard curve for qPCR.** Standard curves gave a mean slope of −3.3392±0.0405, a mean intercept of 39.4547±0.3976, and a mean R^2^ equal to 0.9970±0.0007. As an example, in figure is the standard curve relative to UC samples. On *x* axis, log of 'number of plasmid copies'; on *y* axis, cycle threshold (Ct). (TIF) ###### Click here for additional data file. [^1]: **Competing Interests:**The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: VI SS. Performed the experiments: VI FS AG. Analyzed the data: VI SS LN. Contributed reagents/materials/analysis tools: VI VT MN RDB LN. Wrote the paper: VI SC MPC SS.

Background {#Sec1} ========== Airway remodeling is one of the major features of many obstructive airway diseases, and has been observed both in asthma and in chronic obstructive pulmonary disease (COPD) \[[@CR1], [@CR2]\]. The adverse consequences of airway remodeling could include a decrease in lung function and reduced responsiveness to bronchodilator therapy. Different mechanisms were proposed to explain smooth muscle remodeling, including proliferation (hyperplasia), an increase in myocyte mass and size and an increase in their components as a response to injury or inflammation \[[@CR3], [@CR4]\]. However, many studies have suggested that the migration of ASM cells toward the lumen of the airway might contribute to ASM remodeling \[[@CR5]--[@CR10]\] and plays an important role in the pathophysiology of airway remodeling \[[@CR8], [@CR9], [@CR11]--[@CR13]\]. It has been reported that cigarette smoke or cigarette smoke condensate induced tracheal and lung epithelial cell damage and shedding. Lung epithelial cells are not only the simple structural cells of the lung but also serve as a barrier for inhaled particles and noxious gases (such as cigarette smoke). These cells are also actively involved in inflammatory and repair processes through the release of cytokines and growth factors that maintain normal airway homeostasis and play an important role in the prevention of pulmonary diseases. Airway remodeling may represent the effect of cigarette smoke on the epithelium and the attempts by the airway epithelium to protect itself and repair the injury caused by cigarette smoke. The human alveolar epithelial cell line A549 (A549) is widely acknowledged to be a relevant model cell line for biological evaluations of this mechanism \[[@CR14], [@CR15]\]. Airway smooth muscle (ASM) cells are associated with asthma severity and COPD and have the capacity to migrate. Many chemokines, such as growth factors and cytokines, have been implicated in the initiation and the perpetuation of ASM cell migration \[[@CR11], [@CR16]\]. In fact, a strong relationship between chronic injury or defective repair of the airway epithelium and airway remodeling has been reported in the literature \[[@CR17]\]. Not surprisingly, the epithelium is a major source of the cytokines and growth factors that act as mitogens for ASM cells, such as IL-8 and TGF-β1. More important to the study of epithelial-ASM cell interactions, the receptors for epithelial-derived growth factors and cytokines, such as TGF-β1 and IL-8, were identified on the ASM cell surface \[[@CR18], [@CR19]\]. Increased expression of the TGF-β1 and IL-8 mRNAs and proteins has been observed in the bronchiolar epithelium of smokers with COPD compared to smokers without COPD \[[@CR20]--[@CR22]\]. The expression of TGF-β1 at the injury site may have a central role in tissue homeostasis and repair. IL-8 is a potent neutrophil chemoattractant \[[@CR23]\] and a potential contributor to airway remodeling, as it induces airway smooth muscle proliferation and migration \[[@CR16]\]. Airway epithelial damage exposes the sensory nerve endings to the airway lumen and promotes reflex mechanisms, leading to enhanced vagal release of acetylcholine \[[@CR24]\]. Recent findings suggest that the acetylcholine production in the airways is not restricted to the parasympathetic nervous system, but can also be released from non-neuronal cells and tissues, such as the airway epithelium, to regulate aspects of inflammation and remodeling through its action on mAChRs \[[@CR25]--[@CR29]\]. Dysfunction of the non-neuronal cholinergic system appears to be involved in the pathophysiology of asthma and COPD. Although a number of molecules that are involved in acetylcholine-mediated airway inflammation and remodeling have been identified, the net effect of mAChRs in regulating epithelial-derived chemokine-induced ASM cell migration has not been explored in great detail. The aim of the present study was to characterize the potential role of the non-neuronal components of the cholinergic system on ASM cell migration toward bronchial epithelial cells stimulated with CSE. For this reason, we first evaluated whether the IL-8 and TGF-β1 from the CSE-stimulated epithelial cells might be able to initiate ASM cell migration using the transwell assay and the co-culture model of human ASM cells and A549 cells. Then, we tested whether the ability of CSE to induce the ASM cell migration might be regulated by mAChRs and reproduced by an exogenous mAChRs agonist, and determined their mechanism. Methods {#Sec2} ======= Cell culture {#Sec3} ------------ The A549 cells obtained from the American Type Culture Collection (Manassas, VA, USA) and the normal primary human airway smooth muscle cells (passage 3-8, from a single healthy donor) obtained from Pricell Inc. (Shanghai, China) were cultured in Ham's F-12 medium (Gibco, USA) and Dulbecco's Modified Eagle's Medium (DMEM), respectively, containing 10 % fetal bovine serum, L-glutamine (2 mM), penicillin (100 U/ml) and streptomycin sulfate (100 U/ml, Gibco, USA) in a humidified incubator at 37 °C with 5 % CO~2~. Cigarette Smoke Extract (CSE) {#Sec4} ----------------------------- The CSE was prepared as previously described, with minor modifications \[[@CR30]\]. The CSE was prepared by combusting 1 cigarette (double happiness, the amount of tar was 12 mg), using a pump and passing the smoke through 10 ml of FBS-free culture medium at a rate of 5 min/cigarette. The resulting solution was adjusted to pH 7.4 with 1 mol/L of concentrated NaOH and filtered through a 0.22-μm filter. The obtained solution was referred to as 100 % strength and diluted to the desired concentrations with culture medium. Measurement of the IL-8 and TGF-β1 produced by the A549 cells {#Sec5} ------------------------------------------------------------- After serum deprivation for 24 h before each experiment, the cells were pre-treated with tiotropium (Sigma--Aldrich, USA) for 30 min before 3 % CSE stimulation, or treated with different concentrations of carbachol alone. After 72 h, the concentrations of IL-8 and TGF-β1 in the culture supernatants were immediately measured using an ELISA kit (R&D Systems, Minneapolis, MN) according to the manufacturer's protocol. Cell migration assay {#Sec6} -------------------- We jointly applied transwell migration chamber assay and scratch assay to examine the ASM cell migration induced by epithelial derived supernatants. For transwell migration chamber assay, the A549 cells were cultured in the lower wells (omitted in the negative control) at densities of 100,000 cells/cm^2^ and grown to 80 % confluence, followed by serum-deprivation for 24 h before each experiment. Subsequently, the cells were treated with carbachol or 3 % CSE for 72 h. Tiotropium (Tio, 0.1, 1, 10 μM) and 4-DAMP (10 μM, selective M3 mAChR antagonist) were added to the cells 30 min before CSE stimulation and were present throughout the experiment. Additionally, where appropriate, the A549 cells were pre-incubated with either the IL-8 inhibitor Ac-RRWWCR-NH2 (AnaSpec, Inc., Fremont, Calif) or the TGF-β1 inhibitor SB431542 (Sigma--Aldrich, USA) for 30 min. The ASM cells (30,000 cells/cm^2^) were added to the upper chambers. The vehicle was DMSO and the final concentration of DMSO was 0.1 % in all wells, including the controls. After incubating the cells for 6 h at 37 °C, the upper chamber was removed from the 24-well plates and the residual cells on the upper side of the transwell filters were wiped with cotton swab. The transwell filters were rinsed with PBS and fixed in 4 % formaldehyde in PBS for 15 min. The cells that had migrated to the lower surface were stained with 0.1 % crystal violet for 25 min, visualized and photographed under an inverted fluorescence microscope (Zeiss, Germany). The migration rate was analyzed using the Image Pro Plus 6.0 software. For scratch assay, the confluent A549 cells were stimulated with 3 % CSE for 72 h with the pretreatment of Ac-RRWWCR-NH2 and SB431542. Then, the supernatants were collected as stimulants for ASM cell migration. ASM cells were cultured in DMEM until 100 % confluence and then wounded with a sterile 20 μL pipette tip. Subsequently, the cells were washed with PBS and treated with the epithelial derived supernatants. The wounds were photographed at baseline and 24 h later under an inverted fluorescence microscope (AMG, America). The wound healing width was analyzed using the Image Pro Plus 6.0 software. Determination of the acetylcholine levels by LC-MS/MS {#Sec7} ----------------------------------------------------- The acetylcholine levels in the supernatants from the A549 cells were determined by LC--MS/MS, as previously described \[[@CR31]\]. Real-time quantitative PCR of the mAChRs {#Sec8} ---------------------------------------- The A549 cells were stimulated with 3 % CSE for 24, 48 or 72 h. The expression of the mAChRs was detected by Real-time quantitative PCR (RT-PCR), as previously described \[[@CR31]\]. Western blot analysis {#Sec9} --------------------- Equal amounts of protein were loaded onto 10 % SDS-PAGE gels and transferred to polyvinylidene fluoride membranes. To avoid non-specific binding, the membranes were blocked with 5 % non-fat milk in TBS-T (Tris-HCl 50 mM, NaCl 150 mM, and Tween-20 0.1 %) for 1 h at room temperature. The membranes were then incubated with primary antibodies against phosphorylated-p38 (\#4511; Cell Signaling Technology, USA) or phosphorylated-Akt (\#4060; Cell Signaling Technology, USA) overnight at 4 °C. After washing three times with TBS-T for 10 min, the membranes were incubated with an HRP-conjugated secondary antibody (1:5000) for 1 h at room temperature, followed by three additional washes with TBS-T. The bands were subsequently visualized on film using enhanced chemiluminescent reagents. All bands were normalized against the total p38 (for phospho-p38) and total Akt (for phospho-Akt) levels, as appropriate. In the experiments designed to establish whether the p38 MAPK or PI3K/Akt pathways were involved in the ASM cell migration induced by the mAChRs agonist carbachol, a p38 inhibitor (SB203580: 10 μM) or a PI3K/Akt inhibitor (LY294002: 10 μM) was added to the culture 1 h prior to stimulation. Statistical analysis {#Sec10} -------------------- The data are presented as the means ± SEM. The statistical significance of the differences between groups was assessed with one-way analysis of variance (ANOVA), followed by a Dunnett's test for selected pairs, as appropriate. *p*-values \<0.05 were considered significant. All statistical analyses were performed using Prism version 5.0 (GraphPad Software, San Diego, USA). Results {#Sec11} ======= The role of IL-8 and TGF-β1 from the CSE-stimulated epithelial cells in initiating ASM cell migration {#Sec12} ----------------------------------------------------------------------------------------------------- The A549 cells that had been activated with 3 % CSE exhibited increased IL-8 and TGF-β1 levels in the supernatants, resulting in ASM cell migration toward the A549 cells. The IL-8 or TGF-β1 levels in supernatants were increased by approximately 2.23-fold and 1.89-fold, respectively, in response to CSE stimulation (*p* \<0.01), that effect was markedly inhibited by tiotropium in a concentration-dependent manner (Fig. [1a](#Fig1){ref-type="fig"} and [b](#Fig1){ref-type="fig"}). We also used the transwell assay and the scratch assay to further assess whether CSE-induced ASM cell migration was initiated by IL-8 or TGF-β1. The IL-8 inhibitor Ac-RRWWCR-NH2 or the TGF-β1 inhibitor SB431542 was added to the lower chamber of the transwell system (A549 epithelial cells). As expected, the results from transwell migration assay suggested that Ac-RRWWCR-NH2 and SB431542 decreased ASM cell migration ratio by 81.06 and 75.81 % (*p* \<0.01) compared to the non-treated cells, indicating that the compounds inhibited ASM cell migration (Fig. [1c and d](#Fig1){ref-type="fig"}). As results of transwell assay, scratch assay also demonstrated that supernatants from CSE-stimulated epithelial cells induced ASM cell migration, this phenomenon was significantly inhibited by both Ac-RRWWCR-NH2 and SB431542 (Fig. [1e](#Fig1){ref-type="fig"} and [f](#Fig1){ref-type="fig"}), suggesting that the IL-8 and TGF-β1 from the CSE-stimulated A549 cells have an important role in driving ASM cell migration.Fig. 1The role of IL-8 and TGF-β1 from CSE-stimulated epithelial cells in initiating ASM cell migration. Tiotropium (Tio, 0.1, 1, 10 μM) was added to the A549 cells 30 min before stimulation with 3 % CSE. After 72 h of stimulation, the supernatants were collected and the IL-8 (**a**) and TGF-β1 (**b**) concentrations were determined using an ELISA. ASM cell migration initiated by the A549 cells stimulated with 3 % CSE were reduced by the IL-8 inhibitor Ac-RRWWCR-NH2 and the TGF-β1 inhibitor SB431542. Cell migration was detected by using transwell chamber assay (**c**, ×100) and scratch assay (**e**, ×100). A quantitative analysis of migration used Image Pro Plus 6.0 (**d** and **f**). The values are expressed as the means ± SEM, *n* = 3-5. ^††^ *p* \< 0.01 vs the control group, ^\*^ *p* \< 0.05 and ^\*\*^ *p* \< 0.01 vs the CSE group Blockade of the epithelial cell-derived chemokines by an mAChRs antagonist reduces ASM cell migration {#Sec13} ----------------------------------------------------------------------------------------------------- As airway epithelial cells exhibit all components of the non-neuronal cholinergic system, we investigate the involvement of this system in the control of ASM cell migration in response to the epithelial cell-derived chemokines. The A549 cells were pre-incubated with the mAChRs antagonist tiotropium which was added 30 min before CSE stimulation. As shown in Fig. [2](#Fig2){ref-type="fig"}, the CSE-stimulated A549 cell-induced ASM cell migration was significantly reduced by tiotropium in a concentration-dependent manner. The ability of tiotropium (0.1, 1, 10 μM) to regulate the ASM cell migration was 37.53, 69.32 and 89.96 % lower than the cells treated with 3 % CSE alone, respectively, suggesting that the endogenous acetylcholine exerts its activity by activating the mAChRs on the epithelial cells.Fig. 2Effects of tiotropium on ASM cell migration initiated by the A549 cells stimulated with CSE. The A549 cells were stimulated with 3 % CSE for 72 h. Tio (0.1, 1, 10 μM) was added to the cells 30 min before 3 % CSE stimulation and was present throughout the experiment. ASM cell migration initiated by the stimulated A549 cells was detected after 6 h of co-culture in the transwell system, the migrated ASM cells were stained with crystal violet (**a**, ×100). A quantitative analysis of migration used Image Pro Plus 6.0 (**b**). The values are expressed as the means ± SEM, *n* = 3. ^††^ *p* \< 0.01 vs the control group. ^\*^ *p* \< 0.05 and ^\*\*^ *p* \< 0.01 vs the CSE group Involvement of the non-neuronal cholinergic system in epithelial-derived chemokine-mediated ASM cell migration {#Sec14} -------------------------------------------------------------------------------------------------------------- If the non-neuronal cholinergic system in epithelial cells is indeed involved in the epithelial cell-derived chemokine-mediated ASM cell migration, we hypothesized that the application of exogenous mAChRs agonists would reproduce the effect of endogenous acetylcholine. To verify this hypothesis, the A549 cells were first stimulated with the acetylcholine analogue carbachol. As shown in Fig. [3](#Fig3){ref-type="fig"}, stimulation of the A549 cells with the mAChRs agonist carbachol increased IL-8 (Fig. [3a](#Fig3){ref-type="fig"}) and TGF-β1 (Fig. [3b](#Fig3){ref-type="fig"}) levels and induced concentration-dependent ASM cell migration, which could be inhibited by the IL-8 inhibitor Ac-RRWWCR-NH2 and the TGF-β1 inhibitor SB431542, respectively (Fig. [3c](#Fig3){ref-type="fig"} and [d](#Fig3){ref-type="fig"}). Moreover, carbachol-induced ASM cell migration was significantly reduced by tiotropium (10 μM) (*p* \<0.01) and 4-DAMP (10 μM) (*p* \<0.01) (Fig. [4](#Fig4){ref-type="fig"}). These results strongly support our hypothesis that endogenous acetylcholine may stimulate the M3 mAChR in A549 cells to promote ASM cell migration. We also found that the stimulation of A549 cells with CSE increased the expression of the M3 mAChR mRNA (Fig. [5a](#Fig5){ref-type="fig"}), as well as acetylcholine production. Interestingly, we also found that CSE-induced acetylcholine release was increased by 42.34 % in the presence of a cholinesterase inhibitor neostigmine, which did not affect acetylcholine release when using alone (Fig. [5b](#Fig5){ref-type="fig"}). Because the inducing effect of endogenous acetylcholine could be reproduced by an exogenous mAChRs agonist, we further hypothesized that the stimulatory effect of CSE on ASM cell migration could be enhanced by carbachol. Thus, co-incubation experiments were performed using carbachol (10 μM) and 3 % CSE. CSE-induced ASM cell migration was significantly and cooperatively enhanced by carbachol compared to the cells stimulated with CSE alone. And a synergistic effect of CSE plus carbachol on ASM cell migration was observed in comparison with CSE or carbachol alone (Fig. [6](#Fig6){ref-type="fig"}). The enhancing effect of carbachol prompts us to propose that an autocrine/paracrine loop for acetylcholine and the activation of M3 mAChR regulates ASM cell migration.Fig. 3The carbachol-stimulated A549 cells induced IL-8 and TGF-β1 release and ASM cell migration. The A549 cells were stimulated with carbachol (CCh, 0.1, 1, 10 μM) in the presence or absence of the IL-8 inhibitor Ac-RRWWCR-NH2 and the TGF-β1 inhibitor SB431542 for 72 h. Then, the supernatants were collected and the IL-8 (**a**) and TGF-β1 (**b**) concentrations were determined using an ELISA. ASM cell migration initiated by the stimulated A549 cells was detected after 6 h of co-culture in the transwell system, the migrated ASM cells were stained with crystal violet (**c**, ×100). A quantitative analysis of migration used Image Pro Plus 6.0 (**d**). The values are expressed as the means ± SEM, *n* = 5--8. ^†^ *p* \< 0.05 and ^††^ *p* \< 0.01 vs the control group, ^\*\*^ *p* \< 0.01 vs the CCh (10 μM) group, ^‡‡^ *p* \< 0.01 vs the CCh (10 μM) groupFig. 4Effects of the mAChR antagonists on carbachol-induced ASM cell migration. The A549 cells were stimulated with CCh (10 μM) for 72 h. Tio (10 μM) or 4-DAMP (10 μM) was added to the cells 30 min before CCh stimulation and was present throughout the experiment. ASM cell migration initiated by the stimulated A549 cells was detected after 6 h of co-culture in the transwell system, the migrated ASM cells were stained with crystal violet (**a**, ×100). A quantitative analysis of migration used Image Pro Plus 6.0 (**b**). The values are expressed as the means ± SEM, *n* = 4. ^††^ *p* \< 0.01 vs the control group. ^\*\*^ *p* \< 0.01 vs the CCh groupFig. 5Effects of the CSE on acetylcholine release and the mRNA levels of M1, M2, M3. **a** Neostigmine (10 μM) was added to the A549 cells 1 h before stimulation with 3 % CSE. The acetylcholine (ACh) levels were determined after 72 h co-culture using an LC-MS/MS assay. **b** The A549 cells were incubated with 3 % CSE for 24, 48, 72 h, and the mRNA levels were determined using a quantitative RT-PCR and standardized to that of the GAPDH gene. The values are expressed as the means ± SEM. *n* = 3-5. ^\*^ *p* \< 0.05 and ^\*\*^ *p* \< 0.01 vs the control groupFig. 6Enhancement of ASM cell migration by co-stimulating the A549 cells with CSE and carbachol. The A549 cells were incubated with 3 % CSE and CCh (10 μM) for 72 h. ASM cell migration initiated by the stimulated A549 cells was detected after 6 h of co-culture in the transwell system, the migrated ASM cells were stained with crystal violet (**a**, ×100). A quantitative analysis of migration used Image Pro Plus 6.0 (**b**). The values are expressed as the means ± SEM, *n* = 4. ^\*^ *p* \< 0.05 and ^\*\*^ *p* \< 0.01 vs the control group The role of the p38 and PI3K/Akt signaling pathways in mAChR-mediated ASM cell migration {#Sec15} ---------------------------------------------------------------------------------------- To better understand how mAChRs mediated ASM cell migration, the downstream signaling pathways that were activated by the mAChR agonist or CSE were investigated. We first assessed the relationship between mAChR activation and the p38 and PI3K/Akt pathways on ASM cell migration using an mAChR agonist and antagonist. As shown in Fig. [7a](#Fig7){ref-type="fig"} and [b](#Fig7){ref-type="fig"}, the phosphorylation of both p38 and Akt was increased by the mAChR agonist carbachol (10 μM, *p* \<0.05) and CSE-induced phosphorylation of p38 and PI3K/Akt were enhanced by carbachol (*p* \<0.01). In addition, the carbachol- or carbachol plus CSE-induced phosphorylation of p38 and Akt was significantly inhibited by 4-DAMP, suggesting that there is a relationship between the M3 mAChR and the p38 and PI3K/Akt pathways in ASM cell migration. Specific inhibitors of p38 and PI3K/Akt were used to further confirm the contribution of p38 and PI3K/Akt activation to ASM cell migration. As shown in Fig. [7c](#Fig7){ref-type="fig"} and [d](#Fig7){ref-type="fig"}, SB203580 (10 μM, *p* \<0.01) and LY294002 (10 μM, *p* \<0.01) caused marked reductions in carbachol-induced ASM cell migration, suggesting that the p38 and PI3K/Akt signaling pathways are required for ASM cell migration. These findings indicated that the p38 and PI3K/Akt signaling pathways are involved in carbachol-induced ASM cell migration via the M3 mAChR in the A549 cells.Fig. 7The involvement of the p38 MAPK and PI3K/Akt pathways in ASM cell migration. **a** and **b** The A549 cells were co-incubated with CCh (10 μM) or 3 % CSE for 72 h after pretreatment with 4-DAMP (10 μM). Then the cell lysates were assayed for the levels of total and phosphorylated p38 and Akt; the total p38 and Akt levels were used as a loading control. The A549 cells were incubated with CCh (10 μM) for 72 h after pretreatment with LY294002 (10 μM) or SB203580 (10 μM). ASM cell migration initiated by the stimulated A549 cells was detected after 6 h of co-culture in the transwell system, the migrated ASM cells were stained with crystal violet (**c**, ×100). A quantitative analysis of migration used Image Pro Plus 6.0 (**d**). The values are expressed as the means ± SEM. *n* = 4--6. ^†^ *p* \< 0.05 and ^††^ *p* \< 0.01 vs the control group, ^\*^ *p* \< 0.05 and ^\*\*^ *p* \< 0.01 vs the corresponding group without 4-DAMP. ^‡‡^ *p* \< 0.01 vs the CCh group Discussion {#Sec16} ========== This study demonstrates that the endogenously released acetylcholine stimulates the M3 mAChR in the airway epithelial cells via an autocrine loop and thus leads to the release of IL-8 and TGF-β1, initiating ASM cell migration. Moreover, the p38 and PI3K/Akt signaling pathways are involved in carbachol-induced ASM cell migration via the M3 mAChR in the A549 cells. These findings indicate that the M3 mAChR may be important therapeutic target as it regulates the effects of the epithelial-derived chemokines on ASM cell migration during lung remodeling. Airway remodeling is the result of tissue injury and repair, and involves the interactions between a variety of structural cells, including epithelial cells and ASM cells. Some growth factors and cytokines have been reported as the chemokines for ASM cells \[[@CR12], [@CR13], [@CR32]\]. Cigarette smoking causes direct lung damage, as well as the activation of the lung inflammatory responses and airway remodeling pathways that are important in the etiology of the development of asthma and COPD. The epithelium is a significant source of chemokines, such as cytokines and growth factors. The exposure of epithelial cells to the CSE induced the production of these chemokines, which play an important role in inducing ASM cell migration toward the epithelial surface and contribute to airway remodeling. TGF-β1 and IL-8, important chemokines, are known to be released by epithelial cells following cigarette smoke exposure \[[@CR33]--[@CR35]\]. In the present study, we showed that the CSE-stimulated A549 epithelial cells released IL-8 and TGF-β1, resulting in ASM cell migration toward the epithelial cells. The migratory effect was substantially inhibited by Ac-RRWWCR-NH2 (IL-8 inhibitor) and SB431542 (TGF-β1 inhibitor), suggesting that these chemokines may initiate ASM cell migration. Interestingly, we have shown that the migration of the ASM cells is induced by the chemokines from the stimulated A549 epithelial cells and could be reduced by treatment with non-selective (tiotropium) and selective (4-DAMP) antagonists of the M3 mAChR. To date, five mAChR subtypes have been identified (M1-M5) and are thought to mediate the majority of the actions of acetylcholine in the peripheral and central nervous systems. Of these, M1, M2 and M3 are expressed at the highest levels in the airways and subserve different physiological functions \[[@CR36]\]. The presence of endogenously released acetylcholine stimulates the M3 mAChR via an autocrine loop and thus modulates the functions of the airway epithelial cells, as was previously reported by our and other laboratories \[[@CR31], [@CR37], [@CR38]\]. Our observation of the predominant expression of the M3 mRNA by the A549 epithelial cells is consistent with other reports showing that a human bronchial epithelial cell line (16-HBE) expressed both the proteins and mRNAs for the M1, M2 and M3 mAChRs, with the following levels: M3 \> M1 \> M2 mAChRs \[[@CR39]\]. The CSE increases the activity of the chemokines from the human epithelial cells and promotes the cellular response to acetylcholine by affecting the M3 mAChR. The fact that the carbachol-induced migration of the ASM cells toward the A549 epithelial cells blocked by 4-DAMP confirmed the involvement of the M3 mAChR. A synergistic effect of carbachol and CSE on ASM cell migration was also observed in this study. Furthermore, tiotropium could prevent the carbachol (acetylcholine analogue)- and CSE-induced ASM cell migration toward the epithelial surface. This finding further demonstrates that the repertoire of mAChR subtypes in the A549 epithelial cells is modified by CSE stimulation, and the M3 mAChR is significantly up-regulated. Therefore, it is reasonable to assume that the CSE-induced up-regulation of M3 mAChR expression observed in the present study increases the A549 cells' response to acetylcholine. The application of the mAChRs antagonists tiotropium or 4-DAMP reduced the CSE-induced ASM cell migration in the absence of exogenous mAChRs ligands or cholinergic neurons in the culture preparations. This finding suggests that the levels of endogenous acetylcholine that are released by the CSE-stimulated A549 epithelial cells are sufficient to exert its autocrine effects via the stimulation of the M3 mAChR. The release of endogenous acetylcholine was also demonstrated by ASM migration toward the epithelial cells treated with the cholinesterase inhibitor neostigmine, which increased the concentration of acetylcholine in the supernatant, suggesting that cholinesterase activities are involved in acetylcholine degradation and constitute an active, endogenous component of the non-neuronal cholinergic system in the process of cells migration. In the A549 cells, the concentration of acetylcholine that is secreted in close proximity to the membrane M3 mAChR is likely to be much higher than that measured in the supernatant. However, these values are similar to the concentrations of exogenous acetylcholine that are required to induce IL-8 release from A549 cells \[[@CR31]\] and leukotriene B4 release from 16-HBE cells \[[@CR38]\]. In fact, the epithelial cells express the machinery of the cholinergic system, including the acetylcholine-synthesizing choline acetyltransferase, the mAChRs and the acetylcholine-hydrolyzing enzymes acetylcholinesterase and butyrylcholinesterase \[[@CR31], [@CR37]\]. Furthermore, the acetylcholine analogue carbachol induced ASM cell migration in a concentration-dependent manner and the carbachol-induced cell migration event could be blocked by tiotropium or 4-DAMP, suggesting that M3 mAChR stimulation is indeed involved in ASM cell migration induced by the chemokines from the stimulated A549 epithelial cells, resulting in airway remodeling. The current scenario strongly suggests a local, autocrine or paracrine role of epithelial acetylcholine in regulating various aspects of airway remodeling via the mAChRs. In addition to synthesizing and releasing acetylcholine, the epithelial cells can also be activated by acetylcholine, which exerts its physiological effects via the activation of M3 mAChR. The last part of our work focused on the signaling pathways that jointly or individually regulate the ASM cell migration induced by IL-8 and TGF-β1 from the activated A549 epithelial cells in the presence or absence of CSE. IL-8 expression requires the activation of NF-kB and at least one or two MAPK pathways. In addition to the Smad-mediated pathways, other pathways, such as the MAPK and PI3K/Akt pathways, have been implicated in TGF-β1 signaling \[[@CR40], [@CR41]\]. Several studies have suggested that the p38, extracellular signal-regulated kinase (ERK) and phosphoinositide 3-kinase (PI3K) signaling pathways play important roles in regulating ASM cell migration \[[@CR8]\]. In the present study, we provide evidence regarding the mechanism by which M3 mAChR activation induces IL-8 and TGF-β1 release from the A549 cells through phosphorylation of p38 and Akt, thus causing the ASM cells to migrate toward the epithelial cells. Cell migration was reduced by SB203580 (p38 MAPK inhibitor) and LY294002 (PI3K/Akt inhibitor), respectively, suggesting that p38 and PI3K/Akt signaling pathways were both involved in cell migration. Moreover, the addition of both carbachol and CSE enhanced p38 and Akt phosphorylation and was also inhibited by 4-DAMP, showing that M3 mAChR activation can amplify the signaling pathways that govern the CSE-induced IL-8 and TGF-β1 expression. These findings reinforce and expand other reported experiments showing that the phosphorylation of proteins in the Smad2/3 and ERK pathways is involved in the epithelial-mesenchymal transition triggered by the TGF-β1 from the carbachol-stimulated A549 epithelial cells, resulting in airway remodeling \[[@CR42]\]. Moreover, acetylcholine mediates the release of IL-8 in human bronchial epithelial cells by an NF-kB/ERK-dependent mechanism \[[@CR39]\], demonstrating the involvement of the p38 and PI3K/Akt signaling pathways in CSE-induced ASM cell migration via the activation of the M3 mAChR. Conclusion {#Sec17} ========== In summary, our study provides evidence that the non-neuronal cholinergic system is involved in regulating ASM cell migration. Lung epithelial cells secrete acetylcholine, which may function as an autocrine growth factor via the activation of M3 mAChR, to induce ASM cell migration via the p38 and Akt signaling pathways. These findings demonstrated that the M3 mAChR may be important therapeutic target for obstructive airway diseases, as it regulates the effects of the epithelial-derived chemokines on ASM cell migration, which results in lung remodeling. **Competing interests** The authors declare that they have no competing interests. **Authors' contributions** JJL conducted the experiments, analyzed the data and wrote the manuscript. GNX participated in the design of the study and performed the statistical analysis. PY participated in the experiments and helped to write the manuscript. YS helped in the interpretation of the data and coordination and helped to draft the manuscript. LZ, HZC, YYC conceived of the study, and participated in its design, and edited and revised the manuscript. All authors read and approved the final manuscript. This work was funded by the National Natural Science Foundation of China (Grants 81370144 and 81102491).

All relevant data are within the paper and its Supporting Information files. Introduction {#sec001} ============ With the development of the society and economy, oil products are becoming more and more important for automobile industry. Driven by the great economic benefit, some unscrupulous traders sold low-value or adulterated oil products instead of high-value oil products in recent years. Many oil refinery factories in China are producing adulterated oil to make more profits according to a report by China Central Television (CCTV) in its annual 3.15 Gala program\[[@pone.0146547.ref001]\]. They use 90\# gasoline, naphtha, aromatics and other additives to produce 93\# blend oil. Adulterated oil has not only damaged the consumers' benefits, but also threated people's safety. Therefore, to guarantee and promote oil products' quality, the identification of the qualified oil products and adulterated oil products is extremely essential. High Performance Liquid Chromatography (HPLC) and Mass Spectroscopy (MS) are well known chemical detection methods, and HPLC has advantages in terms of accuracy and sensitivity \[[@pone.0146547.ref002]\].Although the result achieved by HPLC is accurate, it is time consuming, inefficient and destructive, and also requires highly trained and qualified professionals. Moreover, the identification cannot be used on-line in the industrial field. Thus, an effective method based on spectral technique and pattern classification technique has been proposed for the identification of the qualified oil products and adulteration products. Because it's faster, cheaper and nondestructive, it is considered as an alternative method for oil detection. Kim et al were the first to use real-time classification method for petroleum products detection and studied oil products classification of six varieties using near-infrared spectra \[[@pone.0146547.ref003]\]. However there is limited research on the identification of the oil adulteration, especially using hyperspectral imaging technique. Owing to its advantages, hyperspectral imaging (HSI) which integrates imaging and spectral technique together has been studied extensively in many areas. By analyzing sesame oil, Xie et alachieved 95.59% and 98.53% classification performance by SPA-LS-SVM and CARS-LDA using near-infrared hyperspectral imaging\[[@pone.0146547.ref004]\]. As pointed out by Kesslerin *Science* \[[@pone.0146547.ref005]\], oil products samples exhibit bright fluorescence under 365nm ultraviolet light. Actually the mechanism of the phenomenon is much more complicated. Different components and percentage of oil produce different fluorescence. If oil products are adulterated, the color and luminous intensity of fluorescence will be changed. It can be shown in the hyperspectral imaging. Yi et alused wavelet of three-dimensional fluorescence spectrum to classify six oil varieties of four classes under halogen illumination\[[@pone.0146547.ref006]\]. Besides halogen illumination, UV illumination is also a possible excitation way. Atas et al achieved 90% classification rate of examining aflatoxin-infected chili pepper under UV fluorescence using a hyperspectral imaging system\[[@pone.0146547.ref007]\]. This paper aims to find a way to identify the oil products and adulterated ones using HSI technique under compound light, halogen illumination and UV excitation. Four radiation indexes which are extracted from each ROI are used as feature vectors. And then a novel feature set of quantized histogram matrix (QHM) and a novel feature selection method based on improved kernel independent component analysis (iKICA) are proposed. The objectives of this work are: 1. to select effective features using feature selection method by our constructed model; 2 to compare the performance of different feature selection models under different light illumination; 3. to find out the quantitative relationships between the spectral information and the oil adulteration. In the following section, we will describe hyper spectral data capture and preprocessing. And then, the feature extraction and selection methods will be introduced in Section 3. Next, we will present and discuss our experimental results in Section 4. Finally, conclusions will be given in Section 5. Materials {#sec002} ========= Flow of the study {#sec003} ----------------- In the previous studies \[[@pone.0146547.ref002]\]\[[@pone.0146547.ref003]\]\[[@pone.0146547.ref004]\]\[[@pone.0146547.ref006]\],single illumination source was used. Especially, some studies were performed just under halogen illumination, and others were performed only under UV illumination. Because UV illumination is utilized for the fluorescence and halogen excitation is for reflectance phenomena. In this study, we utilized both excitations to investigate their contribution to the classification performance. Figs [1](#pone.0146547.g001){ref-type="fig"} and [2](#pone.0146547.g002){ref-type="fig"} respectively depict the general overview of the hyperspectral imaging system and the flowchart of the proposed system. {#pone.0146547.g001} {#pone.0146547.g002} The main steps are as follows. The samples are divided into calibration set and prediction set in a proportion of 4: 1. In the first step, we acquire the hyperspectral images of the four oil varieties within the wavelength region of 400--720nm. In the second step, the reflectance information is extracted from ROI of the hyperspectral images of each sample and used as feature vectors. These feature vectors include individual band radiation index (RI), difference of consecutive spectral band radiation index (DRI), ratio of consecutive spectral band radiation index (RRI) and normalized of DRI (NDRI). Another set of features called quantized histogram matrix (QHM) are extracted from those features by quantizing the histogram of the image. They can be calculated by Eqs [1](#pone.0146547.e001){ref-type="disp-formula"}--[5](#pone.0146547.e005){ref-type="disp-formula"}. In the third step, we select significant features based on improved kernel independent component analysis (iKICA). For comparison, some recommended algorithms such as plus L reduce R (plusLrR), Fisher, multidimensional scaling (MDS), principle component analysis(PCA) and independent component analysis (ICA) are also used to identify the most significant wavelengths or features. In the next step, an identification model is established based on support vector machine (SVM) and optimal identification model is selected by comparing the identification performance (correct classification rate, CCR). At last, the result whether the oil samples are qualified or not is achieved by the model. Samples {#sec004} ------- Four varieties of oil products including gasoline, diesel, kerosene and engine oil are from different gas stations or shops in Shandong province, China. Most of them are provided by Shengli Oilfield. There are total 64 samples (ROI) of each variety, and 75% of them are adulterated. All the samples were sent to Shengli Oilfield for oil quality analysis and labeled as qualified oil products and adulterated products. Then, 60 ml of each sample is distributed individually in a glassware with the same size(d = 90mm). And each is then captured individually by the HSI system. Hyperspectral imaging system {#sec005} ---------------------------- The image acquisition system consists of a CCD camera with a lens assembly. Hyperspectral image series have been captured under 100W halogen light and UV 365nm LED (LUYOR-3404, USA) illumination sources ranging from 400nm to 720nm with spectral bandwidth 10nm. Size of each image is 1392⤬1020. A raw hyperspectral image (hyperspectral cube) with a dimension of (*x*,*y*,*λ*) which is scanned along the direction of the 33 bands in *λ* dimension is created as the sample. [Fig 3](#pone.0146547.g003){ref-type="fig"} depicts sample images from the hyperspectral image series under halogen and UV illuminations and spectral reflectance curve of different oil samples. {#pone.0146547.g003} Preprocessing and ROI extraction {#sec006} -------------------------------- Default camera software uses histogram equalization to acquire images. By adaptively changing exposure time, histogram equalization not only automatically controls over-saturation and under-saturation, but also modifies original pixels' value. Then, to settle this particular issue, we have to set the value of exposure time and parameter of the camera as a predefined value. Eventually, under saturated and over saturated regions are generated in the hyperspectral image series due to single value of exposure time. Therefore, we multiply a normalization coefficients by reflectance value for all bands. The normalization coefficient is defined as the reciprocal of exposure time. Before feature extraction, the exposure value is normalized by their normalization coefficient. Then we use histogram equalization method to adjust pixel gray values to the range of 0--255. An area that is considered as the region of interest (ROI) with 174⤬130 pixels is obtained from different locations of each corrected hyperspectral image (each sample) and results in a total of 64 samples of each variety of oil product. Reflectance values of all pixels are obtained by ENVI5.1 software. All features are extracted via Matlab2008a software to establish calibration model for the identification of different oil products. Methods {#sec007} ======= Radiation index and quantized histogram matrix {#sec008} ---------------------------------------------- Classification performance is closely related to the features extracted from the images. Ideally, the feature vectors should keep the most concise descriptions of the desired function. These feature vectors specify the distinction between qualified oil and adulterated oil. Nevertheless, it is not a straightforward and trivial process to extract meaningful and discriminative features. It requires domain knowledge and underlying physical phenomena. In these hyperspectral images of the oil samples, morphology features do not correlate with difference of oil, and it is not desirable to rely on solely spectral band mean intensity Therefore, useful features should be considered. In this study, we extract features by calculating radiation indexes of consecutive spectral bands and applying histogram quantization method. Assume the gray value of the pixel located at (*x*,*y*) of the *k*th spectral band is *I*~*k*~(*x*,*y*). The Individual band radiation index (RI)\[[@pone.0146547.ref007]\] is defined by: $$RI_{k} = {\sum\limits_{x}{\sum\limits_{y}{I_{k}\left( {x,y} \right)}}}\mspace{27mu}{k = 1,2,\cdots,33}$$ Then, we extract the following feature vectors by calculating radiation index of consecutive spectral band. Difference of consecutive spectral band radiation index (DRI) is calculated by: $$DRI_{k} = {\sum\limits_{x}{\sum\limits_{y}\left( {I_{k + 1}\left( {x,y} \right) - I_{k}\left( {x,y} \right)} \right)}}\mspace{27mu}{k = 1,2,\cdots,32}$$ Ratio of consecutive spectral band radiation index (RRI) is calculated by: $$RRI_{k} = {\sum\limits_{x}{\sum\limits_{y}\left. {I_{k + 1}\left( {x,y} \right)}/{I_{k}\left( {x,y} \right)} \right.}}\mspace{27mu}{k = 1,2,\cdots,32}$$ Normalized of DRI (NDRI) is calculated by: $${NDRI_{k} = {\sum\limits_{x}{\sum\limits_{y}\left. \left( {I_{k + 1}\left( {x,y} \right) - I_{k}\left( {x,y} \right)} \right)/\left( {I_{k + 1}\left( {x,y} \right) + I_{k}\left( {x,y} \right)} \right) \right.}}}\quad{k = 1,2,\cdots,32}$$ Here, *x* = 1 *to M*, *y* = 1 *to N*, *M* and *N* correspond to the size of the band image. The feature vectors described in Expressions 1 to 4 reduce the information in a given band to a single value. However, valuable information may be provided by the frequency of the difference of the intensity values or the frequency of a particular intensity value, which can be extracted when the difference of the intensity values or the histogram of the intensity values for a given spectral band is used. [Fig 4](#pone.0146547.g004){ref-type="fig"} presents the extracting process of the quantized histogram matrix feature. First, the histogram of the spectral band image is computed with a number of bins predefined which not only limits the size of feature vectors but also promotes a reasonable number of pixels falling in each bin. Within the particular bin the total number of pixels is used as the histogram feature. Then we can construct the quantized histogram matrix (QHM) by using all spectral bands depicted in [Fig 4](#pone.0146547.g004){ref-type="fig"}. For simplicity, we present the extraction process only for 12 bins. {#pone.0146547.g004} Here, we only calculat QHM features for RI and DIR. The QHM features can be described as: $$QHM_{k,n} = {\sum\limits_{x}{\sum\limits_{y}{I_{k,n}\left( {x,y} \right)}}}\mspace{27mu}{k = 1,2,\cdots,33}\quad{n = 1,2,\cdots,12}$$ Where *k* denotes index of spectral band, *n* denotes the bin index and 12 is the number of bins that we want to apply. Consequently, *I*~*k*,*n*~(*x*,*y*) is the RI or DIR of the *n*th bin. Kernel ICA and its improvement {#sec009} ------------------------------ Independent component analysis (ICA) \[[@pone.0146547.ref008]\] can be expressed as the problem that a latent random vector *X* can be recovered from observations of *m* unknown linear functions of that vector. The components of *X* are assumed to be independent of each other. And, an observation *Y* is modeled as: $$Y = AX\qquad\text{here},\, X = (x_{1},x_{2},\cdots,x_{m}),\mspace{9mu} Y = (y_{1},y_{2},\cdots,y_{m}).$$ Where *x* is a latent random vector with independent components, and *A* is a parameter matrix of m⤬m. Given *N* independently, identically distributed observations of *y*, we hope to estimate *A* and thereby to recover the latent vector *x* corresponding to any specific *y* by solving a linear problem. We can obtain a parametric model which can be estimated via maximum likelihood by specifying distribution for the components *x*~*i*~. With *w* = *A*^−1^ as the parameterization, one can easily obtain a gradient or fixed point algorithm that yields an estimate $\hat{W}$ and provide estimates of the latent components via $\hat{X} = \hat{W}Y$. Hyvärinen et alhave proposed an algorithm named fast fixed-point algorithm for independent component analysis\[[@pone.0146547.ref009]\]. Unfortunately, it is difficult to approximate and optimize the mutual information based on a finite sample. In this paper, we provide a new solution to the ICA problem based on an entire function space of candidate nonlinearities instead of a single nonlinear function. Especially, the functions are dealed with in a reproducing kernel Hilbert space, which we can use "kernel trick" to search over efficiently. It is the use of the function space that makes it possible to adapt to all kind of sources and makes algorithms more robust to various source distributions depicted as follows. Bach et al defined a contrast function that can do a rather direct measurement of the dependence of a set of random variables functions from ° to ${\mathbb{R}}'$\[[@pone.0146547.ref010]\]. For simplicity, assume *x*~1~ and *x*~2~ are two univariate random variables and *F* is a vector space of functions from ° to ${\mathbb{R}}'$ and *F*-correlation *ρ*~*F*~ is the maximal correlation between the random variables *f*~1~(*x*~1~) and *f*~2~(*x*~2~), where *f*~1~ and *f*~2~ range over *F*: $$\rho_{F} = \max\limits_{f_{1},f_{2} \in F}corr(f_{1}(x_{1}),f_{2}(x_{2})) = \max\limits_{f_{1},f_{2} \in F}\frac{{cov}(f_{1}(x_{1}),f_{2}(x_{2}))}{\left( {{var}f_{1}(x_{1})} \right)^{1/2}\left( {{var}f_{2}(x_{2})} \right)^{1/2}}$$ It is clear that the *F*-correlation would equal to zero if the variables are independent. Furthermore, the converse is also true if F is big enough. We use the idea of reproducing kernel Hilbert space (RKHS) to get a computationally manipulable implementation of the *F*-correlation. Let *F* be an RKHS on °, *K*(*x*,*y*) be the associated kernel, and Φ(*x*) = *K*(⋅,*x*) be the feature map, where *K*(⋅,*x*) is a function in *F* for each *x*. Then we have the famous reproducing property. {#pone.0146547.e012g} f ( x ) = ⟨ Φ ( x ) , f ⟩ , ∀ f ∈ F , ∀ x ∈ ℝ . This implies: $$corr(f_{1}(x_{1}),f_{2}(x_{2})) = corr\left( {\left\langle \left. {\Phi(x_{1}),f_{1}} \right\rangle \right.,\left\langle \left. {\Phi(x_{2}),f_{2}} \right\rangle \right.} \right).$$ Consequently, between one-dimensional linear projections of Φ(*x*~1~) and Φ(*x*~2~) the *F*-correlation is the maximal possible correlation that is exactly the definition of the first canonical correlation between Φ(*x*~1~) and Φ(*x*~2~), which suggests that the computation of a canonical correlation can be based on an ICA contrast function in a function space. The separated independent components (ICs) are unordered using traditional ICA. The first separated ICs may be not important. Therefore, it needs some criteria to sort these ICs. In this paper, we use negentropy as a criterion to measure the nongaussianity of ICs. Then, the IC with maximum negentropy will be separated first. Negentropy is given by $$N_{g}(Y) = H(Y_{Gauss}) - H(Y)$$ Where, *Y*~*Gauss*~ is a random gauss variable and has the same variance as *Y*, *H*(⋅) is the differential entropy of the random variable. Support vector machine {#sec010} ---------------------- Support vector machine was proposed by Cortes C.& Vapnik V. \[[@pone.0146547.ref011]\]. SVM has been widely used in many fields \[[@pone.0146547.ref012]\]\[[@pone.0146547.ref013]\]\[[@pone.0146547.ref014]\], and can solve both linear and nonlinear multivariate calibration problems. Rather than a quadratic programming (QP) problem, a set of linear equations was used to get the support vectors (SV). Here, we utilize support vector machine (SVM) as the classifier for our problem. The radial basis function (RBF) is used as the kernels in consideration of its excellent performance. The SVM algorithm is presented as below: $$y(x) = {\sum\limits_{k = 1}^{N}{\alpha_{k}K\left( {x,x_{k}} \right) + b}}$$ Where, *α*~*k*~ are Lagrange multipliers, *K*(*x*,*x*~*k*~) is the kernel function, and *b* is the bias value. The regularization parameter *gam*(*γ*) is used to measure the tradeoff between the training error and model complexity, and parameter *sig*^2^(*σ*^2^) is used to define the non-linear mapping from input space to high dimensional feature space. In this study, the optimal parameter values of (*γ*,*σ*^2^) are calculated by grid search and they are calculated by free LIBSVM toolbox(v2.91) \[[@pone.0146547.ref015]\] in matlab2008A. Results {#sec011} ======= Feature construction {#sec012} -------------------- Series of hyperspectral images of each oil samples are acquired at two different illumination modes (halogen and UV) within the spectral region of 400-720nm with 33 spectral bands. The spectral information has the characteristic of high dimensionality with redundancy among contiguous wavelengths. Images of the 64 different locations of each sample generate a total of 16896 images of 1392⤬1020 resolution. If the gray value is used as the feature vector directly, the size of it will be too large. Large feature size causes "curse of dimensionality" problem as known to all. As increasing the dimension of the feature vector results in exponential increase in the data size, the size of feature vector should be reduced to a reasonable level. Fewer features have many advantages, such as improving the classifier performance, providing a faster computation and making the underlying mechanism of the problem better understood. By Eqs ([1](#pone.0146547.e001){ref-type="disp-formula"}) to ([4](#pone.0146547.e004){ref-type="disp-formula"}), feature vectors are extracted with the size of 33 or 32. The other two types of feature sets are extracted according to Eq ([3](#pone.0146547.e003){ref-type="disp-formula"}). They are respectively quantized by RI and DRI. The total number of features in the quantized IR is 33(spectral bands)⤬12(quantization bins) = 396. Similarly, we have 384(32⤬12) features for the quantized DRI.The left figure in [Fig 5](#pone.0146547.g005){ref-type="fig"} depicts the boxplot of the IR and DRI under halogen illustration, and their QHM features are shown in the right figure in [Fig 5](#pone.0146547.g005){ref-type="fig"}. It is obvious that the DRI data have strong separation ability. As seen the right figure in [Fig 5](#pone.0146547.g005){ref-type="fig"}, there exists large number of zero value features in the feature set. These zero feature will be discarded in the first step. {#pone.0146547.g005} Features selection {#sec013} ------------------ Effective feature selection aims to seek for a subset of features as small as possible to cover the full wavelengths. The subsets of features, as the substitution of the full spectral features, are equal or more efficient because reducing the dimensionality of raw data makes the identification less time-consuming. Plus L reduce R (plusLrR) and Fisher algorithms are used in this paper as features selection methods to identify the most significant wavelengths, which can also be used in the development of the multispectral imaging identification system. With the reduced spectral bands, it will be possible to construct a simple machine vision system for oil detection. [Fig 6](#pone.0146547.g006){ref-type="fig"} illustrates the Fisher discrimination ability values of each band. It is obvious that some bands have stronger discrimination ability, such as 400,410nm under halogen illumination and 430 and 520nm under UV illumination. {#pone.0146547.g006} Features optimization methods are to seek a nonlinear mapping from all bands wavelength to a feature space, whose size is smaller than that of bands wavelength. We use multidimensional scaling (MDS) and principle component analysis (PCA) for feature detection. Independent component analysis (ICA) has been used to identify the single component spectra in glucose\[[@pone.0146547.ref016]\]. Kernel independent component analysis (KICA) is a kind of feature detection methods essentially\[[@pone.0146547.ref017]\]. They can be carried out to identify the most significant feature mapping from feature set. The improved kernel independent component analysis (iICA) proposed in this paper uses negentropy as a criterion to measure the nongaussianity of ICs. The IC with maximum negentropy will be first separated. This will be very useful for classification. Classifier results {#sec014} ------------------ In this paper we use K-fold cross validation technique to evaluate the generalization performance. In the machine learning community, Wassenaar et al suggest that the recommended value of *K* is usually 5 or 10 \[[@pone.0146547.ref018]\]. Therefore we set *K* as 5 in this paper. Our data set is randomly divided into five disjoint folds. Four of them are used for training and validation purposes, and the left is used as the test set for our predictive model. The process for each fold is repeated for 5 times to get the average accuracy rate. To evaluate the classification performance of our method, we compared our method with the original features and reduced features using plusLrR, Fisher, MDS and PCA methods. To achieve a fair comparison, wee unify the size of the feature subsets to 12. [Table 1](#pone.0146547.t001){ref-type="table"} shows overall accuracy rates of several feature sets with various feature selection methods under the halogen and UV illuminations. The best accuracy rates of different feature sets are highlighted in bold. The composite illustration is depicted in [Fig 7](#pone.0146547.g007){ref-type="fig"}. As indicated in [Table 1](#pone.0146547.t001){ref-type="table"} and [Fig 7](#pone.0146547.g007){ref-type="fig"}, in most case, iKICA method outperforms the others. Even though Fisher, MDS and PCA can achieve a 100% accuracy rate in two cases (DIR and RRI under UV excitation). iKICA still exhibits the best performance. As can be seen from [Table 1](#pone.0146547.t001){ref-type="table"}, taking operation of consecutive spectral band generally improves the classification performance for both halogen and UV excitations. The DRI is the most perfect. {#pone.0146547.g007} 10.1371/journal.pone.0146547.t001 ###### Result of the extracted features with different feature selection methods by SVM classifier under halogen and UV excitations. {#pone.0146547.t001g} Illumination source Feature sets Org. feature size Feature selection methods(12features) SVM classifier --------------------- -------------- ------------------- ------------------------------------------------------ ------- ----------- ----------- ----------- ----------- Halogen IR 33 81.25 44.64 64.29 46.43 84.82 **95.54** DIR 32 93.75 90.18 90.18 89.29 94.64 **99.11** RRI 32 78.57 64.29 83.04 73.21 73.21 **100.0** NDRI 32 74.11 56.25 66.96 73.21 73.21 **99.11** UV IR 33 99.11 96.43 98.21 93.75 99.11 **99.25** DIR 32 **100.0** 99.11 **100.0** **100.0** **100.0** **100.0** RRI 32 **100.0** 93.75 **100.0** **100.0** **100.0** **100.0** NDRI 32 99.11 86.61 100.0 99.11 100.0 **99.75** [Table 2](#pone.0146547.t002){ref-type="table"} compares the proposed quantized histogram matrix (QHM) features \[[@pone.0146547.ref019]\] with texture features proposed by Wang et al. As shown in this table, our proposed QHM features outperform texture features. We compare the result of SVM with that of artificial neural network (ANN). In the ANN, how to determine the optimal number of neurons in the hidden layer is still an open problem. There exist some empirical rules which are widely used \[[@pone.0146547.ref020]\]\[[@pone.0146547.ref021]\]. Specifically, the Rapid Miner team suggested the number of neurons in the hidden layer could be calculated by\[[@pone.0146547.ref020]\]: $$N_{nodes} = \frac{(N_{features} + N_{classes})}{2} + 1$$ where, *N*~*nodes*~ is the number of nodes in the hidden layer, *N*~*features*~ is the number of features in input nodes, *N*~*classes*~ is the number of expected classes. In our trials, we used this approach due to the satisfactory results. 10.1371/journal.pone.0146547.t002 ###### Generation performance of QHM and texture features of DIR by feature selection iKICA and identification by SVM, ANN and PLS under halogen, UV and light fusion. {#pone.0146547.t002g} Illumination source Feature sets Org. feature size Feature selection methods(12features) SVM classifier --------------------- -------------- ------------------- ------------------------------------------------------ ------- ----------- ----------- ----------- ----------- Halogen IR 33 81.25 44.64 64.29 46.43 84.82 **95.54** DIR 32 93.75 90.18 90.18 89.29 94.64 **99.11** RRI 32 78.57 64.29 83.04 73.21 73.21 **100.0** NDRI 32 74.11 56.25 66.96 73.21 73.21 **99.11** UV IR 33 99.11 96.43 98.21 93.75 99.11 **99.25** DIR 32 **100.0** 99.11 **100.0** **100.0** **100.0** **100.0** RRI 32 **100.0** 93.75 **100.0** **100.0** **100.0** **100.0** NDRI 32 99.11 86.61 100.0 99.11 100.0 **99.75** In addition, we also compared SVM with linear discriminant analysis (LDA) algorithm. [Table 2](#pone.0146547.t002){ref-type="table"} indicates that SVM method outperforms other methods in most cases, and QHM shows higher performance than texture features of DIR. As shown in [Table 2](#pone.0146547.t002){ref-type="table"}, taking both halogen and UV excited at the same time, the CCR will be improved a little. Additionally, the QHM features were more efficient than the band features above, which can be seen from [Table 1](#pone.0146547.t001){ref-type="table"} and [Table 2](#pone.0146547.t002){ref-type="table"}. Method validation {#sec015} ----------------- In order to provide evidence for the efficiency of this new method above, here we use another dataset to do the experiment repeatedly. The data set was also collected by hyperspectral camera under two kinds of illuminations (Halogen and UV). There are two kinds of samples, one is crude oil, and the other is crude oil which has been emulsified. Each sample includes 64 hyperspectral data with 33 spectral bands and the resolution is 1392⤬1020. The emulsified oil is the mixture which is composed of crude oil and different percent of emulsification. Here the crude oil indicates the gasoline, and the emulsified oil indicates the adulterationed gasoline. By the method proposed above, we can get the result as shown in [Table 3](#pone.0146547.t003){ref-type="table"}. As the above conclusion, it is obvious that the proposed method shows the best performance for identifying the crude oil and the oil emulsified. At the same time the DIR provides better feature set than the IR method. 10.1371/journal.pone.0146547.t003 ###### Generation performance on another set (crude oil and emulsified crude oil). {#pone.0146547.t003g} Illumination Feature set Feature selection methods(12features) SVM classifier -------------- ------------- ------------------------------------------------------ ------- ------- ------- ------- Halogen IR 68.21 71.17 91.67 91.67 95.24 DIR 69.05 95.24 92.53 95.24 97.62 UV IR 70.24 98.81 97.62 98.81 98.81 DIR 96.43 100.0 98.81 100.0 100.0 Conclusion {#sec016} ========== This paper aims to evaluate the feasibility of identifying the qualified oil and the adulterated oil using HSI with a spectral range of 400-720nm. Hyperspectral image series of 64 oil sample are acquired under both UV and halogen illumination conditions. DIR, RIR and DTIR feature set are extracted based on IR. And then, the most discrimination features QHM are constructed with 12 bins quantization. Besides this, a novel feature selection method has been proposed based on the maximum negentropy of ICs separated by kernel independent component analysis. Compared with plusLrR, Fisher, MDS and PCA methods, our approach achieves a 100.0% accuracy rate under the UV illumination with DIR feature set and KICA feature selection method. UV illumination is superior to halogen. Experimental results demonstrate that SVM outperform ANN in terms of classification accuracy. Robustness of our proposed method is verified by QHM of DIR features under UV excitation with a classification accuracy of 100.0%. Supporting Information {#sec017} ====================== ###### Hyperspectral Imaging Features Dataset. (ZIP) ###### Click here for additional data file. [^1]: **Competing Interests:**The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: ZH. Performed the experiments: JW KL. Analyzed the data: ZH. Contributed reagents/materials/analysis tools: ZH. Wrote the paper: LD KL.

{#sp1 .93} {#sp2 .94} {#sp3 .95} {#sp4 .96} {#sp5 .97} {#sp6 .98} {#sp7 .99} {#sp8 .100} {#sp9 .101} {#sp10 .102} {#sp11 .103} {#sp12 .104} {#sp13 .105} {#sp14 .106} {#sp15 .107} {#sp16 .108}

Introduction {#s1} ============ Genome-wide association studies (GWAS) have become ubiquitous in complex disease genetics. While the tools to conduct these studies have improved substantially, the cost of conducting them remains expensive. This is despite the plummeting cost per genotype, and is a result of the increasing number of markers being interrogated with each successive generation of genotyping chip. Identifying efficient study designs thus remains important. One popular efficient design for GWAS is the two stage GWAS (2S-GWAS), which has been used for investigations of a wide range of diseases, such as type 2 diabetes [@pone.0042367-Scott1], schizophrenia [@pone.0042367-ODonovan1], lupus erythematosus [@pone.0042367-Graham1], psoriasis [@pone.0042367-Nair1], and breast cancer [@pone.0042367-Thomas1]. In the 2S-GWAS, a full sample of cases and controls is divided between a first stage that employs a whole genome genotyping platform and tests all available markers for association with the disease, and a second stage that uses a more expensive custom genotyping platform to follow up those markers exhibiting sufficiently strong association with the disease in stage 1. The evidence of association from both stages is then considered jointly to reach a final determination of association between marker and disease. The 2S-GWAS was shown to be more efficient than one stage analyses in which all samples are genotyped on the whole genome platform by Satagopan and Elston [@pone.0042367-Satagopan1] and Thomas et al [@pone.0042367-Thomas2]. Since these early investigations, continued attention has been paid to the theoretical properties of two stage designs [@pone.0042367-Kitamura1], [@pone.0042367-Gail1], and efforts have been made to explicitly tie these theoretical properties to the problem of computing experimental designs [@pone.0042367-Skol1]--[@pone.0042367-Scherag1]. Recent summaries of methodological and practical issues pertaining to 2S-GWAS are provided by Thomas et al [@pone.0042367-Thomas3] and Van Steen [@pone.0042367-VanSteen1]. In this paper, we are concerned with computing experimental designs for 2S-GWAS. Our work is based upon that of Skol et al [@pone.0042367-Skol1], [@pone.0042367-Skol2], who 1) demonstrated that joint analyses that combine information on case/control allele frequency differences across stages are substantially more powerful than those based on replication, although slightly less powerful than more expensive one stage analyses; and 2) developed a software package (CaTS) to compute minimum cost designs for 2S-GWAS, subject to both an explicitly stated constraint on the minimum level of acceptable study power, and an implicitly stated equality constraint on the proportion of cases and controls allocated to the stages. We extend this work in two ways. First, we improve the cost efficiency of the 2S-GWAS by defining a procedure that allows different proportions of cases and controls to be assigned to stages, and developing software to compute minimum cost, power constrained designs for the unrestricted 2S-GWAS. We demonstrate that the unrestricted 2S-GWAS can be substantially more cost effective than designs that restrict case and control allocation proportions to be equal. In the studies we present here, which use relatively modest differences of case and control sample sizes, we achieve up to a 40% relative cost advantage as compared to the 2S-GWAS designs computed by CaTS and 80% compared to one stage designs. Second, and based upon our success in improving the cost effectiveness of 2S-GWAS designs relative to a power constraint, we consider 2S-GWAS designs that maximize power with respect to a cost constraint. Calculating such designs may be useful for maximizing the utility of studies that are cost constrained rather than designed to meet a given level of power, or are subject to reductions in funding relative to that required to achieve a given level of power in a cost minimizing 2S-GWAS. We examine the latter case, and demonstrate that substantial amounts of power can be retained by recalculating power maximizing experimental designs subject to cost constraints, even for large reductions in planned cost. To support our results and assist applied researchers, our 2S-GWAS design software is included with this paper (Code S1) and available at <http://www.bioinformatics.org/stanhope/2SGWASdesign/>. Methods {#s2} ======= Defining a two stage GWAS with different allocations of cases and controls to stage 1 {#s2a} ------------------------------------------------------------------------------------- Let the total number of cases and controls be and with defined as the ratio of controls to cases, let be the total number of biallelic markers to be genotyped in stage 1, and let and be stage 1 and stage 2 per genotype costs. We define the risk allele frequency at a hypothetical disease marker in cases and controls as and respectively, and we assume Hardy-Weinberg equilibrium within the population. Let and be the respective proportions of cases and controls allocated to stage 1, and let be the expected proportion of markers selected for follow-up in stage 2 if no markers are associated with disease. (Note that is not selected to control the type I error rate, but to reduce cost by ensuring that uninteresting markers are not genotyped in stage 2.) We suppose that the risk allele frequencies of the cases and controls assigned to stages 1 and 2 ( and where the second term in the subscript corresponds to stage) are equal to and respectively. That is, there is no population heterogeneity. Stage 1 of the 2S-GWAS proceeds by comparing allele frequencies at each marker, using the allocated cases and controls. For each marker showing significant differences in allele frequencies between cases and controls in stage 1 (where stage 1 significance is determined by ), a stage 2 test of allele frequencies is calculated using the remaining cases and controls. The stage 1 and 2 test statistics are then combined according to their Fisher informations, and a joint statistic is used to evaluate the total evidence of association with the disease. For clarity, in [Fig. 1](#pone-0042367-g001){ref-type="fig"} we provide a flowchart of the steps in this 2S-GWAS. The following section provides technical details. Some of the presented results have already been established (e.g. Theorem 1). However, and for clarity, we choose to err on the side of completeness. {#pone-0042367-g001} ### The stage 1 test statistic and its asymptotic behavior {#s2a1} In stage 1, differences between the estimated case and control allele frequencies and are evaluated using the statistic:Formally, we wish to evaluate vs. by comparing stage 1 case and control allele frequencies, and we do so by using the asymptotic distribution of . **Theorem 1:** where and . (Proof provided in [Appendix S1](#pone.0042367.s001){ref-type="supplementary-material"}.) ### Stage 1 critical value {#s2a2} Under , follows from Theorem 1. The critical value for the stage 1 test is therefore determined by , and is defined as:Note that under the null the test is expected to pass markers from stage 1 to stage 2. ### Stage 1 power {#s2a3} Under the alternative, the power of the stage 1 test is:where and are as in Theorem 1, and we have stated as a function of . ### The stage 2 test statistic and its asymptotic behavior {#s2a4} Stage 2 analysis proceeds for markers rejecting in stage 1 by estimating case and control allele frequencies based on stage 2 genotypes, and , and calculating the statistic:The asymptotic distribution of under either the null or the alternative is analogous to that of . ### Constructing the joint test statistic {#s2a5} After calculating , the markers under consideration are reevaluated using a joint analysis of stage 1 and stage 2 allele frequencies based on a null model Fisher information-averaged test statistic. Letting and be the weights given to and , we compute the joint analysis test statistic as:where , and is defined:(see [Appendix S1](#pone.0042367.s001){ref-type="supplementary-material"} for details). ### Stage 2 joint test critical value calculation {#s2a6} Let and be critical values for the stage 1 and joint tests respectively. To be significantly associated with disease a marker must be rejected at both the first and second stages. Let and be stage 1 and 2 test rejection indicators; is the indicator that the marker is genome-wide significant. The probability of this event under the null can be evaluated by conditioning:where is the probability under the null model. To achieve a marker-wise type I error equal to , is to be maintained. by construction, therefore must be such that . For example, if (e.g. a type I error equal to that of a Bonferroni-controlled 5% test), then is to be set such that:We compute by integrating over the conditional distribution of and decomposing into its stage-specific portions, and numerically solve for (see [Appendix S1](#pone.0042367.s001){ref-type="supplementary-material"} for details). ### Stage 2 power {#s2a7} At the susceptibility marker, and are and distributed (where and are defined in Theorem 1, and analogously for and ). Letbe the stage 2 power stated in terms of . Because of its complexity, we omit providing an explicit form for here, but do number the equation to correspond to its reference in [Fig. 1](#pone-0042367-g001){ref-type="fig"}. Obtaining the power of the joint test conditional on is done with a computation analogous to that used to compute type I error (see [Appendix S1](#pone.0042367.s001){ref-type="supplementary-material"} for details and the explicit statement of the equation). Defining constrained minimum cost and maximum power two stage designs {#s2b} --------------------------------------------------------------------- We define an optimal two stage design as that which achieves a specified power at the least cost or, alternatively, that which maximizes power for a given experimental cost. The genotyping cost incurred when performing a 2S-GWAS iswhere and are the per marker genotyping costs for stages 1 and 2, and the total power of the 2S-GWAS is . The optimal cost minimized design is that having power of at least with the lowest cost . That is, the following constrained optimization problem is to be solved: The optimization problem determining a power maximizing, cost constrained 2S-GWAS design is defined analogously: Implementation {#s2c} -------------- We developed algorithms in C to identify optimal 2S-GWAS designs as a function of . In our methods, the integration used to calculate the joint statistic\'s critical value and its power are computed using rectangular cubature; stage 2 critical values are obtained using the bisection method; and the three parameter constrained cost minimization and power maximization problems are solved using grid search. These algorithms are provided in Code S1. Evaluating the cost advantages of two stage designs allowing unequal proportions of cases and controls allocated to stage 1 {#s2d} --------------------------------------------------------------------------------------------------------------------------- We examined the cost benefits and optimal design parameters for 2S-GWAS when not restricting case-control proportions in stage 1 to be equal (i.e. ) under an array of experimental conditions. Each condition was characterized by several factors that influence the optimal design and its cost: the ratio of controls to cases (); stage 2 per marker genotyping cost (, where stage 1 per marker genotyping cost is taken to be 1); disease prevalence (); and the population frequency of the risk allele (). We considered studies with cases and markers, and determined case and control disease allele frequencies such that a one stage GWAS would have 80% power under a multiplicative model of genetic effects with experiment-wise type I error rate of 5% and controlling for multiple testing with a Bonferroni correction using the CaTS software ([@pone.0042367-Skol1], [@pone.0042367-Skol2]). The full set of experimental conditions is outlined in [Table S1](#pone.0042367.s003){ref-type="supplementary-material"}. For each experimental condition, we identified cost minimizing 2S-GWAS designs that would maintain 78% power both with and without the restriction, using CaTS and the unrestricted methods described in this paper respectively. (As suggested by [@pone.0042367-Skol1] and [@pone.0042367-Skol2], 2S-GWAS designs typically target slightly lower power levels than one stage analyses, and so we reduced our target power by 2% from the one stage baseline.) Cost minimizing designs found using the unrestricted method were determined to the nearest 1% allocation of cases and controls to stage 1 () and 0.1% proportion of markers to be passed from stage 1 to stage 2 (). We compared the costs of the one stage and optimal 2S-GWAS designs and verified the powers of the 2S-GWAS designs and critical values computed by the unrestricted methods using 100000 sets of sampled risk allele data. Finally, we evaluated the power sensitivities of the cost minimizing 2S-GWAS designs determined by both CaTS and our unrestricted method to batch effects or genetic heterogeneity between stages by assuming the second stage case and control disease allele frequency to be 90% of that used to calculate design parameters. Using the new second stage disease allele frequency, we then re-computed the powers of the original 2S-GWAS designs and critical values using 100000 sets of sampled risk alleles. We repeated this process assuming the second stage disease allele frequency was 110% of that specified. Evaluating how much power is recovered by recomputing experimental designs after reductions in study funding {#s2e} ------------------------------------------------------------------------------------------------------------ To examine how much power could be recovered by recomputing experimental designs after a hypothetical reduction in the funding available to a previously planned study, we focused on the set of experimental conditions defined in [Table S1](#pone.0042367.s003){ref-type="supplementary-material"} with disease prevalence () of 10% and disease allele frequency () of 10%. For each experimental condition, we determined the minimum cost 78% power unrestricted 2S-GWAS design, and then calculated maximum power unrestricted 2S-GWAS designs that were constrained to cost 50%, 75% and 90% of that. We compared the maximum obtainable study power after cost constraint to the original 78% target, verified the power of the computed designs and critical values using 100000 sets of sampled risk allele data, and determined the performance sensitivity of the power maximizing designs to batch effects as we did in our analogous study in cost minimizing designs. Results {#s3} ======= Two stage designs with unequal proportions of cases and controls allocated to stage 1 are optimal when controls outnumber cases {#s3a} ------------------------------------------------------------------------------------------------------------------------------- Minimum cost experimental design parameters and performance characteristics for the full set of experimental conditions described in [Table S1](#pone.0042367.s003){ref-type="supplementary-material"} are provided in [Tables S2](#pone.0042367.s004){ref-type="supplementary-material"} and [S3](#pone.0042367.s005){ref-type="supplementary-material"}. Here, we provide plots of our results for three sets of conditions: , letting  = 1, 2, 4, and 8; , letting  = 10, 25 and 50%; and letting  = 1, 10 and 100 (where is disease prevalence, the population disease allele frequency, the stage 2 genotyping cost and the ratio of controls to cases). [Figures 2](#pone-0042367-g002){ref-type="fig"} and [3](#pone-0042367-g003){ref-type="fig"} describe the cost advantages of our unrestricted methods and characteristics of its computed design parameters respectively. {#pone-0042367-g002} {#pone-0042367-g003} As in previous work [@pone.0042367-Satagopan1], [@pone.0042367-Thomas2], two stage designs computed by both CaTS and our unrestricted algorithm have substantial cost advantages relative to the one-stage design ([Fig. 2](#pone-0042367-g002){ref-type="fig"}). More important from the perspective of this paper are the cost advantages of unrestricted 2S-GWAS designs relative to those computed by CaTS. This advantage increases as the ratio of controls to cases or the cost of stage 2 genotyping increases, and as population disease allele frequency decreases. For the experimental conditions plotted in [Fig. 2](#pone-0042367-g002){ref-type="fig"}, the unrestricted algorithm shows up to a 40% cost advantage in comparison to CaTS. These results are consistent with those provided in [Tables S2](#pone.0042367.s004){ref-type="supplementary-material"} and [S3](#pone.0042367.s005){ref-type="supplementary-material"}, which show that although there is little difference in cost performance in 2S-GWAS designs when the number of cases equal that of controls (), when there is a 10--40% cost advantage gained by using unrestricted designs (taken across all other experimental conditions). For more modest differences between the number of cases and controls (), gains in cost efficiency can be observed as disease allele frequency decreases. For example, when we see a cost advantage to unrestricted designs of 10--15% relative to those of CaTS. Given the cost advantages obtained by removing the equality constraint on the proportion of controls and cases ( and respectively) allocated to stage 1, it would be expected that optimal design parameters computed using the unrestricted procedure and CaTS will differ. For the experimental conditions plotted in [Fig. 3](#pone-0042367-g003){ref-type="fig"}, it can be observed that designs computed with the unrestricted algorithm assign lower and higher proportions of controls and cases (respectively) to stage 1 than those computed by CaTS, and often reduce the proportion of markers passed to stage 2 (). Additionally, it is clear that the degree of difference in design specification between the two methods can be influenced by each of the ratio of controls to cases, population disease allele frequency and stage 2 genotyping cost. The differences in design parameters shown in [Fig. 3](#pone-0042367-g003){ref-type="fig"} are again consistent with the results presented in [Tables S2](#pone.0042367.s004){ref-type="supplementary-material"} and [S3](#pone.0042367.s005){ref-type="supplementary-material"}. We note that the powers of the unrestricted 2S-GWAS designs and critical values are verified in our simulation studies, with no systematic deviation from the 78% target level ([Tables S2](#pone.0042367.s004){ref-type="supplementary-material"} and [S3](#pone.0042367.s005){ref-type="supplementary-material"}). In terms of differences between stage specific powers of the design, both unrestricted 2S-GWAS designs and those proposed by CaTS generally have higher stage 1 specific power than stage 2 power. The influence of batch effects on power were similar for experiments designed with and without the sample allocation constraint; in both types of design, scaling stage 2 case and control disease allele frequencies to 90 or 110% of those in stage 1 reduces or increases the power of a proposed design by 5--10% respectively. As the cost of stage 2 genotyping increases (holding the ratio of controls to cases constant) the sensitivity of a proposed design to batch effects is reduced. Although it is intuitive that removing an equality constraint for the proportion of cases and controls allocated to stage 1 should improve the performance of an optimal design, it is useful to illustrate why this occurs. In [Fig. 4](#pone-0042367-g004){ref-type="fig"}, we plot power (red) and cost (blue) curves as functions of the proportions of controls and cases allocated to stage 1 for the experimental conditions , and , holding the expected proportion of markers to be passed to stage 2 at the cost minimizing values of 8.5% and 0.7% respectively ([Tables S2](#pone.0042367.s004){ref-type="supplementary-material"} and [S3](#pone.0042367.s005){ref-type="supplementary-material"}). In these plots, the green identity line represents designs where the constraint holds. For the conditions with equal number of controls and cases () the power surface is symmetric about and has cost constrained maxima on the identity line. That is, in this case, the optimal design should have equal proportions of cases and controls allocated to stage 1. When the number of controls increases (), both the power surface and cost curves become asymmetric with the optimal design having . {#pone-0042367-g004} Power maximizing designs with unequal proportions of cases and controls allocated to stage 1 can compensate for cost reductions {#s3b} ------------------------------------------------------------------------------------------------------------------------------- Power maximizing experimental designs and their performance characteristics are provided in [Table S4](#pone.0042367.s006){ref-type="supplementary-material"}. In [Fig. 5](#pone-0042367-g005){ref-type="fig"}, we describe the results calculated for a disease with 10% prevalence (), population disease allele frequency () of 10%, and stage 2 genotyping cost () of 10, with control/case ratios () of 1 and 8 (blue and green lines respectively), as a function of the degree of relative cost restriction (50--100% of that of the cost minimizing 78% power design). As the experimental cost constraint is decreased from 100% of the minimum cost 78% power experimental design to 50% of that design, the maximum power obtainable by a cost constrained experimental design decreases from 78% to 44% and 68% for respectively. In comparison to the baseline minimum cost 78% power design, power maximizing cost limited designs typically pass a lower proportion of markers to stage 2, and then increase the relative power of the stage 2 test by allocating a greater proportion of both cases and controls to it. {#pone-0042367-g005} The effects of diminishing experimental cost on study design and power for the experimental conditions plotted in [Fig. 5](#pone-0042367-g005){ref-type="fig"} are consistent with those for the other conditions reported in [Table S4](#pone.0042367.s006){ref-type="supplementary-material"}. Additionally, [Table S4](#pone.0042367.s006){ref-type="supplementary-material"} reports the results from validation studies of experimental power, and analyses of the sensitivity of the performance of power maximizing designs to batch effects. The levels of power calculated for the power maximizing two stage GWAS designs were verified by simulation. We observed a reduced sensitivity of the performance of power maximizing designs to batch effects as increases (holding constant). This is evidenced by tightening ranges of power estimates over 90 and 110% scaling of second stage disease allele frequencies. Increases in (holding constant) did not have any such effect. Discussion {#s4} ========== In many analyses of two stage genome-wide association studies (2S-GWAS) experimental design, the proportions of cases and controls allocated to stages are implicitly constrained to be the same. In this paper we have expanded the framework for 2S-GWAS originally proposed by Skol et al [@pone.0042367-Skol1], [@pone.0042367-Skol2] to remove this restriction. Using the expanded framework, we then demonstrated that in fact, cost-minimizing designs computed with respect to a desired level of statistical power often allocate different proportions of cases and controls to each stage. Relative to those computed under an equality constraint for proportions of cases and controls allocated to stage 1, unrestricted designs typically allocate fewer controls and more cases to stage 1, often pass fewer markers from stage 1 to stage 2, and have higher stage 1 and lower stage 2 power than those under the equality constraint. As would be expected, such designs offer substantial cost advantages relative to a 2S-GWAS that imposes equal allocation proportions. Such performance improvements become larger as the ratio of controls to cases increase, as the stage 2 per genotype cost increases, and as the population disease allele frequency decreases. Based on this result, we extended our analysis to the problem of computing maximum power 2S-GWAS designs, subject to a cost constraint. We demonstrated that when a study budget is reduced below that of a minimum cost 2S-GWAS design meeting a targeted level of power, recomputing a maximum power 2S-GWAS design subject to the new cost constraint can retain much of the desired power. Relative to the original minimum cost design, power retention is achieved by allocating fewer cases and controls to stage 1, and passing fewer markers from stage 1 to stage 2. That is, the reduction in cost is compensated for by collecting less information in stage 1, focusing on fewer markers in stage 2, and then using a more powerful stage 2-specific test. We note that the results achieved here are in some respects obvious - removal of a constraint from an optimization problem always weakly results in improvements in performance. However, the extent to which this is true for 2S-GWAS has not been made explicit in previous studies. Additionally, many questions pertaining to such improvements, such as why the optimal designs changed as experimental parameters changed, could only be understood by investigating the problem geometry. Related to such investigations, our studies of the sensitivity of 2S-GWAS designs to batch effects or genetic heterogeneity between stages demonstrated that our unconstrained 2S-GWAS designs are not substantially different (in that respect) from those that constrain case and control sample proportions to be equal. The gains in efficiency related to removing the sample proportion constraint do not come at the cost of higher sensitivity to batch effects. Because most 2S-GWAS designs constrain sample allocation proportions to be the same across cases and controls, we suggest the results presented here may have implications beyond our particular study. For example, in [@pone.0042367-Pahl1], it was demonstrated that for experiments using the same numbers of cases and controls, it is in principle possible to obtain greater cost efficiencies by using three or four stages rather than two. However, the use of differential case/control allocation proportions for problems in which the numbers of cases and controls differ was not considered. Likewise, in [@pone.0042367-Wang1], two stage GWAS designs using false discovery rate criteria were considered, again while imposing that equal numbers of cases and controls assigned to each stage. It is possible that using techniques analogous to those described here could yield greater levels of cost efficiency and power performance in such designs. We recommend that when designing a 2S-GWAS, investigators think carefully about the relative number of controls to cases, genotyping costs, and disease allele frequencies. To assist in doing do, our programs for identifying optimal two-stage GWAS designs are provided in Code S1 or alternatively at <http://www.bioinformatics.org/stanhope/2SGWASdesign/>. Supporting Information {#s5} ====================== ###### **Supporting derivations.** This appendix provides mathematical details omitted in the main text, including the proof of Theorem 1; the Fisher information stage weighting calculation; and the stage 2 critical value and power calculations. (PDF) ###### Click here for additional data file. ###### **Supporting software.** This file contains all codes necessary to calculate cost minimizing and power maximizing two-stage GWAS designs with unequal proportions of cases and controls allocated to stages. Instructions are provided for their compilation and use, as well as example calculations. (GZ) ###### Click here for additional data file. ###### **Experimental conditions for 2S-GWAS design calculations.** Experiments are described in terms of disease prevalences ; disease allele frequencies ; ratio of controls to cases ; stage 2 genotyping costs ( is held constant); and case/control allele frequencies . Numbers of cases and markers are constant at and . (PDF) ###### Click here for additional data file. ###### **Cost minimizing 2S-GWAS designs and their performance characteristics,** **.** For experimental conditions with in [Table S1](#pone.0042367.s003){ref-type="supplementary-material"}, [Table S2](#pone.0042367.s004){ref-type="supplementary-material"} reports two-stage 78% power designs computed from both CaTS and the unrestricted method. The costs of two-stage designs are compared to those of 80% power one-stage designs, and the costs of the unrestricted two-stage designs are compared to those of CaTS. Verification of the power levels of unrestricted 2S-GWAS designs is performed by Monte Carlo. (PDF) ###### Click here for additional data file. ###### **Cost minimizing 2S-GWAS designs and their performance characteristics,** **.** For experimental conditions with in [Table S1](#pone.0042367.s003){ref-type="supplementary-material"}, [Table S3](#pone.0042367.s005){ref-type="supplementary-material"} reports two-stage 78% power designs computed from both CaTS and the unrestricted method. The costs of two-stage designs are compared to those of 80% power one-stage designs, and the costs of unrestricted two-stage designs are compared to those of CaTS. Verification of the power levels of unrestricted 2S-GWAS designs is performed by Monte Carlo. (PDF) ###### Click here for additional data file. ###### **Power maximizing two stage GWAS designs and their performance characteristics,** **.** For all experimental conditions with in [Table S1](#pone.0042367.s003){ref-type="supplementary-material"}, [Table S4](#pone.0042367.s006){ref-type="supplementary-material"} reports two-stage maximum power designs and the powers they attain, with respect to a cost constraint expressed as a percentage of the cost of the minimum cost designs reported in [Table S2](#pone.0042367.s004){ref-type="supplementary-material"}. (PDF) ###### Click here for additional data file. We would like to thank Mark Abney for his support. [^1]: **Competing Interests:**The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: SAS ADS. Performed the experiments: SAS. Analyzed the data: SAS ADS. Contributed reagents/materials/analysis tools: SAS ADS. Wrote the paper: SAS ADS. Development of statistical software: SAS.

Introduction ============ Breast-conserving surgery is a standard treatment for stage I and II breast cancer; however, 5--10% of patients treated with breast-conserving surgery are diagnosed with ipsilateral breast tumor recurrence (IBTR) within 10 years ([@b1-ol-0-0-5797],[@b2-ol-0-0-5797]). IBTR following breast-conserving surgery is associated with an elevated risk of mortality or of developing distant recurrence ([@b3-ol-0-0-5797]--[@b7-ol-0-0-5797]). The time interval between the initial surgery and the occurrence of IBTR is defined as the disease-free interval (DFI), which is a predictor of disease recurrence following IBTR ([@b3-ol-0-0-5797]--[@b6-ol-0-0-5797],[@b8-ol-0-0-5797]--[@b12-ol-0-0-5797]), and patients with early IBTR have a poorer prognosis, compared with those with late IBTR ([@b8-ol-0-0-5797],[@b10-ol-0-0-5797]--[@b12-ol-0-0-5797]). However, irrespective of the DFI, the standard treatment for patients with IBTR is surgery is mastectomy. This treatment strategy must be modified if a subgroup of patients with early IBTR, with an equally poor prognosis as that of patients with regional or distant recurrence, is present ([@b13-ol-0-0-5797]). Therefore, it is important to estimate the risk of disease recurrence in such patients, as risk factors following early IBTR have not yet been elucidated. In the present study, the risk factors for distant recurrence following early IBTR were examined. Patients and methods ==================== ### Patients The medical records of 3,793 patients with breast cancer who underwent breast-conserving surgery between January 1989 and December 2013 at the Osaka Medical Center for Cancer and Cardiovascular Diseases (Osaka, Japan) were reviewed. Of these patients (ages 28--89), 180 (4.7%) developed IBTR as the first event with no evidence of synchronous metastatic disease, and subsequently underwent salvage surgery. Within this group, the exclusion criteria were as follows: Patients with non-invasive tumors present in IBTR tissue specimens and patients who received neoadjuvant therapy as the initial treatment. A total of 153 patients with IBTR were eligible for the present study. A previous study examined the same patient group, focusing on patients with IBTR that occurred 5 years following the initial surgery ([@b14-ol-0-0-5797]), whereas, in the current study, 40 patients with IBTR that occurred within 3 years of the initial surgery were analyzed. The present study was approved by the local ethics committee of the Osaka Medical Center of Cancer and Cardiovascular Diseases, with waiver of informed patient consent. Patients received a physical examination (palpation for breast, chest wall and regional lymph nodes) every 3--6 months for 5 years following primary or salvage surgery and annually thereafter, and also underwent mammograms annually following primary or salvage surgery. The estrogen receptor (ER) status of the surgical specimens obtained from patients was determined using immunohistochemistry ([@b15-ol-0-0-5797]), and tumors were classified as positive for ER expression if ≥10% of cells exhibited positive nuclear staining with monoclonal rabbit anti-human ERα (clone EP1, Dako; Agilent Technologies, Inc., Santa Clara, CA, USA). The human epidermal growth factor receptor 2 (HER2) status of patients\' tissues was considered positive if the immunohistochemistry was 3+ or if the fluorescence *in situ* hybridization ratio (HER-2/chromosome 17) was \>2.0 ([@b16-ol-0-0-5797]). ### Statistical analysis Distant disease-free survival (DDFS) rate was defined as the period of time between the date of surgery for patients with IBTR and the date of the appearance of distant recurrence, and was calculated using the Kaplan-Meier method. Log-rank tests were performed to evaluate the differences in DDFS among various patient subgroups. Univariate and multivariate analyses were performed using the Cox proportional hazards model. All statistical tests were performed using SPSS version 21.0 (IBM SPSS, Armonk, NY, USA). All statistical tests and P-values were two tailed, and P\<0.05 was considered to indicate a statistically significant difference. Results ======= ### Patient characteristics Patients\' clinical characteristics are presented in [Table I](#tI-ol-0-0-5797){ref-type="table"}. Some data was missing (such as HER2 status of primary tumor and IBTR). Within a median follow-up period of 2.2 years (range, 0.1--20.8 years) following salvage surgery for IBTR, distant recurrence occurred in 15/40 patients (37.5%), and the 3-year DDFS rate was 64.3%. ### Association with DDFS Various clinical and pathological factors associated with DDFS among patients with early IBTR are listed in [Table II](#tII-ol-0-0-5797){ref-type="table"}. The nodal status at primary surgery and the use of adjuvant chemotherapy treatment following primary surgery were significantly correlated with DDFS (P=0.001 and P=0.002, respectively). Patients who were node-positive at primary surgery had a significantly poorer DDFS than node-negative patients (3-year DDFS, 33.5 vs. 93.3%, respectively; P=0.001; [Fig. 1](#f1-ol-0-0-5797){ref-type="fig"}). Patients who received adjuvant chemotherapy (n=13; mainly anthracycline and/or taxane) following primary surgery exhibited a significantly poorer DDFS than those who did not receive chemotherapy (3-year DDFS, 34.4 vs. 77.9%, respectively; P=0.002; [Table II](#tII-ol-0-0-5797){ref-type="table"}). No significant differences were observed between any of the following groups: Negative or positive margin at primary surgery (P=0.58), radiotherapy or no radiotherapy following primary surgery (P=0.57) and basal (both ER- and HER2-negative) or non-basal type primary tumors (P=0.27) ([Table II](#tII-ol-0-0-5797){ref-type="table"}). Multivariate analyses demonstrated that the nodal status at primary surgery was an independent predictive factor of distant recurrence (P=0.050; [Table III](#tIII-ol-0-0-5797){ref-type="table"}). Discussion ========== The present study demonstrated that the nodal status at the time of primary surgery and the use of adjuvant therapy subsequent to primary surgery were risk factors for distant recurrence following early IBTR. It was hypothesized that the nodal status at primary surgery may interact with adjuvant therapy following primary surgery. Node-positive breast cancer patients have poorer prognosis compared with patients with negative lymph node metastasis. Therefore, patients with positive lymph node metastasis are more likely to be recommended for adjuvant chemotherapy compared with those with negative lymph node metastasis. Therefore, multivariate analysis incorporating these two factors was performed, which revealed that the nodal status at primary surgery was an independent prognostic factor in the present study group. At present, the risk factors that follow IBTR and are associated with the DFI require further investigation ([@b8-ol-0-0-5797]--[@b10-ol-0-0-5797],[@b14-ol-0-0-5797]) and, to the best of our knowledge, no previous studies have been conducted to examine the risk factors following early IBTR. The nodal status at primary surgery and the use of adjuvant therapy following primary surgery, which were demonstrated to be prognostic factors among patients with early IBTR in the current study, were also associated with primary surgery, but not with recurrent tumors. By contrast, a previous study identified that the prognostic factors among patients with late IBTR were the ER and HER2 status of IBTR tissue specimens, which were associated with recurrent tumors, but not with primary surgery ([@b14-ol-0-0-5797]). Taken together, these findings suggest that early IBTR is associated with true recurrence, whereas late IBTR is associated with the presence of new primary tumors. The 3-year DDFS rate in the present study was 33.5% among patients with early IBTR and a positive nodal status at the time of primary surgery. This DDFS rate is concordant with that reported by Wapnir *et al* ([@b5-ol-0-0-5797]), in which the 3-year DDFS was 44.9% among patients with early IBTR and a positive nodal status at the time of primary surgery. Furthermore, this DDFS rate is similar to that observed in patients with ipsilateral supraclavicular node recurrence ([@b17-ol-0-0-5797]) or lung metastases ([@b18-ol-0-0-5797]). Pergolizzi *et al* ([@b17-ol-0-0-5797]) reported that the median time to progression was 28 months in 44 patients with ipsilateral supraclavicular node recurrence from breast cancer (as a part of recurrent regional disease and without distant metastases) who received combined chemotherapy and radiotherapy treatment. Ludwig *et al* ([@b18-ol-0-0-5797]) observed that, during a retrospective analysis, the median DDFS following resection of lung metastatic tumors was 27.6 months. The results of the current study suggest that patients with early IBTR and positive axillary nodes at the diagnosis of the primary tumor possess a high risk of distant recurrence and, therefore, should potentially receive more aggressive treatment compared with conventional treatment, including novel (neo)adjuvant systemic therapy or regional radiotherapy. In addition to the DFI, previous studies have demonstrated that the nodal status at the time of primary surgery was a prognostic factor among patients with IBTR ([@b4-ol-0-0-5797],[@b19-ol-0-0-5797]). The association between the DFI and the nodal status of the primary tumor, and its prognostic relevance among patients with IBTR, has yet to be elucidated. In addition, the small sample size, short follow-up period and high frequency of missing data, particularly for the HER2 status of patients \[primary tumor, 30.0% (12/40); IBTR, 22.5% (9/40)\] were limitations of the present study. For ER-positive tumors, the annual breast cancer mortality rates are similar during years 0--4 and 5--14 ([@b20-ol-0-0-5797]). In conclusion, the nodal status at primary surgery was demonstrated to be an independent predictive factor of distant recurrence among patients with early IBTR in the current study; however, further studies are required to support this association. The present study was supported in part by the Osaka Foundation for the Prevention of Cancer and Cardiovascular Diseases (grant no. 1601079183). {#f1-ol-0-0-5797} ###### Characteristics of patients. Characteristics of patients No. of patients(n=40) -------------------------------------------------------------------------------------------------- ----------------------- Median age at initial diagnosis (range), years 54 (30--81) p-T stage of primary tumor   *In situ* 3   1 7   2 30 Grade of primary tumor   1 0   2 18   3 19   Unknown 3 Lymphovascular invasion of primary tumor   Negative 19   Positive 20   Unknown 1 Histological type of primary tumor   DCIS 3   Invasive ductal 35   Invasive lobular 1   Other 1 No. of positive lymph nodes of primary tumor   0 18   1--3 12   ≥4 4   Unknown 6 ER status of primary tumor   Positive 17   Negative 22   Unknown 1 HER2 status of primary tumor   Positive 10   Negative 18   Unknown 12 Adjuvant chemotherapy following primary   surgery   Yes 13   No 27 Adjuvant hormonal therapy following primary surgery^[a](#tfn1-ol-0-0-5797){ref-type="table-fn"}^   Yes 11   No 6 Adjuvant trastuzumab following primary surgery^[b](#tfn2-ol-0-0-5797){ref-type="table-fn"}^   Yes 0   No 10 Median time interval between initial surgery and IBTR (range), years 1.9(0.1--2.1) Median age at IBTR diagnosis (range), years 56.5(32.0--82.0) p-T stage of IBTR   *In situ*   0   1 26   ≥2 13   Unknown   1 Grade of IBTR   1   3   2 10   3 21   Unknown   6 Lymphovascular invasion of IBTR   Negative 19   Positive 17   Unknown   4 Histological type of IBTR   DCIS   0   Invasive ductal 37   Invasive lobular   1   Other   1   Unknown   1 ER status of IBTR   Positive 17   Negative 20   Unknown   3 HER2 status of IBTR   Positive   9   Negative 22   Unknown   9 Adjuvant chemotherapy following salvage surgery   Yes 15   No 22   Unknown   3 Adjuvant hormonal therapy following salvage surgery^[a](#tfn1-ol-0-0-5797){ref-type="table-fn"}^   Yes   9   No   5   Unknown   3 Adjuvant trastuzumab following salvage surgery^[b](#tfn2-ol-0-0-5797){ref-type="table-fn"}^   Yes   4   No   5 Including only patients with ER-positive tumors. Including only patients with HER2-positive tumors. DCIS, ductal carcinoma *in situ*; ER, estrogen receptor; HER2, human epidermal growth factor receptor 2; IBTR, ipsilateral breast tumor recurrence; p-T, pathological tumor. ###### Three-year DDFS rates according to various clinicopathological factors among patients with early IBTR (n=40). Characteristics of patients 3-year DDFS rates, % P-value -------------------------------------------------------------------------------------------------- ---------------------- --------- Age at initial diagnosis, years   \<50 48.9 0.870   ≥50 70.3 p-T stage of primary tumor   *In situ* or 1 80.2 0.110   2 50.3 Margin of primary tumor   Negative 66.0 0.58   Positive 53.3 Grade of primary tumor   1 or 2 66.5 0.770   3 58.0 Lymphovascular invasion of primary tumor   Negative 74.9 0.190   Positive 51.9 Lymph node status of primary tumor   Negative 93.3 0.001   Positive 33.5 ER status of primary tumor   Positive 72.2 0.400   Negative 55.9 HER2 status of primary tumor   Positive 71.1 0.220   Negative 50.2 Basal type of primary tumor   Yes 43.8 0.27   No 66.2 Radiotherapy following primary surgery   Yes 66.5 0.57   No 61.4 Adjuvant chemotherapy following primary surgery   Yes 34.4 0.002   No 77.9 Adjuvant hormonal therapy following primary surgery^[a](#tfn1-ol-0-0-5797){ref-type="table-fn"}^   Yes 71.6 0.460   No 75.0 Age at IBTR diagnosis, years   \<50 48.9 0.870   ≥50 70.3 p-T stage of IBTR   1 67.3 0.450   ≥2 54.9 Grade of IBTR   1 or 2 75.0 0.490   3 55.1 Lymphovascular invasion of IBTR   Negative 69.1 0.170   Positive 52.1 ER status of IBTR   Positive 56.4 0.540   Negative 64.7 HER2 status of IBTR   Positive 77.8 0.270   Negative 58.0 Adjuvant chemotherapy following salvage surgery   Yes 55.8 0.210   No 69.2 Adjuvant hormonal therapy following salvage surgery^[a](#tfn3-ol-0-0-5797){ref-type="table-fn"}^   Yes 64.8 0.071   No 26.7 Adjuvant trastuzumab following salvage surgery^[b](#tfn4-ol-0-0-5797){ref-type="table-fn"}^   Yes 75.0 0.800   No 80.0 Including only patients with ER-positive tumors. Including only patients with HER2-positive tumors. DDFS, distant disease.free survival; IBTR, ipsilateral breast tumor recurrence; ER, estrogen receptor; HER2, human epidermal growth factor receptor 2; p.T, pathological tumor. ###### Multivariate analysis of predictors of distant recurrence following early ipsilateral breast tumor recurrence. Characteristics of patients HR 95% CI P-value ------------------------------------------------------------------------- ------- --------------- ---------------------------------------------------- Lymph node status of primary tumor (positive vs. negative) 5.281 1.002--27.002 0.050^[a](#tfn5-ol-0-0-5797){ref-type="table-fn"}^ Adjuvant chemotherapy following primary surgery (positive vs. negative) 2.983 0.750--11.750 0.120 P\<0.05 indicates a statistically significant difference. HR, hazard ratio; CI, confidence interval.

1. Introduction {#sec1-polymers-11-01291} =============== Conventional oil and gas resources are steadily diminishing worldwide \[[@B1-polymers-11-01291],[@B2-polymers-11-01291]\]. Most of the remaining oil and gas resources are unconventional oil and gas reservoirs, such as coalbed methane, shale gas, and tight oil and gas reservoirs. \[[@B3-polymers-11-01291],[@B4-polymers-11-01291]\]. Therefore, the efficient and economic development of unconventional reservoirs has become an important research area \[[@B5-polymers-11-01291],[@B6-polymers-11-01291]\]. Hydraulic fracturing technology is one of the most effective techniques to increase the productivity of unconventional reservoirs \[[@B7-polymers-11-01291],[@B8-polymers-11-01291],[@B9-polymers-11-01291]\]. In this technique, a particular viscoelastic fluid injected into the formation is referred as a fracturing fluid. The type and performance of fracturing fluids is the key to successful application of fracturing processes. Polymer fracturing fluid (HPAM solution) and viscoelastic surfactant (VES) fluids are the two most representative fracturing fluid systems in low permeability reservoirs \[[@B10-polymers-11-01291],[@B11-polymers-11-01291],[@B12-polymers-11-01291],[@B13-polymers-11-01291]\]. Filtration loss of fracturing fluid is one of the most important indexes for evaluating the performance of fracturing fluids. \[[@B14-polymers-11-01291]\]. Compared with polymer fluids, the VES fluids have a much higher filtration loss to reservoir matrix. Excessive filtration loss increases development costs and reduces economic benefits \[[@B15-polymers-11-01291],[@B16-polymers-11-01291]\], which is a serious problem for further application. It was found that the flow characteristics of the fracturing fluid in the porous media affect the amount of filtration loss. However, research in this field is still insufficient. Therefore, further experimental studies are needed to clarify the flow characteristics of fracturing fluids in pore throat structures \[[@B17-polymers-11-01291],[@B18-polymers-11-01291]\]. VES fluids are composed of low molecular weight surfactants. The surfactants are weakly bound by non-covalent bonds that repeatedly undergo scission and recombination in dynamic equilibrium. As the surfactant concentration increasing, the spherical micelles will transform into giant wormlike micelles and fluid viscosity will increase \[[@B19-polymers-11-01291],[@B20-polymers-11-01291]\]. What's more, the viscoelastic properties can recover from micellar degradation in strong shearing and stretching deformation \[[@B21-polymers-11-01291],[@B22-polymers-11-01291],[@B23-polymers-11-01291],[@B24-polymers-11-01291],[@B25-polymers-11-01291],[@B26-polymers-11-01291],[@B27-polymers-11-01291]\]. In recent years, microfluids have become a common method for dealing with fluids at micro-length scale, especially for generating and operating complex fluids with controllable size and customized structure \[[@B28-polymers-11-01291],[@B29-polymers-11-01291]\]. Furthermore, it has been developed as microfluidic rheometers or tools to study new flow characteristics, providing a powerful and efficient way to simulate the flow patterns of the VES fluids in geometric structures similar to porous media \[[@B30-polymers-11-01291]\]. However, previous study of the microfluids are focusing on single abrupt contraction--expansion geometry, while the research on the flow characteristics in the continuous pore-throat structure are insufficient and unclear \[[@B31-polymers-11-01291]\]. In this work, we present the flow patterns of the VES fluids and the HPAM solutions in two continuous pore-throat structures with different length to observe the deformations of the flow patterns. The evolution process of all solutions with increasing injection rates is captured and recorded by microparticle image velocimetry (μ-PIV) in the first place. The three-dimensional velocity fields and velocity magnitude maps are illustrated in detail to further reveal the feedback effect of downstream structure on upstream flowing. 2. Materials and Experimental Methods {#sec2-polymers-11-01291} ===================================== 2.1. Microchannel Design and Fabrication {#sec2dot1-polymers-11-01291} ---------------------------------------- The microfluidic devices were fabricated on a plate (100 × 80 × 10 mm) of polymethyl methacrylate (PMMA) by precisely milling and sealed with another PMMA plate (100 × 80 × 3 mm). The layout of the device is displayed in [Figure 1](#polymers-11-01291-f001){ref-type="fig"}. All the microfluidic devices consisted of an injection point, an outlet point, a liner microchannel and a continuous pore-throat structure with two pressure taps located 10 mm away at upstream and downstream. The continuous pore-throat structure had a contraction--expansion structure at upstream and an expansion--contraction structure at downstream ([Figure 1](#polymers-11-01291-f001){ref-type="fig"}b). The solution at the inlet was connected to an independent syringe (10 mL, Hamilton, Germany) driven by a syringe pump (Harvard Apparatus, PHD2000, Plymouth Meeting, PA USA) to obtain accurate flow rates ranging from 0.5 mL/h to 12 mL/h. The microfluidic devices with three different continuous pore-throat structures were fabricated as mentioned to investigate the effect of flow length. Pore width w~p~ = 800 μm, throat width of w~t~ = 100 μm as well as the pore length *l* are illustrated in [Figure 1](#polymers-11-01291-f001){ref-type="fig"}a. 2.2. Fluid Rheological Characterizations and Dimensionless Number {#sec2dot2-polymers-11-01291} ----------------------------------------------------------------- Two kinds of viscoelasticity solutions were prepared in this study, and a Newtonian fluid was used as the contrast. The viscoelasticity fluids were 0.2% hydrolyzed polyacrylamide (HPAM) and water solutions with HPAM molecular weights of 9.6 × 10^6^ g/mol. The VES fluid composed of 25 mMol/L cetyltrimethylammonium bromide (CTAB) and 12 mMol/L sodium salicylate (NaSal). In contrast, the Newtonian fluid was 93% glycerol-water solution to match the approximate viscosity of 95 mPa s. All the solutions were mixed for at least 24 h under temperature 25 °C. HAAKE MARS 60 Rheometer were equipped with a cylinder rotor at 25 °C to test the rheological characteristics of all solutions. Before testing, samples were kept stable for 5 min to reach equilibrium. The oscillatory measurements were proceeded with shear rates rising from 0.01 s^−1^ to 1000 s^−1^. Due to the rheological characteristics of shear thinning, the viscosities (*η*) of both solutions were similar while shearing rates γ \> 5 s^−1^ (shown in [Figure 2](#polymers-11-01291-f002){ref-type="fig"}). The relaxation times were 1.41 s for the VES fluid and 0.51 s for the HPAM solution. In contrast, the Newtonian fluid was 83% glycerol--water solution with a viscosity of 75 mPa s. The dimensionless numbers were used in this investigation to quantitate the deformation of the macroscopic flow field observed in different pore-throat structures. The Reynolds number was defined by the average flow velocity $\overset{-}{V_{p}}$ in pore channel, in which $D_{h} = 2w_{p}h/(w_{p} + h)$:$$Re = \frac{\rho\overset{-}{V_{p}}D_{h}}{\eta_{p}} = \frac{2\rho Q}{(w_{p} + h)\eta_{p}}$$ Deborah number (De) is the ratio of a solution's relaxation time to residence time in the flow as:$$De = \frac{\lambda_{1}}{w_{p}lh/Q} = \frac{\lambda_{1}Q}{w_{p}lh}$$ where *Q* is the injection rate, *h* is the depth of the microchannels and *l* is the length of the pore-throat structures. 2.3. Flow Visualization {#sec2dot3-polymers-11-01291} ----------------------- In order to obtain and analyze flow fields of contraction and expansion structures, micro-particle image velocimetry, μPIV (LaVision. Ltd) was used to perform quantitative measurements on the flow kinematics. For this purpose, mono-disperse fluorescent polystyrene microspheres, with 0.1 mL per 50 mL of bulk fluids, were blended into the fluids. A quite low magnification Zeiss Plan-Neofluar objective lens was used to capture the full width of the flow channel and a reasonable length of both upstream and downstream. Using a light sheet, formed by passing a double pulsed laser beam through an optical arrangement including cylindrical lenses, the fluorescent particles in the flow are illuminated twice within a small time separation between (dt). The images of particles displacement between the laser pulse were captured and recorded by the CCD (charge coupled device) camera. The whole velocity vectors were calculated by a conventional cross-correlation PIV algorithm. 3. Result and Discussion {#sec3-polymers-11-01291} ======================== 3.1. Flow Characteristics in Infinite Length Pore-Throat Structure {#sec3dot1-polymers-11-01291} ------------------------------------------------------------------ The flow characteristics of all solution were first examined in the microchannel with single pore-throat structure at upstream. Here, the downstream pore structure was treated as an infinite straight channel, which eliminates the impact of downstream structure on flow field in the pore. As displayed in [Figure 3](#polymers-11-01291-f003){ref-type="fig"}, the flow characteristics of all three solutions exhibited similar flow patterns with different Reynolds numbers (Re). After flowing through the entrance at upstream, the streamlines consisted of velocity components on both x-axis and y-axis directions. The velocities from the centerlines and the boundaries were different in the beginning. Afterwards, the velocity at centerlines gradually reduced until arriving at a relatively stable value. Finally, the streamlines tended to follow the stable linear flow expected behaviors, that is to say the velocity component was along the x-direction and barely changed. The difference of the divergent degree was visible among three solutions. The HPAM solution exhibited the most convergent patterns, while the VES fluid was observed as the most divergent patterns. 3.2. Flow Characteristic in Long Pore-Throat Structure {#sec3dot2-polymers-11-01291} ------------------------------------------------------ The velocity fields of the Newtonian fluid in long continuous pore-throat structures were illustrated in [Figure 4](#polymers-11-01291-f004){ref-type="fig"}a. When Newtonian fluid flowed through the continuous pore-throat structure, the streamline underwent three stages including the divergent flow, the stable flow and the contractive flow. The maximum velocity was observed at the entrance due to the gathering of streamlines. After that, the streamlines performed to be divergent from the centerlines to boundaries. This flow stage was signed as the divergent flow. Then the streamlines tended to follow the stable linear flow expected behaviors, in other words, entered the stable flow stage. Afterwards, the streamlines got gathering from surrounding to centerlines causing the increase of velocity near the exit, which was classified as the contractive flow. It was found that all the velocity fields of the Newtonian fluid in continuous pore-throat at different flow rates (Re number increasing from 2.59 × 10^−2^ to 0.621) presented the same flow pattern. Namely, the rising of flow rate has little effect on the transformation of the flow patterns. In addition, the upstream and downstream velocity fields of microchannels were basically symmetrical. When it came to the HPAM solution, at low Re number (Re = 1.79 × 10^−3^), the streamlines were still divergent to the boundaries at upstream. The streamlines tended to gather to the centerlines at downstream. However, unlike the flow pattern of Newtonian fluid, the gathering of streamlines was observed farther away from the exit of microchannel, about 250 μm before expansion--contraction structure. As the Re number rising to 4.29 × 10^−2^ (shown in [Figure 4](#polymers-11-01291-f004){ref-type="fig"}b), the flow of HPAM solution only experienced one stage that the flow patterns kept stable from the entrance to the exit, and there was no dispersion of streamlines from the centerlines to the boundaries. Additionally, the arrows of velocity in flow fields showed that there were only velocity components along the flow direction on the x-axis direction. The velocities at boundaries were quite small compared with the centerlines. As shown in [Figure 4](#polymers-11-01291-f004){ref-type="fig"}c, the velocity fields of the VES fluid were measured with flow rates spanning 0.5 \< *Q* \< 12 mL/h. The flow patterns of the VES fluid was similar with the HPAM solution at low Re number rates (1.52 × 10^−3^), but transformed into new patterns as Re number larger than 3.04 × 10^−3^. The velocity fields of VES fluid at high flow rates were illustrated as triangle shape, which was distinct from the HPAM solution. When flow rates were higher than 2 mL/h, there was almost no velocity component on *y*-axis direction along the streamline of the HPAM solution. However, the streamlines of VES fluid still maintained the characteristics of Newtonian-like flow at upstream of the long continuous pore-throat structures. The streamlines of VES fluid were observed to have the contractive behavior far away from the expansion--contraction structure. The streamlines were accelerated to flow from the boundary to the exit at about 300 μm from the entrance, leading to the contractive behavior of flow patterns. In addition, the velocities of streamlines outside the triangle shape flow fields were close to zero. In those areas, flow rates were very minimal compared with the main flow areas. 3.3. Flow Characteristic in Short Pore-Throat Structure {#sec3dot3-polymers-11-01291} ------------------------------------------------------- The results of the Newtonian fluid, HPAM solution and VES fluids in short continuous pore-throat structure are shown in [Figure 5](#polymers-11-01291-f005){ref-type="fig"}. The velocity fields indicated that the flow patterns of Newtonian fluid were barely affected by the length of the pore-throat structure and flow rates. Similar with the flow characteristics from [Figure 4](#polymers-11-01291-f004){ref-type="fig"}a, the process of Newtonian flow was separated into the divergent flow, the stable flow and the contractive flow. Even though the length of the pore-throat structure were shortened to 400 μm. As shown in [Figure 5](#polymers-11-01291-f005){ref-type="fig"}b, in the case of the HPAM solution, at low flow rates (Re = 1.79 × 10^−3^) there was an existence of the dispersion of streamlines observed in the microchannel at upstream. Compared with the μ-PIV velocity field in [Figure 4](#polymers-11-01291-f004){ref-type="fig"}b at Re = 1.79 × 10^−3^, the streamlines in short microchannel performed a more distinct contractive behavior. This phenomenon was generated by the influence of the length between the structure entrance and exit. According to the former study on the Newtonian fluid and HPAM solution in long continuous pore-throat structures, the streamlines would experience contractive flow stage before going through the exit at downstream. The elastic of HPAM solution was capable to result in the stronger gathering of streamlines. The distinct contractive behavior in [Figure 5](#polymers-11-01291-f005){ref-type="fig"}b was caused by the length of the microchannel. Before the streamlines were able to show the Newtonian-like divergent characteristic, streamlines have already been influenced by the contractive behavior at downstream. However, the flow patterns were similar with [Figure 4](#polymers-11-01291-f004){ref-type="fig"}b at higher flow rates (Re = 1.43 × 10^−2^ and 4.29 × 10^−2^). The more distinct contractive behavior of streamlines was also observed as VES fluid flowing through short microchannel shown in [Figure 5](#polymers-11-01291-f005){ref-type="fig"}c. Flow patterns of VES fluid were more similar with Newtonian fluid compared with HPAM solution, particularly at Q = 0.5 mL/h. The streamlines were observed with divergent behavior at all flow rates increasing from 0.5 mL/h to 12 mL/h, while the flow patterns of HPAM solution were contraction flow at Q = 1 mL/h and 12 mL/h. However, there were obvious differences between the flow pattern of VES fluid in long and short pore-throat structure, with the streamlines performed as triangle shape velocity fields as flowing through the long microchannel. While the flow length was shortened to 400 μm, the arrows of velocity in short microchannel were mostly along the x-direction. This is different from the velocity components shown in [Figure 4](#polymers-11-01291-f004){ref-type="fig"}b. Because of lacking enough velocity components along the vertical flow direction, the divergent flows of VES fluid were not obvious compared with the triangle shape flow patterns visualized in the long microchannel. As non-Newtonian fluids flowing through the continuous pore-throat structures, the flow patterns were influenced by the structure of upstream entrance, downstream exit and the flow length. There were divergent flows at upstream and contractive flows at down, due to the sudden change of width in pore-throat structures. However, the elasticity of both HPAM solution and VES fluid affected the streamlines distribution near the entrance and exit. With the influence of fluids elasticity, there was a process of streamlines to establish the flow patterns after flowing out or before entering the width changing structures of pore-throat. The length between the entrance and exit determined the process space to generate the divergent flow or contractive flow. In the short microchannel, the flow patterns at upstream were influenced by the downstream structure to come into being the insufficiency divergent flow. 3.4. Velocity Distribution in Continuous Pore-Throat Structure {#sec3dot4-polymers-11-01291} -------------------------------------------------------------- To quantify the flow pattern differences among those three solutions in both pore-throat structures, matrix tool was used to establish 3D velocity fields at *Q* = 12 mL/h. The heights of 3D-velocity fields represented the velocity at each point (located by x-position and y-position), which were marked with different colors as shown in [Figure 6](#polymers-11-01291-f006){ref-type="fig"} and [Figure 7](#polymers-11-01291-f007){ref-type="fig"}. The velocity contour in those 3D velocity fields primly explained the velocity distributions seized by μ-PIV equipment. The flow pattern of the Newtonian fluid was nearly symmetrical in [Figure 6](#polymers-11-01291-f006){ref-type="fig"}a with the ellipsoid shape, which was similar to the former study of Rodd \[[@B32-polymers-11-01291]\]. The streamlines of the HPAM solution has shown a strong contractive behavior at high flow rates as the velocity component of each point was almost always along *x*-axis. The 3D velocity field of the HPAM solution was observed as a banded shape where most of the streamlines were gathering (shown in [Figure 6](#polymers-11-01291-f006){ref-type="fig"}b). Thus, other area outside the banded shape owned little velocity. At the upper part in [Figure 6](#polymers-11-01291-f006){ref-type="fig"}c, the velocity distribution of the VES fluid was similar with the Newtonian fluid that the streamlines diverged gently to each area (from x = 400 μm to x = 200 μm). After that, two high velocity isoline parts were generated at both sides causing the gathering of streamlines, accelerating the contractive stage of the VES fluid flow patterns. Furthermore, the 3D velocity fields and velocity magnitude lines of all solutions flowing through short continuous pore-throat structure at *Q* = 12 mL/h are illustrated in [Figure 8](#polymers-11-01291-f008){ref-type="fig"}. All the data of the Newtonian fluid are used as the standard comparison. It can be seen that the length of pore channel has little effect on the flow characteristics of the Newtonian fluid and the HPAM solution. However, the VES fluid presented a different flow pattern. As mentioned in [Figure 4](#polymers-11-01291-f004){ref-type="fig"}c, the behavior of VES fluid was divergent flow and similar with the Newtonian fluid at upstream. But it was more contractive than the Newtonian fluid, which was in contrast with the result in the long continuous pore-throat structure ([Figure 6](#polymers-11-01291-f006){ref-type="fig"}c). Comparing [Figure 8](#polymers-11-01291-f008){ref-type="fig"}b,e, there was no appearance of the velocity along the y-direction and 'approximate inviscid flow area' near the contraction--expansion structure in the short pore-throat structure. At the middle area of the structure, flow patterns developed into double-peaks velocity line in the long structure ([Figure 8](#polymers-11-01291-f008){ref-type="fig"}c), while kept a single-peak shape as well in [Figure 8](#polymers-11-01291-f008){ref-type="fig"}f. After flowing near the exit at downstream, they reappeared as a similar shape ([Figure 8](#polymers-11-01291-f008){ref-type="fig"}d,g). In general, the structure of downstream was able to affect the development of flow pattern upstream. 4. Conclusions {#sec4-polymers-11-01291} ============== Viscoelastic fluids underwent shear and extensional flow when flowing in porous media as characterized by pore-throat structures. Because of the feedback effect, the pore length has significant influence on the distribution of velocity fields. In continuous pore-throat structures, the distance between the upstream and downstream structures determined the development degree of flow patterns, especially for VES fluid. For infinite length distance, the extensional deformations of both VES fluid and HPAM were negligible without the contraction at downstream. As the flow length shortened to 800 μm, there were double-peak velocity patterns obtained for VES fluid. However, this pattern vanished at a microchannel length of 400 μm. Through this work, we have clearer knowledge about the characteristics of the viscoelastic surfactant fracture fluid in porous media, which is beneficial for developing hydraulic fracturing technology. Formal analysis, X.Y. and Y.L.; Investigation, X.Y.; Resources, Y.L.; Validation, Y.Y.; Writing--original draft, X.Y.; Writing--review & editing, Y.W. The work was supported by the National Natural Science Foundation of China (U1663206, 51704313, 51425406, 21706284) and Fundamental Research Funds for the Central Universities (18CX02028A, 24720182146A). The authors no conflicts of interest to declare. {#polymers-11-01291-f001} {#polymers-11-01291-f002} ###### The development of flow patterns for the flow of (**a**) Newtonian fluid, (**b**) HPAM solution and (**c**) VES fluid in infinite length pore-throat structure, flow rates range from 0.5 mL/h to12 mL/h.   ###### The development of flow patterns for the flow of (**a**) Newtonian fluid, (**b**) HPAM solution and (**c**) VES fluid in long pore-throat structure, flow rates range from 0.5 mL/h to12 mL/h.   ###### The development of flow patterns for the flow of (**a**) Newtonian fluid, (**b**) HPAM solution and (**c**) VES fluid in short pore-throat structure, flow rates range from 0.5 mL/h to12 mL/h.   ###### The 3D velocity fields of the (**a**) Newtonian fluid; (**b**) HPAM solution and (**c**) VES fluid in long continuous pore-throat structure at *Q* = 12 mL/h.   ###### The 3D velocity fields of the (**a**) Newtonian fluid; (**b**) HPAM solution and (**c**) VES fluid in long continuous pore-throat structure at *Q* = 12 mL/h.   {#polymers-11-01291-f008}

INTRODUCTION ============ Medical librarians have played a crucial role in patient education for nearly a century. References to terms such as bibliotherapy, social hygiene, and consumer health are easily found in the literature to give a historical perspective \[[@b1-jmla-106-38]--[@b3-jmla-106-38]\]. However, the role of the medical librarian today has expanded to become embedded in both the academic and medical setting. In some settings, librarians are a part of the patient safety and quality improvement teams, and in provision of patient-and-family centered information to patients \[[@b4-jmla-106-38], [@b5-jmla-106-38]\]. In many settings, nurses depend on medical librarians to teach them the art of evaluating health information for patient care \[[@b6-jmla-106-38]\]. Librarians also impact patient care by teaching health care providers how to evaluate patient education material and by providing services to determine if the reading grade level of material is appropriate for the patient population. Printed patient education materials are important aspects in patients' recovery in that they reinforce verbal communication and vital care instructions that patients receive from health care providers \[[@b7-jmla-106-38]\]. Goal two of the US Office of Disease Prevention and Health Promotion's "National Action Plan to Improve Health Literacy" promotes changes in health care systems that "improve health information, communication, and informed decision-making" \[[@b8-jmla-106-38]\]. Distributing patient education materials that are comprehensible to all patients is essential to achieving this national goal. Research has indicated that patient education materials influence patient compliance. Mundt and colleagues studied the impact of patient education on patients' adherence to instructions and medication regimes. They found that the compliance rate for patients who completed the follow-up assessments were higher than for those who received normal care \[[@b9-jmla-106-38]\]. Another study concluded that patients who received educational materials on their aromatase inhibitor treatment for a specific form of breast cancer had a higher rate of compliance to physician recommendations for treatment than those who did not receive any educational material \[[@b10-jmla-106-38]\]. Lastly, a longitudinal study on asthma medication adherence, in which patients received an audiotape and educational booklet, determined that this intervention had a beneficial effect on asthma medication adherence over time \[[@b11-jmla-106-38]\]. Given the impact of patient education materials on patient compliance, the grade level at which the materials are written is crucial. The average American reads at the eighth- to ninth-grade level, while one of five reads at the fifth-grade level \[[@b12-jmla-106-38]\]. The National Institute of Health (NIH) and the American Medical Association (AMA) both recommend patient education materials be written at or below a sixth-grade reading level \[[@b13-jmla-106-38], [@b14-jmla-106-38]\], while the Joint Commission \[[@b15-jmla-106-38]\] recommends materials be written at a fifth-grade level or lower. Despite these recommendations, patient education materials continue to be written at grade levels beyond the average American's literacy skills \[[@b16-jmla-106-38]\]. Numerous readability formulas are available when assessing reading grade level of patient education materials. Simple Measure of Gobbledygook (SMOG) is an assessment tool that utilizes a hand-scored method. SMOG allows an evaluator to determine the grade level of patient education by counting 10 sentences at the beginning, the middle, and the end of a document. The evaluator counts every word of 3 or more syllables in those 30 sentences \[[@b17-jmla-106-38]\]. The number of syllables in each section is then totaled and converted to a corresponding reading grade level score \[[@b18-jmla-106-38]\]. Previous research suggests that SMOG is a useful tool when completing the reading grade level and predicts 100% comprehension \[[@b19-jmla-106-38]\]. The computerized Flesch-Kincaid (F-K) grade level method is another commonly used reading grade level assessment tool. F-K grade level is added to Microsoft Word through the Spelling and Grammar tool, in which readability statistics are displayed. Once the document is put through Microsoft's spell check feature, the readability statistics are automatically calculated and displayed. Patient education material providers often use this tool because of its ease of use and computerized platform. However, F-K grade level tends to predict lower reading grade level scores \[[@b20-jmla-106-38]\], and research has consistently found this to be true. Walsh and Volsko evaluated the reading grade level of Internet-based patient education on the top 5 causes of death in the United States and found that F-K grade level scored 2 to 3 grade levels lower than SMOG \[[@b21-jmla-106-38]\]. Additionally, Freda found that F-K grade level consistently scored patient education brochures 2 to 3 grade levels lower than SMOG \[[@b22-jmla-106-38]\]. Hand-scoring tends to be more reliable than computer scoring, since computerized scoring can be misleading based on the text provided in each document \[[@b23-jmla-106-38]\]. Computer programs use different methods to count sentences, words, and syllables, which can cause discrepancies in grade level \[[@b20-jmla-106-38]\]. By hand-scoring, evaluators are working directly with the text, which alerts evaluators to longer sentences and multisyllabic words as well as sentence structure and ease of reading \[[@b23-jmla-106-38]\]. Wang and colleagues found that the F-K grade level was the most commonly used readability formula, but SMOG was the most consistent in performance and the most practical in the health care setting \[[@b24-jmla-106-38]\]. Appropriate reading grade levels for patient education materials are of extreme importance in low literacy areas of the United States, including Appalachia. The Appalachian region has historically lagged behind average US literacy levels, with nearly 30% of adults living in the region considered "functionally illiterate" \[[@b25-jmla-106-38]\]. In addition, the educational attainment in the Appalachia region is lower than the national level, and fewer students obtain a bachelor's degree \[[@b26-jmla-106-38]\]. The study location for this project, the University of Tennessee Medical Center (UTMC), is located in the heart of Appalachia, Tennessee. In the 21-country region serviced by UTMC ([Figure 1](#f1-jmla-106-38){ref-type="fig"}), 10%--19% of the population lacks basic prose literacy skills \[[@b27-jmla-106-38]\]. Additionally, UTMC, like many US hospital systems, outsources its patient education materials to a third party provider, ExitCare, a patient education material provider owned by Elsevier. ExitCare markets their patient education materials as written "at a fifth-to eighth-grade reading level, with easy-to-read education written at a fourth-grade reading level or below" \[[@b28-jmla-106-38]\]. {#f1-jmla-106-38} The purpose of this study was to compare and contrast reading grade level scores of UTMC's patient education materials using hand-scored and computerized methods and to determine if these materials adhered to both the vendor's published reading grade levels and the nationally recommended sixth-grade reading level. Although studies have addressed F-K grade level versus SMOG in evaluating patient education, this study reviewed the importance of using a reliable readability tool that would accurately portray the reading grade level and highlighted the differences in using a computerized method versus a hand-scoring method when evaluating patient education. Additionally, this study focused on the Appalachian population and how librarians could play a role in advocating for quality patient education. METHODS ======= Both SMOG and F-K grade level assessment methods were used in this study. To assess the readability of patient education materials distributed at UTMC, the authors received a list from the Patient Education Committee of the 150 most distributed patient education documents from January 2016 through May 2016. These documents, which covered a broad base of health topics, were then downloaded from UTMC's internal system. Each document was assigned a unique identification number to be used for tracking and data entry. These materials included 137 documents created by the vendor and 13 custom documents created internally by UTMC staff. Twenty-two of the 137 vendor documents were labeled "easy-to-read." However, F-K grade level could not be obtained on 2 of the documents because they were images; therefore, the sample size for SMOG versus F-K grade level was n=148. Graduate students from the University of Tennessee--Knoxville, who were trained in SMOG assessment, reviewed the patient education materials. Each document was analyzed by three independent reviewers. Results from each reviewer were collected, reviewed, and tallied by the principle investigator (PI), a librarian. F-K grade level scores were determined by the PI, who entered each document into the existing F-K tool available in Microsoft Word. Data were checked for coding and data entry errors by the librarian to ensure data validity. Data from both the SMOG and F-K analyses were entered into SPSS version 21 (Armonk, NY: IBM) for further analysis. The SMOG score for each document was calculated by averaging the 3 independent reviewers' SMOG scores. The 3 independent SMOG ratings were analyzed for inter-rater reliability using 1-way random intra-class correlation coefficients (ICCs). These continuous variables from the SMOG analysis were tested for normality using skewness and kurtosis statistics. Any skewness or kurtosis statistic above an absolute value of 2.0 assumed a non-normal distribution. The distribution of differences between the SMOG and F-K grade level scores were also checked for normality and outliers. A paired *t*-test was used to compare the SMOG scores and F-K grade level scores for all documents. Statistical significance was assumed at an alpha value of 0.05. RESULTS ======= We found excellent inter-rater reliability between SMOG reviewers (ICC=0.95). The average SMOG scores were normally distributed. For the 148 assessed documents, SMOG produced a significantly higher mean grade reading level (mean \[M\]=9.6, SD=1.3) than F-K grade level (M=6.5, standard deviation \[SD\]=1.3) (*t*(147)=56.56, *p*\<0.001, 95% confidence interval \[CI\] of difference 2.96--3.17). SMOG test results showed that reading grade levels were higher than the national recommended reading grade level average for both custom and vendor documents. Using SMOG, 13 of 13 (100.0%) of the custom-designed patient education materials scored above the nationally recommended sixth-grade reading level. All of the "easy-to-read" documents (n=22) created by the vendor scored above the company's promoted maximum fourth-grade reading level. For standard vendor documents, 115 of 115 documents (100.0%) scored above the company's promoted maximum eighth-grade reading level, and 77% (n=115) scored above the nationally recommended sixth-grade reading level. When assessing the same materials using F-K grade level, 9 of 11 custom documents (81.8%) scored above the nationally recommended sixth-grade reading level. For the vendor's "easy-to-read" documents, 17 of 22 (77.3%) scored above the company's stated maximum fourth-grade reading level. For standard vendor documents, only 12 of 115 (10.4%) scored above the company's stated maximum eighth-grade reading level, while 92 of 115 (80.0%) scored above the nationally recommended sixth-grade reading level ([Figure 2](#f2-jmla-106-38){ref-type="fig"}). A scatterplot of SMOG versus F-K grade level scores is shown in [Figure 3](#f3-jmla-106-38){ref-type="fig"}. {#f2-jmla-106-38} {#f3-jmla-106-38} DISCUSSION ========== The patient education documents evaluated in this study covered a wide spectrum of topics in the most commonly distributed materials at UTMC. Assessing patient education materials at UTMC produced significantly different results when using a hand-scored method, SMOG, and a computerized method, F-K grade level. When SMOG was used, 77% of materials were above the nationally recommended sixth-grade reading level, while 58% scored above eighth grade. When using the computerized F-K grade level, documents produced a significantly lower grade level score as opposed to SMOG. This result is consistent with D'Alessandro and colleagues, who concluded that patient education materials distributed on the Internet produced significantly higher reading grade level scores when they used SMOG assessments as opposed to F-K grade level \[[@b29-jmla-106-38]\]. The findings are also consistent with Walsh and Volsko \[[@b21-jmla-106-38]\] and Freda \[[@b22-jmla-106-38]\], who found SMOG to be 2 to 3 grade levels higher than F-K grade level. Additionally, the Centers for Medicare & Medicaid Services, US Department of Health and Human Services, warns that "Flesch-Kincaid scores tend to underestimate actual reading grade level because they are often several grade levels below results obtained using other measurements" \[[@b30-jmla-106-38]\]. One potential explanation for the discrepancy between SMOG and F-K grade level is that the formula used in F-K grade level only allows a maximum of a twelfth-grade reading level score \[[@b20-jmla-106-38]\]. This systematic limitation could lessen the actual reading grade level score, as values of thirteen and higher are not possible. Hand-scoring patient education materials allows evaluators to work directly with the text, alerting them to multisyllabic words and long sentences \[[@b23-jmla-106-38]\]. Computerized-scoring tends to be easier to use; however, as previously mentioned, patient education materials tend to score lower in reading level than they are. Computerized-scoring is highly dependent on period placement. For example, document number 30 in this study had only 1 period at the very end of the document, causing the F-K grade level to be 6.7. However, once periods were added to the end of each sentence, the score decreased to 5.8, further proving the use of periods are of the utmost importance when using the computerized method. Therefore, researchers found the hand-scoring method SMOG to be the better method to assess patient education at UTMC. A lack of comprehension of print patient education materials can negatively impact a patient's health literacy \[[@b31-jmla-106-38]\]. Patients with low health literacy have increased readmission rates, frequently return to the emergency department for the same conditions, and are less able to manage their chronic conditions \[[@b32-jmla-106-38]--[@b34-jmla-106-38]\]. To help increase patient understanding and help combat these issues, the quality of written communication needs to be improved. Research has shown that "using materials that are written in a manner that facilitates the uptake and use of patient education content has great potential to improve the ability of patients and families to be partners in care and to improve outcomes, especially for those patients and families with limited general literacy or health literacy skills" \[[@b35-jmla-106-38]\]. Therefore, print communication needs to be written at a reading grade level that is understandable to people of all literacy levels. With this in mind, librarians can advance their role and help to ensure that patient education materials are written at a grade level that is understandable to all. As teachers, educators, and researchers in a health care setting, it is important that librarians understand the different reading grade level tools and be able to teach nurses and health care providers how to accurately find the reading grade level of patient education materials. Medical librarians' role is constantly changing in that they are not only providing support for staff, but also providing support for patients who need education \[[@b36-jmla-106-38]\]. Librarians can take a more active role in evaluating patient education distributed at their medical centers to determine if it is at the required reading grade level. This can be done through participating on a hospital's patient education committee or teaching classes on evaluating grade level to nurses, hospital staff, medical students, and residents. At the UTMC, a hospital librarian evaluates in-house created materials for grade level due to involvement in the Patient Education Committee \[[@b37-jmla-106-38]\]. Through working with the Patient Education Committee, the librarian can advocate for the patient's right to have patient education materials written at a lower reading grade level. Medical librarians have the potential to initiate research on lowering the reading grade level of patient education materials, such as this study has done. Research can include using focus groups of lay people to review patient education materials for understandability and readability. For example, Blanck and Marshall discussed how a librarian was able to get their Patient Education Committee to test patient education materials with a panel of lay people \[[@b38-jmla-106-38]\]. Future work is being conducted by librarians who worked on this study to edit the patient education materials that they reviewed and, as a result, lower the reading grade level. There were several limitations to this study that should be noted. First, by design, the most utilized patient education materials were selected for analysis in a nonrandomized fashion. This limited our ability to infer "causal effect" due to lack of document randomization. It cannot be said with certainly that the results of this nonrandomized sample would be duplicated if all documents distributed by UTMC were analyzed. Future research should consider using a randomized sample of patient education materials from various geographical regions and institutions around the United States. The patient education materials in this study that are being distributed at UTMC do not meet the current NIH, AMA, or Joint Commission recommendations for reading grade levels in the United States. These results will interest libraries whose institutions have chosen this vendor to provide patient education materials. Results produced by computerized methods produced significantly lower reading grade levels than hand-scored methods. While computerized methods may be more cost-effective and efficient than hand-scored methods, third-party patient educational material providers and health care providers should be cautioned that the scores produced by this method may be manufacturing reading grade levels that seem lower than they actually are. Health care providers in low literacy areas, like Appalachia, should be particularly concerned about reading grade levels of materials distributed, given the region's high percentage of individuals with low literacy. Medical librarians have the potential to make an impact on lowering the reading grade level of patient education materials through teaching, being on a patient education committee, and conducting future research. **Kelsey Leonard Grabeel, MSIS, AHIP,** <kgrabeel@utmck.edu>, Assistant Director of the Health Information Center, Preston Medical Library/Health Information Center, University of Tennessee Graduate School of Medicine/University of Tennessee Medical Center, Knoxville, TN **Jennifer Russomanno, MPH, CHES,** <JRussomanno@utmck.edu>, Continuing Education Coordinator, Preston Medical Library/Health Information Center, University of Tennessee Graduate School of Medicine/University of Tennessee Medical Center, Knoxville, TN **Sandy Oelschlegel, MLIS, AHIP,** <SOelschl@utmck.edu>, Associate Professor and Director, Preston Medical Library/Health Information Center, University of Tennessee Graduate School of Medicine/University of Tennessee Medical Center, Knoxville, TN **Emily Tester, BA,** <epollar2@vols.utk.edu>, Graduate Research Assistant, Preston Medical Library/Health Information Center, University of Tennessee Graduate School of Medicine/University of Tennessee Medical Center, Knoxville, TN **Robert Eric Heidel, PhD,** <RHeidel@utmck.edu>, Assistant Professor, Department of Surgery, Division of Biostatistics, University of Tennessee Graduate School of Medicine/University of Tennessee Medical Center, Knoxville, TN

Data are from the Maternal Depression Online study and can be located in the fourth supporting information file. This file contains the minimal data set underlying the findings in this manuscript (i.e. the data necessary to replicate experimentation, as well as the values behind statistics, graphs, figures, and tables). Additional data are not needed to replicate experimentation. The data are not from a third-party. Introduction {#sec002} ============ Postpartum depression (PPD) impacts approximately 8 to 15% of women within the first year of childbirth \[[@pone.0149186.ref001]\]. The disorder affects a woman's daily functioning, ability to care for her infant, the mother-infant bond, and overall quality of life \[[@pone.0149186.ref002]\]. PPD may also result in short- and long-term developmental consequences for the infant \[[@pone.0149186.ref003]\]. While there is growing evidence in support of psychotherapy for PPD \[[@pone.0149186.ref004]\], participation rates in treatments are extremely low \[[@pone.0149186.ref005]\]. Stigma associated with receiving mental health treatment, difficulty arranging childcare, transportation challenges, and time and financial constraints are factors that contribute to the under-treatment of PPD \[[@pone.0149186.ref006]\]. Internet-Delivered Therapy {#sec003} -------------------------- Internet delivery systems could overcome many of these face-to-face treatment barriers. Receiving treatment from the comfort and convenience of a woman's home would reduce mobility and childcare concerns. Relatively anonymous Internet-delivered therapy may circumvent women's concerns of stigma \[[@pone.0149186.ref007]\]. Further, evidence is accumulating to suggest that Internet-delivered treatments are as efficacious and more cost effective in terms of the provider's time and resources as compared to face-to-face treatments \[[@pone.0149186.ref008]\]. Ample empirical evidence supports the efficacy of Internet-delivered Cognitive Behavior Therapy (ICBT) in the treatment of major depression and sub-threshold depression. Particularly with therapist-assistance, ICBT is an effective method for reducing depressive symptoms when compared to treatment as usual (TAU), control groups, and treatment from a general practitioner \[[@pone.0149186.ref009]--[@pone.0149186.ref011]\]. Meta-analytic data from 21 studies indicate that effects obtained by supported Internet-delivered therapy for major and sub-threshold depression is analogous to those obtained in face-to-face therapy \[[@pone.0149186.ref009]\]. Moreover, supported Internet-delivered interventions evidence both greater program adherence \[[@pone.0149186.ref012]\] and stronger effects than self-directed interventions \[[@pone.0149186.ref013]\], with more frequent support resulting in greater symptom change \[[@pone.0149186.ref014]\]. Internet-Delivered Therapy for Postpartum Depression {#sec004} ---------------------------------------------------- Preliminary research suggests that Internet-delivered treatment is efficacious for women in the postnatal period. In one study, researchers tested the efficacy of an Internet-delivered Behavioral Activation (iBA) program compared with TAU for PPD \[[@pone.0149186.ref015]\]. The 11-session program consisted of psychoeducation and behavioral exercises. Although the iBA program was self-directed, it included access to an Internet chat room that was moderated by parent supporters and supervised by specialist health visitors. Findings from 910 women with PPD indicated that participants in the iBA group reported greater reductions in depressive symptoms compared to the TAU group; attrition rates were high, however, with less than one-third of the participants completing the program. In an effort to improve treatment understanding and adherence rates, these same researchers recently created an 11-module iBA program and added a guided support component (i.e., telephone calls; \[[@pone.0149186.ref016]\]). Again, attrition rates were high. Only 5% of the women completed more than eight modules, with less than 2% completing all iBA modules. Women who reported lower perceived support, who were working or studying, and who reported lower socio-economic status completed fewer modules. Another Internet-delivered intervention, Mom-Net, provided Therapist-Assisted ICBT (TA-ICBT) to economically disadvantaged mothers of young children \[[@pone.0149186.ref017]\]. Consisting of eight modules, Mom-Net included cognitive behavioral skills with adaptations relevant to mothers. Weekly phone calls were arranged between coaches and participants to assist with understanding and applying the materials. Women (*n* = 70) with elevated levels of depression, a child under five years of age, and attending a low-income program were randomly assigned to TA-ICBT or TAU. When compared to the TAU participants, Mom-Net participants reported improvements on measures of depression, parental satisfaction and efficacy, as well as a reduction on observational indices of harsh parenting behavior \[[@pone.0149186.ref017]\]. For the Mom-Net intervention, there was less attrition, with nearly two-thirds of women completing all Internet-delivered modules \[[@pone.0149186.ref017]\]. While preliminary research supports the efficacy of TA-ICBT for women of young children \[[@pone.0149186.ref017]\] and Internet-delivered behavioral intervention for PPD \[[@pone.0149186.ref015],[@pone.0149186.ref016]\], researchers have not yet determined whether ICBT is an efficacious modality for treating sub-threshold and clinical PPD within the first post-natal year. Prior to implementation of TA-ICBT within community settings, it is important to ensure that the approach is more efficacious than symptom reduction over time as PPD is often a spontaneously remitting condition \[[@pone.0149186.ref018]\]. The purpose of the present investigation was to explore the efficacy of a specialized TA-ICBT program for women with symptoms of sub-threshold and clinical PPD. It was hypothesized that TA-ICBT participants would demonstrate greater symptom improvement than participants who received the delayed intervention on primary and secondary outcome measures (i.e., PPD, postnatal anxiety, general stress, parental stress, quality of life). These changes were expected to be clinically significant and maintained at follow-up. Materials and Methods {#sec005} ===================== Participants {#sec006} ------------ Using the GPower program \[[@pone.0149186.ref019]\], to achieve power at 80% (alpha = 0.05), *n* = 23 participants per group were required to detect a medium effect size for a mixed model ANOVA. A medium effect size is consistent with previous meta-analyses of therapist-assisted ICBT \[[@pone.0149186.ref011],[@pone.0149186.ref013]\]. A total of 50 participants were recruited to account for participant attrition. Participant recruitment and follow-up occurred between March, 2012 and February, 2013. Recruitment efforts included notifying physicians, perinatal nurses, and the public through the use of posters, information cards, Internet advertisements, and media interviews. Participant recruitment concluded when data collection was complete. Eligibility criteria were: (a) 18 years of age or older; (b) gave birth to an infant within the past year; (c) residing in Saskatchewan; (d) self-reported access to and comfort using a computer and the Internet; (e) score of ≥ 10 on the Edinburgh Postnatal Depression Scale (EPDS \[[@pone.0149186.ref020]\]); (f) consent to notify a physician of their participation; (g) not receiving other psychotherapy; (h) if taking medication, stable dose for more than a month; and (i) no past or present psychotic mental illness (schizophrenia), bipolar disorder, or current suicide plan or intent. [S1 Table](#pone.0149186.s005){ref-type="supplementary-material"} provides demographic details for the participants in the study. As displayed in [S1 File](#pone.0149186.s002){ref-type="supplementary-material"}, a total of 56 adult women responded to recruitment efforts. Of these, six participants were considered ineligible for reasons including: (1) they did not meet PPD symptom inclusion criteria, *n* = 2; (2) they declined to participate, *n* = 1; and (3) they recently had a change in their medication, *n =* 3. [S1 File](#pone.0149186.s002){ref-type="supplementary-material"} Flow of process for research participants {#sec007} ---------------------------------------------------------------------------------------------------------- Participants who were interested in taking part in the study contacted the researcher. Participants were informed of the need to conduct an initial telephone screening interview to determine eligibility for the study. Verbal informed consent for the telephone screening was obtained at this stage and documented on each participant's file. Participants then completed the telephone screening. Following this initial verbal consent and telephone screening, eligible participants then completed an electronic informed consent form followed by the online questionnaire battery. Participants were not paid for their participation in the study. The informed consent process was approved by the research ethics boards of the institutions involved. Verbal and electronic consent were deemed sufficient for participation. Ethics approval was granted by the Universities of Regina and Saskatchewan and the Regina Qu'Appelle Health Region in March, 2012. The trial was registered with the International Standard Randomized Controlled Trials (ISRCTN: 85456371) in June 2013. There was a delay in registering the trial as we were made aware of the benefits of registering the trial after data collection had begun. No changes were made to the study protocol between the start of the trial and the time of registration. The authors confirm that all ongoing and related trials for this intervention are registered. The project was partially supported by funding provided by the Canadian Institutes of Health Research (\#101526) and the Saskatchewan Health Research Foundation. Procedure {#sec008} --------- ### Screening {#sec009} Initial assessment included a telephone screen using the Edinburgh Postnatal Depression Scale (EPDS \[[@pone.0149186.ref020]\]) along with collection of background information including age of infant, computer and Internet access, and any current psychological or pharmacological treatment. Elevated PPD symptomatology was operationalized as a score of ≥ 10 on the EPDS. A threshold of ≥ 10 is commonly used to screen for subclinical PPD to increase sensitivity, and has been recommended for community settings to ensure that all potential cases of PPD are identified \[[@pone.0149186.ref021]\]. The mean participant EPDS score at pre-screening was 15.96 (*SD* = 3.89). Participants reporting sub-threshold PPD were included given that sub-clinical PPD symptoms are associated with parenting difficulties and adverse child outcome \[[@pone.0149186.ref022]\]. Following the initial assessment, participants then completed a telephone interview consisting of the administration of the MINI International Neuropsychiatric Interview (MINI \[[@pone.0149186.ref023]\]). The MINI confirmed sub-clinical and clinical symptoms of PPD and ruled out psychotic mental illnesses, bipolar disorder, alcohol or substance disorders, and those at elevated risk of suicide. Participants were also required to complete a battery of self-report measures administered over the Internet as described below. ### Randomization {#sec010} Eligible and willing participants were individually randomized (allocation 1:1) to either TA-ICBT/ Maternal Depression Online or WLC conditions. Participants were allocated in equal numbers to each intervention using blocked randomization. Random allocation was conducted via an Internet computer program by a researcher who was not involved with the research study. The researchers were blind to the randomization process and were notified through e-mail regarding participant allocation following the screening process. Following the notification, the primary researcher (N.E.P.) contacted the participants over the telephone and informed them of their assigned treatment group (i.e., TA-ICBT or WLC). Due to the nature of the conditions (i.e., TA-ICBT or WLC), it was not possible for the study therapists and participants to be blind to the treatment assignment. ### Post-Intervention and Follow-Up Assessment {#sec011} Post intervention/delay period measures were administered to all participants over a secure Internet website (T2; 7--10 weeks after T1). Once the T2 measures were completed by the WLC participants, they were contacted and offered TA-ICBT. Four-week follow-up measures were administered over the Internet for participants assigned to the TA-ICBT condition only (T3; 4 weeks after T2). Measures {#sec012} -------- ### Primary Assessment Measure {#sec013} The EPDS \[[@pone.0149186.ref020]\] is a 10-item self-report measure that assesses emotional and cognitive symptoms of PPD. Each EPDS item is scored from 0 to 3 in relation to the past seven days. Total scores range from 0 to 30 with higher scores indicating more severe symptomology. The EPDS has demonstrated excellent psychometric properties \[[@pone.0149186.ref020]\], is sensitive to change over time \[[@pone.0149186.ref021]\], and the computerized version has been validated \[[@pone.0149186.ref024]\]. In the current study, the EPDS demonstrated acceptable internal consistency, α = .80. ### Secondary Assessment Measures {#sec014} The Depression Anxiety Stress Scale- Short Form \[[@pone.0149186.ref025]\] (DASS) is a 21-item measure assessing dysphoric mood, fear and autonomic arousal, and general nervousness and agitation. With reference to the past week, participants respond to items on a 4-point Likert scale. The computerized version of the DASS has been validated \[[@pone.0149186.ref026]\] and the measure has been assessed in a population of postnatal women with and without PPD \[[@pone.0149186.ref027]\]. In the current study, the DASS-21 demonstrated acceptable internal consistency for each subscale: Depression α = .81; Anxiety α = .77; and Stress α = .78. The Parenting Stress Index-Short Form (PSI-SF \[[@pone.0149186.ref028]\]) comprises three subscales assessing parental distress, parent-child dysfunctional interaction, and perception of a difficult child. Each subscale consists of 12 items rated from 1 (*strongly disagree*) to 5 (*strongly agree*). The PSI-SF has demonstrated excellent psychometric properties and is appropriate for use with parents of children as young as one month \[[@pone.0149186.ref028]\]. In the current study, the PSI-SF exhibited acceptable internal reliability: Parental Distress α = .77; Parent-Child Dysfunction α = .87; and Difficult Child α = .90. The World Health Organization Quality of Life Assessment BREF (WHOQOL-BREF \[[@pone.0149186.ref029]\]) is a 26-item measure assessing quality of life in four domains: Physical health (e.g., sleep, pain); Psychological health (e.g., self-esteem, concentration); Social relationships (e.g., support, personal relationships); and Environment (e.g., physical safety, financial resources). Participants respond on a 5-point Likert scale ranging from 1 (*not at all*) to 5 (*an extreme amount*). The computerized version of the WHOQOL-BREF has demonstrated excellent psychometric properties \[[@pone.0149186.ref030]\] and the measure has been validated on women with and without PPD \[[@pone.0149186.ref031]\]. In the current study, the four WHOQOL-BREF domains demonstrated acceptable internal consistency: Physical α = .65; Psychological α = .67; Social α = .65; and Environment α = .87. ### Treatment Relevant Outcome Measures {#sec015} The Therapeutic Alliance Questionnaire (TAQ) consists of 17 items assessing the perceived helpfulness of the ICBT therapeutic relationship and has been used in multiple TA-ICBT studies \[[@pone.0149186.ref032]\]. The TAQ was adapted from the Helping Alliance Questionnaire (HAQ-II \[[@pone.0149186.ref033]\]). Statements are rated from 1 (*strongly disagree*) to 6 (*strongly agree)* with higher scores indicating the participant perceived the ICBT relationship as helpful. The HAQ-II has demonstrated acceptable psychometric properties \[[@pone.0149186.ref034]\], but the psychometric properties of the TAQ have not been reported. In the current study, the TAQ demonstrated excellent internal consistency, α = .94. Treatment satisfaction was assessed with two questions: "How much did you like the treatment program?", and "How much did you enjoy communicating with your Internet therapist?" Each item was rated on a 0 (*not at all*) to 7 (*very much so*) Likert scale. These questions were adapted from the Treatment Satisfaction Questionnaire-Modified (TSQ \[[@pone.0149186.ref035]\]). In this study, the responses to the two items were summed to create a total TSQ score. The responses to the two items were highly correlated, *r* = 0.77. The Credibility/Expectancy Questionnaire (CEQ \[[@pone.0149186.ref036]\]) assessed perception of treatment credibility with four items rated on a scale ranging from 1 (*not at all*) to 9 (*very much*), while treatment expectancy was measured with two items rated using a 0 to 100% scale. Psychometric properties are strong and factor analysis has confirmed the subscales of the CEQ \[[@pone.0149186.ref036]\]. Treatment Conditions {#sec016} -------------------- ### Waitlist Control {#sec017} Participants randomized to the WLC condition were provided with an information pamphlet that included psycho-education on PPD and websites to access provincial mental health support services. They were requested to inform the researchers of any treatment they received during the wait period. ### TA-ICBT for PPD {#sec018} Maternal Depression Online was adapted from a TA-ICBT program for depression offered through the Online Therapy Unit for Service Education and Research ([www.onlinetherapyuser.ca](http://www.onlinetherapyuser.ca/) \[[@pone.0149186.ref037]\]), located in Saskatchewan, Canada. The original program content was licensed from Swinburne University of Technology National eTherapy Centre \[[@pone.0149186.ref038]\]. Maternal Depression Online maintained much of the original program's content, but incorporated adaptations relevant to mothers of young infants based on Milgrom et al.'s treatment for PPD \[[@pone.0149186.ref039]\]. Given that treatment studies for in-person CBT for PPD typically range from six to eight sessions \[[@pone.0149186.ref040]\] and TA-ICBT studies for major depression often range from five to eight modules \[[@pone.0149186.ref013]\], the depression program was shortened from 12 modules to seven modules. Material that is not traditionally part of CBT was removed such as information on nutrition and mindfulness. Each module included a range of media (e.g., text, graphics, animation, audio, video) given research to suggest that multimedia options enhance the effectiveness of Internet-delivered treatment and are preferred over static treatments \[[@pone.0149186.ref041]\]. Maternal Depression Online was designed so that each page of the program was viewed prior to proceeding to the next page. Check-in questions focusing on the module content were presented at the beginning of each module while homework exercises were assigned at the end of each module. Participants were encouraged to progress at a pace of one module per week although more time was often taken. Most TA-ICBT and WLC participants completed T2 measures at a standard 10-week follow-up time. Measures were administered earlier, however, if the participants withdrew from the program (*n* = 3) or completed seven modules in a shorter period of time (*n* = 6). Participants were provided with a username and password that allowed them to access the site and message their therapist over a secure private messaging system. An industry-standard encryption algorithm was used for all data, emails, and check-in questions, and stored on a dedicated university server \[[@pone.0149186.ref037]\]. The Internet therapists included two doctoral students in Clinical Psychology (including co-author N.E.P.) who were supervised by a registered psychologist and expert in TA-ICBT (co-author H.D.H.). Internet therapists completed a training workshop in the provision of TA-ICBT \[[@pone.0149186.ref042]\]. Therapists emailed their assigned participant on a set day each week to provide support and encouragement, and to answer questions. E-mails took a variable length of time to compose depending on the information submitted to the therapist with the average email taking 15 to 20 minutes to compose. While the e-mail content overlapped substantially with the treatment material, different components were emphasized and individualized depending on the problems presented by the participant. The e-mails were not pre-prepared. If a participant reported significant distress or failed to log into the program for over seven days, the therapist telephoned the participant. Statistical Analysis {#sec019} -------------------- All outcome data were analyzed on an intention-to-treat basis such that the analyses included every participant who was randomized according to randomized treatment assignment. Between-group intervention effects on the EPDS from baseline (screening) to T2 (post-intervention/wait-time) were compared using longitudinal mixed effects models \[[@pone.0149186.ref043]\]. The longitudinal mixed model uses all data on each subject, is unaffected by randomly missing data, and it can flexibly model time affects. The more familiar mixed model ANOVA (condition as the between subject factor; time as the within subject factor) is a special case of a longitudinal mixed effect model. Random intercept models were computed and a maximum likelihood method with covariance type (based on the variance components) was employed to provide the estimates. Chi-square analysis will determine if the difference between the two conditions is statistical significant. Clinically significant change on the EDPS was calculated using procedures outlined by Matthey \[[@pone.0149186.ref044]\] in their study of 181 women with PPD. Matthey followed Jacobson and Truax's \[[@pone.0149186.ref045]\] calculations to determine the magnitude of change on the EPDS required to be considered clinically significant. Participants whose EPDS score of more than 10 at baseline, decreased by at least four points were classified as *recovered* based on calculations of Matthey's Reliable Change Index (RCI) for the EPDS. Those who demonstrated a decrease in their EPDS score of four or more points but still scored 10 or more after treatment were classified as *improved but not recovered*. Those participants who showed a four point increase in their EPDS score after treatment were classified as *deteriorated*. Finally, participants whose pre-post difference score was less than four points were classified as *no change* regardless of whether or not their post-treatment scores fell below the cut-off score of ten. The procedure and recommendations outlined by Matthey \[[@pone.0149186.ref044]\] have been utilized in other Internet therapy research (e.g., \[[@pone.0149186.ref016]\]) Multiple regression analyses were used to determine if group differences (TA-ICBT versus WLC) predicted secondary outcome measure scores (i.e., DASS, PSI, WHO-QOL subscales) when controlling for pre-treatment scores on each respective secondary outcome measures. The effects of missing follow-up data were explored by imputing missing data using chained equations \[[@pone.0149186.ref046], [@pone.0149186.ref047]\] that impute data for all relevant variables. All multiple regressions were analyzed twice---once with the original data set and once with the imputed data set. Unstandardized coefficients were used as a measure of effect size to assess the proportion of variance associated with or accounted for by each of the regressions. The results of the two regression analyses were compared to determine if the missing data impacted the results. A within-subjects analysis for participants receiving TA-ICBT was conducted comparing four-week follow-up (T3) EPDS scores with post-treatment (T2) and baseline EPDS scores (screening). Finally, descriptive statistics are reported for program engagement, satisfaction, and therapeutic alliance reported by the TA-ICBT group (i.e., TSQ, TAQ, CEQ). Results {#sec020} ======= Preliminary Analyses {#sec021} -------------------- The TA-ICBT and WLC groups did not differ with respect to demographic characteristics (see [S1 Table](#pone.0149186.s005){ref-type="supplementary-material"}). Descriptive statistics for primary and secondary outcome measures administered at T1 and T2 are presented in [S2 Table](#pone.0149186.s006){ref-type="supplementary-material"}. There were no differences between completers and non-completers on the baseline EPDS: *t*(48) = 1.87, *p* = .067, *d* = .54. The distribution of the EPDS was normal and similar variance was revealed between the two groups. Attrition rates for TA-ICBT and WLC groups were 16% (*n* = 4) and 12.5% (*n* = 3), respectively between T1 and T2. Attrition rates for the TA-ICBT group between T2 and T3 was 28.5% (*n* = 5). Treatment Received Reported by WLC Group {#sec022} ---------------------------------------- WLC participants reported receiving various support/care during the waiting period. Close to 58% reported regular contact with their family physician, 32% sought psychotherapy for maternal depression, and 25% of participants reported using psychotropic medication. Primary Outcome Measure {#sec023} ----------------------- Results from the longitudinal mixed model analysis indicated no effect for condition on EPDS scores, *F*(1, 48.07) = .35, *p* = .56, a statistically significant change in EPDS scores over time, *F(*1,20.99) = 16.23, *p* = .001, but most importantly and consistent with our primary hypothesis, a condition by time interaction, *F*(1, 11.82) = 5.15, *p* = .02. Following the criteria outlined by Matthey \[[@pone.0149186.ref044]\], approximately 20% of participants receiving TA-ICBT were classified as improved, while over 62% were recovered. On the other hand, while 38% of the participants allocated to WLC condition were classified as recovered, over 50% exhibited no reliable change. Chi-square analysis showed the differences between the two conditions approached statistical significance: χ^2^(1) = 2.93, *p* = .08, Cramer's *V* = .026. Secondary Outcome Measures {#sec024} -------------------------- In order to determine whether the treatment influenced participants' stress, anxiety, depression, parental stress, and quality of life, scores on these variables at T2 were the outcome variables in a series of multiple regressions. The predictor variables were condition (TA-ICBT or WLC) and scores on the secondary outcome variables at T2. The results for these analyses are presented in [S3 Table](#pone.0149186.s007){ref-type="supplementary-material"}. When compared to the WLC, the TA-ICBT at T2 (controlling for scores at T1) produced lower stress (β = -.41), lower parental stress (β = -.25) and distress (β = -.41), and better psychological health (β = .34) and environmental quality of life (β = .31). A comparison of the results from the original data set to the pooled results indicated no statistically significant differences. Follow-Up for the TA-ICBT Condition {#sec025} ----------------------------------- We conducted within-subjects analyses comparing follow-up scores (T3) with post-treatment scores (T2) for participants in the TA-ICBT condition. The EPDS collected at follow-up had statistically significant and large gains compared with post-treatment scores, *t*(14) = 4.13, *p* \< .01, *d* = 1.10, indicating that intervention effects not only were maintained, but also that the TA-ICBT participants continued to improve. Program Engagement and Satisfaction {#sec026} ----------------------------------- [S4 Table](#pone.0149186.s008){ref-type="supplementary-material"} presents the descriptive statistics for program engagement. TA-ICBT participants completed on average 5.92 of the seven modules (60% of the participants completed all seven modules). Less than 18% of participants dropped out before the fourth module; the greatest attrition occurred during the 5^th^ of 7 modules (11.8%). The website was extensively used by participants as indicated by the mean number of program visits (*M* = 26.88; *SD* = 11.63), and the number of emails sent (*M* = 5.4; *SD* = 4.15) and received by the participants (*M* = 10.52; *SD* = 3.95). Over 80% of the participants reported that they liked the overall program and enjoyed communicating with their therapist (TSQ). Participants also reported a high level of therapeutic alliance, giving a rating of 86.42% (TSQ). Discussion {#sec027} ========== PPD is an undertreated condition that may result in both short and long-term consequences for the infant \[[@pone.0149186.ref003]\]. The results of this study suggest that a therapist-assisted Internet-delivered intervention is efficacious for women with PPD. While we were not able to compare the findings to an active treatment condition, when compared to WLC TA-ICBT not only reduced symptoms of PPD, but also reduced stress, parental distress and improved domains of quality of life. Reductions in PPD were clinically significant and symptom reduction on the EPDS was maintained at follow-up. Our findings were in accordance with evidence to support the efficacy of Internet-delivered behavioral activation for PPD \[[@pone.0149186.ref015],[@pone.0149186.ref016]\], TA-ICBT for economically disadvantaged mothers of young children \[[@pone.0149186.ref017]\], and face-to-face CBT for PPD \[[@pone.0149186.ref004]\]. Further, approximately 20% of TA-ICBT participants were considered improved, while over 62% were considered recovered. This high rate of symptom recovery was promising given not only the morbidity and consequences associated with PPD for the mother, but also the pervasive effects PPD has on the family. Moreover, resulting from the multiple face-to-face treatment barriers reported for women with PPD, this is the first study providing evidence that TA-ICBT is an efficacious treatment for women struggling with PPD and this novel approach may overcome face-to-face treatment barriers \[[@pone.0149186.ref006]\]. In addition to the efficacy of TA-ICBT, our results indicated significant program adherence, with participants completing, on average, 5.92 of the seven modules (60% of completed the entire program). Participants not only completed weekly check-ins with each module, but they also messaged their therapist on average five times. Therapists, in turn, provided considerable support to participants sending, on average, eleven messages over the course of treatment. A strong therapeutic alliance and high satisfaction with the overall program was also reported. Findings are consistent with a qualitative investigation that explored participants' perceptions of this program \[[@pone.0149186.ref048]\]; the majority of women expressed that the TA-ICBT program was flexible, accessible, convenient, and afforded them increased feelings of anonymity and privacy. The degree of engagement was promising given past research findings of lower participant involvement in Internet-delivered treatment. O'Mahen et al. \[[@pone.0149186.ref015]\], for instance reported only a 39% completion rate for an Internet-delivered behavioral activation intervention for PPD that included weekly automated email reminders and no therapist assistance. On the other hand, for a TA-ICBT program offered to mothers of young children that included weekly telephone calls from a therapist, similar to the present study, 63% of participants completed all treatment modules \[[@pone.0149186.ref017]\]. Collectively, these findings are in accordance with a meta-analysis suggesting that for the treatment of major depression, when compared with self-administered Internet--delivered interventions, some form of guidance is superior (i.e., TA-ICBT \[[@pone.0149186.ref013]\]). While weekly Internet therapeutic contact has been validated in a sample of participants diagnosed with panic disorder \[[@pone.0149186.ref032]\], research has yet to establish the ideal therapist-client contact frequency for women with PPD. Given the multiple demands and challenges of parenting, often coupled with low levels of social support \[[@pone.0149186.ref049]\], women with PPD may reap additional benefits from more frequent therapeutic contact. We did not find any group differences for the secondary outcome measures relating to general depression, general anxiety, overall parental stress, and select domains of quality of life. Interestingly, group differences were revealed for depression when assessed with the EPDS, but not when measured with using the DASS-21. This may have been the result of the sensitivity of the DASS-21 to assess postnatal depression. Indeed, general depression measures often rely on somatic symptoms of depression, which overlap with physical symptoms associated with the postpartum period \[[@pone.0149186.ref050]\]. While statistically significant group differences were revealed with respect to parental distress, group differences were not identified for the other parental stress subscales. The Parental Stress Inventory may not have adequately tapped into the parental stress reported by women with PPD as the measure was originally validated on a sample of mothers of young children (mean age under four years) and a variety of questions focused on parental stress related to caring for an older child \[[@pone.0149186.ref028]\]. Given the dearth of literature in the area of TA-ICBT for PPD, future research directions are abundant. To begin, additional well-designed trials of TA-ICBT for PPD are required. It is suggested that these studies include a larger, more diverse sample, as well as longer-term follow-up times. Varying Internet therapeutic contact and styles could also be compared, such as weekly versus bi-weekly therapist contact or telephone versus email versus audio/video correspondence. Additionally, given the efficacy and participant satisfaction of TA-ICBT for PPD, future research exploring TA-ICBT for the treatment of other perinatal clinical disorders is warranted, such as a transdiagnostic program to concurrently address symptoms of perinatal anxiety and depression. The comparison of TA-ICBT to an active treatment condition such as face-to-face CBT as well as other Internet-delivered treatments (e.g., Internet behavioral activation) would also be beneficial, as not all participants responded favorably to TA-ICBT. An economic evaluation study comparing the cost-effectiveness and cost utility of TA-ICBT for PPD compared with face-to-face treatment is also warranted. Finally, future implementation research is required to determine whether different health care professionals are equally effective in delivering TA-ICBT for PPD and how this type of novel programming can be effectively implemented and utilized within community settings. Limitations {#sec028} ----------- There were a number of limitations that should be considered. The sample utilized was relatively small and prohibited an examination of other factors. The results, however, are promising as the sample included participants with sub-threshold PPD. An additional limitation was the absence of an active control condition, which may have exaggerated our results. However, the study demonstrates considerable feasibility and acceptability of TA-ICBT for PPD. The limited sample with respect to demographic characteristics was an additional drawback. Moreover, given that more than half of the sample reported some university education, additional research utilizing a more heterogenous sample is required. A group comparison at follow-up was not conducted, which would have been informative given that PPD is often a spontaneously remitting condition \[[@pone.0149186.ref018]\] and, therefore, may improve over time irrespective of treatment. For ethical reasons, however, we did not unduly extend the waiting period for WLC participants. Additionally, depressive symptoms were assessed using self-report measures, which are known to produce false positives. While we conducted a clinical interview (i.e., MINI) to screen participants, a post-treatment clinical interview would have provided a more objective measure of changes in depressive symptoms. Finally, these findings are specifically related to one TA-ICBT program (i.e., Maternal Depression Online) delivered by Internet therapists who were trained graduate students in clinical psychology. A recent literature review suggests that ICBT outcomes do not differ when delivered by experienced versus inexperienced therapists \[[@pone.0149186.ref051]\]. It is possible that the structure of TA-ICBT (e.g., modules) results in less variation in the delivery of TA-ICBT than is observed in face-to-face therapy; therefore, similar outcomes may arise regardless of therapist experience \[[@pone.0149186.ref052]\]. Nevertheless, it is possible that outcomes could have been more impressive had a different TA-ICBT program been used. It would be valuable to conduct further research on which type of allied health provider will obtain the best outcomes with TA-ICBT for PPD and who is best positioned to deliver TA-ICBT if delivered in community settings. Conclusion {#sec029} ========== Despite some limitations to the research design, we demonstrated that TA-ICBT was an efficacious, well-utilized, and desirable intervention for women with PPD. When compared to a WLC condition, TA-ICBT appeared more efficacious at reducing symptoms of PPD, parental stress, and improved psychological and environmental quality of life. Further investigation for TA-ICBT for PPD is required, particularly comparing TA-ICBT to an active control condition, utilizing a larger and more heterogeneous sample, with longer-term follow-up. Additionally, an examination of the cost-effectiveness of TA-ICBT when delivered in community settings would also be desirable alongside the study of factors that constrain or enhance the implementation of TA-ICBT in community practice. Supporting Information {#sec030} ====================== ###### Consort List. (DOC) ###### Click here for additional data file. ###### Consort Fig. (TIFF) ###### Click here for additional data file. ###### Database. (SAV) ###### Click here for additional data file. ###### Ethics Protocol. (DOCX) ###### Click here for additional data file. ###### Demographic Characteristics by Group. (DOC) ###### Click here for additional data file. ###### Means and Standard Deviations for Primary and Secondary Outcomes. (DOC) ###### Click here for additional data file. ###### Regression Analyses of Secondary Outcome Variables. (DOC) ###### Click here for additional data file. ###### Descriptive Statistics for Program Engagement in the TA-ICBT condition (*N* = 24). (DOC) ###### Click here for additional data file. Program materials used by the Online Therapy Unit were licensed from Anxiety Online at Swinburne University of Technology in Melbourne Australia through Drs. David Austin and Britt Klein. We would like to extend our thanks to the staff, students, and faculty associated with the Online Therapy Unit for Service, Education, and Research at the University of Regina. The authors also thank Marcie Nugent, Chantalle Fuchs, and Hugh McCague for their contributions to this study. [^1]: **Competing Interests:**The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: NEP HDH. Performed the experiments: NEP. Analyzed the data: NEP. Contributed reagents/materials/analysis tools: NEP HDH. Wrote the paper: NEP HDH DD. [^3]: Current address: Department of Psychology, Vancouver Coastal Health, Vancouver, BC, Canada

All relevant data are within the manuscript and its Supporting Information files. Introduction {#sec001} ============ By sequestering dopamine (DA) into presynaptic neurons, the dopamine transporter (DAT) regulates spatio-temporal components of DA transmission. As a critical regulator of DA transmission, DAT contributes to voluntary movement, reward and mnemonic functions of the brain and modulates the efficacy of therapeutic drugs targeted to this plasma membrane protein. DAT expression is highly circumscribed in discrete regions throughout the brain and the expression of the human DAT gene (*SLC6A3*) varies among subjects\[[@pone.0218129.ref001]--[@pone.0218129.ref005]\]. Thus DNA sequence variations in the regulatory regions specially the promoter of *SLC6A3* may contribute to altered expressional patterns in the brain, dopamine-related individuality as well as diseases\[[@pone.0218129.ref006], [@pone.0218129.ref007]\]. The essential roles of DAT in brain function have mandated extensive studies of *SLC6A3* associations with behaviors and diseases. During the last twenty five years *SLC6A3* has been extensively studied for genetic associations but the association studies with different markers located in *SLC6A3*'s 3' regions obtained mixed results. On chromosome 5 (chr5), *SLC6A3* spans 70 kilobase (kb) from the 5' promoter to 3'untranslated region (3'UTR, located in Exon 15). Sporadic genetic markers in several regulatory regions throughout *SLC6A3*, including the 5' promoter, Intron 8 and 3'UTR, have been used in hundreds of association studies on more than eight different diseases and a number of human behaviors. Vast majority of these studies used a classical variable number tandem repeat located in the 3'UTR (3'VNTR/rs28363170) and another more recently used VNTR in Intron 8 (Int8VNTR/rs3836790)\[[@pone.0218129.ref008], [@pone.0218129.ref009]\]. As a result, findings from these association studies, especially those with 3'VNTR, in various populations were largely inconclusive or showed small effect sizes, such as studies on schizophrenia and bipolar disorder among others\[[@pone.0218129.ref010]--[@pone.0218129.ref015]\]. In particular, *SLC6A3* has been well implicated in the etiology and treatment of attention deficit hyperactivity disorder (ADHD)\[[@pone.0218129.ref016]--[@pone.0218129.ref022]\] but human genetic association studies with the 3'VNTR could not obtain consistent positive signals\[[@pone.0218129.ref023]--[@pone.0218129.ref025]\]. Another example was pharmacogenetics of Parkinson's disease (PD): 3'VNTR or In8VNTR was associated with differential response to pharmacotherapy of PD\[[@pone.0218129.ref026]--[@pone.0218129.ref030]\] but not based on other studies \[[@pone.0218129.ref031], [@pone.0218129.ref032]\]. Moreover, *SLC6A3* genotype was found to modulate the risk of pesticide exposure for PD by some studies \[[@pone.0218129.ref033]\]but not by others\[[@pone.0218129.ref034]\]. The lack of evidence-based selection of markers resulted in the mixture, unfortunately causing little motivation to add more association studies with any markers in the *SLC6A3* genetic field. Importantly, the unreliable human genetic findings are inconsistent with ample positively-related evidence for *SLC6A3* activity *versus* phenotypes from other approaches such as pharmacology, imaging and animal genetics\[[@pone.0218129.ref020], [@pone.0218129.ref035]--[@pone.0218129.ref039]\]. In contrast to 3'VNTR, its promoter markers were more consistently associated with ADHD in various populations \[[@pone.0218129.ref040]--[@pone.0218129.ref047]\]. Consistently, we and others have shown that the 5' promoter regions display varying regulatory activity and also in a haplotype-dependent manner *in vitro*\[[@pone.0218129.ref048]--[@pone.0218129.ref051]\]. Findings from rodent genetic studies have demonstrated the causality of reduced DAT activity on various phenotypes\[[@pone.0218129.ref037], [@pone.0218129.ref052]--[@pone.0218129.ref056]\]. These findings suggest that polymorphism- or haplotype-dependent *SLC6A3* promoter activity may confer risk for related diseases and that genetic association studies should have resulted in consistent positive findings. One explanation for the current elusiveness of the association findings with the 3'VNTR was that this marker is far away from upstream regulatory regions, including Int8VNTR and the 50 kb-away 5' promoter, and unable to capture the related information due to high recombination rates or weak linkage disequilibrium (LD). Other explanation is that different populations carry different frequencies of the same markers or even different disease loci. In either case, these already used genetic markers were unlikely the underlying disease loci. *In vitro* studies have shown polymorphisms in the 5' core promoter, Int8VNTR and 3'VNTR all regulated promoter activity\[[@pone.0218129.ref049], [@pone.0218129.ref057]--[@pone.0218129.ref061]\]. This information suggests that it be critical for association studies to use genetic markers in all distinct regulatory regions, in order to capture variable *SLC6A3* expression as a whole and identify the underlying haplotypes and signaling pathways\[[@pone.0218129.ref051]\]. To clarify these possibilities, it is necessary to deeply sequence the regulatory regions for a better understanding of the *SLC6A3* genetic structure including the 3' VNTRs, given the implications of *SLC6A3* in a spectrum of diseases and other behavioral characteristics. This task requires systemic discovery of polymorphisms and haplotypes in the regulatory regions, through targeted deep-sequencing that is helpful in discovery of novel functional loci or mutations in different fields\[[@pone.0218129.ref062]--[@pone.0218129.ref069]\]. In other words, the presence of multiple *SLC6A3* regulatory regions mandates mechanistic studies of the *SLC6A3* genetics, to help delineating functionally distinct *SLC6A3* haplotypes. In this targeted deep-sequencing study, we uncover unique and common polymorphisms and haplotypes, and recombination hotspots for two major U.S. populations, African Americans (AAs) and European Americans (EAs), for finding generalization. The findings may help explain the mixed association findings\[[@pone.0218129.ref025], [@pone.0218129.ref070], [@pone.0218129.ref071]\] and instruct our future strategy for identifying disease loci in *SLC6A3*\[[@pone.0218129.ref029], [@pone.0218129.ref072], [@pone.0218129.ref073]\]. Materials and methods {#sec002} ===================== Subjects {#sec003} -------- Sixty unrelated subjects were selected from the Collaborative Studies on Genetics of Alcoholism (COGA) pedigrees\[[@pone.0218129.ref074]\] and their de-identified genomic DNAs were provided by COGA through the Coriell Institute (NJ, U.S.A.) with the approval of National Institute on Alcohol Abuse and Alcoholism (NIAAA). Subjects gave their informed consent to the COGA study. From each of the COGA pedigrees, we selected the grandparents and their offspring's spouses that came from outside the pedigree as unrelated subjects and the unrelatedness was verified by genomic control\[[@pone.0218129.ref075]\]. They included two cohorts: 30 AAs and 30 EAs. Each cohort consisted of 15 controls and 15 patients with substance use disorders (SUDs) (see [Table 1](#pone.0218129.t001){ref-type="table"}). The 30 control subjects were all unaffected based upon the Diagnostic and Statistical Manual of Mental Disorders III Revision (DSM-IIIR), Feighner Criteria and International Classification of Diseases, Tenth Revision (ICD-10). All of the 30 affected subjects met at least two of the DSM-IIIR, Feighner Criteria and ICD-10 criteria for alcohol dependence. This study was approved by McLean Hospital Institutional Review Board. 10.1371/journal.pone.0218129.t001 ###### Demographic information on 60 COGA subjects used. {#pone.0218129.t001g} AA (30) EA (30) ------------- ------------ ----------- ---- ------------ ----------- ---- Male 7 7 14 8 8 16 Female 8 8 16 7 7 14 Average age 40.4 ± 4.3 44.3 ±3.6 49.0 ± 3.8 48.1 ±3.6 Age range 21--72 27--70 24--77 24--63 DNA sequencing {#sec004} -------------- A two-step \"boost/nest\" polymerase chain reaction (PCR) strategy was used to sequence the 18 kb promoter regions at Polymorphic DNA Technologies, Inc (Alameda, CA). We first did a boost reaction for a larger PCR amplicon and then used this amplicon as a template for the nest reaction, followed by sequencing of the nest product. The conditions for the boost PCR reaction were identical to the nest with the following exceptions: 10 ng of genomic DNA was used for the boost, then 1 μL of boost product as template for the nest reaction. The two reactions used two different pairs of different primers. PCR cycle was: 94°C for 4 min, 25 cycles of 94°C for 20 sec, 55°C for 25 sec and 72°C for 1 min, followed by an extension of 72°C for 7 min. Double-stranded DNA sequencing was carried out by using the Applied Biosystems 3730/3730*xl* chemistry in a 384-well format. Genotyping {#sec005} ---------- A multiplex PCR-restriction fragment length polymorphism (RFLP) method was used for VNTR genotyping. PCR followed standard protocols, with Fast PCR Master Mix (Fermentas, Glen Burnie, MD, U.S.A.) as described before\[[@pone.0218129.ref076]\]. For -14kb-VNTR, the PCR product was 913-bp, 733-bp, 652-bp, and 771-bp for 1-, 2-, 3-, and 4-repeat respectively. 5'VNTR was 426-bp, 486-bp, 546-bp, and 606-bp for 6-, 7-, 8-, and 9-repeat respectively. Int8VNTR was 291-bp and 321-bp for 5- and 6-repeat, and 3'VNTR was 441-bp, 481-bp and 521-bp for 9-, 10- and 11-repeat. These VNTR's PCR products were subject to agarose electrophoresis directly. The allele sequences of -10kb-pA (polyadenine) and simple sequence length polymorphism (SSLP) were verified by TA cloning (Invitrogen, Carlsbad, CA, U.S.A.) and DNA sequencing of TA clones (five to six per subject for a 95% confidence on biallelic polymorphisms). All Chromas sequencing graphs were refereed manually by two researchers independently for double-verification of sequence accuracy. All primers used are listed in [S1 Table](#pone.0218129.s006){ref-type="supplementary-material"}. Genetic analysis {#sec006} ---------------- Linkage disequilibrium (LD), expressed as D' and r^2^\[[@pone.0218129.ref077]\], was analyzed by using Haploview (<http://www.broadinstitute.org/haploview/haploview>) for biallelic polymorphisms and SHEsis (<http://analysis.bio-x.cn/myAnalysis.php>) for multiallelic polymorphism\[[@pone.0218129.ref078], [@pone.0218129.ref079]\]. Haplotyper (<http://www.people.fas.harvard.edu/~junliu/Haplo/click.html>) and PHASE (<http://www.stat.washington.edu/stephens/phase/download.2.0.2.html>) softwares were used for haplotype inference\[[@pone.0218129.ref080]\]. Recombination fraction was estimated by LDhat (<http://www.cecalc.ula.ve/BIOINFO/servicios/herr1/LDhat/readme.html>). To evaluate general chromosomal recombination, five populations of European ancestry in the 1000 Genomes Project (1KGP)\[[@pone.0218129.ref081], [@pone.0218129.ref082]\], including US Caucasians, Great Britain, Italy, Spain, and Finland, were combined for reliability to reveal recombination hotspots in *SLC6A3* by using the published FastEPRR protocol\[[@pone.0218129.ref083]\]. To localize genome wide association study (GWAS) markers in this gene, three SUDs GWAS datasets, all past their embargo periods, were downloaded from the dbGaP,\[[@pone.0218129.ref084]\] including Collaborative Study on the Genetics of Alcoholism \[[@pone.0218129.ref085]\] (COGA, phs000125.v1.p1), Study of Addiction: Genetics and Environment (SAGE, phs000092.v1.p1) and the Australian twin-family study of alcohol use disorder (OZALC, phs000181.v1.p1). Datasets were cleaned or quality-controlled extensively by using a published protocol,\[[@pone.0218129.ref086]\] followed by imputation as described before.\[[@pone.0218129.ref087]\] Basic manipulations of datasets used PLINK\[[@pone.0218129.ref088]\]. To estimate Tajima's D statistic in 30 unrelated COGA subjects, we calculated nucleotide diversity θ as the number of segregating sites, S, divided by a~1~, where $a_{1} = {\sum\limits_{i = 1}^{59}\frac{1}{i}} = 4.6632$, divided by the number of nucleotides sequenced. Heterogeneity π was estimated by $k = {\sum\limits_{j = 1}^{S}{2p_{j}}}(1 - p_{j})$, divided by 1--(1/59) = 0.983051, divided by the number of nucleotides sequenced, where *p*~j~ was the observed frequency of the j^th^ diallelic polymorphism. Statistic D was calculated using the θ and π, according to Tajima\[[@pone.0218129.ref089]\]. Phylogenic analysis was carried out by ClustalX (<http://www.bioinformatics.ubc.ca/resources/tools/index.php?name=clustalx>), with neighbor joining method for clustering and phylogenetic tree was displayed by TreeView (<http://taxonomy.zoology.gla.ac.uk/rod/treeview.html>) for cladistic analysis\[[@pone.0218129.ref090], [@pone.0218129.ref091]\]. Validations used MEGA7\[[@pone.0218129.ref092]\] under the Kimura 2 parameter with Gamma Distributed as rates among sites model for nucleotide substitutions. The relatedness results were shown in a radiation graphic, rather than in traditional trees, for better visualization. Results {#sec007} ======= Unique variation and ethnic differences {#sec008} --------------------------------------- We sequenced the 18 kb promoter region including the 16 kb 5' region, Exon 1, and the 2 kb Intron 1 and further genotyped the two VNTRs in Intron 8 and 3'UTR among 30 AAs and 30 EAs. A sample size of 30 allowed 80% confidence of detecting an allele with a frequency of 2.6% or 95% confidence of detecting an allele with a frequency of 4.9%. Based on more than 7,300 PCR reactions and sequencing of more than 2.4 mega-bases (Mb), the 60 subjects were found to carry 134 polymorphisms in the 18 kb 5' promoter regions and 20.1% of them were novel ([S2 Table](#pone.0218129.s007){ref-type="supplementary-material"}). The 5' promoter had two VNTRs, one novel at -14280 (-14kb-VNTR) and another at -11000 (5'VNTR/rs70957367); a novel -15.0kb-indel (insertion/deletion), one novel variable poly A at -10331 (-10kb-pA) and one novel SSLP at +1531 (use of the first base of Exon 1 as +1 and negative numbers for upstream of the promoter). All of these polymorphisms were all confirmed by TA cloning and DNA re-sequencing. The novel -14kb-VNTR had four alleles that were formed by multiple indels, di-nucleotide polymorphisms (DNPs) and single nucleotide polymorphisms (SNPs) (see [S1 Fig](#pone.0218129.s001){ref-type="supplementary-material"} *upper panel*). 5'VNTR had 7--9 repeats of imperfect 60 bp, whose primary sequence was reported previously\[[@pone.0218129.ref076]\]. -10kb-pA had 9--11 As. SSLP in Intron 1 had nine different length polymorphisms ([S1 Fig](#pone.0218129.s001){ref-type="supplementary-material"} *lower panel*). Int8VNTR had two alleles, 5 and 6 repeats of 30 bp and 3'VNTR had 9--11 repeats of 40 bp as previously discovered\[[@pone.0218129.ref008], [@pone.0218129.ref060], [@pone.0218129.ref093], [@pone.0218129.ref094]\]. For the sample size of 30 subjects, the AA group had 108 promoter polymorphisms and the EA group, 79 polymorphisms (see [Fig 1A](#pone.0218129.g001){ref-type="fig"} for distribution). The polymorphism density was 6.1/kb in AA and 4.5/kb in EA. Between AA and EA, 53 of the 134 polymorphisms were shared; the former carried 55 additional unique polymorphisms and the later carried 26 additional EA-specific ones. +24G/T (rs45611137) was the only Exon 1 mutant (minor allele frequency (MAF): 0.0167), present in the AA, not in the EA cohort. {#pone.0218129.g001} Recombination rate (*ρ*) {#sec009} ------------------------ The core promoter region at -2.3 kb and Intron 1 at +1.6 kb both displayed higher recombination rates than upstream promoter regions ([Fig 1B](#pone.0218129.g001){ref-type="fig"}). Two other hot spots were at -10.4 kb and -12.6 kb. In the core promoter region, ***ρ*** was 3.57 in AA and 2.55 in EA, representing the hottest region for recombination in the *SLC6A3* promoter. This rate decreased sharply as the distance from the -2.3 kb spot increased towards either side (see *Insert* in [Fig 1B](#pone.0218129.g001){ref-type="fig"}). In this core promoter region, the average ***ρ*** value was 1.48 in the AA cohort, 1.4-fold higher of the ***ρ*** value in the EA cohort. The Intron 1 had another hotspot. The largest difference in recombination rate between the two cohorts was three-fold at -15.7 kb, the 5' end of the promoter regions where ***ρ*** value was 1.29 in the AA and 0.43 in the EA cohort. The average ***ρ*** value was 1.26 in the AA, 1.3-fold higher of the ***ρ*** value in the EA cohort. Although we have not stratified the analysis by sex, the findings persistently pointed to four recombination hotspots in the *SLC6A3* promoter of the COGA cohorts. LD {#sec010} -- The overall AA LD was low across the entire regulatory regions (average D' = 0.8214 and square of the Pearson correlation coefficients r^2^ = 0.1283). There was a weak 6 kb block from -6234A/G ([rs1354139](http://www.ncbi.nlm.nih.gov/SNP/snp_ref.cgi?rs=1354139)) to -68T/A ([rs2975226](http://www.ncbi.nlm.nih.gov/SNP/snp_ref.cgi?rs=2975226)) (D' = 0.8487 and r^2^ = 0.1621, see [Fig 1C](#pone.0218129.g001){ref-type="fig"} *left panel*). Intron 1 displayed relatively weak LD within the Intron (D' = 0.8414 and r^2^ = 0.0964). In particular, the 5' end of Intron 1 from +24G/T (rs45611137) to +579G/A ([rs28382214](http://www.ncbi.nlm.nih.gov/SNP/snp_ref.cgi?rs=28382214)) represented a major subregion of weak LD (D' 0.8938 and r^2^ 0.0438). Int8VNTR displayed low LD (D' 0.6126 and r^2^ 0.06524) with the upstream polymorphisms. Int8VNTR had perfect LD with 1787G/A ([rs11564757](http://www.ncbi.nlm.nih.gov/SNP/snp_ref.cgi?rs=28382219)) (D' and r^2^ both = 1) and high LD with -15.0kb-indel, -10250C/T (rs72717506), -9701C/T ([rs10063727](http://www.ncbi.nlm.nih.gov/SNP/snp_ref.cgi?rs=10063727)), -4913A/G ([rs10079467](http://www.ncbi.nlm.nih.gov/SNP/snp_ref.cgi?rs=10079467)), -1675T/C ([rs11564751](http://www.ncbi.nlm.nih.gov/SNP/snp_ref.cgi?rs=11564751)) and -1479G/T ([rs6413429](http://www.ncbi.nlm.nih.gov/SNP/snp_ref.cgi?rs=6413429)) (D' = 1 and r^2^ = 0.4). The overall EA LD (D' = 0.9119, r^2^ = 0.1876) was bit higher than the AA LD, consistent with the lower recombination rate in EA than in AA. There was a 10 kb block covering from -12499 to -2315, which was located towards the 5' end, approximately 4 kb up compared to the location of the AA weak block (indicated by asterisks in [Fig 1C](#pone.0218129.g001){ref-type="fig"} *right panel*). The average LD within this block was D' = 0.9458 and r^2^ = 0.2985. Polymorphisms at -5487, -6731, -3182, and -2600 displayed weak LD, D' = 0.5333 and r^2^ = 0.06459 on average within this block. There is no LD between -10397G/A ([rs6860992](http://www.ncbi.nlm.nih.gov/SNP/snp_ref.cgi?rs=6860992)) and -4825T/C (rs188332761). Weak LD regions surrounded the transcription start site (TSS), covering the regions from -2296 to +1298 (D' = 0.9507, r^2^ = 0.0835). Similar to what was observed in the AAs, the 5' end of the EA Intron 1 also displayed weak LD. Different from the AAs was the 3' end of Intron 1 that displayed strong LD within the region or with upstream regions. Again, Int8VNTR displayed weak LD (D' = 0.6255, r^2^ = 0.08734) with upstream polymorphisms. However, Int8VNTR displayed strong LD with -15.0kb-indel, -10250C/T, -1675T/C ([rs11564751](http://www.ncbi.nlm.nih.gov/SNP/snp_ref.cgi?rs=11564751)) and +1787G/A (rs11564757) (D' = 1 and r^2^ = 0.4). When we stratified the 30 subjects by phenotypes, the control LD (D' = 0.8774) was stronger than the case (D' = 0.7992) in the AAs and differences in multifocal pairs were quite significant, by comparing the *upper panels* of [S2 Fig](#pone.0218129.s002){ref-type="supplementary-material"}. This phenotype-related difference in LD was also observed in EAs, especially for core promoter-Intron 1 *versus* the upstream regions ([S2 Fig](#pone.0218129.s002){ref-type="supplementary-material"}). Therefore, SUDs-related changes in LD might represent a major difference in *SLC6A3* genetics between AA and EA. In addition to the biallelic polymorphisms, we also analyzed LD by using the five multiallelic polymorphisms, including -14kb-VNTR, 5'VNTR, -10kb-pA, SSLP and 3'VNTR. Consistently, the AA LD was weaker than the EA LD ([S3 Fig](#pone.0218129.s003){ref-type="supplementary-material"} *far right panels*). 3'VNTR displayed weak LD with upstream markers: r^2^ = 0.049--0.170 in the AAs and 0.115--0.200 in the EAs. Stratification with phenotypes showed that LD in these patients was stronger than in the controls for both populations, based on color intensity ([S3 Fig](#pone.0218129.s003){ref-type="supplementary-material"} *left panels*). Genetic selection of polymorphisms {#sec011} ---------------------------------- Genetic selection may indicate functionality of polymorphisms. We utilized Tajima's statistic D to evaluate these genetic processes. The average D values for the entire 18 kb promoter were slightly positive (0.48) for EA and slightly negative (-0.29) for AA. However, when we stratified the polymorphisms by variation types, significant D values were revealed. In the AA cohort, VNTRs and SSLP both had significantly positive D values. In particular, the Intron 1 SSLP (heterogeneity or h = 0.7556), 5'VNTR (h = 0.5915) and -10kb-pA (h = 0.6228) in the 5' region had D values of 3.28, 2.67 and 2.48. Between two DNPs, the 5' DNP had a negative D value but the Intron 1 DNP (DNPi) had a positive D (-1.09 and 1.47). The Intron 1 SSLP and DNP were the only types with positive D values in this region. Neither Int8VNTR nor 3'VNTR had significant D values ([S4A Fig](#pone.0218129.s004){ref-type="supplementary-material"}). In the EA cohort, all types of variations (-10kb-pA, DNP, VNTR and SSLP) except SNPs had significantly positive D values (2.37, 2.32, 2.45, 4.57 *versus* -0.02857 for SNPs). This EA SSLP displayed the highest heterogeneity of 0.9647 among all of the polymorphisms. The positive DNP D was attributable to both DNPs in the 5' region and in Intron 1. The positive VNTR D was attributable to 5'VNTR (h = 0.6395) in the 5' region. Unlike the AA D values, SNPs and -15.0kb-indel were the only types that showed negative D values in the EA. The two VNTRs at the 3' side of the gene showed positive but insignificant D values ([S4B Fig](#pone.0218129.s004){ref-type="supplementary-material"}). It was noticed that SSLP, 5'VNTR and -10kb-pA were all positively selected in both populations. Haplotypic relatedness {#sec012} ---------------------- The AA and EA cohorts carried 108 and 79 polymorphisms plus Int8VNTR and 3'VNTR, and these polymorphisms constituted 57 different AA haplotypes (\#1--57; \#1-\#29 from the patients and \#30-\#57 from the controls; \#2, \#34 and \#40 each occurred twice in the controls) and 57 different EA haplotypes (\#58-\#114; \#58-\#87 from the patients and \#88-\#114 from the controls; \#62 and \#65 each occurred once in a control and once in a patients and \#107 occurred twice in the two controls). That is, three haplotypes each occurred twice and 54 other haplotypes each occurred only once in each cohort of 30 subjects. None of the AA haplotypes were present in the EA and *vice versa*, suggesting that *SLC6A3* carries great diversity not only between different ethnicities but also within the same population. Co-analyses of the AA and EA haplotypes for relatedness could help understand whether or not the *SLC6A3* haplotypes co-evolved independently between AA and EA during human history. Therefore, we generated a phylogenic tree containing all 114 haplotypes ([Fig 2](#pone.0218129.g002){ref-type="fig"}). It turned out that the AA and EA haplotypes were mixed up in terms of relatedness. Overall, the top half of the phylogenetic tree represented a major mosaic of the AA and EA haplotypes, including pairs of 10/61 and 2/68. {#pone.0218129.g002} Haplotype analysis of three of the EA polymorphisms -14kb-VNTR~2/4~, Int8VNTR~5/6~ and 3'VNTR~9/10/11~ showed that 4-6-10 occurred eight times in the patients but did not occur in controls, with a nominal *p* value of 0.0046 by Fisher's Exact Tests (odds ratio (OR) 23.0; 95% confidence interval (CI) 1.26--420.39). The consistencies in allele-specificity of association tendency and the suggestive haplotypic association warrant future investigation of these potential risk factors in large samples. Recombination hotspots in general populations of European ancestry {#sec013} ------------------------------------------------------------------ To confirm the recombination results from the COGA cohorts, we consulted with the 1KGP and analyzed the entire 70 kb chromosomal region of *SLC6A3*. We combined five European ethnicities, including EA (99 persons), Italy (108 persons), Spain (107 persons), Finland (99 persons), Great Britain (92 persons), for a total of 505 persons. Several hotspots were revealed but the most significant one was in Intron 2 and next to Exon 2. The ones in the core promoter and in Intron 1, revealed by the COGA subjects, were confirmed. Others were located in Introns 4, 7, 8, 11 and 14. None of them were localized to any coding regions. The distal promoter regions were relatively quiet, with six minor ones ([Fig 3](#pone.0218129.g003){ref-type="fig"} *upper part*). {#pone.0218129.g003} Discussion {#sec014} ========== It is important to uncover novel, potentially functional polymorphisms and distinct haplotypes, and recombination hotspots because *SLC6A3* activity can be haplotype-dependent partly due to *cis*-antagonism between 5' and 3' sides of *SLC6A3*\[[@pone.0218129.ref095]\]. The most significant findings from this deep-sequencing study included discovery of novel and selected polymorphisms and the presence of recombination hotspots throughout *SLC6A3*. Although association analysis of SUDs and *SLC6A3* haplotypes was not a primary purpose here, five implications are highlighted as follows. Great diversity {#sec015} --------------- The levels of variation including novel polymorphisms in the regulatory regions have never been expected. In either of the two cohorts, 60 chromosomes were represented by 57 regulatory haplotypes. The haplotypic diversity was attributable to both high density of polymorphisms and several recombination hotspots. Because of the genetic diversity, the overall LD was generally low, based on the r^2^ values of \< 0.2 across the regulatory regions and the previously observed balancing selection of promoter haplotypes disappeared here\[[@pone.0218129.ref096], [@pone.0218129.ref097]\]. Such information explains the inconsistent findings on associations between *SLC6A3* and ADHD from genetic studies that used markers located in "random" regions for different populations\[[@pone.0218129.ref098]--[@pone.0218129.ref101]\]. The rationale of choosing *SLC6A3* markers in most of the studies suffered from the lack of a mechanistic understanding of regulatory genetics in ethnic *SLC6A3* and because of that, the obtained results were more marker- and sample-dependent than phenotype-dependent. Our systemic observations suggest that use of markers that cover more regulatory regions could result in more consistent and positive associations, a notion that has been supported by two studies\[[@pone.0218129.ref042], [@pone.0218129.ref046]\]. Convergence {#sec016} ----------- Phylogenetic analysis indicated for the first time the convergence of the *SLC6A3* regulatory regions between the AAs and the EAs, which means that DNA sequence variation occurred in this promoter independently in the two ethnicities. Despite one main EA clade and three localized minor AA clades, the majority of the haplotypes between two populations were mixed by being localized to many subclades of various sizes. Given the observations that two thirds of the EA polymorphisms were shared by the AA and that the four 5' recombination hotspots were shared between the two ethnicities, it is a reasonable assumption that *SLC6A3* evolves in a largely context-dependent manner. In addition, the mosaic patterns could also be attributable to history of gene flow between European male and African female ancestry \[[@pone.0218129.ref102]\]. Finally, this phylogenetic tree indicates a tendency of bidirectional diversion among the 114 18kb-haplotypes. By analyzing twenty six 1KGP populations together (thousands of haplotypes), a much stronger bidirectional diversion of the 18kb promoter was observed (without geographic correlation, data not shown), validating the COGA subjects-based finding and implying again a functional correlation between genetics of the *SLC6A3* promoter and related phenotypes\[[@pone.0218129.ref103]\]. Ethnicity {#sec017} --------- As [Fig 1C](#pone.0218129.g001){ref-type="fig"} shows, the AAs had clearly overall lower LD than the EAs. Specifically, there were three lines of evidence showing genetic differences in *SLC6A3* between the AAs and the EAs. First, the AA gene had many more common polymorphisms ([Fig 1A](#pone.0218129.g001){ref-type="fig"} *upper panel*) than the EA gene ([Fig 1A](#pone.0218129.g001){ref-type="fig"} *lower panel*), noticing that most polymorphisms appeared clustered around the recombination hotspots (indicated as inverted triangles). Second, the AA gene lacked polymorphisms with MAF \> 0.3 but the EA gene lacked polymorphisms with 0.3 \< MAF \< 0.45 (see gray areas in [Fig 1A](#pone.0218129.g001){ref-type="fig"}), suggesting that the AA mutations were less likely to be selected and that the commonest EA variants had been selected. The lack of mutation fixation might have been compensated by the higher mutation frequency in the AA *SLC6A3*. The third evidence was that a haplotype block covered the core promoter region in the AA gene but was larger and covered the center of the 5' region in the EA gene. These genetic differences between the two populations are consistent with the fact that dopamine-related brain function has ethnic differences and suggest that association study needs to use different sets of markers for different populations. These potentially functional polymorphisms might have contributed to the observed genetic selection of particular polymorphisms (see upper braces in [Fig 1A](#pone.0218129.g001){ref-type="fig"}). Selection {#sec018} --------- Despite the haplotypic diversity, selections of polymorphisms were observed. In both populations, polymorphisms with MAF of 0.31--0.40 were de-selected. For the most common or ethnicity-specific polymorphisms (MAF of \> 0.40), de-selection was the most significant in the AAs whereas selection was the most significant in the EAs, representing the largest difference between the two ethnicities. Tajima's D statistic has been used in trying to identify functional polymorphisms\[[@pone.0218129.ref104], [@pone.0218129.ref105]\] and indicates here the selection of some polymorphisms in both cohorts. We now report indeed that Tajima's D results are consistent with the functionality of DNPi because the EA DNPi was positively selected and this DNPi indeed mediates a long non-coding RNA (lncRNA) regulation of the *SLC6A3* promoter\[[@pone.0218129.ref075]\]. Significantly, SSLP in Intron 1, 5'VNTR and -10kb-pA were all selected in both populations, consistent with the fact that *SLC6A3 cis*-acting elements such as the 5 kb super enhancer (5KSE) can be localized to distal 5' regions\[[@pone.0218129.ref095]\] and that 5'VNTR was positively correlated with the mRNA levels in postmortem dopamine neurons\[[@pone.0218129.ref076]\]. These positive selections thus generate testable hypotheses for functionality and associations. Localized recombination {#sec019} ----------------------- *SLC6A3* is located at 1.4 Mb, near a telomere of chr5, and harbors several hotspots, consistent with previous observations that regions near telomeres tend to have higher recombination rates than those near centromeres\[[@pone.0218129.ref106]--[@pone.0218129.ref109]\]. The identified recombination hotspots not only help explain negative findings on *SLC6A3* in previous GWAS but also guide selection of new association markers. [Fig 3](#pone.0218129.g003){ref-type="fig"} *lower part* indicates genetic markers used by three previous GWAS, in reference to the distribution of the recombination hotspots. It is noticed that all of these markers are separated from the core promoter by recombination hotspots. The two widely used VNTRs, located in last exon and Intron 8, are each separated from the promoter regions (also as previously suggested \[[@pone.0218129.ref110]\]) by several recombination hotspots, suggesting that these markers are unable to capture information effectively on varying promoter activity\[[@pone.0218129.ref051]\]. As such, the previous association studies have not effectively used genetic markers for this gene yet. The recently identified functional DNPi (indicated by an asterisk)\[[@pone.0218129.ref075]\] is buried in two hotspots, suggesting that none of the previously used markers could capture the DNPi-related information. We have tried to use either the 1KGP or the COGA genotype as the templates to impute DNPi in the other GWAS datasets but failed completely. This is partly explained by the fact that DNPi has low LD with other markers and is located in a small, isolated haplotype block ([S5 Fig](#pone.0218129.s005){ref-type="supplementary-material"}). When sample sizes get large enough and using the 1KGP with strong LD as the template ([S5 Fig](#pone.0218129.s005){ref-type="supplementary-material"}), DNPi could be imputed but the results would not be reliable for cohorts of interest\[[@pone.0218129.ref111]\]. Therefore, this marker should be experimentally typed for accurate association information. We acknowledge that the main limitations of this study include small sample sizes which aimed at common variants only, limited number of ethnicities and mainly the promoter regions. Including the patients might affect any ethnic comparisons and LD might cause overestimation of distances in the phylogenetic tree. The results, however, should provide guidance for rational selection of genetic markers for functional and association studies, as we have done before\[[@pone.0218129.ref051], [@pone.0218129.ref076]\], and for in-depth interrogations of the entire gene in more ethnicities. Conclusions {#sec020} =========== Extensive DNA sequence variations not only around the core promoter but also in other distal regulatory regions may work in concert or in haplotypes and influence dopamine-related individuality and diseases. Such genetic diversity of *SLC6A3* may help explain the elusiveness of previous association findings with the classical markers in the 3' side. The findings also lay a foundation for a better understanding of the roles that the polymorphic *SLC6A3* plays in human brain. Supporting information {#sec021} ====================== ###### Details of multiple alleles of -14kb-VNTR (*upper panel*) and Intron 1 SSLP (*lower panel*). (PDF) ###### Click here for additional data file. ###### Haploview-based linkage disequilibrium (LD) in *SLC6A3* regulatory regions by phenotypes of the two COGA cohorts. (PDF) ###### Click here for additional data file. ###### SHEsis-based LD by multi-allelic polymorphisms in the COGA cohorts. (PDF) ###### Click here for additional data file. ###### Genetic selection of polymorphisms by Tajima's statistic D for AA (a) and EA (b). (PDF) ###### Click here for additional data file. ###### Significant difference in 18kb *SLC6A3* promoter LD (D') between 1KGP CEU (upper) and COGA sample (30 controls and 30 patients with SUDs). (PDF) ###### Click here for additional data file. ###### PCR primers used. (XLSX) ###### Click here for additional data file. ###### Common and unique alleles in COGA samples. (PDF) ###### Click here for additional data file. We thank NIH for granting ZL access to dbGaP, under project \#1542: \"Functional genomics of dopamine-related diseases\", all the subjects and investigators for contributing to the COGA projects and NIAAA for approving the use of de-identified genomic DNAs. We also thank Drs. Haipeng Li and Zi-Qian Hao for help on the use of FastEPRR. [^1]: **Competing Interests:**The authors have declared that no competing interests exist.

Background {#Sec1} ========== The effects of climate change are expected to result in frequent and extreme weather events causing major damage to crop production \[[@CR1], [@CR2]\]. Plants respond to these changes (abiotic stress) by developing different resilience mechanisms at the phenotypic, physiological and molecular levels \[[@CR3]\]. To improve plant response, microRNAs provide an opportunity to mend alfalfa traits \[[@CR4]\]. MicroRNAs are small RNAs of approximately 16--26 nucleotides in length that regulate gene expression at the posttranscriptional level in a sequence-specific manner \[[@CR5]\]. Of the hundreds of microRNAs \[[@CR6]\], microRNA156 (miR156) is highly conserved in plants, where it functions by down-regulating a group of SQUAMOSA-PROMOTER BINDING PROTEIN-LIKE (SPL*)* transcription factors \[[@CR7]--[@CR9]\]. There are at least eight members (a to h) of miR156 in *Arabidopsis thaliana*, with g and h unique to this species. A smaller number of miR156 members (a to f) have been discovered in other plant species, including *Medicago truncatula* \[[@CR10]\]. SPLs regulate a network of downstream genes affecting plant development and physiology by binding to gene promoters at a consensus DNA sequence NNGTACR (where N = any nucleotide, R = A or G) known as the SPL Binding Domain (SBD) \[[@CR11]--[@CR14]\]. In *Arabidopsis,* miR156 regulates 10 out of 16 *SPL*s, affecting various aspects of plant growth and development \[[@CR15]\], whereas in alfalfa, miR156 regulates at least seven *SPL*s (*SPL2,3,4,6,9,12* and *13*) \[[@CR8]\]. Despite the conservation of miR156 among plant species, some of its regulation outputs are species-specific \[[@CR9], [@CR13], [@CR16]\]. We previously showed that overexpression of miR156 in alfalfa delays flowering time, enhances root nodulation, and improves vegetative and root growth \[[@CR7], [@CR13]\]. Many of these traits are associated with abiotic stress tolerance \[[@CR17], [@CR18]\]. Moreover, overexpression of miR156d was shown to improve alfalfa's tolerance to heat \[[@CR19]\], salinity \[[@CR20]\] and drought stress \[[@CR21]\]. miR156-mediated silencing of *SPL2, SPL9* and *SPL11* improved heat, salt and drought stress resilience in *Arabidopsis* and rice \[[@CR22], [@CR23]\]. *Arabidopsis* mutants with increased miR156 expression silenced *SPL9,* and enhanced expression of *DIHYDROFLAVONOL-4-REDUCTASE* (*DFR*) and *PRODUCTION OF ANTHOCYANIN PIGMENT 1* (*PAP1*), which resulted in increased anthocyanin accumulation and improved stress tolerance \[[@CR22]\]. The enhancement of anthocyanins and proanthocanidins is regulated by transcription factors such as WD40, MYB and bHLH \[[@CR24], [@CR25]\]. These secondary metabolites scavenge free radicals during plant abiotic stress \[[@CR26]--[@CR28]\] and function in a coordinated manner with transient stress-related primary metabolites such as proline, galactinol, raffinose and gamma-aminobutyric-acid (GABA) to alleviate stress symptoms \[[@CR26], [@CR29]\]. We recently reported that drought stress enhances miR156 expression to improve alfalfa's resilience to this stress by increasing leaf gas exchange and abscisic acid (ABA), while reducing water loss \[[@CR21]\]. Despite these findings, our understanding of how the miR156/SPL network regulates downstream genes such as *DFR* and *WD40--1* to affect stress tolerance in alfalfa is unknown, especially as it relates to drought stress and secondary metabolism. In this study, we investigated the mechanism of how miR156 regulates drought stress response in alfalfa. To that end, we analyzed miR156 over-expressors, *SPL13*-silenced genotypes, *WD40--1* over-expressors and *WD40--*1 RNAi silenced genotypes at the metabolomic, transcriptomic and physiological levels. Moreover, we investigated the binding of SPL13 to the *DFR* promoter to regulate flavonoid biosynthesis. The findings from this report would be useful to understand the mechanisms deployed by miR156 in regulating drought stress and could be used as a tool in marker-assisted breeding to improve alfalfa and potentially other crops. Results {#Sec2} ======= Enhanced miR156 expression improves drought tolerance by altering root architecture and water holding capacity {#Sec3} -------------------------------------------------------------------------------------------------------------- To determine drought stress regulation by miR156, we used one-month-old miR156OE alfalfa plants with low (A8a = 0.5), moderate (A8 = 1.5) and higher (A11 = 2.5) relative miR156 expression levels than the empty vector (EV) \[[@CR13]\] grown under drought and well-watered conditions. Root weight, root length, stem basal width and fresh root-to-shoot weight ratios were affected by drought stress depending on the genotype (Fig. [1](#Fig1){ref-type="fig"}, Additional file [2](#MOESM2){ref-type="media"}: Table S5.1). Relative to EV, A8a had significantly longer roots and increased root biomass (Fig. [1](#Fig1){ref-type="fig"}a), with increases of root length up to 1.8-fold (Fig. [1](#Fig1){ref-type="fig"}b) and 1.7-fold in root weight (Fig. [1](#Fig1){ref-type="fig"}c). The increment of root biomass in A8a was the result of longer roots rather than short and thicker roots (Fig. [1](#Fig1){ref-type="fig"}b,c). To understand if the improved root architecture affected plant water potential, we measured leaf water potential \[[@CR30]\] and changes in the lower stem diameter before and after drought \[[@CR31]--[@CR33]\]. MiR156OE genotypes, A8a and A8, maintained a higher leaf water potential (Fig. [1](#Fig1){ref-type="fig"}f) and also either maintained or increased basal stem diameter (Fig. [1](#Fig1){ref-type="fig"}d) while EV plants showed a reduction over the 2 weeks of stress. The unchanged basal stem diameter was accompanied by an increase in root/shoot biomass ratio in A8a and A8 (Fig. [1](#Fig1){ref-type="fig"}e). Fig. 1Effects of miR156 overexpression on drought tolerance and physiological responses in alfalfa. **a** Roots of EV and miR156OE plants under drought stress; **b** root length; **c** root weight; **d** stem basal diameter change under drought; **e** root/shoot biomass ratio; **f** leaf water potential; **g** *Vcmax*, maximum rate of rubisco carboxylase activity; **h** *Jmax*, maximum rate of photosynthetic electron transport; **i** dark adapted chlorophyll florescence, Fv/Fm, and **j** photosynthetic assimilation rate in well-watered (control) and drought stressed plants. Values are sample means ± SE, *n* = 4 individual plants except in '**d**', '**e**', '**f**', '**i**', '**j**' where *n* = 5. ANOVA *p* values are provided in Additional file [2](#MOESM2){ref-type="media"}: Table S5.1. Significant difference in Post hoc Tukey multiple comparisons test is indicated with different letters. Letters in multiple time point data of '**i**' and '**j**' is analyzed separately miR156 overexpression affects photosynthesis parameters {#Sec4} ------------------------------------------------------- Since drought stress negatively affects photosynthesis parameters \[[@CR34]\], we investigated this effect in miR156OE and EV plants. Accordingly, photosystem II (PS II) chlorophyll fluorescence, Fv/Fm ratio, was measured. Fv/Fm was significantly affected by genotype, drought exposure time, and a combination of both (Additional file [2](#MOESM2){ref-type="media"}: Table S5.1). MiR156OE plants maintained higher levels of Fv/Fm ratio (0.75) at later stages (day 11 and 14) comparable to unstressed plants while EV plants showed a gradual reduction to 0.69 after 14 days of drought (Fig. [1](#Fig1){ref-type="fig"}i). Furthermore, photosynthesis assimilation rate was significantly affected by genotype and the duration of drought exposure (Additional file [2](#MOESM2){ref-type="media"}: Table S5.1). Our data revealed that during drought stress the photosynthetic assimilation rate was higher in A8, gradually decreased in A8a, and further decreased in A11except on day 14 when it was greater than in EV (Fig. [1](#Fig1){ref-type="fig"}j). Moreover, the maximum rate of rubisco carboxylase activity *Vcmax* was maintained at a relatively higher level in A8a and A8 plants while a comparably significant reduction (64--75%) was observed in EV and A11 plants during drought stress (Fig. [1](#Fig1){ref-type="fig"}g). In line with this, maximum photosynthetic electron transport rate *Jmax* was also maintained at higher levels in A8a and A8 during drought stress while it was reduced (64%) in EV and A11 plants (Fig. [1](#Fig1){ref-type="fig"}h). miR156OE plants accumulate anthocyanin and other stress-related secondary metabolites under drought {#Sec5} --------------------------------------------------------------------------------------------------- Using more than 4000 metabolite features, a Principal Component Analysis (PCA) plot of LCMS-based metabolite profiles depicted a distinct difference between drought-treated EV and miR156OE stem tissues (Fig. [2](#Fig2){ref-type="fig"}a). These metabolite features are spectral data generated from metabolites \[[@CR35], [@CR36]\]. Principal component-1 (PC-1) contributed 32.7% of the variance and clearly separated EV and miR156OE genotypes stem samples while principal component-2 (PC-2) accounted for 13% of the variance. Fig. 2LCMS-based metabolite profiling illustrates distinct profile in miR156OE genotypes during drought stress. **a** Principal component analysis of metabolite profile in stem, **b** leaf, and **c** root tissues under drought stress; **d** metabolite features that are significantly different at *p* \< 0.01 from EV plants in tissues of stem, **e** leaf, and **f** root tissues; **g** proportion of metabolite features that are significantly increased (≥ 1.5 log 2 fold change) or decreased (≤ − 1.5 log 2 fold change) relative to EV under drought stress; **h** relative levels of anthocyanin metabolites of peonidin 3-O-glucoside, PG, and **i** delphinidin 3-O-(6″-acetyl)-glucoside, DAG. The relative abundance of metabolites is normalized to an internal standard. Values are sample means ± SE, *n* = 4 individual plants. ANOVA p values are provided in Additional file [2](#MOESM2){ref-type="media"}: Table S5. 4. Significant difference in Post hoc Tukey multiple comparisons test is indicated with different letters Unlike stem tissues (Fig. [2](#Fig2){ref-type="fig"}a), roots possessed a differential metabolite features profile for all genotypes with PC-1 and PC-2 variance of 19.21 and 11.05%, respectively (Fig. [2](#Fig2){ref-type="fig"}c). On the other hand, leaves of A8a and EV were metabolically closer (Fig. [2](#Fig2){ref-type="fig"}b), whereas the higher miR156 expressor, A11, possessed a different metabolic profile, with PC-1 and PC-2 variance of 18.85 and 12.96%, respectively. Based on their significance level and fold change relative to EV, the numbers of metabolite features common or different in stem, leaf and root of miR156OE genotypes under drought stress are presented in Fig. [2](#Fig2){ref-type="fig"}d, e and f, respectively. Figure [2](#Fig2){ref-type="fig"}d reveals a communal relatively high number of differentially abundant metabolite features (770) between stems of miR156OE and EV plants. The majority (85.1, 81.1, and 73.4% for A8a, A8, and A11, respectively) of the differentially abundant stem metabolites are significantly increased in comparison to EV stem (Fig. [2](#Fig2){ref-type="fig"}g). The differential metabolite feature between miR156OE and EV is likely associated with the commonly observed pigmentation of the stem basal internode in miR156OE plants (Additional file [2](#MOESM2){ref-type="media"}: Figure S1). Drought stress induces production of reactive oxygen species (ROS) \[[@CR37]\], and plants employ several strategies, including secondary metabolite antioxidants to decrease ROS \[[@CR38]\]. Of the many secondary metabolites used by plants as antioxidants, anthocyanins are well documented \[[@CR39], [@CR40]\]. Here, levels of anthocyanins such as peonidin 3-O-glucoside (PG) and delphinidin 3-O-(6″-acetyl)-glucoside (DAG) were significantly affected by genotype and tissue (Additional file [2](#MOESM2){ref-type="media"}: Table S5.4). LCMS-based metabolite profiling showed anthocyanins and other ROS scavenging phenolic metabolites were increased mainly in stems of low-to-medium miR156 expressors (A8a and A8), although PG was also increased in A11 (Fig. [2](#Fig2){ref-type="fig"}h, i and Additional file [2](#MOESM2){ref-type="media"}: Table S2). Acylation of the sugar moiety in anthocyanins increases metabolite stability \[[@CR41], [@CR42]\]. It remains to be determined whether such acylation is a factor in leaves of A8 having higher levels of DAG relative to A11 and EV resulting in improved drought tolerance (Fig. [2](#Fig2){ref-type="fig"}i). Alfalfa plants expressing moderate levels of miR156 accumulate stress-related primary metabolites under drought {#Sec6} --------------------------------------------------------------------------------------------------------------- Plants coordinate primary and secondary metabolites for tight metabolite regulation and stress response \[[@CR27], [@CR28], [@CR43]\]. Hence, we used GCMS for analysis of primary metabolites to determine their levels during drought stress. Results indicated that metabolite levels were governed by tissue and genotype (Additional file [2](#MOESM2){ref-type="media"}: Table S5.5). In general, the relative abundance of proteinogenic amino acids was higher in leaf tissues of moderate miR156OE plants, but reduced in highly overexpressing A11 plants (Fig. [3](#Fig3){ref-type="fig"} and Additional file [2](#MOESM2){ref-type="media"}: Table S3). With the exception of valine, which showed no significant differences among stem, root and leaf tissues, levels of proteinogenic amino acids were significantly affected by tissue type and a combination of genotype and tissue (Additional file [2](#MOESM2){ref-type="media"}: Table S5.5). Alanine, asparagine, glycine and tryptophan showed a relatively higher abundance in leaves of A8 (Fig. [3](#Fig3){ref-type="fig"}a). Interestingly, proline, which functions as an osmolyte to maintain plant water potential \[[@CR26]\], was significantly increased in root tissues of A8a, comparable in A8 but was reduced in leaf, stem and root tissues of A11 compared to EV plants (Fig. [3](#Fig3){ref-type="fig"}b). Fig. 3GCMS-based primary metabolite profiling demonstrates drought stress tolerance strategies by miR156. **a** Relative levels of proteinogenic amino acids in leaf tissues during drought stress: alanine, asparagine, aspartate, glycine, isoleucine, serine, threonine, tryptophan and valine; **b** relative levels of metabolites from the γ-aminobutyric acid (GABA) shunt in leaf, stem and root tissues of proline, and **c** GABA; **d** relative levels of sugars from tissues of leaf, stem and root as fructose, and **e** arabinose under drought stress. Values are sample means ± SE, *n* = 4 individual plants. ANOVA p values are provided in Additional file [2](#MOESM2){ref-type="media"}: Table S5.5. Significant difference in Post hoc Tukey multiple comparisons test is indicated with different letters Levels of gamma-aminobutyric acid, GABA, a stress-responsive metabolite that mediates carbon to nitrogen balance between glutamate and succinate in the TCA cycle \[[@CR29]\], were enhanced in root tissues of A8 and A8a (Fig. [3](#Fig3){ref-type="fig"}c). The higher miR156 over-expressor, A11, on the other hand, reduced GABA levels in all tissues as compared to EV (Fig. [3](#Fig3){ref-type="fig"}c). An increased level of fructose, one of the main sugar sources for the carbon skeleton of downstream metabolites and a source of energy, was observed in leaf tissues of A8 while its levels were unchanged in stems and roots (Fig. [3](#Fig3){ref-type="fig"}d). On the other hand, A11 had variable levels of fructose (Fig. [3](#Fig3){ref-type="fig"}d), with levels being reduced in stems but comparable in roots. Conversion of carbon sources from sugars into the downstream pathways including glycolysis and pentose phosphate pathway (PPP) is of great importance in stress response and tolerance \[[@CR44], [@CR45]\]. Arabinose, an important component of cell wall polysaccharides, PPP, and a major component of glycoproteins and arabinogalactan proteins, had enhanced levels in stems while it was unchanged in leaf and roots of A8a and A8 while reduced in roots of A11 compared to EV (Fig. [3](#Fig3){ref-type="fig"}e and Additional file [2](#MOESM2){ref-type="media"}: Table S5.5). A complete list of annotated metabolites using GCMS analysis is presented in Additional file [2](#MOESM2){ref-type="media"}: Table S3. Overexpression of miR156 affects expression of photosynthesis and flavonoid genes {#Sec7} --------------------------------------------------------------------------------- Our physiological and metabolite profiling analysis showed that alfalfa plants overexpressing miR156 at low-to-moderate levels (A8a and A8) have higher anthocyanin levels (Fig. [2](#Fig2){ref-type="fig"}h,i) and maintained higher photosynthetic efficiency during drought stress (Fig. [1](#Fig1){ref-type="fig"}g-k). We, therefore, investigated if these are regulated at the molecular level by determining relative transcript levels of genes involved in the flavonoid and photosynthetic pathways. Genotype, tissue and their interaction have a significant impact on the transcript levels of flavonoid biosynthesis *DFR* and *MYB112* genes, although *MYB112* showed little difference between tissues (Additional file [2](#MOESM2){ref-type="media"}: Table S5.6). Accordingly, higher transcript levels of *DFR* and *MYB112* were observed in stem and leaf tissues of at least some miR156OE plants. DFR*,* which catalyses the conversion of dihydroflavonol to leucoanthocyanidin, had two- to 15-fold higher transcription in miR156OE leaf tissues compared to EV (Fig. [4](#Fig4){ref-type="fig"}a). *DFR* transcription was also 25 to 35-fold higher in miR156OE root samples. *MYB112* encodes a transcription factor that regulates flavonoid biosynthesis \[[@CR46]\]. Its transcript level was five- to 19 times higher in leaf tissues of miR156OE compared to EV while a four-fold higher expression level was observed in miR156OE stem tissues regardless of genotype (Fig. [4](#Fig4){ref-type="fig"}b). A slight increment in the expression level of *WD40--1* (1.9-fold), a transcription factor in the phenylpropanoid pathway, was observed in A8 root tissues while it was decreased in stem and leaf tissues (Fig. [4](#Fig4){ref-type="fig"}c). Moreover, *FLAVONOID GLUCOSYLTRANSFERASE2* (*FGT2*), which catalyses the transfer of a glycosyl group onto flavonoids, was significantly increased up to six-fold in leaves of A8a while a 19-fold increment was observed in roots (Fig. [4](#Fig4){ref-type="fig"}d). Fig. 4Differential transcript levels of selected genes in the phenylpropanoid pathway and photosystems during drought stress. **a** qRT-PCR based transcript levels of leaf, stem and root tissues of * DIHYDROFLAVONOL-4-REDUCTASE, DFR*; **b** *MYB112*; **c** *WD40--1*; **d** *FLAVONOID GLUCOSYLTRANSFERASE2, FGT2*; **e** *PHOTOSYSTEM I p700 CHLOROPHYLL A APOPROTEIN APS I*, *PSI*; **f** *PHOTOSYSTEM II Q(b)*, *PSII*, *n* = 4 individual plants, values are sample means ± SE. Transcript abundance is relative to empty vector after being normalized to acetyl-CoA carboxylase, *ACC1*, and *ACTIN* housekeeping genes. ANOVA p values are provided in Additional file [2](#MOESM2){ref-type="media"}: Table S5.6. Significant difference in Post hoc Tukey multiple comparisons test is indicated with different letters Photosynthesis efficiency related *PHOTOSYSTEM I p700 CHLOROPHYLL A APOPROTEIN APS I* (*PSI*) and *PHOTOSYSTEM II Q(b)* (*PSII*) transcript levels were affected by genotype and tissue type (Additional file [2](#MOESM2){ref-type="media"}: Table S5.6). *PSI* and *PSII* transcripts were five and four-fold higher in A8a leaves and roots, respectively (Fig. [4](#Fig4){ref-type="fig"}e, f). On the other hand, these two genes were significantly decreased in stems of miR156OE plants (Fig. [4](#Fig4){ref-type="fig"}e, f). Stems of miR156OE plants had pigmentation at the basal internode consistent with enhanced anthocyanin accumulation, which interferes with typical green chlorophyll colouration (Additional file [2](#MOESM2){ref-type="media"}: Figure S1) \[[@CR47]--[@CR49]\]. SPL13 regulates physiological responses and anthocyanin accumulation during drought stress in alfalfa {#Sec8} ----------------------------------------------------------------------------------------------------- Since miR156 functions in alfalfa by downregulating *SPL* genes, including *SPL13* \[[@CR8], [@CR13]\], we investigated the effect of drought on the physiological and phenotypic parameters of alfalfa plants having RNAi-silenced *SPL13.* We previously reported that a green, normal appearing phenotype accompanied enhanced root development in *SPL13*RNAi genotypes under drought \[[@CR21]\]. In the current study, leaf water potential was significantly affected by genotype under drought stress (Additional file [2](#MOESM2){ref-type="media"}: Table S5.2). In line with this, *SPL13*RNAi-5 and *SPL13*RNAi-6 plants maintained higher midday leaf water potential during drought stress (Fig. [5](#Fig5){ref-type="fig"}a). Moreover, photosynthesis efficiency parameters showed that *SPL13*RNAi-5 and *SPL13*RNAi-6 with moderate *SPL13* silencing \[[@CR21]\] maintained a higher Fv/Fm ratio of 0.74 (Fig. [5](#Fig5){ref-type="fig"}b) after 8 days of drought stress. The level of Fv/Fm is significantly affected by genotype, length of drought exposure and a combination of both (Additional file [2](#MOESM2){ref-type="media"}: Table S5.2). As a stress tolerance strategy, plants use flavonoids such as anthocyanin to scavenge ROS, and in our study we observed that *SPL13*RNAi-6 plants had a significantly higher basal monomeric anthocyanin level under a well-watered condition (Fig. [5](#Fig5){ref-type="fig"}c). Interestingly, all *SPL13*RNAi genotypes accumulated a higher level of total monomeric anthocyanin during drought stress while levels in EV did not change (Fig. [5](#Fig5){ref-type="fig"}c). A comparable total polyphenol content was mainatined by all genotypes regardless of whether the plants were under well-watered or drought conditions (Fig. [5](#Fig5){ref-type="fig"}d). Fig. 5SPL13 silencing regulates drought by coordinated metabolite, transcript, and physiological adjustments. **a** Leaf water potential in *SPL13*RNAi and EV plants; **b** dark adapted chlorophyll florescence, Fv/Fm, during drought stress; **c** total monomeric anthocyanin expressed as cyanidin-o-glucoside equivalent (CG); and **d** total polyphenol content expressed as gallic acid equivalent (GAE); **e** transcript levels of *PHENYLALANINE AMMONIA-LYASE*, *PAL*, and *DIHYDROFLAVONOL-4-REDUCTASE*, *DFR*; **f** *FLAVONOID GLUCOSYLTRANSFERASE2*, FGT2, and *DEHYDRATION RESPONSIVE RD-22-LIKE*, *DRR*; **g** *MYB112* and *WD40--1* transcription factor genes from the phenylpropanoid pathway in stems of *SPL13*RNAi and EV genotypes; **h** transcript levels of *PHOTOSYSTEM I p700 CHLOROPHYLL A APOPROTEIN APS I, PSI*, and *PHOTOSYSTEM II Q(b), PSII* under drought stress; **i** schematic representation of potential SPL13 binding sites in the promoter region of *DFR*, **j** ChIP-qPCR based fold enrichment analysis of SPL13 in p35S:SPL13-GFP and WT plants from means of n = three individual plants where *LATERAL ORGAN BOUNDARES-1, LOB1,* is used as a negative control. Values are means ± SE, light gray bars in '**a**', '**c**' and '**d**' represent values under well-watered condition while dark gray bars represent values under drought stressed conditions. Relative transcript levels in '**e**', '**f', 'g'** and '**h'** are shown relative to EV after being normalized to acetyl-CoA carboxylase, *ACC1*, and *ACTIN* housekeeping genes. ANOVA p values are provided in Additional file [2](#MOESM2){ref-type="media"}: Table S5.2, S5.7 and S5.8. Significant difference in Post hoc Tukey multiple comparisons test is indicated with different letters. Letters in multiple time point data of '**b**' is analyzed separately Flavonoid- and photosynthesis-related genes are enhanced in *SPL13*-silenced plants {#Sec9} ----------------------------------------------------------------------------------- To understand whether the observed increase in the level of total monomeric anthocyanin and maintenance of photosynthesis efficiency under drought stress is regulated at the transcript level we analysed the expression levels of anthocyanin-related and dehydration responsive genes. Our results showed that there were significant differences between genotypes under drought and control conditions (Fig. [5](#Fig5){ref-type="fig"}e-h and Additional file [2](#MOESM2){ref-type="media"}: Table S5.7). As expected, the transcript level of *PHENYLALANINE AMMONIA-LYASE*, *PAL,* the first committed step in the phenylpropanoid pathway, was significantly higher in two out of three *SPL13*RNAi genotypes (Fig. [5](#Fig5){ref-type="fig"}e). Similarly, *DFR* and *FGT2* were also higher in two out of three *SPL13*RNAi genotypes (Fig. [5](#Fig5){ref-type="fig"}e,f). These consistently higher levels of *PAL*, *DFR* and *FGT2* transcripts suggest that the induction of flavonoid biosynthesis in response to drought stress is regulated by *SPL13*. In addition, the *DEHYDRATION RESPONSIVE RD-22-LIKE* (*DRR*) gene, which is regulated by MYB and MYC transcription factors and induced by drought and ABA \[[@CR50], [@CR51]\], was also expressed four- to 17-fold higher in *SPL13*RNAi plants (Fig. [5](#Fig5){ref-type="fig"}f). In line with that, the transcription factor *WD40--1* was increased three- to 14-fold in *SPL13RNAi* plants during drought stress (Fig. [5](#Fig5){ref-type="fig"}g). For photosynthesis-related genes, we analysed the transcript levels of *PSI* and *PSII* and found a two- to 10-fold and six to 43-fold increase in expression levels, respectively, in *SPL13*RNAi plants relative to EV (Fig. [5](#Fig5){ref-type="fig"}h), consistent with results in A8a and A8 genotypes (Fig. [4](#Fig4){ref-type="fig"}e, f). SPL13 is a direct regulator of *DFR* {#Sec10} ------------------------------------ miR156 regulates the expression level of *SPL*s including *SPL13* in alfalfa \[[@CR8]\]. Given that *DFR* has four putative SBD binding motifs with core GTAC sequence in the promoter region (Fig. [5](#Fig5){ref-type="fig"}i and Additional file [2](#MOESM2){ref-type="media"}: Figure S3), we studied the occupancy of SPL13 in the promoter region of *DFR* using ChIP-qPCR in p35S:SPL13-GFP plants. The transgenic (p35S:SPL13-GFP) alfalfa plants were developed previously by our group \[[@CR52]\]. We selected three regions (I, II & III) with the conserved SBD core sequences located at 750, 544 and 260 bp, respectively, upstream of the translation start codon of *DFR* as potential SPL13 binding sites, and we tested them for SPL13 occupancy. *LATERAL ORGAN BOUNDARIES-LIKE1*, *LOB1,* was used as a negative control for ChIP-qPCR due to the low SPL13 binding ability to this gene despite the presence of a putative SBD sequence \[[@CR52]\]. Compared to WT, p35S:SPL13-GFP plants were significantly higher in SPL13 binding to the *DFR* promoter region (Fig. [5](#Fig5){ref-type="fig"}j and Additional file [2](#MOESM2){ref-type="media"}: Table S5.8). There is a preferential binding of SPL13 towards the two most downstream putative SBD regions (II & III) in the *DFR* promoter while region I did not show strong binding (Fig. [5](#Fig5){ref-type="fig"}i, j and Additional file [2](#MOESM2){ref-type="media"}: Figure S3). Of the three regions, region III showed the strongest binding to SPL13 (Fig. [5](#Fig5){ref-type="fig"}i, j), indicating that SPL13 could bind directly to *DFR* to regulate its expression. WD40--1 positively regulates *DFR* expression and drought tolerance {#Sec11} ------------------------------------------------------------------- With the observed higher expression level of *WD40--1* and flavonoid accumulation in miR156OE genotypes during drought stress and a finding from literature regarding the involvment of WD40--1 in the phenylpropanoid pathway \[[@CR53]\], we aimed to investigate whether miR156 or SPL13 directly regulate the expression of *WD40--1*. Hence, we investigated the presence of conserved SPL binding (SBD) motifs in the promoter region of *WD40--1*. We used genome walking (GenomeWalker Clonetech Laboratories, Inc.) to obtain the promoter region sequence of *WD40--1*. However, we could not find either a miR156 target sequence or a SBD motif and thus concluded an indirect regulation of *WD40--1* by miR156 or SPL13 (Additional file [2](#MOESM2){ref-type="media"}: Figure S4). To further understand the potential role of WD40--1 in alfalfa drought tolerance, we generated plants with overexpressed (OE) or silenced (RNAi) *WD40--1* and exposed these plants to drought stress. We used four different event-derived plants of WD40--1OE (OE04, OE09, OE14 and OE15) and *WD40--1*RNAi (RNAi03, RNAi04, RNAi10 and RNAi11) in comparison to WT plants (Fig. [6](#Fig6){ref-type="fig"}a, b). *WD40--1* overexpressing genotypes were drought tolerant while the RNAi silenced *WD40--1* genotypes were susceptible to drought stress (Fig. [6](#Fig6){ref-type="fig"}a, Additional file [2](#MOESM2){ref-type="media"}: Table S5.3). We investigated phenotypic and physiological responses such as root development, cholorophyll concentration and leaf water potential during drought stress and well-watered conditions. Fig. 6WD40--1 enhances drought tolerance in alfalfa. **a** above ground phenotypes of WT, four *WD40--1R*NAi and four WD40--1OE genotypes during drought stress; **b** transcript levels of *WD40--1* in WT, *WD40--1*RNAi WD40--1OE genotypes used for the study; **c** leaf water potential in WT and WD40--1OE genotypes under well-watered and drought stress condition; **d** root weight in drought stressed WT, *WD40--1*RNAi and WD40--1OE plants; **e** root length in well-watered and drought stressed WT, *WD40--1*RNAi and WD40--1OE plants; and **f** chlorophyll concentration in well-watered and drought stressed WT, *WD40--1*RNAi and WD40--1OE plants. Values are means ± SE; *n* = 4 individual plants for '**b**' to '**e**' while *n* = 20 in '**f**'. ANOVA p values are provided in Additional file [2](#MOESM2){ref-type="media"}: Table S5.3. Significant difference in Post hoc Tukey multiple comparisons test is indicated with different letters WD40--1OE genotypes maintained a higher leaf water potential during drought stress (Fig. [6](#Fig6){ref-type="fig"}c) as compared to WT and *WD40--1*RNAi genotypes (data not shown). WD40--1OE genotypes developed longer roots and associated root weight (Fig. [6](#Fig6){ref-type="fig"}d, e, Additional file [2](#MOESM2){ref-type="media"}: Table S5.3). Moreover, WD40--1OE genotypes also maintained higher level of leaf chlorophyll concentration during drought stress (Fig. [6](#Fig6){ref-type="fig"}f, Additional file [2](#MOESM2){ref-type="media"}: Table S5.3). To understand the role of WD40--1 in regulating drought stress through possible interaction with *DFR* and other genes in the phenylpropanoid/flavonoid pathway \[[@CR24]\], we measured transcript levels of phenylpropanoid-assosciated genes under drought and well-watered conditions in *WD40--1* silenced and over-expressing genotypes. Accordingly, an increase in *WD40--1* expression enhanced *DFR*, *PAL* and *FGT2* transcripts during drought stress while levels similar to that of WT were observed when plants were kept under well-watered condition (Fig. [7](#Fig7){ref-type="fig"}a, b, c, Additional file [2](#MOESM2){ref-type="media"}: Table S5.8). Moreover, the ABA-related dehydration responsive gene, *DRR*, and photosynthesis related genes, *PSI* and *PSII*, were increased in WD40--1OE genotypes compared to *WD40--1*RNAi and WT plants (Fig. [7](#Fig7){ref-type="fig"}d, e, f, Additional file [2](#MOESM2){ref-type="media"}: Table S5.8). Fig. 7*WD40--1* regulates transcript levels of genes in the phenylpropanoid pathway and photosystem during drought stress. **a** Transcript levels of *PHENYLALANINE AMMONIA-LYASE***,** *PAL*; **b** *DIHYDROFLAVONOL-4-REDUCTASE*, *DFR*; **c** *FLAVONOID GLUCOSYLTRANSFERASE2*, *FGT2*; **d** *DEHYDRATION RESPONSIVE RD-22-LIKE*, *DRR*; **(e)** *PHOTOSYSTEM I p700 CHLOROPHYLL A APOPROTEIN APS I, PSI*; **f** *PHOTOSYSTEM II Q(b), PSII*. Transcript levels are shown relative to EV after being normalized to acetyl-CoA carboxylase, *ACC1*, and *ACTIN* housekeeping. Values are means ± SE, n = 4 individual plants, ANOVA p values are provided in Additional file [2](#MOESM2){ref-type="media"}: Table S5.9; **g** schematic representation of miR156-based alfalfa drought resilience model system**.** Solid line represents an experimentally confirmed mechanism while broken lines are hypothesized functions. Arrow heads indicate positive regulation while line heads indicate negative regulation. Significant difference in Post hoc Tukey multiple comparisons test is indicated with different letters Discussion {#Sec12} ========== Drought is one of the main factors that impair plant growth and development \[[@CR54]\]. Plants respond to drought by showing deleterious effects, or by engaging in adaptive responses involving various molecular, biochemical and physiological strategies \[[@CR55]--[@CR57]\]. In this study, we used miR156OE, *WD40--1*OE, *WD40--1*RNAi, *SPL13*RNAi and GFP-tagged SPL13 genotypes to investigate the molecular and physiological strategies used by miR156 to regulate drought stress in alfalfa. Moderate levels of miR156 overexpression, *WD40--1* overexpression or *SPL13* silencing are sufficient to induce phenotypic and physiological drought tolerance strategies in alfalfa {#Sec13} ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Of the different plant organs that respond to soil water deficit, roots are first to encounter changes in the rhizosphere. Findings in model plants showed initiation and elongation of lateral roots in drought tolerant genotypes to improve water uptake \[[@CR58], [@CR59]\]. In this study, we observed a significant increase in root length accompanied by higher root biomass in plants moderately over-expressing miR156 (A8a and A8) and *WD40--1*. This is associated with a reportedly enhanced level of ABA \[[@CR21]\] in miR156 overexpressing genotypes under drought stress. ABA enhances primary and lateral root development by regulating the expression of *LATERAL ROOT ORGAN DEFECTIVE* (*LATD*) gene \[[@CR60]\]. Moreover, miR156 contributes to root development by silencing *AtSPL10* to decrease the expression of *AGAMOUS-LIKE MADS box protein 79* (*AGL79)* in Arabidopsis \[[@CR61]\]. Accordingly, the enhanced root development under drought stress helps alfalfa plants to better access water from deeper soil surface. This finding is consistent with our previous report that showed increased root length in miR156OE and *SPL13*RNAi genotypes under drought conditions \[[@CR21]\]. Moreover, moderate miR156OE, *SPL13*RNAi and WD40--1OE genotypes had higher leaf water potential despite their exposure to drought conditions. The observed drought tolerance in miR156OE (A8a and A8), WD40--1OE and *SPL13*RNAi genotypes suggests this trait is at least partially negatively regulated by SPL13 and positively by miR156 and *WD40--1*. Photosynthesis is negatively impacted by drought stress in alfalfa and other plant species \[[@CR34], [@CR62]\]. Of the many photosynthesis efficiency parameters, Fv/Fm reflects the maximum quantum efficiency of PSII photochemistry possible in a dark-adapted state, and is considered a good stress indicator in plants \[[@CR63]--[@CR67]\]. Therefore, maintaining a higher Fv/Fm was one of the parameters used in selecting abiotic stress tolerant cultivars of tomato and wheat \[[@CR64], [@CR68], [@CR69]\]. The observed higher level of Fv/Fm in A8a and A8 genotypes in the current study suggests that their leaves may have a functional photosynthetic unit, in agreement with the observed maintained photosynthesis assimilation rate under drought. The observed higher *Vcmax* (Rubisco carboxylase activity) and *Jmax* (electron transport rate) in A8a and A8 under drought further illustrate the maintenance of their photosystem despite drought stress. Such physiological adjustments were low to absent in A11 plants which showed susceptibility to drought stress. We also observed a higher Fv/Fm ratio in *SPL13*RNAi-5 and *SPL13*RNAi-6, which is consistent with our previously reported finding of increased photosynthetic assimilation rate in drought-treated *SPL13*RNAi genotypes \[[@CR21]\]. This suggests that the maintenance of a higher photosynthetic assimilation rate, *Vcmax*, *Jmax* and high Fv/Fm ratio during drought stress in miR156OE and WD40--1OE genotypes may be regulated at least in part by SPL13 and *WD40--1*. miR156 overexpression enhances accumulation of stress-related metabolites {#Sec14} ------------------------------------------------------------------------- The impact of environmental perturbations on plant metabolism varies among plant species, cultivars, and tissues considered \[[@CR70]\]. Accumulation of specific secondary and transient primary metabolites (primary metabolites that are direct precursors of secondary metabolites) in various tissues are used in part to mitigate drought stress \[[@CR27], [@CR28], [@CR71], [@CR72]\]. Naya et al. \[[@CR73]\] indicated the role of carbon metabolism and oxidative damage on nitrogenase activity reduction during moderate and higher drought stress levels in alfalfa. Other studies in *M. truncatula*, have shown a decrease in symbiotic nitrogen fixation under drought stress resulting in low levels of nitrogen-based metabolites \[[@CR74]\]. In our study, alfalfa with a moderately enhanced expression of miR156 caused accumulation of anthocyanins, flavonols, and proteinogenic amino acids in leaf and stem tissues. The accumulation of these metabolites may help the plant to scavenge ROS produced during drought stress \[[@CR40]\]. Moreover, these metabolites could help the plants to reduce water loss, and further absorb any remaining tightly bound water from the soil by lowering the osmotic balance in the root tissues. The high level of GABA in leaf, stem and root tissues of A8a and A8 should maintain a carbon-to-nitrogen balance through a GABA shunt bypassing the decarboxylation part of the TCA cycle \[[@CR29]\]. The importance of GABA in mediating abiotic stress has been well documented in various plant species, including *Arabidopsis* \[[@CR75]\], black pepper \[[@CR76]\] and bentgrass \[[@CR77]\]. Proline was also increased in A8a and A8 but not in A11 roots to regulate osmotic homeostasis as reported in other studies \[[@CR21], [@CR26]\]. The relatively lower level of proline abundance in roots of the highest miR156 overexpressor, A11, might have prevented these plants from maintaining high water levels in their system (Fig. [1](#Fig1){ref-type="fig"}g). The higher level of fructose and arabinose in leaf and stem tissues of drought-treated moderate miR156 expressors respectively could provide an energy source and/or an osmolyte. The higher sugar level suggests an actively functioning photosynthetic assimilation with the potential to supplement a carbon source for downstream metabolites. This is consistent with a previous finding that drought-stressed alfalfa plants accumulate sugars \[[@CR78]\]. Moreover, the increased total monomeric anthocyanin and comparable total polyphenol levels in *SPL13*RNAi genotypes illustrated a targeted enhancement of flavonoids at least partially governed by silencing SPL13 in alfalfa to scavenge ROS during drought stress. miR156, WD40--1 and SPL13 regulate phenylpropanoid and photosystem genes under drought {#Sec15} -------------------------------------------------------------------------------------- Due to the various roles that polyphenols play in stress response, efforts have been made to increase their levels in many plants, including alfalfa \[[@CR79]\]. Enhanced accumulation of flavonoids and proanthocyanidins (PA) in alfalfa has important quality implications for animal feed, as moderate amounts of PA tend to reduce bloating in ruminant animals \[[@CR80]--[@CR82]\]. In our study, we found that phenylpropanoid pathway-related genes are enhanced in moderately overexpressing miR156 alfalfa plants, which is consistent with the increase in anthocyanin and flavonol levels in these plants. *DFR, WD40--1* and *MYB112* were higher in A8a and A8 during drought, contributing to anthocyanin accumulation. Similarly, *SPL13*RNAi genotypes showed enhanced levels of *DFR, FGT2* and *PAL* transcripts associated with enhanced level of total monomeric anthocyanin, indicating enhancement of the phenylpropanoid/flavonoid pathway. In another study, *Arabidopsis* plants overexpressing miR156 accumulated anthocyanin in response to salt and mannitol (mimicking drought) treatments by increasing *DFR* expression \[[@CR23]\]. The enhanced *DFR* expression level in *Arabidopsis* was regulated by silencing *SPL9* \[[@CR23]\]. Our findings suggest that accumulation of anthocyanins and other polyphenols may be regulated via SPL13 in alfalfa. Moreover, the enhanced level of *DFR* in WD40--1OE plants and reduced in *WD40--1*RNAi plants suggests that *DFR* is positively regulated by the WD40--1 to promote flavonoid biosynthesis, but the mechanism of this regulation remains to be investigated. To investigate whether the higher photosynthetic assimilation rate during drought stress in *SPL13*RNAi \[[@CR21]\] and also WD40--1OE, *WD40--1*RNAi and miR156OE genotypes (current study) are regulated at the transcriptional level, we investigated expression of genes mediating photosynthesis. We found that *PSI* and *PSII* were significantly increased in moderately overexpressing miR156OE genotypes and *SPL13*RNAi genotypes upon drought. Previously, we reported an increased abundance of ABA, which regulates stomatal aperture by active hydrolysis during drought stress in miR156OE A8 plants \[[@CR21]\]. In the current study, we examined expression of the ABA-induced dehydration responsive gene (*RD22*) and found it to be significantly increased in *SPL13*RNAi plants during drought stress. The consistent observation of higher polyphenols and photosystem assimilation rate with associated transcripts during drought stress in moderate miR156OE and *SPL13*RNAi genotypes suggests a drought regulation strategy of miR156. SPL13 binds to *DFR* to regulate its expression and flavonoid biosynthesis {#Sec16} -------------------------------------------------------------------------- To investigate whether the increased flavonoid accumulation and expression of phenylpropanoid-associated genes, especially *DFR*, are directly regulated by the miR156-SPL13 module, we conducted a ChIP-qPCR analysis to determine binding of SPL13 to *DFR*. DFR catalyses flavonoid biosynthesis by reducing dihydroflavonols to leucoanthocyanidins playing a critical role in anthocyanin biosynthesis \[[@CR83]\]. A previous report showed SPL9 directly regulates the expression level of *DFR* to enhance accumulation of anthocyanin in response to NaCl and mannitol treatment in *Arabidopsis* \[[@CR23]\]. In the current study, we showed increased *DFR* expression during drought stress in moderately overexpressing miR156 and *SPL13*RNAi plants. Accordingly, we selected *DFR* to test for SPL13 binding, given the presence of multiple potential SBD core GTAC sequences in the *DFR* promoter. The fold enrichment from ChIP-qPCR showed the strongest SPL13 binding was observed at region III of the *DFR* promoter, which is located closest (260 bp) to the *DFR* coding sequence. This is in line with reports that showed the conserved core SBD element is not by itself sufficient for SPL binding, but also determined by the position of the SBD and the flanking DNA sequences \[[@CR11], [@CR84], [@CR85]\]. SPL13 acts as a transcriptional suppressor of *DFR* during drought stress as confirmed by higher expression of *DFR* in *SPL13*RNAi and miR156OE plants. Conclusions {#Sec17} =========== We recently reported that miR156 regulates drought tolerance in alfalfa by silencing *SPL13* \[[@CR21]\]. Understanding the mechanisms deployed by miR156 in drought tolerance could be exploited as a tool in crops for marker-assisted breeding. In the current study, we investigated metabolomic, physiological and molecular mechanisms to show how low- to moderate levels of miR156 expression is sufficient to induce drought tolerance in alfalfa. Moderate levels of miR156 in genotypes of A8a and A8 induced accumulation of stress mitigating metabolites, such as anthocyanins, flavonols, GABA, proline and others in the leaf, stem and root tissues. These metabolites could help the plants to scavenge ROS, reduce water loss and further absorb any remaining tightly bound water from the soil by lowering the osmotic balance in the root tissues. In addition, the plants showed physiological adjustments such as improved photosynthetic assimilation rate, maintained Fv/Fm ratio, and enhanced root growth and development. The relatively low levels of stress mitigating metabolites and reduced physiological adjustments may have resulted in drought susceptibility in the highest miR156 overexpressor (A11). We also determined direct binding of *SPL13* to the *DFR* promoter*. SPL13* acts as a transcriptional suppressor of *DFR* during drought stress as confirmed by higher expression of *DFR* in *SPL13*RNAi and miR156OE plants. Similar observation of SPLs suppressing the expression of *DFR* has been reported in *Arabidopsis* \[[@CR86]\] where *SPL9* silenced *DFR* in response to salt and mannitol treatment \[[@CR23]\]. Moreover, we detected an increase in expression of genes involved in the phenylpropanoid and photosynthetic pathways, including *DFR, MYB112, PSI* and *PSII* in miR156OE plants under drought. *DFR, FGT2, PSI* and *PSII* were also increased in *SPL13*RNAi plants under drought stress. We propose a model for a drought tolerance mechanism regulated by moderate levels of miR156 over-expression (Fig. [7](#Fig7){ref-type="fig"}g). The diagrammatic representation shows the central role of miR156 in regulating drought stress in alfalfa. MiR156 is induced by drought stress, which in turn silences *SPL13* \[[@CR21]\]. Reduced expression of *SPL13* driven by miR156 and increased levels of *WD40--1* enhance *DFR* resulting in accumulation of anthocyanins. In moderate miR156OE plants, primary metabolites such as GABA, proline and sugars also accumulate for carbon-to-nitrogen balance and osmotic homeostasis. Induction of miR156 during drought stress also enhances phenotypic plasticity, such as longer roots and higher biomass to access more water from the rhizosphere. With reduced *SPL13* expression, miR156OE and *WD40--1*OE*,* higher photosynthesis efficiency is also achieved during drought stress. We conclude that moderate levels of miR156 expression silence *SPL13* and increase WD40--1 expression to fine-tune *DFR* expression for anthocyanin biosynthesis and regulate various developmental, physiological and biochemical processes in alfalfa leading to improved drought resilience. Methods {#Sec18} ======= Genetic material {#Sec19} ---------------- ### miR156 overexpressing and *SPL13*RNAi plants {#Sec20} *Medicago sativa* L. N4.4.2 plants \[[@CR87]\] were obtained from Dr. Daniel Brown (Agriculture and Agri-Food Canada) and used as wild-type genotypes. Plants over-expressing miR156 (miR156OE) at different levels (A8a, A8 and A11) and an empty vector control (EV) were generated previously in our laboratory and used in this experiment \[[@CR13]\]. miR156 is slightly (0.5) elevated in A8a, but it is moderate (1.5) to higher (2.5) relative transcript level in A8 and A11, respectively \[[@CR13]\]. The plants were grown in a fully automated greenhouse with 16-h light (380--450 W/m^2^), relative humidity (RH) of 70% and temperature of 25 ± 2 °C at the Agriculture and Agri-Food Canada London Research and Development Center, London, Ontario, Canada. Given that alfalfa is an obligatory outcross, we used vegetative cuttings for propagation according to Aung et al \[[@CR13]\] to maintain genotypes throughout the study. Since miR156 down-regulates seven *SPL* genes (including *SPL13*) to regulate a network of downstream genes, we used *SPL13*RNAi genotypes (*SPL13*RNAi-2, *SPL13*RNAi-5 and *SPL13*RNAi-6) \[[@CR21]\] selected for their low *SPL13* expression levels relative to wild-type alfalfa and other *SPL13*RNAi transgenic alfalfa plants. ### Generation of WD40--1 overexpressing and WD40--1RNAi alfalfa plants {#Sec21} Four WD40--1OE (OE04, OE09, OE14 and OE15) and four WD40--1RNAi (R03, R04, R10 and R11) genotypes were generated to investigate the role of WD40--1 in drought tolerance. WD40--1 overexpression and downregulated genotypes were generated using constructs made from alfalfa homolog *WD40--1* (Medtr3g074070) using Gateway cloning system (Thermo Fisher Scientific, Mississauga ON). For overexpression studies, full-length *WD40--1* was amplified from alfalfa (*Medicago sativa*) cDNA using primers with AttB sites attached, forward (B1-WD40--1) and reverse (B2-WD40--1) (Additional file [1](#MOESM1){ref-type="media"}: Table S1) and cloned into the pDONR/Zeo entry vector. For downregulation studies, a 253 bp putative *WD40--1* fragment was amplified from alfalfa cDNA using AttB sites attached forward (B1-WD40--1-RNAi) and reverse (B2-WD40--1-RNAi) (Additional file [1](#MOESM1){ref-type="media"}: Table S1) primers and cloned into pDONR/Zeo entry vector. After PCR screening and confirmation by sequencing, LR reactions were performed for the overexpression and RNAi constructs to recombine the fragments into the pMDC83 (overexpression) and pHELLSGATE12 (RNAi) vectors. Subsequently, overexpression and RNAi constructs were used to transform *Agrobacterium tumefaciens* strain EHA105 which was then used to transform alfalfa. QRT-PCR was then used to analyze WD40--1 gene expression in WD40--1-OE *WD40-1-*RNAi genotypes using primers WD1-qPCR-F and WD1-qPCR-R (Additional file [1](#MOESM1){ref-type="media"}: Table S1). Imposing drought stress {#Sec22} ----------------------- Drought stress was imposed on alfalfa plants devoid of water for 2 weeks at 30 days post vegetative propagation (juvenile vegetative stage) during which time plants were kept in a completely randomized design. Equal soil moisture levels were maintained before starting the experiment using a SM 100 soil moisture sensor (Spectrum Technologies Inc., Jakarta, Indonesia). At least four biological replicates were used per genotype per treatment for transcript and metabolite analysis, while 4 to 10 plants were used for physiological analysis (each replicate being an individual plant). The entire experiment was repeated under the same growth and drought stress conditions to test the repeatability of results. Leaves (newly developed upper leaves), stems (lower 5 cm internode close to soil) and roots (7.5 cm of main and auxiliary root tips) were harvested from miR156OE, *SPL13*RNAi, *WD40--1*OE, *WD40--1*RNAi, EV and wild-type plants depending on the experiment. Samples were flash frozen with liquid nitrogen and kept at − 80^0^ C for later metabolomic and transcriptomic analyses. Metabolite extraction for parallel LCMS and GCMS analysis {#Sec23} --------------------------------------------------------- To explore miR156-related regulation of secondary metabolites and transient primary metabolites, extracts of stem, leaf and root tissues of drought-stressed miR156OE and control plants were subjected to Liquid Chromatography-Mass Spectrometry (LCMS) and Gas Chromatography-Mass Spectrometry (GCMS) analysis. Extraction of samples was performed according to Ayenew et al. \[[@CR28]\] for parallel LCMS and GCMS analysis. Unless stated otherwise, chemicals used for the analysis were obtained from Sigma-Aldrich, Canada. Briefly, frozen 50 mg tissues were crushed with a RETCH-mill (Retsch Gmbh, 42,787 Haan, Germany) and stainless-steel beads. One milliliter prechilled extraction solution, methanol/chloroform/water (2.5/1/1 v/v/v), was added containing an internal standard Ribitol/adonitol 0.225 mg/mL for GCMS analysis while ampicillin (Sigma, and Saint Luis, Missouri, USA) and corticosterone at 1 mg/mL for LCMS to normalize extraction variability. The mixture was vortexed and ultra-sonicated for 10 min. Following centrifugation at 14000 rpm for 10 min (at 4^0^ C), supernatant was collected and mixed with equal volumes of 300 μL water and chloroform. The mixtures were vortexed briefly and centrifuged at 14000 rpm for 5 min to collect the upper aqueous phase for parallel LCMS and GCMS analyses. LCMS analysis was performed using an Agilent 1290 Infinity LC system coupled with a Thermo Q-Exactive Quadrupole-Orbitrap mass spectrometer. Analytes were separated with an Agilent Eclipse Plus C18 ZORBAX Rapid Resolution High Definition (RRHD) 1.8 μm particle 2.1 i.d. X 50 mm column. The instrument was equipped with electrospray ionization (ESI) interface operating in a negative and positive ion mode for better metabolite identification. Metabolites were identified based on mass to charge ratio (*m/z*), retention time and fragmentation pattern in comparison to commercial standards, ChemSpider and ReSpect phytochemical databases \[[@CR28], [@CR71]\]. MZmine2 software \[[@CR88]\] was also used for LCMS metabolite mass detection, chromatogram building, and the separation of overlapping peaks. In parallel, transient primary metabolites were explored using 75 μL aliquots of the extracted samples for LCMS using an Agilent 5975C Triple-Axis Detector MSD and 7890A GC system in splitless mode. The aliquots were dried using an Eppendorf Vacufuge™ concentrator (Hamburg, Germany), derivatized by 40 μL *O*-methylhydroxylamine hydrochloride in pyridine with 7 μL standard alkane mixture (0.029% v/v C10-C20 of each 50 mg/l) for 2 h at 37 °C followed by 70 μL *N*-methyl-*N*-\[trimethylsilyl\] trifluoroacetamide (MSTFA) for silylation. Metabolites from GCMS were identified using the retention time of the standard alkane mixture with their mass spectra and a NIST 2011 mass spectral library \[[@CR27], [@CR28], [@CR72]\]. Total monomeric anthocyanin and polyphenol determination {#Sec24} -------------------------------------------------------- Total monomeric anthocyanin, TMA, and total polyphenol, TPP, were determined using a pH deferential extraction method \[[@CR89], [@CR90]\]. Briefly, flash-frozen in liquid nitrogen samples were crushed with mortar and pestle under liquid nitrogen and 500 mg tissue were used for the combined analysis of TMA and TPP. Samples were treated with 2 ml acidified methanol (MeOH with 1% HCL), vortexed and sonicated at 20 KHz for 15 min. Homogenate was stirred at 3000 RPM for 1 h and centrifuged (at 4 °C) for 10 min at full speed (14,000 RPM). The supernatant was collected, added 2 ml chloroform, vortexed and centrifuged at full speed for 10 min. The upper aqueous phase was collected, filtered with Whiteman 0.2 um filters, and divided into three equal aliquots for TMA (pH 1.0 and 4.5) and TPP analysis. The first aliquot was mixed with an equal volume of 0.025 M KCl at pH 1.0 while the second is mixed with equal volumes of 0.4 M sodium acetate at pH 4.5 and measured absorbance at 520 and 700 nm with water as a blank. TPP was analysed by mixing an equal volume of the third aliquot with Folin-chiocalteu reagent (diluted 1:10 with water) and vortexed for 3 min. Four ml of sodium carbonate (7.5% w/v) was added to the mixture, which was then vortexed and incubated for 30 min in the dark. TPP was determined as gallic acid equivalent (GAE) after measuring absorbance of the aliquot at 765 nm with acidified methanol as blank. TMA level is expressed as mg cyanidin-3-o-glucoside (CG) equivalent. Physiological and phenotypic data measurement {#Sec25} --------------------------------------------- To determine drought mitigating strategies, we investigated phenotypic and physiological parameters. Midday photosynthesis assimilation rates and dark-adapted chlorophyll fluorescence (Fv/Fm) were measured in newly growing upper unshaded leaves using a LI-6400XT portable photosynthesis meter coupled with Fluorescence System (LI-COR Biosciences, Lincoln, Nebraska, USA). Photosynthetic assimilation rate responses across a gradient of CO~2~ level (A/Ci) in the mesophyll cells to determine the maximum rate of rubisco carboxylase activity (*Vcmax*) and maximum photosynthetic electron transport rate (*Jmax*) was calculated to determine photosynthetic efficiency using the R statistical software plantecophys package \[[@CR91]\]. Chlorophyll concentration index (CCI) of newly growing upper leaves were also determined using an Apogee MC100 instrument (Apogee instruments, Logan, Utah, USA) \[[@CR92]\]. To determine plant water status, the midday leaf water potential was measured using a SAPS II Portable Plant Water Status Console (Soilmoisture Equipment Corp., Santa Barbara, CA, USA) in dark-adapted leaves by covering leaves with a polyethylene bag and aluminium foil for 20 min. In addition, above and below ground phenotypic parameters were measured, such as stem number and shoot weight, root length and weight according to Aung et al. \[[@CR13]\], and stem basal diameter at 1 cm above stem-soil interface. RNA extraction and qRT-PCR analysis {#Sec26} ----------------------------------- Stem, leaf and root samples were collected and flash frozen in liquid nitrogen and kept in a -80 °C freezer until further use. Approximately 50 mg fresh weight was used for total RNA extraction using a PowerPlant® RNA isolation kit (Cat \# 13500) for leaf samples, a QIAGEN RNeasy® Plant mini kit for stem and root tissues (Cat \# 74904), and a PowerLyzer®24 bench top bead-based homogenizer (Cat \# 13155) following manufacturers protocols. The extracted RNA was treated with Ambion®TURBO DNA-*free™* DNase (Cat \# AM1907) followed by iScript™ cDNA synthesis (Cat \# 1708891). Transcript levels of selected genes involved in secondary metabolite biosynthesis and photosynthesis were investigated in this study. Using publicly available transcriptomics data of two miR156OE alfalfa genotypes under control (unstressed) conditions \[[@CR8]\] and *M. truncatula* genome sequence Mt4.0 V2 (<http://www.medicagogenome.org/downloads>), transcripts of differentially expressed genes with the SBD core GTAC sequence within 2.5 kb of their promoter regions were identified. Among those, genes shown by Gene Ontology analysis to be involved in flavonoid biosynthesis, photosynthetic efficiency and stress tolerance were chosen for expression analysis by qRT-PCR. Primers specific to the above genes (Additional file [1](#MOESM1){ref-type="media"}: Table S1) were designed using *M. truncatula* genome sequence and amplified product was sequenced for an identity check (Additional file [2](#MOESM2){ref-type="media"}: Figure S2). Publicly available Primer3 software (<http://primer3.ut.ee/>) was used to design primers, and their efficiency was verified at different concentrations with gradient annealing temperature PCR before using for qRT-PCR analysis. QRT-PCR was performed using the CFX96™ Real-Time PCR detection system and SsoFast™ EvaGreen® Supermixes (Bio-Rad Cat \# 1725204). Specifically, 2 μL cDNA (equivalent to 200 ng cDNA), 1 μL forward and reverse gene-specific primers (10 μM each), 5 μL SsoFast Eva green Supermix, and 2 μL of nuclease-free water was used to make the final reaction volume of 10 μL. PCR amplification was performed at: cDNA denaturation at 95 °C for 30 s followed by 40 cycles of 95 °C for 10 s, 58 °C for 30 s and 72 °C for 30 s (denaturation, annealing and extension, respectively) followed by a melting curve that runs from 65 °C to 95 °C with a gradual increment of 0.5 per 5 s. All reactions were performed with three technical and four biological replicates. Transcript levels were analysed relative to acetyl-CoA carboxylase (*ACC1*) and *ACTIN* housekeeping genes designed based on alfalfa sequence \[[@CR13], [@CR21]\]. ChIP-qPCR analysis of SPL13-DNA binding {#Sec27} --------------------------------------- Shoot tips of alfalfa plants overexpressing *SPL13* tagged with GFP driven by the CaMV35S promoter (p35S:SPL13-GFP) \[[@CR52]\] were used to understand the occupancy of SPL13 on promoters of downstream genes contributing to drought tolerance. One-month-old SPL13-GFP overexpressing genotypes and WT control plants were used for ChIP-qPCR analysis based on previously published protocol \[[@CR93]\] with some modifications. Briefly, 500 mg of shoot tips from WT and p35S:SPL13-GFP plants were collected, washed, proteins bound to DNA were cross-linked using 1% formaldehyde and mixtures were ground with liquid nitrogen. Extraction reagents and buffers are listed in Additional file [2](#MOESM2){ref-type="media"}: Table S4. Powdered tissues were homogenized with 15 ml of prechilled Extraction Buffer 1 and filtered with two layers of Miracloth (Millipore, Canada). Subsequently, the filtered mixture was centrifuged at 3000 *g* for 20 min and supernatant was discarded while the pellets were resuspended in 1 ml of prechilled Extraction Buffer 2 and centrifuged at 12000 *g* for 10 min. Afterwards, pellets were resuspended in 300 μL prechilled Extraction Buffer 3 and centrifuged at 16000 *g* for 1 h. The supernatant was removed, and chromatin pellets were resuspended in 300 μL of Nuclei Lysis Buffer by gentle pipetting and sheared twice at power 3 for 15 s on ice using a Sonic Dismembrator (Fisher Scientific, USA). Twenty microliter of supernatant aliquots were kept aside for later use as an input DNA control while using the remaining solution for immunoprecipitation. Chromatin solution was brought to 1.5 mL using a ChIP dilution buffer and divided into two equal parts for chromatin immunoprecipitation and a negative control. To each tube, 30 μL of protein A-agarose beads (Millipore, Canada) were added and the mixture was gently agitated, centrifuged (3500 *g*) for 1 min, and supernatant was transferred for immunoprecipitation while discarding the beads. Five μL of Ab290 GFP antibody was added to one of the chromatin solutions (keeping the second one as a no-antibody negative control) for an overnight gentle agitation at 4 °C. After 12 h, 40 μL of protein A-agarose beads were added and immune complexes were recovered by centrifugation and washed with cycle of low normality salt, high salt, LiCl and TE buffer. Immunocomplexes were eluted from beads using 250 μL of Elution Buffer and cross linking was reversed with 20 μL of 5 M NaCl incubated at 65^0^ C for 5 h. To each sample 10 μL 0.5 M EDTA, 20 μL 1 M Tris-HCl (pH 6.5) and 2 μL of 10 mg/mL proteinase K (Sigma-Aldrich, Canada) were added. DNA was extracted using phenol: chloroform (1:1, v:v), recovered by precipitation with ethanol and 0.3 M sodium acetate (pH = 5.2) and 2 μL glycogen carrier 10 mg/mL (Sigma-Aldrich, Canada) after overnight incubation at -20 °C. After 12 h, the solution was centrifuged at full speed for 20 min to pellet the DNA and pellet was then washed with 70% ethanol, resuspended with 16 μL of distilled water, and DNA was used for ChIP-qPCR analysis. To obtain the *DFR* promoter region sequence from *M. sativa*, proDFR1-MTR primers (Additional file [1](#MOESM1){ref-type="media"}: Table S1) were designed using a close relative *M. truncatula* sequence and amplified region was cloned into TOP10 competent *E. coli* cells using CloneJET (Thermo Scientific) and sequenced. Subsequently, proDFR ChIP-qPCR primers (Additional file [1](#MOESM1){ref-type="media"}: Table S1) were designed based on alfalfa sequences. QRT-PCR was performed using ChIP-precipitated DNA as described above while fold enrichment was calculated by dividing Ct values of p35S:SPL13-GFP to WT and comparing with *LOB1* reference gene \[[@CR52]\]. Genome walking for WD40--1 promoter nucleotide sequence {#Sec28} ------------------------------------------------------- Due to lack of alfalfa genome sequence, we used Clonetech GenomeWalker™ (California, USA Cat No. 638904) to obtain nucleotide sequence of the *WD40--1* promoter region. In brief, we extracted genomic DNA from wild-type alfalfa plants using a Nucleospin®Tissue DNA extraction kit (MACHEREY-NAGEL Gmbh & Co. KG Germany, Cat. No. 740952). GenomeWalker "libraries" were prepared by digesting the DNA with four different restriction enzymes (*Dra*I, *Eco*RV, *Pvu*II and *Stu*I) at 37 °C for 2 h to generate blunt ends. Subsequently, two nested PCR amplifications were performed sequentially for each library using gene specific primers (GSP1 and GSP2) and adapter primers (AP1 and AP2) from the kit (Additional file [1](#MOESM1){ref-type="media"}: Table S1). PCR products were analyzed on a 1.5% agarose gel followed by cloning into a pJET1.2 cloning vector to facilitate sequencing. Subsequently, sequences obtained from the four libraries were aligned together to generate the consensus promoter region sequence of WD40--1 in alfalfa. Statistical data analysis {#Sec29} ------------------------- Shapiro-Wilk test were used for checking the normal distribution of data before proceeding to analysis of variance (ANOVA). Subsequently, Tukey post hoc multiple comparison were done on molecular (qRT-PCR and ChIP-qPCR), metabolomics (LCMS and GCMS), physiological and phenotypic data. Pair-wise t-test comparison was implemented between WD40--1OE and wild-type plants and with WD40--1RNAi plants for WD40--1 transcript abundance. Metabolite profile data were subjected to pareto scaling before principal component analysis (PCA) in which metabolites were mean-centred followed by dividing with the square root of the standard deviation. All statistical data analyses were undertaken using R-software environment 3.2.5. Supplementary information ========================= {#Sec30} **Additional file 1:** **Table S1.** List of primers used and their nucleotide sequences. **Additional file 2:** **Table S2.** LCMS-based metabolite profiles of drought stressed alfalfa plants. **Table S3** GCMS-based relative metabolite abundance in drought stressed alfalfa plants. **Table S4** Buffers used in ChIP assay and their components. **Table S5.1** Analysis of variance, ANOVA, *P* values of data for phenotype and physiological responses in miR156OE genotypes and EV plants. **Table S5.2** Analysis of variance, ANOVA, P values of data for phenotype, physiological and metabolite responses in SPL13RNAi genotypes and EV plants. **Table S5.3** Analysis of variance, ANOVA, P values of data for phenotype and physiological responses in WD40--1OE, WD40--1RNAi and wild type plants. **Table S5.4** Analysis of variance, ANOVA, P values of data for LCMS-based metabolite profiling in miR156OE genotypes and EV alfalfa plants. **Table S5.5** Analysis of variance, ANOVA, P values (P \> F) of data for GCMS-based metabolite profiling in miR156OE genotypes and EV alfalfa plants. **Table S5.6** Analysis of variance, ANOVA, P values of data for qRT-PCR based transcript level in miR156OE genotypes and EV alfalfa plants. **Table S5.7** Analysis of variance, ANOVA, P values of data for qRT-PCR based transcript level in SPL13RNAi genotypes and EV alfalfa plants. **Table S5.8** Analysis of variance, ANOVA, P values of data for ChIP-qPCR based transcript level in p35S:SPL13-GFP genotypes and Wild-type alfalfa plants. **Table S5.9** Analysis of variance, ANOVA, P values of data for qRT-PCR based transcript level in WD40--1RNAi silenced and WD40--1over expressing plants. **Figure S1** Stem colour development in miR156OE plants during drought stress. **Figure S2** Alignment of sequences of amplified by q-PCR from *Medicago sativa* with those of their counterparts in *Medicago truncatula.* **Figure S3** Promoter sequence of the alfalfa *DIHYDROFLAVONOL-4-REDUCTASE (DFR)* gene with putative SBD binding elements. **Figure S4** Nucleotide sequence of the alfalfa *WD40--1* promoter region. ABA : Abscisic acid DAG : Delphinidin 3-O-(6″-acetyl)-glucoside *DFR* : *DIHYDROFLAVONOL-4-REDUCTASE* *DRR* : *DEHYDRATION RESPONSIVE RD-22-LIKE* EV : Empty vector *FGT2* : *FLAVONOID GLUCOSYLTRANSFERASE2* GABA : Gamma-aminobutyric-acid GCMS : Gas chrmotaography mass spectrometry LCMS : Liquid chromatography mass spectrometry *LOB1* : *LATERAL ORGAN BOUNDARIES-LIKE1* miR156 : microRNA156 PA : Proanthocyanidins *PAL* : *PHENYLALANINE AMMONIA-LYASE* *PAP1* : *PRODUCTION OF ANTHOCYANIN PIGMENT 1* PCA : Principal Component Analysis PG : Peonidin 3-O-glucoside PPP : Pentose phosphate pathway *PSI* : *PHOTOSYSTEM I p700 CHLOROPHYLL A APOPROTEIN APS I* gene PSII : Photosystem II *PSII* : *PHOTOSYSTEM II Q(b)*gene ROS : Reactive oxygen species SBD : SPL binding domain SPL : SQUAMOSA-PROMOTER BINDING PROTEIN-LIKE WT : Wild-type **Publisher's Note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Supplementary information ========================= **Supplementary information** accompanies this paper at 10.1186/s12870-019-2059-5. The authors acknowledge Dr. Justin Renaud for his help with LCMS and GCMS. BAF and MA developed materials; BAF performed the experiments; SEK and AH supervised the project; BAF and AH designed the research; BAF, MA, MYG, SEK and AH wrote, revised and approved the manuscript. The research is funded through grants from Agriculture and Agri-Food Canada and the Natural Science and Engineering Research Council of Canada to AH. The funding agencies had no role in the design of the study; collection, analysis, and interpretation of data; and in writing the manuscript. Data used in this study are provided as 'additional file.xlsx' as a supplementary file. Not applicable. These requirements are not applicable to the current manuscript. The authors declare that they have no competing interests.

Introduction {#Sec1} ============ Best practice in health-related quality of life (HRQL) assessment evolved to require inclusion of patient input in the development of measures \[[@CR1]\], and for many conditions, patient self-report is the standard for assessment of disease and treatment impact \[[@CR2]\]. Over the last four decades, the role of the patient in health services research has also expanded. Qualitative and quantitative methods for incorporating the "patient voice" in research are now established \[[@CR3]\]. Patients have become increasingly active partners in research, with contributions that extend beyond participation as study subjects. Patient-based advocacy in health research was energized by the ACT-UP movement in the 1980s, changing the norm from investigator-only control of research toward models of shared control. Disability rights activists contributed to this changing research norm through the 1990s, bringing the concept of "nothing about us without us" to public policy and research-based advocacy \[[@CR4]\]. The advocacy movement overtly combined political action aims with research, and community-based participatory research emerged as a new model of health research \[[@CR5]\]. In the early 2000s, best practice in health outcomes research, which included capturing the patient perspective directly from patients, coalesced with the ethos of patient engagement in research in roles beyond that of study participation only. At the same time, the rise of formal health technology assessment (HTA) programs in the UK, Germany, and elsewhere was coupled with a focus on healthcare stakeholder buy-in in the process and outcomes of HTA, leading to new models of inclusion of patients and others involved in health care. During this time, several countries, including the UK, Canada, Germany, France, and Sweden, initiated or expanded patient and public involvement programs \[[@CR6]--[@CR10]\]. While several hallmarks of HTA have been explicitly rejected in the US, patient and stakeholder engagement have been increasingly embraced as worthwhile \[[@CR10]--[@CR12]\]. The establishment of the Patient-Centered Outcomes Research Institute (PCORI) in 2010 to fund comparative clinical effectiveness research (CER) energized the focus on engagement of patients and other stakeholders as research partners as a new way of pursuing clinical research. PCORI recognizes that in addition to patients, relevant stakeholder groups include clinicians, hospitals and health systems, industry, training institutions, healthcare purchasers, payers, and policymakers. PCORI's requirement, unique for the large US funders of clinical research, that funded projects engage patients and other stakeholders as partners in the production of the evidence, presents a novel opportunity to address the question of how partner engagement impacts research. Even in countries in which engagement has been incorporated into health research and HTA for years (for example, Canada, UK), little is known about the specific effects that engaging research partners have on the research process or its outcomes, and collecting this evidence is part of the planning for CIHR and NIHR \[[@CR13], [@CR14]\]. The current evidence base is recognized as extremely limited thus far (see for example \[[@CR15]\]).We review the evaluation framework developed to guide understanding of PCORI's work and a conceptual model of patient-centered outcomes research (PCOR) \[[@CR16]\], and we present quantitative and qualitative findings from the PCORI experience in partner-engaged research to describe engagement in PCORI projects and characterize its effects. Results are compared against the evaluation framework and conceptual model. Implications for ongoing evaluation of engagement in research are presented. Methods {#Sec2} ======= Conceptual and practical basis for research engagement questions {#Sec3} ---------------------------------------------------------------- This work was guided by the Evaluation Framework \[[@CR17]\], part of PCORI's evaluation plan, developed with input from several groups representing diverse healthcare stakeholders, including the PCORI Board of Governors, Methodology Committee, and Advisory Panel on Patient Engagement. The full framework addresses all aspects of PCORI's work and operationalizes questions about PCORI's work in practice. The section focusing on the impact of engagement in research is the source of the research questions addressed here and is organized into four areas (Fig. [1](#Fig1){ref-type="fig"}a):Fig. 1**a** PCORI evaluation framework for engagement in research. Note: to view the full evaluation framework regarding all of PCORI's work, see <http://www.pcori.org/research-results/evaluating-our-work/planning-our-evaluation-reporting-results>. **b** Conceptual model of patient-centered outcomes research. Reproduced with permission from Frank et al. \[[@CR16]\] description of engagement approaches (who, when, how, etc.), including perceived influence of the research partners and application of the PCOR principles (trust, transparency, honesty, reciprocal relationships, etc.) \[[@CR16]\];effect of engagement on research processes and intermediate outcomes reflective of studies that matter to patients (e.g., research questions, outcomes selected, study design, dissemination of results);longer-term effects of engagement on achievement of PCORI's strategic goals (<http://www.pcori.org/about-us/what-we-do/pcoris-strategic-plan>) to increase the quantity and quality of useful information for health decision making, speed uptake of evidence-based information in health decision making, and influence other research to be more patient-centered; andimpact of engagement in research on better health (health decisions, health care, health outcomes). The conceptual model of PCOR \[[@CR16]\] Fig. [1](#Fig1){ref-type="fig"}b specifies foundational elements, actions, and outcomes relevant for patient-centered outcomes research and provides a theoretical foundation for the evaluation framework. The qualitative and quantitative findings presented here provide a check on the accuracy of that model.Fig. 2Investigator report sample Sample and data collection {#Sec4} -------------------------- From the first funding cycle in December 2012 to June 30, 2016, PCORI funded 434 projects, most for \$3 M each and 3 years, but 11 projects were 3- to 5-year "targeted" projects funded for approximately \$7 M each, and three pragmatic clinical studies were funded for approximately \$12 M each and 5-year duration. As part of annual project progress reporting, PCORI investigators answer closed- and open-ended questions about their experiences with patient and stakeholder engagement in their PCORI-funded projects. Responses are not anonymous. Data in these analyses are from Year 1 or Year 2 progress reports submitted from July 1, 2015 to June 30, 2016 (reports from the third year of the project are excluded due to small sample size) (Fig. [2](#Fig2){ref-type="fig"}). Four projects (\<1.5% of sample) reported no engagement and were excluded (two projects that did not include engagement in their research plan and two projects that did not engage with partners during the reporting period).Fig. 3Ratings of partner influence across study phases (investigator-reported) Upon completion of each annual progress report, investigators are asked to nominate up to 10 patient and other stakeholder research partners to provide feedback on their experience with the project. Partners answer closed- and open-ended items about engagement via the Ways of Engaging-Engagement Activity Tool (WE-ENACT), offered via web survey with an option to complete via phone interview. PCORI emails invitations to partners, with up to three e-mail reminders. Data included in the analyses are from partner reports on their engagement in Year 1 or Year 2 of projects, collected from July 1, 2015 to June 30, 2016. Unlike investigator data, research partner completion of the WE-ENACT is voluntary, so the research partner sample represents a smaller pool of funded projects. MaGil Institutional Review Board approved the protocol for collection of information from research partners and for secondary analysis of project or administrative data. Although more than half (53%) of the projects in this sample include reports by both the investigators and partners, within-project comparisons between investigators and partners are not made for several reasons. Projects represented in the investigator data may not be represented in the partner data (overall, about two-thirds of investigators nominate partners); partner reporting is voluntary (overall 51% response rate); and due to the time lag between partner nominations and partner reporting, reports from those nominated near the end of the sampling timeframe may not have been available at the time of this analysis. Multiple partners may report on a single project (range 1--8). Partners report on their individual participation and influence and, while most participate in multiple phases of the project, investigators are responsible for the entirety of the project and report on the contributions of all partners collectively. Measures {#Sec5} -------- The closed- and open-ended engagement items of the annual progress report for PCORI investigators and the WE-ENACT for research partners (Appendix A & B, online) were developed by PCORI staff based on past data collection efforts \[[@CR18]\], PCORI's Evaluation Framework \[[@CR17]\], a conceptual model of PCOR \[[@CR16]\], and the published literature. Over time, changes have been made to both data collection tools based on cognitive testing, feedback from investigators and partners, PCORI needs, and standard survey practices (e.g., retiring items that have reached saturation, resting and rotating items to minimize respondent burden, and adding/modifying items to capture new information). Investigators report on multiple aspects of partner engagement including the communities represented by research partners, the study phases in which partners are engaged, and engagement approaches used. For each study phase with partner engagement, investigators rate the influence of partners (on a 4-point Likert scale from "None" to "A great deal") and complete an open-ended item on partner engagement activities and the effect of these activities ("Describe what patients and/or other stakeholders actually did and any impact this had on the project"). Investigators also quantitatively rate the partner influence on how the team works together and on research projects other than the specific PCORI-funded project. Each research partner who completes the WE-ENACT reports the primary community he/she represents on the project, the study phases in which he/she has been engaged, and demographic information. For each relevant study phase, the partner is asked to "Describe what you did and how it made a difference." Analysis {#Sec6} -------- Cross-sectional analyses were conducted separately for the investigator and partner samples. Closed-ended item responses from investigators were quantitatively analyzed using descriptive statistics (e.g., proportions, means). Open-ended item responses from both investigators and partners were analyzed via content and thematic analysis \[[@CR19], [@CR20]\]. Hierarchical codebooks were developed using deductive (generated through prior work) and inductive approaches (based on the current analytic samples). Codes were applied to relevant text using NVivo v11 software. To ensure coder agreement, three coders independently coded 10% of the same data and met to reconcile discrepancies. Coded text occurrence queries were generated using NVivo. A frequentist approach to explain the patterns in the data and establish the prominence of themes was deemed appropriate considering the highly structured data collection tools \[[@CR21]\]. The minimum threshold for theme identification was 10% of the relevant responses from both respondent samples. Results are presented separately for the two main aims: description of engagement and characterization of effect of engagement. Results {#Sec7} ======= Investigator report sample {#Sec8} -------------------------- These analyses include 235 reports from investigators: 91 reports on Year 1 and 144 reports on Year 2 (Fig. [2](#Fig2){ref-type="fig"}). The projects represent PCORI's five program areas (*n* = 221; 94%), including "targeted" projects (*n* = 11; 5%) and pragmatic projects (*n* = 3; 1%). Three quarters of these investigators had 10 or more years of research experience and 52% were male (Table [1](#Tab1){ref-type="table"}). The majority (96%) of the investigators were reporting on their first PCORI award; 42% had previously been principal investigators on more than 10 research studies awarded by other funders.Table 1Project reports: investigator characteristicsCharacteristicYear 1 reports (*n* = 91)Year 2 reports (*n* = 144)Total (*N* = 235)Gender (*n*, %) Female44 (48%)68 (47%)112 (48%) Male47 (52%)76 (53%)123 (52%)Research experience^a^ (*n*, %) 0--4 years7 (8%)5 (4%)12 (5%) 5--9 years16 (18%)29 (20%)45 (19%) 10+ years68 (75%)108 (76%)176 (76%) Missing022Previous projects as PI^b^ (*n*, %) 03 (3%)1 (\<1%)4 (2%) 1--528 (31%)53 (37%)81 (35%) 6--1025 (27%)27 (19%)52 (22%) 11--1514 (15%)16 (11%)30 (13%) 16--207 (8%)14 (10%)21 (9%) 21+14 (15%)32 (22%)46 (20%) Missing011^a^Based on question: How many years of research experience do you have related to this field of research?^b^Based on question: Approximately how many grants/contracts have you had funded as the PI or project lead? Partner report sample {#Sec9} --------------------- These analyses include 123 reports from Year 1 and 137 reports from Year 2, a total of 260 reports from partners, from 124 different projects, with one to seven partners reporting per project. (mean 2.1 ± 1.3). Partners in this reporting sample were mostly female (70%) and white (78%), mean age of 54 (±13) (Table [2](#Tab2){ref-type="table"}). The partners in the projects most commonly represented the patient/consumer (29%), caregiver (12%), or clinician (15%) communities.Table 2Project reports: partner characteristicsCharacteristicsYear 1 reports (*n* = 123)Year 2 reports (*n* = 137)Total (*N* = 260)Age (mean ± SD years)55 (±13) (*n* = 115)54 (±13) (*n* = 128)54 (±13) (*n* = 243)Gender (*n*, %) Female79 (68%)96 (73%)175 (70%) Male37 (32%)36 (27%)73 (29%) Transgender1 (\<1%)0 (0%)1 (\<1%) Missing6511Race (*n*, %) American Indian/Alaska Native0 (0%)3 (2%)3 (1%) Asian4 (3%)5 (4%)9 (4%) Black or African American12 (10%)20 (15%)32 (13%) Native Hawaiian or other Pacific Islander1 (\<1%)1 (\<1%)2 (\<1%) White95 (80%)98 (75%)193 (78%) Other7 (6%)3 (2%)10 (4%) Missing4711 Ethnicity (*n*, % Hispanic/Latino)7 (6%) (*n* = 118)5 (4%) (*n* = 131)12 (5%) (*n* = 249)Primary partner community represented (*n*, %) Patient/consumer35 (32%)37 (28%)72 (29%) Clinician18 (16%)14 (11%)32 (13%) Caregiver/family member of patient12 (11%)18 (14%)30 (12%) Patient/caregiver advocacy organization17 (16%)7 (5%)24 (10%) Community-based organization5 (5%)12 (9%)17 (7%) Subject matter expert7 (6%)8 (6%)15 (6%) Clinic/hospital/health System representative5 (5%)7 (5%)12 (5%) Payer (public or private insurance)0 (0%)4 (3%)4 (2%) Policy maker (government official)0 (0%)2 (2%)2 (\<1%) Other^a^11 (10%)23 (17%)34 (14%) Missing14518Educational attainment (*n*, %) Less than high school0 (0%)1 (\<1%)1 (\<1%) High school graduate or GED2 (2%)3 (2%)5 (2%) Post high school training other than college (vocational or technical)3 (3%)4 (3%)7 (3%) Some college16 (13%)25 (19%)41 (16%) College graduate28 (23%)31 (23%)59 (23%) Postgraduate71 (59%)69 (52%)140 (55%) Missing347Previously partnered on other research project^b^ (*n*, % yes)64 (54%) (*n* = 119)Previously partnered with current investigators^b^ (*n*, % yes)46 (42%) (*n* = 109)Time worked with current investigators^b,c^ (mean ± SD)4.3 (± 3.0) (*n* = 45)Study phase(s) in which engaged Researcher understanding of patient and stakeholder needs96 (86%)102 (77%)198 (81%) Research topics and/or research questions43 (38%)37 (28%)80 (33%) Interventions and/or comparators44 (39%)34 (26%)77 (32%) Outcomes and/or measurement62 (55%)56 (42%)118 (48%) Recruitment: Training research staff on how to recruit and work with patients35 (31%)23 (17%)58 (24%) Recruitment and retention: Finding and/or retaining participants49 (44%)43 (33%)92 (38%) Data collection23 (21%)20 (15%)43 (17%) Data analysis and/or results review39 (35%)56 (42%)95 (39%) Data application to real world settings34 (30%)42 (32%)76 (31%) Dissemination22 (20%)40 (30%)62 (25%) Missing11516^a^Includes: Advisory panel member; Community-based organization and free clinic/pharmacy; Chair, parent advisory board; Clinical informaticist; Clinical researcher; Clinical social worker; Community advisor; Community partner intermediary and cultural broker; Disparity expert; Executive director of patient foundation; Long-term and post-acute care provider trade association; Parent; Parent and leader of advocacy organization; Patient advisor x 2; Patient advisor/co-author; Patient advocate x 2; Patient and caregiver; Patient and research advocate; Patient and subject matter expert; Patient/consumer/caregiver/family member of patient; Patient family and child advocate; Peer group facilitator; Practice-based co-PI; Previously a patient; Professional society representative; Project consultant x 2; Research assistant with lived experience; Research expert x 2; Survivor of child abuse^b^Item only asked at Year 1^c^Item only asked of respondents who indicated they previously partnered with the current investigators Description of engagement {#Sec10} ------------------------- ### Quantitative findings {#Sec11} A majority of the investigators reported engaging with research partners that were patients (88%) and/or clinicians (89%), and more than half reported engaging with clinic or health system representatives (60%), patient or caregiver advocacy organizations (57%), and caregivers (51%) (Table [3](#Tab3){ref-type="table"}). Investigators reported engaging an average (±SD) of 4.9 ± 2.0 communities (range 1--11). Common approaches to engaging research partners were via advisory groups (82%), and as research team members (81%); fewer investigators endorsed using opinion polls/interviews/surveys (39%). More than half (56%) of the investigators who reported engaging partners as research team members identified them as engaging at the most active level, as co-investigators on the project. On average, investigators reported using 2.6 ± 1.1 different approaches to engaging partners (range 1--5). Investigators reported that partners were engaged across eight possible study phases (from identifying research topics to disseminating research results; mean 4.9 ± 1.9 phases, range 1--9 when "other" is included). Outcomes and measurement identification were the most common phase with engagement (75%). As expected, the proportions of investigators reporting that partners were engaged in the later aspects of a project, including data collection, data analysis/results review, and dissemination were higher for Year 2 reports.Table 3Characteristics of engagement in research (investigator-reported)Year 1 reports (*n* = 91)Year 2 reports (*n* = 144)Total (*N* = 235)Partner communities engaged^a^ Clinician83 (91%)126 (88%)209 (89%) Patient/consumer82 (90%)125 (87%)207 (88%) Patient/caregiver advocacy organization56 (62%)84 (58%)140 (60%) Clinic/hospital/health System representative53 (58%)81 (56%)134 (57%) Caregiver/family member of patient43 (47%)77 (53%)120 (51%) Subject matter expert43 (47%)78 (54%)121 (51%) Training Institution representative (non-research health professions educator)15 (16%)22 (15%)37 (16%) Policy maker (government official)10 (11%)28 (19%)38 (16%) Payer (public or private insurance)13 (14%)22 (15%)35 (15%) Life sciences industry representative2 (2%)9 (6%)11 (5%) Purchaser (small or large employers)0 (0%)5 (3%)5 (2%) Other^b^26 (29%)68 (47%)94 (40%)Approaches to engaging partners^a^ (*n*,  %) Patient/stakeholder research team members74 (81%)118 (82%)192 (82%)  Team members as co-investigators^c^44 (59%)63 (53%)107 (56%) Advisory groups72 (79%)123 (85%)195 (83%) Opinion polls or interviews39 (43%)53 (37%)92 (39%) Other^d^4 (4%)13 (9%)17 (7%)Study phases in which partners were engaged^a^ (*n*, %) Research topics and/or research questions54 (59%)90 (63%)144 (61%) Interventions and/or comparators62 (68%)101 (70%)163 (69%) Outcomes and/or measurement71 (78%)106 (74%)177 (75%) Other aspects of study design61 (67%)94 (65%)155 (66%) Recruitment and/or retention53 (58%)97 (67%)150 (64%) Data collection29 (32%)64 (44%)93 (40%) Data analysis and/or results review34 (37%)98 (68%)132 (56%) Dissemination24 (26%)77 (53%)101 (43%)^a^Not mutually exclusive^b^Includes biostatisticians, case managers, clinical investigators, community health worker organizations, community-based organizations, community residents, dietitians, educational institutions, National Institutes of Health, nurses, professional organizations/societies, regulatory/compliance professionals, support group organizations, and technology advisors^c^Asked only to those reporting patient or stakeholder partner research team members^d^Includes "conference presentations", "conversations", "peer buddies", "pilot study participants", and "webinars" ### Qualitative findings {#Sec12} Investigators and partners described a wide range of engagement activities (Table [4](#Tab4){ref-type="table"}). Partners commonly described how they s*hared personal perspectives* in early study phases. These perspectives were grounded in partners' lived experiences (e.g., living with or caring for someone with a health condition among patients and caregiver research partners) and professional expertise (e.g., priorities for clinical care among clinician partners or reimbursement decisions among payer partners), and provided insights on how projects could best address the needs and preferences of the priority patient population(s) and thereby ensure patient-centeredness. For example, one patient/consumer was able to share "how different cultures, genders, and age groups of patients value medical communications and attribute meaning to the end of life" (see Table [4](#Tab4){ref-type="table"} for additional examples).Table 4Partner engagement activities---illustrative quotations by study phase (*N* = 235 investigator reports; *N* = 260 partner reports) Both investigators and partners commonly reported that partners *provided guidance and feedback* or *shared decision* *making* about research questions, design, processes, materials, and outcomes. Across relevant phases, between 20 and 72% of investigator and partner responses described how partners *provided guidance and feedback*. Fewer described how partners *shared decision* *making* (but exceeded the 10% theme identification threshold). Both respondent groups also described how partners *participated directly in study conduct* (from 36 to 77% of responses in relevant phases). Specifically, partners participated in study participant recruitment (e.g., speaking directly to patients, training research team staff to interact with specific patient groups) and data collection (e.g., conducting interviews, co-facilitating focus groups, administering surveys to study participants, tracking study participant visits). Despite none of the projects being complete yet, investigators and partners reported partner involvement in dissemination activities (e.g., co-presenting at scientific meetings, developing manuscripts, determining avenues to share findings, writing newsletters, participating in media interviews, speaking with public health officials about the project). Effects of partner engagement {#Sec13} ----------------------------- ### Quantitative findings {#Sec14} Investigators indicated that research partners exerted influence in multiple ways, with more than two-thirds indicating at least a moderate influence (Fig. [3](#Fig3){ref-type="fig"}). Most investigators (73%) indicated that partners had a moderate or great deal of influence on how the team works together. More investigators noted influence at Year 2 relative to Year 1 on research projects beyond the current PCORI-funded (53 vs. 34% rating of moderate or greater influence, (*x* ^2^ = 8.08, *p* \< 0.01). ### Qualitative findings {#Sec15} As a result of partner engagement, multiple aspects of the projects were refined and made more patient-centered (Table [5](#Tab5){ref-type="table"}). Across relevant phases, between 11 and 52% of investigator and partner responses described *enhanced patient*-*centeredness of study processes and outcomes*, and 20--81% described *enhanced study design, conduct, or efficiency*. As described below, research partners had an impact on selection of research topics and/or research questions, interventions and/or comparators, and outcomes and/or measures used. Both investigators and research partners describe participant recruitment and retention and data collection as more efficient as a result of partner engagement. Although few projects in this sample were nearing completion, there is evidence of research partner influence on affecting data analysis and/or results review and dissemination of study results. These qualitative findings correspond to the investigators' quantitative report of influence of research partners across study phases (Fig. [3](#Fig3){ref-type="fig"}).Table 5Effects of partner engagement---illustrative quotations by study phase (*N* = 235 investigator reports; *N* = 260 partner reports) Both investigators and partners reported that partner input confirmed the importance of research topics they were pursuing, inspired pursuit of specific research questions, and/or refined the research questions to be relevant and aligned with patient or other stakeholders' priorities. For example, one patient/consumer noted "Anxiety became a study topic when it had not been considered before" (see Table [5](#Tab5){ref-type="table"} for additional examples). Partners also contributed to refining interventions and/or comparators to be more patient-centered, adapting materials or interventions to be culturally/linguistically appropriate, and modifying the intervention to be less burdensome to participants. Partner contributions to outcomes and/or measurement phase include selection of specific primary and secondary outcomes that matter to patients and other information users, and identification and/or refinement of measures of these constructs. Partners often noted that outcomes of interest to them would have been otherwise overlooked and remained unmeasured. Investigators and partners reported changes to recruitment strategies such as adding or changing recruitment locations, refining inclusion/exclusion criteria, and use of culturally appropriate ways to recruit specific populations. Partner input shaped materials and consent forms (e.g., streamlining, adding more information about risks and benefits). Partners also contributed to participant retention through guidance on the best ways to communicate and suggesting new modes of data collection. Both investigators and partners recognized that partner input contributes to greater perceived value of trial participation among enrolled patients/caregivers, enhanced enrolment rates, and/or improved retention throughout project follow-up periods. Effects of partner contributions on data collection include selection of specific modes of data collection (e.g., electronic vs. phone), informed decisions about timing such as the appropriate follow-up periods, changes as part of clinic work flow, and increased sensitivity around data collection (e.g., insights on why racial/ethnic minorities may be hesitant to share personal information). While these projects are not complete, early signals indicate effects of engagement on data analysis and/or results review, dissemination, new ways to share results, new audiences to reach, improved communication with different audiences, and increasing credibility of the findings. Discussion {#Sec16} ========== PCORI's requirement that awardees engage patients and/or other stakeholders as partners in the research it funds presents a unique opportunity to describe engagement in research as it is implemented across a large portfolio of CER. PCORI's effort to understand engagement in research is a unique, systematic data collection initiative, using data from both investigators and their research partners. This quantity of rich information on the engagement experience from multiple perspectives is not available from any other research funder. PCORI investigators reported engaging a greater number of different types of stakeholder communities in more phases of the project and through more active approaches (e.g., partner co-investigator) than has been previously documented in the literature. For example, Concannon et al. found that most projects in the published literature that reported involving research partners engaged with patients, about half engaged with clinicians, and a few involved other stakeholders \[[@CR22]\]. In contrast, nearly all the PCORI reports in this sample indicated engagement with both patients and clinicians. Engagement with caregivers, patient and caregiver advocacy organizations, and health systems representatives occurred in more than half of the reports. Inclusion of these diverse perspectives in production of research evidence is unprecedented on this scale and presents the opportunity to learn about the ways in which such inclusion changes research process, research quality, and speed of uptake. In the data reported here, investigators and partners recognized similar effects of partner engagement, including refinements to research questions, design, study processes, and outcomes selection. These effects may address historic challenges in clinical research that limit the value of research for its end-users \[[@CR23]\] by increasing the relevance and importance of the research for those end-users. Both investigators and partners report that engagement aided recruitment and retention, particularly noteworthy given that failures in recruitment and retention are a major factor in clinical trial failure \[[@CR24]\]. Improvements in data collection efforts are also noteworthy, and suggest engagement may enhance other strategies to reduce missing data \[[@CR25]\]. The findings here reinforce findings from smaller samples that did not include the partner view \[[@CR8], [@CR26]--[@CR30]\]. PCORI expects engagement to inform key aspects of its funded projects but not necessarily every phase of every project. The role of engagement for a given project depends on the project content and context, and goals and past work in that research area. More work is underway to identify optimal engagement models by understanding intensity (e.g., in the number of partner types, phases and methods of engagement,), as well as effects of engagement such as the number and type of outcomes selected for study, recruitment rates, time to study completion, and study quality. Investigators reported that partners influence the way the team works together. The implications of this should be further explored given the reported challenges of fully including diverse partner types and managing different perspectives \[[@CR31]\]. Investigators, particularly those in the second year, also noted that partners influence other projects beyond the PCORI project. More research is needed to determine whether the number of investigators reporting such influence grows as the PCORI projects progress beyond the second year. The investigator ratings of research partner influence suggest a shift from more transactional approaches to engagement (e.g., discrete, one-way interactions) to more relational approaches, and qualitative analyses will continue to aid full understanding. Longitudinal examination will permit capture of the potential for transforming programs of research, affecting researchers' career trajectories, and changing the culture of how research is conducted \[[@CR16]\]. Understanding the challenges of research engagement and strategies for overcoming those challenges is also critical to supporting a culture of engaged research. Although a detailed analysis of challenges and facilitators of engagement is beyond the scope of this paper, both investigators and partners identify key challenges, such as barriers to scheduling and logistics, limits on engagement due to health problems, and difficulty identifying and fully involving diverse partners \[[@CR31]\]. Across study phases, investigators report more partner engagement than do partners in this sample. This may be due to limited visibility responding partners have of the extent of partner engagement in the project, since partners are asked to report on their individual contributions and many are engaged in limited parts of the project, while investigators report on the collective contributions of all partners across the entire project. Further, partners have recognized a need for investigators to more frequently and clearly report back to partners the ways that their input has contributed to the study \[[@CR31]\]. While the discrepancy may also reflect over-reporting by investigators, this possibility is mitigated by ongoing relationships between investigators and PCORI project officers. While this study represents significant advancements in knowledge about engagement in research, several limitations exist. These self-report data require respondents to recall their experiences over the past year, and likely capture the most salient, but not all, effects of engagement. Furthermore, respondents, particularly investigators, may overestimate their positive experiences with engagement given that this information is reported to the research funder. Although similar themes were identified among investigators and partners, partner reports may underreport the impacts of engagement because some research teams are more effective than others at communicating with partners about the effects of their contributions. Furthermore, the partner views reported here may not represent all PCORI research partners; only a subset of projects have partner respondents in the sample, and partners with more positive experiences may have been more likely to respond. The potential selection bias represents an important limitation to interpretation of partner report particularly. Further, the data were collected in English only, and the low proportion of Hispanic respondents suggests that the current sample may not fully represent all target study populations. Demographic differences between respondents and non-respondents are unknown. The data also do not yet include large enough sub-samples for meaningful comparison by stakeholder type. Additionally, although these items were refined through cognitive interviewing, additional measure refinement and item performance evaluation are needed. Self-report is an important source of information about engagement in research, but other complementary methods, including observational approaches, would enhance understanding and overcome limitations of self-report. Moreover, findings may not generalize to studies funded by others under different requirements, in different contexts, and in other fields. Finally, data reported here are based on the first two years of funded projects. Some effects of engagement, such as impact on patient trust in findings, are not measurable until after project completion. Conclusion {#Sec17} ========== As PCORI projects mature towards dissemination and implementation of findings, the ultimate measure of the impact of engagement in research will be the usefulness of CER information, the use of that information in clinical decision making, and the impact on better health decisions, health care, and health outcomes, as noted in the PCORI Evaluation Framework. These current findings suggest that PCORI-funded projects are on the path towards desired impacts of engagement in research, with engagement of patients and other stakeholders as partners affecting research questions, design, processes, and outcome selection, as well as recruitment strategies and enrollment rates. Comparing results to the conceptual model of PCOR \[[@CR16]\] shows that several hypothesized actions to facilitate PCOR are evident among the projects, including initiating and maintaining research partnerships, capturing and using partner perspectives, facilitating cross-communication with the research team (supported by open-ended feedback), and ensuring meaningful influence (supported by influence ratings) The near-term outcome of a culture of patient-centeredness is supported by the qualitative findings. Ongoing data collection should inform whether longer-term outcomes as specified in the conceptual model (Fig. [1](#Fig1){ref-type="fig"}) are realized and the extent to which specific intermediate outcomes specified in the evaluation framework (Fig. [2](#Fig2){ref-type="fig"}) are evident. Comparing results to the conceptual model of PCOR \[[@CR16]\] shows that several hypothesized actions to facilitate PCOR are evident among the projects, including initiating and maintaining research partnerships, capturing and using partner perspectives, and facilitating cross-communication with the research team. Future directions {#Sec18} ----------------- Future examinations should explore how engagement affects PCORI's large, multi-site pragmatic clinical studies and other differences based on study type (e.g., interventional vs. observational, treatment comparisons vs. health system comparisons) and population (e.g., for hard to reach populations). Future research is needed to understand the unintended consequences of engagement (e.g., on project budget and timeline) and how to mitigate them, as well as other effects of engagement in research hypothesized in the conceptual model \[[@CR16]\], including the effects on partners, investigators, and their institutions in pursuit of establishing a culture of patient-centeredness in research. Several longer-term questions remain to be addressed as the PCORI-funded projects are completed. The extent of concordance, and causes of any discordance, in views between investigators and research partners bears examination, particularly as such concordance may vary by type of study and engagement approach. Additionally, overcoming the potential selection bias in research partner report requires outreach to more research partners and PCORI is exploring ways to obtain a wider range of perspectives on research engagement from the engaged research partners. Further, understanding differences in the effects of engagement by partner type could ultimately inform strategies for how to engage with different stakeholder communities at various parts of the research process. The evaluation framework and conceptual model that guide this work are applicable to research in which health care stakeholders are actively engaged. They can guide collection of data beyond PCORI and provide a foundation for the accumulating evidence about engagement in research, providing a means to improve the efficiency and effectiveness of engagement itself. As the evidence base about research engagement expands, the role of engagement in research in improving public health must be re-examined, to support the incorporation of new knowledge into practice through a feedback loop from theory to research and back. Electronic supplementary material ================================= {#Sec19} Below is the link to the electronic supplementary material. Supplementary material 1 (PDF 367 kb) Supplementary material 2 (PDF 199 kb) **Electronic supplementary material** The online version of this article (doi:10.1007/s11136-017-1581-x) contains supplementary material, which is available to authorized users. PCORI thanks the investigators and research partners who have contributed to the data collected in this report. The authors also gratefully acknowledge Emily Elstad, PhD; Heather Ma; Andrew Amolegbe, MPH; and Tsion Ghedamu from the American Institutes for Research and Ray Maietta, PhD; Jeff Petruzzelli; and Lucinda Hudson, PhD from Research Talk for their contributions to the qualitative analysis of open-ended responses; Alan Carr, PhD and Corina Owens, PhD for their quantitative analysis of the closed-ended items, as well as other PCORI staff contributing to the collection and analysis of these data, including Lauren Fayish, MPH; Ninma Fearon, MPH; Kellie Hall, BA; Rachel Hemphill, PhD; and Beth Nguyen, MPH.

Background {#Sec1} ========== Hepatitis C is an infectious disease that is widely spread geographically. It has been reported that about 180 million people are infected with the hepatitis C virus (HCV) worldwide, accounting for about 3% of the current population \[[@CR1]\]. The incidence of infection can be easily neglected and thus may develop into cirrhosis and even HCV-related hepatocarcinoma and liver failure, which pose serious threats to human health. More than 0.35 million people have died from HCV-related liver disease \[[@CR2]\]. The application of pegylated interferon (PEG-IFN)-alpha combined with ribavirin (PEG-IFN-α/R) had been considered to be the most popular and effective therapy in blocking virus replication before 2015 \[[@CR3]\]. While the novel sofosbuvir-ledipasvir opened a new time for treatment of eligible HCV-infected patients with 90--100% efficiency \[[@CR4], [@CR5]\]. Unfortunately sofosbuvir is much more expensive, with an estimated cost of an additional \$65 billion during the next 5 years \[[@CR6]\]. It should be carefully thought about the necessity and how much people have to take DAA drugs. And it is not the time to ignore the application of PEG-IFN-α/R, especially in China. Patients taking PEG-IFN-α/R from different ethnic groups have different probabilities of reaching a sustained viral response (SVR). In the United States, the rate of obtaining SVR in the black population is almost 50% less than of white patients, and the probability of obtaining SVR in white patients with HCV genotype 1 treated by PEG-IFN-α/R is approximately 42--53% \[[@CR7]\]. In China, the SVR rate was a light higher, could reach 65.3% \[[@CR8]\]. We recently found that northwest patients of China seem to have higher SVR that was 72.6%. Because there were seldom immigrant in the northwest population, the high SVR is reasonable caused by geographic specificity, such as genotype of HCV and polymorphisms of IL-28B. Determination of HCV genotype is important to predict response to antiviral therapy and time of treatment \[[@CR9]\]. Early clinical trials have reported that SVR rates for patients with genotype 1 (42--46%) are lower than rates for patients with non-type 1 genotype (76--82%) \[[@CR10]\]. On the other hand, genome-wide association studies have suggested that host IL-28B single-nucleotide polymorphisms (SNP) rs12979860 CC and rs8099917 TT are significantly correlated with SVR in patients receiving PEG-IFN-α/R treatment \[[@CR11]\]. The probability of obtaining SVR is further reduced in vivo if IL-28B is mutated in patients with HCV genotype 1 infection \[[@CR12], [@CR13]\]. Liver fibrosis had been suggested being closely associated with the risk of HCC development in chronic hepatitis C patients \[[@CR14]\]. The eradication of HCV with antiviral therapy will prevent the progression of chronic hepatitis and associated complications \[[@CR15]\]. But it was never paid more attention in China \[[@CR8]\]. To determine the favorable patients for treatment with PEG-IFN-α/R, HCV genotype, IL-28B SNP, Fibrosis 4 (FIB-4) index and aminotransferase-to-platelet ratio index (APRI) score we retrospectively analyzed with SVR in this study. Methods {#Sec2} ======= Patients {#Sec3} -------- All patients with CHC were enrolled for this research with signed informed consents following the protocols approved by the Institutional Review Board of the Fourth Military Medical University (Table [1](#Tab1){ref-type="table"}). Inclusion criteria: patients were firstly diagnosed as HCV infection since August 2013, naive-treatment from October 2013 and HCV RNA \> 1000 IU/mL. Exclusion criteria: Patients with recurrence of hepatitis C, hepatitis infected with HAV, HBV, HDV, HEV, EBV or CMV; HIV infection, Diabetes, autoimmune liver disease and HCC etc. were excluded. ###### Clinical characteristic of patients with different HCV genotypes HCV Genotype No. Age (y), range Sex (No. of Patients, %) Viral load (10^6^ IU/mL), range ALT (U/L), range AST (U/L), range -------------- ----- ---------------- -------------------------- --------------------------------- -------------------- ------------------ ----------------- 1a 12 39.67 (24--60) 7 (58.33) 5 (41.67) 2.33 (0.01--29.90) 58.17 (19--249) 37.08 (10--157) 1b 111 46.81 (20--79) 52 (47.27) 58 (52.73) 1.62 (0.02--46.1) 50.15 (13--392) 44.96 (16--453) 2a 89 50.45 (22--80) 40 (44.94) 49 (55.06) 2.60 (0.00--18.10) 49.38 (12--554) 35.25 (15--205) 2b 1 54.00 (54--54) 0 (100.00) 1 (0.00) 1.08 286.00 351.00 3a 11 37.91 (25--53) 8 (72.73) 3 (27.27) 5.52 (0.21--29.5) 91.27 (37--340) 50.36 (32--160) 3b 5 38.00 (23--45) 3 (60.00) 2 (40.00) 1.71 (0.10--6.88) 52.40 (58--204) 37.00 (44--141) 4a 0 5a 1 43.00 (43--43) 2 (100.00) 0 (0.00) 0.00 0.00 0.00 6a 1 25.00 (25--25) 1 (100.00) 0 (0.00) 0.00 0.00 0.00 *Z* = 3.6464 chi-squared =0.643 *Z* = 1.8797 *Z* = 0.6820 *Z* = 0.6820 *P* value .067 .836 .113 .605 .605 The low number of patients with 2b, 4a, 5a and 6a did not meet the statistical requirements; therefore, these were excluded for follow-up statistics We collected data on 230 patients with chronic HCV infection who were seen at Xijing Hospital from October 2013 to February 2016. The average age was 46.71 years (range, 20--80 y), with 112 male and 118 female patients. Patients had been treated with standard of care for 24--48 weeks. PEG IFN-α/RBV: The recommended dose of PEG IFN-α-2a (Pegasys Roche Shanghai) for chronic hepatitis C was 180 ug per time, once a week, subcutaneous injection of the abdomen or thigh. The dose of RBV was determined by the genotype of virus: the dose for genotype 2 or 3 was 800 mg a day for 24 weeks, and the dose of genotype 1 was 1000--1200 mg daily according to body weight, for 48 weeks. We mainly investigated outcomes after 24 weeks. There are many serological markers for HCV or evidence of liver disease including HCV RNA was used to understand the activity of virus replication, ALT, AST, Total bilirubin, direct bilirubin, indirect bilirubin, albumin, globulin, choline esterase, alkaline phosphatase, phosphatase and abdominal ultrasound were used for evaluation of liver damage, CT or MRI was used to be clear of the extent of liver damage. Liver biopsy is the gold standard for evaluation of liver inflammation and fibrosis staging in patients. In addition, there were also patient compliance issues. After careful consideration we chose these common and easy to get markers as ALT and AST for liver damage, FIB-4 and APRI to evaluate liver fibrosis. Use of viral content to evaluate curative effect was according to the 2014 European Association for the Study of the Liver Recommendations on Treatment of Hepatitis C and the Guideline of Prevention and Treatment of Hepatitis C \[[@CR16], [@CR17]\]. We evaluated SVR using quantitative real-time fluorescence polymerase chain reaction (ViiA7 OX, Life Technology) of HCV RNA (less than 15 IU/mL) for at least 24 weeks of follow-up at the end of treatment. Samples were collected and separated from the peripheral blood and serum and then stored at −20 °C until analyses. DNA/RNA extraction {#Sec4} ------------------ For DNA extraction for IL-28B gene detection, we used a blood genomic DNA extraction kit (Tiangen, Beijing, China). For HCV RNA extraction, we used the MinElute column QIAamp method and the virus genome DNA/RNA extraction kit (Tiangen). All extracted DNA/RNA were then immediately used for gene detection or stored at −80 °C. Quantitative real-time fluorescence PCR {#Sec5} --------------------------------------- Primer sequences were designed by using DNAMan 6.0.3.99 (LynnonBiosoft, San Ramon, CA, US) for IL-28B gene amplification, including IL-28B rs8099917, IL-28B rs12979860 and IL-28B rs12980275 genes (Table [2](#Tab2){ref-type="table"}). Among them, 5'of 3reversed primers were labeled with biotin. Amplification included heating at 95 °C for 5 min, 95°Cfor 20 s, and 60 °C for 20 s, with steps repeated for 40 cycles. Amplification was run on a 7500 fast real-time PCR system (Applied Biosystems, Foster City, CA, USA). HCV genotype was measured by using an HCV genotyping PCR kit (Qiagen, Beijing, China), following the manufacturer's protocol.Table 2Primers for genotype analysis of IL-28BPrimerIL-28BSequenceRT-PCR primersrs12989760 forward:5′-GTCGTGCCTGTCGTGTACTGA-3′rs12989760 reverse:5′-AGCGCGGAGTGCAATTCA-3′rs8099917 forward:5′-CTCCTTTTGTTTTCCTTTCTGTGA-3′rs8099917 reverse:5′-ACATAAAAAGCCAGCTACCAAACT-3′rs12980275 forward:5′-ACATGAGGTGCTGAGAGAAGTCAA-3′rs12980275 reverse:5′-TACCCCGGCAAATATTTAGACAC-3′Sequencing primersrs129897605′-AGCTCCCCGAAGGCG-3′rs8099917:5′-TCCTTTCTGTGAGCAAT-3′rs12980275:5′- GAAGTCAAATTCCTAGAAA −3′ Gene sequencing {#Sec6} --------------- IL-28B gene polymorphism was detected with the use of a pyrosequencing method on Q24MDX (Qiagen, Hilden, Germany). Sequencing primers were designed by the Q24 PyroMark. HCV polymorphism sequencing primers were also provided by Qiagen. Liver fibrosis staging {#Sec7} ---------------------- The degree of liver fibrosis was evaluated with APRI score, which can be used for the assessment of liver cirrhosis \[[@CR18]\]. APRI scores \> 2 in adults indicate that the patient has already had liver cirrhosis. The APRI score is calculated as follows: (aspartate aminotransferase \[AST\]/ULN) × 100/platelets (10^9^/L), where ULN is the upper limit of normal value. Fibrosis-4 index was based on values of ALT, AST, platelet count and patient age. This index can be used to diagnose liver fibrosis (similar to least significant fibrosis using METAVIR F scoring system ≥2) \[[@CR19]\]. A significant liver fibrosis has occurred if a patient shows aFIB-4 index of \>3.25. FIB-4 is calculated as follows: age × ALT (IU/L)/(platelet count \[10^9^/L\] × AST \[IU/L\]) ^1/2^. Statistical analyses {#Sec8} -------------------- We used Pearson chi-squared and Kruskal-Wallis tests to analyze the qualitative data with SPSS 19.0 (IBM, USA.). A logistic regression model was used to analyze the correlation between SNPs (IL-28B rs8099917, IL-28B rs12979860 and IL-28B rs12979860) and SVR of patients. Odds ratio (OR) was used to describe the correlating degree of disease and exposed factors. OR tests were two-sided tests, in which *P* \< 0.05 was considered to be statistically significant. Results {#Sec9} ======= HCV genotype and distribution {#Sec10} ----------------------------- HCV genotypes of 230 patients with chronic HCV infection were sequenced, with results shown in Table [2](#Tab2){ref-type="table"}. One patient had mixed infection of 1b and 5a. For statistical analyses, this patient was analyzed in both the 1b and 5a genotype groups. Statistical results of 1a, 1b, 2a, 2b, 3a, 3b, 4a, 5a and 6a genotypes are presented in Fig. [1](#Fig1){ref-type="fig"}. The ratio of genotypes were as follows: 5.19% with 1a (12/230),48.05% with 1b (111/230),38.53% with 2a (89/230),0.43% with 2b (1/230),4.76% with 3a (11/230),2.16% with 3b (5/230),0% with 4a (0/230),0.43% with 5a (1/230), and 0.43% with 6a (1/230). There were no significant differences in age, sex, viral load and alanine aminotransferase (ALT) and aspartate transaminase (AST) levels (*P* \> 0.05).Fig. 1Distribution of HCV genotype. 86% of the patients with HCV infection were genotype 1 and 2, with 1b accounting for 48.05% and 2a accounting for 38.53%, followed by 1a (5.19%), 3a (4.76%), 3b (2.16%), 2b (0.43%), and 6a (0.43%) IL-28B genotype in patients with chronic HCV infection {#Sec11} ------------------------------------------------------ Fifty-one patients with HCV were included in the IL-28B gene polymorphism loci analyses (Table [3](#Tab3){ref-type="table"}). Of the IL-28B genotypes, 76.47% of patients were rs8099917 TT, 76.47% were rs12979860 CC, and 72.55% were rs12980275 AA. We also found that 5.88% of the IL-28B genotypes were rs8099917 GG, rs12979860 TT or rs12980275 GG. Details of clinical features are presented in Table [3](#Tab3){ref-type="table"}. Statistical results showed no statistical differences in age, sex, viral load and ALT and AST levels (*P* \> 0.05).Table 3IL-28B gene polymorphism in patients with chronic HCV infection and clinical characteristics of patients with different IL-28B SNPsGenotypeSNPPercentAge (y)Sex (No. of Patients, %)Viral Load (10^6^ IU/mL), rangeALT (U/L), rangeAST (U/L), range(%)MaleFemaleIL-28B rs8099917TT\ (*n* = 39)76.4748 (25--64)54 (52.9)52 (47.1)7.98 (0.24--72.46)34 (18--179)38 (17--203)TG + GG\ (*n* = 12)23.4347 (22--58)21 (55.3)16 (44.7)9.25 (0.89--59.71)36 (28--150)40 (24--138)IL-28B rs12979860CC\ (*n* = 39)76.4749 (22--64)57 (52.3)52 (47.7)10.23 (1.02--68.52)35 (21--112)32 (19--143)CT + TT\ (*n* = 12)23.4349 (20--58)18 (52.9)16 (47.1)8.72 (0.48--65.32)37 (19--135)34 (17--178)IL-28B rs12980275AA\ (*n* = 37)72.5547 (28--64)51 (51.5)48 (48.5)11.25 (1.37--66.68)38 (22--144)31 (20--123)AG + GG\ (*n* = 14)27.4546 (25--59)24 (54.5)20 (45.5)9.67 (0.69--71.26)32 (19--179)29 (17--112)*Z* = 0.125chi-squared = 0.413*Z* = 0.579*Z* = 0.378*Z* = 2.143*P* value.912.648.825.617.346 Correlation between HCV genotypes and antiviral efficacy {#Sec12} -------------------------------------------------------- HCV genotypes 1b and 2a accounted for 86% of all samples; therefore, we mainly evaluated the correlation between these 2 genotypes and antiviral efficacy. SVR rates in patients with HCV genotype 1b and HCV genotype 2a were 52.25% (58/111) and 75.28% (67/89), with differences being statistically significant (chi-squared =21.56; *P* \< 0.01). In patients with HCV genotype 1b, 70.67% (53/75) did not reach SVR (Fig. [2a](#Fig2){ref-type="fig"}).Fig. 2**a** Correlation between HCV genotype distribution and SVR. Sustained viral response (SVR) rates by HCV genotypes 1b and 2a were 52.25% (58/111) and 75.28% (67/89), statistically significant at *P* \< 0.01. Rates of patients with HCV genotype 1b and 2a who did not reach SVR were 70.67 and 29.33%. Relationship between IL-28B genotype and SVR was showed in (**b**). SVR rates in patients of chronic HCV infection with different IL-28 single-nucleotide polymorphisms (SNP) (rs8099917 TT, rs12979860 CC and rs12980275 AA) were 92.41% (25/27), 92.86% (26/28), and 88.89% (24/27), respectively. In contrast, only 7.41, 7.14 and 11.11% of patients with IL-28B rs8099917 TG + GG, rs12979860 CT + TT, and rs12980275 AG + GG obtained SVR. **c** showed correlation among IL-28B SNP, HCV genotype and SVR Role of IL-28B gene polymorphism in antiviral efficacy {#Sec13} ------------------------------------------------------ We used SNP Stats Software (<http://bioinfo.iconcologia.net/snpstats/start.htm> \[[@CR13]\]) for correlation analyses. After adjusting for confounding factors (including age, sex, viral load, ALT and AST), we found that there was a correlation between patients with SVR and IL-28B genotype (*P* \< 0.05), with OR (95% confidence interval) of11.02 (3.98--23.16) for rs8099917, 10.20 (5.23--22.14) for rs12979860 and 10.08 (2.98--18.59) for rs12980275. Figure [2b](#Fig2){ref-type="fig"} show that patients with chronic HCV infection the SVR rate of IL-28B rs8099917 TT was (25/27,92.41%), IL-28B rs12979860 CC was (26/28, 92.86%) and IL-28B rs12980275 AA was (24/27, 88.89%). In contrast, only 7.41, 7.14 and 11.11% reached SVR with IL-28B SNP rs8099917 TG + GG, rs12979860 CT + TT, and rs12980275 AG + GG, respectively. Patients with protective genotypes were more likely to obtain SVR. Role of IL-28B SNP and HCV genotype in antiviral efficacy {#Sec14} --------------------------------------------------------- We combined the results of IL-28B SNP and HCV genotype and analyzed the correlation between these factors. We found that 12/51 patients with HCV genotype 1b infection obtained SVR, in which 8/12 cases were rs12979860 CC (shown in Fig. [2c](#Fig2){ref-type="fig"}). We also found that the 7 of10 patients infected with HCV genotype 2a who obtained SVR were rs12979860 CC. In addition, 2 in 17 patients with HCV genotype 1b infection and SNP rs12979860 CC did not obtain SVR, and only 1/3 patients with HCV genotype 2a infection and SNP rs12979860 CC did not obtain SVR. Patients HCV genotype 2a infection with IL-28B SNP rs12979860 CC were more likely to reach SVR. Association of liver fibrosis staging and cirrhosis in patients with chronic HCV infection and antiviral efficacy {#Sec15} ----------------------------------------------------------------------------------------------------------------- Because there were only 190 out of 231 patients did PLT detection, and APRI and FIB-4 were calculated according to results of patients' PLT. So APRI score and FIB-4 index were calculated in 190 of 231 patients with chronic HCV infection. Sixty patients had FIB-4 index \> 3.25, including 32 patients with genotype 1b infection and 21 patients with genotype2a infection. In this patient group, 31.0% of patients with HCV genotype 1b (19/32 × 58/111) and 39.4% of patients with genotype 2a (11/21 × 67/89) reached SVR (not statistically significant; *P* \> 0.05). Thirty-nine patients had APRI score \> 2, including 11 patients with genotype 1b and 8 patients with genotype 2a infection. In this patient group, 30.3% with HCV genotype 1b (11/19 × 58/111) and 50.2% with genotype 2a (8/12 × 67/89) reached SVR (*P* \< 0.05). Although we could not prove that genotype affected the ability to reach SVR in patients with cirrhosis (APRI \> 2), we did observe that APRI score \> 2 significantly affected SVR. Details are shown in Tables [4](#Tab4){ref-type="table"} and [5](#Tab5){ref-type="table"}.Table 4Characteristics of patientsVariablesNo. of Patients (%of Total)Total number of patients231Age ± standard deviation, y46.71 ± 14.34Male113Female118APRI data190 (82.3)FIB-4 data190 (82.3) Table 5FIB-4 index, APRI scores and SVR of patientsGenotypeNo. of Patients (% of Total)No. of Patients with FIB-4 \> 3.25 (% of Total)No. of Patients with APRI \> 2 (% of Total)Total SVR (%)No. of Patients with FIB-4 \> 3.25 and SVR (%)No. of Patients with APRI \> 2 and SVR (%)1b90 (81.1)32 (16.8)19 (10.0)58 (64.4)19 (31.0)11 (30.3)2a76 (85.4)21 (11.1)12 (6.3)^\*^67 (88.1)11 (39.4)8 (50.2)Other24 (77.4)7 (3.7)8 (4.2)13Total190 (82.3)60 (31.6)30 (20.5)138 (72.6)^\*^ *P* \< 0.05 Discussion {#Sec16} ========== China has shown a new trend in HCV infection, with epidemic levels of genotypes1, 2, 3 and 6 and no genotypes 4 or 5 found. The most common genotype in China is1b and 2a, with incidence rates of 73.1 and 18.5%, followed by genotypes 3a, 6a, 3b and 1a. Genotypes 3 and 6 are geographically distributed more widely \[[@CR20]\]. In our group, which included 230 patients with chronic HCV infection, sequencing results showed that the incidence of HCV in the Shaanxi region was \> 86% with genotypes 1 and 2, with much lower numbers with genotypes 3 and 6, and none with genotype 4. More patients had genotype 1b (48.05%) and 2a (38.53%), followed by 1a, 3a, 3b, 2b and 6a. The high percent of genotype 2a might be one important reason for high SVR. And we truly found that patients with genotype 2a had greater SVR rates (75.28%) than patients with genotype 1b (52.25%). The results have a slight discrepancy versus the previous reports, which SVR rates for patients with genotype 1 were 42--46% and non-type 1 genotype were 76--82% \[[@CR10], [@CR21]--[@CR23]\]. Furthermore, our SNP analysis results of 51 patients with chronic HCV infection (IL-28B rs8099917, IL-28B rs12979860 and IL-28B rs12980275 SNP) and sequencing results of 230 patients with chronic HCV infection showed that more patients had IL-28B rs8099917 TT versus rs8099917 GG, more had IL-28B rs12979860 CC versus TT, and more had IL-28B rs12980275 AA versus GG in Northwest China, similar to some previous reports \[[@CR24], [@CR25]\]. Presence of rs8099917 is one of the independent predictors in HCV 1b patients treated with PEG-IFN-α/R or interferon-α 2 only. Presence of rs12980275 has great relevance with SNP of rs12979860 \[[@CR26]\], which plays an important role in the prediction of SVR in HCV patients treated with PEG-IFN-α/R, particularly in patients with HCV genotype 1 or 4 \[[@CR10], [@CR27]\]. So rs8099917TT, rs12979860CC, and rs12980275AA were considered to be protective genotypes \[[@CR28]\]. In this study, SVR rates of patients with IL-28B rs8099917 TT, rs12979860 CC and rs12980275 AA were 92.41% (25/27), 92.86% (26/28) and 88.89% (24/27) separately. SVR rates in patients with protective genotypes accounted for more than sixty percent of total, whereas the SVR rates of non-protective genotypes were very low. This suggested that the protective IL-28B genotypes were more likely to result in patients with chronic HCV infection obtaining SVR, which is consistent with previous reports \[[@CR26]--[@CR29]\]. In addition, we corrected for age, sex, AST and ALT levels, HCV genotype and other factors with the use of SNPStats software \[[@CR30]\] and found that patients with chronic HCV infection who obtained SVR 24 weeks after standard of care antiviral treatment were closely related to HCV genotype and IL-28B SNP (*P* \< 0.05). That means more than sixty percent of patients are suitable for PEG-IFN-α/R treatment. Well, the major limitation in this part of study is that the sample amount was relatively too small to determine the role of IL-28B gene polymorphism in antiviral efficacy. However, our findings were in accordance with previous data \[[@CR26]--[@CR29]\]. Interestingly, we found that differences in cirrhosis progression between 1b and 2a were statistically significant, confirming that HCV genotype can influence the progression of liver cirrhosis. FIB-4 index \> 3.25 or APRI score \> 2 indicated that patients had significant liver fibrosis or even cirrhosis \[[@CR31]\]. In this study, the SVR rates of patients with FIB-4 index \> 3.25 and genotypes1b and 2a were 31.0 and 39.4%, respectively, which were much lower than that shown in patients with genotype 1b (62.06%) and genotype 2a (74.54%) with FIB-4 index ≤ 3.25. This indicated that patients with obvious liver fibrosis were less likely to reach SVR, which was associated with HCV genotype (1b or 2a). In addition, SVR rates of patients with APRI score \> 2 were 30.3% for genotype 1b and 50.2% for genotype 2a. The large difference indicated that patients with cirrhosis had greater difficulty reaching SVR, for both HCV genotype 1b and 2a, and the progression of cirrhosis (APRI \> 2) could be influenced by genotype. The clearance of HCV in patients with advanced-stage liver fibrosis can reduce the incidence of decompensated liver cirrhosis. Conclusions {#Sec17} =========== There were high proportion of HCV genotype 1b and 2a in Northwest China. In both HCV 1b and 2a genotypes, patients with protective-genotype of IL-28B were more likely to obtain SVR. However, those with significant fibrosis or cirrhosis were less likely, no matter their genotype. Combined factors of HCV genotype, IL-28B genotype, FIB-4 and ARPI potentially have a very high prediction and clinical value regarding treatment with PEG-IFN-α/R and prognostic evaluation of chronic hepatitis C patients. APRI : Aspartate aminotransferase-to-platelet ratio index CHC : Chronic hepatitis C FIB-4 : Fibrosis 4 index HCV : Hepatitis C virus IL-28B : Interleukin (IL)-28B PEG-IFN-α/R : Pegylated interferon-alpha combined with ribavirin SNP : Single-nucleotide polymorphisms SVR : Sustained viral response We would like to thank Medical Editor for our language correction. Funding {#FPar1} ======= This work was supported by grant from Science and Technology Project of Shaanxi Province (no. 2013 K12-05-10), and National instruments major projects of China (2012YQ03026107). Authors' contributions {#FPar2} ====================== YL and JW performed experiments, analyzed the data, designed the Figures and drafted the manuscript. YX and LY contributed to experiments and helped data analysis. JW and HL were also involved in performing experiments. BX assisted with analyzing and interpreting data, and provided technical support. XH and YM designed the research, analyzed the data and critical revision of manuscript for important intellectual content. All authors read and approved the final manuscript. Competing interests {#FPar3} =================== The authors declare that they have no competing interests.

1. Introduction =============== Attempts have been made to develop a portable gastrointestinal endoscope.^\[[@R1],[@R2]\]^ The advantages of a portable endoscope with high visual quality and maneuverability are evident; equipment with high cost and space requirements would be unnecessary and emergency endoscopy could be available anywhere. The development of a portable disposable ultrathin endoscope (DUE) would increase patient satisfaction, prevent unnecessary sedation, and reduce infection.^\[[@R3]--[@R5]\]^ Percutaneous endoscopic gastrostomy (PEG) is commonly performed to provide long-term enteral feeding.^\[[@R6],[@R7]\]^ PEG via the introducer method is feasible, safe, and can be performed when the pull-through technique is difficult.^\[[@R8]\]^ This method has the advantage of reducing postprocedural peristomal infection.^\[[@R9],[@R10]\]^ The use of an ultrathin endoscope improves the patient\'s comfort, reduces cardiopulmonary risk and has a high success rate.^\[[@R3],[@R4],[@R10]\]^ PEG insertion via the introducer method requires accurate gastric visualization, making it a potential test procedure for DUE evaluation. The E.G.Scan™ II (IntroMedic Co., Ltd, Seoul, S. Korea) is a portable DUE. It allows esophagoscopy in outpatient clinics and is safe and well-tolerated.^\[[@R2]\]^ Its effectiveness has been demonstrated in esophageal disorders and gastrointestinal bleeding.^\[[@R2],[@R11]--[@R13]\]^ The advantages of this system include portability and disposability. The E.G.Scan™ II has an air insufflation function which allows visualization of the gastric walls and external compression site. Our aim was to verify if PEG can be performed under portable DUE guidance. 2. Methods ========== 2.1. Patients ------------- This was a prospective, open-labeled study conducted at Seoul St. Mary\'s Hospital, College of Medicine, The Catholic University of Korea, Seoul, Republic of Korea from August 2013 to December 2014. Consecutive patients referred for PEG were enrolled and compared with gender and indication-matched historical controls who had undergone PEG under conventional ultrathin endoscope (CUE) guidance. Patients who had a history of esophageal or gastric surgery, were at high risk of gastric bleeding, had mechanical ileus, had a history of PEG insertion, or did not provide written informed consent were excluded. This study was approved by the institutional review board of Seoul St. Mary\'s Hospital (KC13DISI0255) and registered with Clinicaltrial.gov (NCT02183207). 2.2. Portable DUE ----------------- The stomach was visualized during PEG by the E.G.Scan™ II, which comprises a disposable probe, controller, and viewing program. The disposable probe specifications of this system are as follows: an optic capsule with a 6-mm diameter and 3- to 50-mm view depth, light source of 4 white light-emitting diodes, a 3.5-mm-diameter tube with a length of 109 mm and weight of 42.5 g, a 125° view angle, and up- and down-bending angles of 160° each. The controller has free, capture, and air insertion functions. The processor is able to displace room air at a rate of 5 L/minute with a maximal air pressure of 0.45 kg/cm^2^. A real-time view is available via a viewing program that can record both video and still images and be set up in normal desk or laptop computers. In the present study, images were recorded at 30 frames per second with a 400 × 400-pixel resolution (Fig. [1](#F1){ref-type="fig"}). {#F1} 2.3. PEG via the introducer method ---------------------------------- All PEG procedures were performed by either B.M.K. or L.C.H., both certified as endoscopy experts by the Korean Society of Gastrointestinal Endoscopy. All procedures were performed free of preprocedural antibiotic administration unless the patient was taking antibiotics for another underlying disease. The procedures were performed at the endoscopy center of Seoul St. Mary\'s Hospital under continuous patient monitoring. PEG was carried out using an introducer PEG kit (Cliny PEG Kit; Create Medic, Yokohama, Japan) according to previously described methods.^\[[@R9]\]^ Before insertion of the endoscope, 0.01 to 0.03 mg/kg of midazolam and 50 mg of pethidine were administered intravenously. Upon verification of mild to moderate sedation, 2% lidocaine jelly was applied to the nasal cavities and the DUE was inserted via the transnasal route. Upon entering the stomach, optimal PEG tube placement at the anterior wall of the lower body was determined by depressing the abdominal wall with a finger. About 10 mL of 1% lidocaine was subcutaneously injected into the region marked for PEG tube insertion. The stomach was punctured with a double-lumen gastropexy device under DUE visualization. Two sutures were placed 2 to 3 cm apart, and a 5-mm incision was made between the 2 suture points. A trocar with a plastic peel-away sheath was inserted into the stomach through the incision site. The trocar was verified with the DUE; the trocar was then removed, and a 15-Fr PEG tube was inserted via the peel-away sheath. The balloon at the end of the PEG tube was inflated with 5 mL of sterile water, and the PEG tube was pulled until appropriate approximation of the balloon to the gastric wall was achieved (Fig. [2](#F2){ref-type="fig"}, see Video, Supplemental Digital Content 1 which shows the PEG insertion). The peel-away sheath was removed and the retaining plate put in place. If the PEG tube insertion failed, the DUE was changed to a conventional ultrathin endoscope (CUE) (GIF-XP260N; Olympus Optical Co., Ltd, Tokyo, Japan). All cases had the gastric insertion state verified by CUE, immediately after completion of PEG insertion. {#F2} 2.4. Follow-up -------------- Enteral nutrition was initiated through the inserted PEG tube in all patients 1 day after the procedure. Laboratory parameters including the hemoglobin level, white blood cell count, and C-reactive protein level were measured on the morning of the procedure and the day afterward. On the morning after the procedure, chest and abdominal radiographs were taken to check for signs of adverse events. Body temperature was checked every 6 to 8 hours for 3 days after the procedure or until discharge. The PEG site was checked by the medical team or the home-visiting nurse for 30 days; peristomal infection was defined as pus or pus-like discharge at the insertion site. 3. Results ========== Twenty-five patients were enrolled in the DUE group and compared with 25 controls. The baseline characteristics are shown in Table [1](#T1){ref-type="table"}. Successful gastric insertion was achieved in 23 patients in the DUE (92.0%) and 24 in the CUE group (96.0%). Of these, PEG insertion was successfully performed in 21 patients in the DUE (91.3%) and 23 in the CUE group (95.8%). Mean PEG insertion time was 22.7 ± 9.3 minutes (9--35 minutes) in the DUE and 17.1 ± 5.7 minutes (8--26 minutes) in the CUE group (*P* = 0.044). ###### Baseline characteristics of study subjects.  There were 3 cases of failed gastric insertion, 2 in the DUE and 1 in the CUE group. All were in esophageal cancer patients with a blocked lumen. PEG failed despite gastric insertion in 3 cases, all due to failure in identifying the insertion site. In the DUE group, the DUE was changed to a stiffer and brighter CUE, and PEGs were successfully achieved. In the CUE group, all failed cases were successfully converted to percutaneous gastrostomy insertion under fluoroscopy guidance. There were 3 complications in the DUE group: 1 case of bleeding and 2 cases of pneumoperitoneum. Postprocedural bleeding occurred in an 89-year-old man who underwent PEG for aspiration due to dementia. Fresh blood was noted in the PEG tube 12 hours after insertion with a fall in the hemoglobin level from 9.6 to 6.3 mg/dL. Emergency endoscopy was performed, which found the insertion site at the anterior wall of the lower body and oozing some fresh blood. The PEG balloon was pulled up to compress the bleeding site, resulting in reduction of the oozing. After conservative treatment, enteral feeding via the PEG tube commenced on the third day without any further events. There were 2 cases of pneumoperitoneum discovered by plain chest and abdominal X-rays. In 1 case, 1 pexy suture was cut during trocar insertion. Although the other suture was pulled up during the procedure, X-rays showed the presence of pneumoperitoneum. The other case of pneumoperitoneum was discovered by routine follow-up X-rays after the procedure. Neither patient complained of any abdominal symptoms, and no fever or peritoneal signs developed. Enteral feeding was started on the second day after PEG tube insertion. The pneumoperitoneum resolved in 2 weeks in 1 patient and 3 weeks in the other without any further events. The CUE group had one case of fever due to a liver abscess. This was determined to be nonrelated to PEG insertion and was conservatively managed with antibiotics. 4. Discussion ============= The results of this study show that PEG via the introducer method may be possible under portable DUE-guided visualization. Successful visualization of the PEG tube insertion site at the anterior wall of the lower body was achieved in 91.3% (21/23) of the cases in which the DUE reached the stomach. These results suggest that endoscopic examination of the stomach could be achieved via a portable DUE. This study also verified several advantages of the E.G.Scan™ II system. First, the anterior wall of the lower body, external compression of the gastric wall, and insertion of the pexy and trocar could be verified through the DUE. Thus, evaluation of the stomach could be possible through a DUE. Second, PEG was successful in 91.3% of the attempted cases in which the DUE reached the stomach. Successful endoscopy via a disposable portable system suggests that this procedure could be performed in centers without dedicated endoscopy and disinfection facilities such as outpatient clinics, nursing care centers, emergency rooms, and field hospitals. Third, though we did not compare the cost-effectiveness between DUE and CUE, the DUE has the potential to be financially more attractive. The DUE probe costs around \$100 while a CUE costs about \$700.^\[[@R13],[@R14]\]^ In case of facility costs, the DUE processing system costs about \$5000 (company marketing information) while the added cost of an upper gastrointestinal endoscope, video endoscope processors/light source, and automated endoscope reprocessors range from \$49,000 to 117,000.^\[[@R15],[@R16]\]^ Considering the substantial investment required for conventional endoscopy facilities, we believe that the portable and inexpensive nature of the DUE has merit in locations where funds are lacking. In the DUE group, the stomach was not reached in 2 cases and the PEG tube insertion site was unable to be identified in another 2 cases. These occurrences suggest that improvements should be made to the E.G.Scan™ II system. The cases of failure to reach the stomach were due to esophageal cancer blocking the esophageal lumen. The DUE is very flexible, similar to a conventional nasogastric tube. As such, it is difficult for the DUE to pass through narrow obstructions, which resulted in the failure to reach the stomach in our study. Because PEG tube insertion was successfully accomplished under conversion to a CUE, which is much stiffer than a DUE, we believe that improvements in the stiffness of the DUE would result in better outcomes. One case of failure was due to the low light for visualization and the lack of a channel for suction. Although the DUE has 2 light settings, 1 bright and 1 dark, the bright setting was insufficient for observing the opposite (anterior) wall of the stomach. The patient in this case was obese and had a large stomach. When the stomach was fully distended, the pexy site could not be verified. If the light source had been brighter or a suction channel was present to reduce the distended stomach, PEG should have been possible, as was done after conversion to the CUE. The other case of failure was due to loss of gastric spatial orientation during the study. The DUE does not have up--down and left--right reference points. Because of its flexible, the scope readily rotates during insertion through the nasal and oral cavities. Upon reaching the stomach, it may be difficult to accurately assess the anatomical position. This patient had a history of a head and neck cancer operation that led to anatomical changes in the mouth and pharynx. Gastric insertion required twisting of the scope which resulted in loss of spatial orientation. This case was also successfully managed after conversion to the CUE. There were 3 cases of adverse events in the DUE group: 1 of bleeding and 2 of pneumoperitoneum. Emergency endoscopy revealed the PEG tube insertion site at the anterior wall of the lower body, the ideal location site, where the risk of bleeding from the epiploic arteries is low. There were no signs of gastric trauma at or opposite the pexy and trocar sites. We believe that the bleeding was not due to poor visibility, but that it may have occurred regardless of type the PEG tube insertion performed. The 2 other adverse events in our study were asymptomatic pneumoperitoneum. Pneumoperitoneum has been reported in 5% to 50% of patients who underwent PEG.^\[[@R17]--[@R21]\]^ While surgical intervention should be considered in patients with clinical signs of intraabdominal adverse events,^\[[@R17],[@R22]\]^ asymptomatic "benign pneumoperitoneum" after PEG is relatively common and should not warrant further intervention.^\[[@R18]--[@R21]\]^ The current portable DUE system requires some improvements. First, as mentioned above, the scope is too flexible, resulting in difficulties in scope manipulation such as passing through areas of resistance or providing torsion and rotation. Second, visibility is limited by both low light power and the lack of a cleansing channel. Insertion of the DUE through the nasal cavity can result in mucus material covering the lens. We found that copious application of a clear jelly to the lens before insertion somewhat mitigated this problem. However, too much mucus resulted in scope withdrawal and reinsertion, which could lead to loss of gastric spatial orientation. Increasing the light power or changing the light source to an optic cable as well as adding a water channel for cleansing would help to increase visibility. Third, the lack of a suction channel caused problems in controlling the gastric distension. Moreover, bloating could develop if the procedure fails or takes too long to complete. Successful insertion of the PEG tube may result in fewer bloating symptoms because PEG tubes have been used for gastrointestinal decompression.^\[[@R23]\]^ Attempts to develop a portable endoscope for evaluation of the gastrointestinal tract have been partially successful.^\[[@R1],[@R2]\]^ Most such studies have involved the esophageal tract because it is easy to access and does not require high optical quality.^\[[@R1],[@R2],[@R12]\]^ However, the use of a portable endoscope with high visual quality and easy maneuverability has advantages: dedicated equipment with high cost and space requirements would not be needed, the endoscope would be available in cases of emergency, and examinations could be performed without moving patients. To date, studies beyond the esophagus have been limited to checking for gastrointestinal bleeding and the present study.^\[[@R13]\]^ A recent study carried out with the previous version of the DUE reported that high-quality images of the duodenum were obtainable.^\[[@R2]\]^ PEG was successfully carried out with the added improvements to this version, such as air insufflation and the increased bending angle. Further improvements including increased stiffness, a brighter light source, and the presence of water/suction channels are underway; these improvements would allow for the performance of true portable diagnostic endoscopy regardless of the surroundings. This would have the potential to save space and costs, limit the need to move patients, and reduce the risk of infection caused by repeated scope use. In conclusion, PEG via the introducer method under portable DUE guidance is feasible and relatively safe, opening up the possibility of the use of the portable DUE in facilities without endoscopy equipment. Further advancements should be made to the DUE in terms of the scope stiffness, light source, and water/suction channels to improve safety and success rates. We believe that this study opens up the possibility of portable DUE becoming a clinically useful tool to endoscopists. Acknowledgment ============== We would also like to thank nurses Ae Kyung Gu, Ho Jin Jang, and Won Jin Choi for their assistance with the PEG tube insertions. This research was supported by the Medical Device Development Center under the direction of the Korea Evaluation Institute of Industrial Technology (Grant number 10049769) and the program of Global Research and Development Center through the National Research Foundation of Korea, funded by the Ministry of Science, ICT and Future Planning (Grant number NRF-2011-00316441). Supplementary Material ====================== ###### Supplemental Digital Content Abbreviations: CUE = conventional ultrathin endoscope, DUE = disposable ultrathin endoscope, PEG = percutaneous endoscopic gastrostomy. Funding: This study was supported by the Medical Device Development Center under the direction of the Korea Evaluation Institute of Industrial Technology (grant number 10049769). The authors have no conflicts of interest to disclose. Supplemental Digital Content is available for this article.

Background ========== Gait analysis is usually used to directly aid patient treatment and to better understand locomotion \[[@B1]-[@B3]\]. Several methods of gait analysis are available, including visual analysis, which depends on the ability and perspicacity of the investigator, and specific analysis, which requires specialised equipment \[[@B2]-[@B4]\]. Force platforms and pressure distribution sensors, such as pressure-sensing walkways, may be used to measure forces, moments, and accelerations \[[@B2],[@B3]\]. Force platforms measure the orthogonal ground-reaction forces (vertical, mediolateral and craniocaudal) that result from limb positioning during locomotion, whereas pressure-sensing walkways measure vertical ground-reaction forces only \[[@B1]-[@B7]\]. Vertical forces have been commonly studied in animals due to their low variability and their magnitude relative to other orthogonal forces \[[@B1],[@B4],[@B6],[@B8],[@B9]\]. Craniocaudal and mediolateral forces may be affected by external factors and have smaller magnitudes \[[@B1],[@B9]\]. Different pressure-measuring devices have been used to perform gait analyses in animals \[[@B7],[@B10]-[@B14]\]. Some validation studies using dogs and horses have shown that certain kinetic measurements may differ when using a force platform rather than a pressure-sensing walkway \[[@B7],[@B10],[@B11]\], and highly accurate absolute force values may not be obtainable using a pressure plate \[[@B12]\]. However, several pressure-sensing walkway systems have been proven to be useful in clinical settings and for repeated gait assessments \[[@B11]-[@B13]\]. Additionally, some authors have suggested that experiments using pressure-sensing walkways are less time consuming, as walkways capture multiple sequential steps and as the contralateral limbs may be evaluated in the same trial \[[@B7],[@B10],[@B11],[@B13],[@B15]\]. Understanding normal gait requires understanding that normal patterns may vary by sex, age \[[@B2]\], and the species in question. Sheep have been used as an *in vivo* experimental model for orthopaedic research studies due to their size, availability and ease of handling, and they can sometimes provide an alternative to dogs \[[@B16]-[@B18]\]; some aspects of gait analysis of the sheep require more comprehension, however. For example, a previous study using a pressure-sensing walkway reported that female Suffolk-mix sheep (with body masses ranging from 69.3 to 103 kg) are not suitable for gait analysis assessments, largely due to their flight zone and flocking behaviour \[[@B19]\]. In another study, however, female Merino-mix sheep (mean body mass 63 kg, range 46--90 kg) were trained to walk over a pressure-sensitive platform. The platform was used for nine weekly evaluations of a healing tibial defect \[[@B20]\]. Therefore, the aim of this study was to use a pressure-sensing walkway to evaluate kinetic and temporospatial parameters in clinically healthy female sheep from three different age groups. The hypothesis was that the older sheep would have altered kinetic parameters due to aging. Methods ======= This study followed the guidelines for the care and use of laboratory animals and was approved by the Institutional Ethics Committee. Twenty-one clinically healthy intact female Santa Ines sheep were divided into three groups: G1 -- seven animals, aged from 8 to 12 months and weighing 19.5-33 kg (mean ± SD, 25.78 ± 4.75 kg); G2 - seven individuals, aged from 2 to 4 years and weighing 26.5-42 kg (mean ± SD, 31.52 ± 4.88 kg); and G3 -- seven animals, aged more than 5 years and weighing 37.3-45 kg (mean ± SD, 42.21 ± 3.11 kg). The sheep were judged to be healthy based on complete physical and orthopaedic examinations and had no history of lameness. Before the data collection, the sheep were trained in being led by a halter twice each day for a period of approximately three weeks. Subsequently, the sheep were trained to walk across the walkway twice each day for a period of one week. In addition, hoof trimming was performed one week before the exams. Food was used to motivate the sheep to walk in all of the training sessions. Each sheep was weighed on a single electronic scale immediately before the data collection. All of the sheep were handled by a single handler. Data collection --------------- The kinetic and temporospatial gait parameters were measured using a 1.951 mm x 447 mm pressure-sensitive walkway (Walkway High Resolution HRV4; Tekscan, South Boston, Massachusetts, USA) containing 33,408 pressure-sensing cells. The sensors of the pressure-sensing walkway were equilibrated and calibrated using a phantom according to the manufacturer's specifications. During each trial, the sheep were led in a straight line over the pressure-sensing walkway by a handler using a halter. The animals were handled from two different directions: first on the left side of the handler and then on the right side. Food was used to motivate the animals. Forty trials (20 in each direction) were captured for each animal. The data from the first five valid trials in each direction were collected for each sheep and analysed using the designated software (Walkway 7.0; Tekscan). A trial was considered valid if the sheep walked within the correct velocity (1.1-1.3 m/s) and acceleration (from −0.15 to 0.15 m/s^2^) ranges without head movement or pulling on the halter and if all four limbs had contact with the surface of the walkway during each walk cycle. The temporospatial gait cycle time (s), stance time (s) swing time (s) and stride length (m) parameters were determined for each limb. The stance time percentage was calculated as follows: (stance time/gait cycle time) x 100. The swing time percentage was calculated as (swing time/gait cycle time) x 100. The stride corresponded to the distance between two consecutive ground contacts of the same limb. The peak vertical force (PVF) and vertical impulse (VI) kinetic parameters were also determined. Both were normalised to the sheep's body weight and represented as percentages of body weight (%BW and %BW\*s). The percentage body weight distribution among the four limbs was calculated by (PVF of the limb/total PVF of the four limbs) x 100. Limb lengths and relative velocity ---------------------------------- The limb lengths of the sheep were measure using retroreflective spherical markers (1.8 cm in diameter) that were tagged using a 3-camera kinematic system (Vicon MX-3+; Vicon, Oxford Metrics Group, Oxford, UK). The markers were placed on the skin by a single investigator using quick-drying glue. The forelimb markers were placed on the point of the cranial angle of the scapula, acromium of the scapulohumeral joint, lateral epicondyle of the humerus, styloid ulnar process, and distal lateral aspect of the III and IV metacarpus. On the hind limbs, the markers were placed on the greater trochanter of the femur, femorotibial joint between the lateral epicondyle of the femur and the fibular head, lateral malleolus and distal lateral aspect of the III and IV metatarsus. The length of each limb was determined by the sum of the distances between each pair of markers on that limb. The relative velocities for the forelimbs and hind limbs were determined by the "Froude number", which was defined as v^2^/gh, where v is the velocity, g is the gravitational acceleration and h is the height \[[@B12],[@B37]\]. Statistical method ------------------ The data were analysed using a linear repeated measurements model. For the temporospatial parameters and kinetic data, the animal side and the direction on the pressure-sensing walkway were considered to be the intra-sheep factors, while the group served as the inter-sheep factor. For the limb lengths, the animal's side was considered to be the intra-sheep factor, and the group was the inter-sheep factor. The inter-sheep factor for body mass was the group. The sequential Bonferroni adjustment procedure was applied to our contrasts. The values were expressed as the mean ± standard deviation**,** and the coefficients of variation (CV) were calculated. A one-way ANOVA followed by Tukey\'s test was performed to compare the relative velocities of the forelimbs and hind limbs. Differences were considered significant at p \< 0.05. Pearson\'s correlation coefficients (r) were used to evaluate the linear relationships between the limb lengths and the PVF. The correlations were deemed significant at the 5% probability level. Results ======= After the training procedure, the sheep were able to walk appropriately on the pressure-sensing walkway. No significant differences were found between the kinetic and temporospatial parameters of the left and right forelimbs or the left and right hind limbs when the group and the direction were not considered. None of the kinetic or temporospatial parameters, in either the forelimbs and or the hind limbs, differed significantly between the left and right sides or between directions when the group and the animal side were not considered. The temporospatial values did not differ significantly among the groups in either the forelimbs or hind limbs (Tables [1](#T1){ref-type="table"} and [2](#T2){ref-type="table"}). The forelimbs had significant differences in the PVF (G1 \> G3) and VI (G1 \> G3) (Table [3](#T3){ref-type="table"}), while the PVF differed significantly (G1 \> G3) in the hind limbs (Table [4](#T4){ref-type="table"}). ###### Comparison of the temporospatial parameters of the forelimbs among three groups (G1, G2 and G3) of healthy sheep   **G1** **G2** **G3** **P value** ------------------- --------------- -------- --------------- ------------- --------------- ------ ------ --------- ------ Stance Time (sec) 0.409 ± 0.020 5.07 0.426 ± 0.038 8.91 0.444 ± 0.025 5.65 0.30 0.04\* 0.26 Swing Time (sec) 0.287 ± 0.008 3.02 0.301 ± 0.021 7.21 0.307 ± 0.015 5.16 0.12 0.03 0.51 Stride Time (sec) 0.701 ± 0.034 4.91 0.718 ± 0.053 7.49 0.751 ± 0.040 5.36 0.48 0.04\* 0.17 Stride Length (m) 0.805 ± 0.032 3.97 0.848 ± 0.052 6.15 0.872 ± 0.057 6.52 0.11 0.019\* 0.38 \% of Stance 58.40 ± 0.83 1.42 59.33 ± 1.77 3 59.09 ± 0.730 1.24 0.20 0.36 0.71 %of Swing 41.07 ± 1.18 5.22 42.04 ± 0.99 2.36 40.91 ± 0.418 1.02 0.06 0.76 0.03 \*Not statistically significant after alpha error correction. ###### Comparison of the temporospatial parameters of the hind limbs among three groups (G1, G2 and G3) of healthy sheep   **G1** **G2** **G3** **P value** ------------------- --------------- -------- --------------- ------------- --------------- ------ -------- -------- -------- Stance Time (sec) 0.430 ± 0.021 4.93 0.446 ± 0.040 9.14 0.466 ± 0.035 7.66 0.39 0.06 0.27 Swing Time (sec) 0.267 ± 0.008 3.04 0.285 ± 0.018 6.34 0.278 ± 0.015 5.45 0.03\* 0.15 0.39 Stride Time (sec) 0.689 ± 0.033 4.78 0.727 ± 0.056 7.78 0.742 ± 0.040 5.5 0.14 0.04\* 0.53 Stride Length (m) 0.810 ± 0.040 5.04 0.842 ± 0.051 6.79 0.857 ± 0.057 6.7 0.25 0.10 0.58 \% of Stance 62.20 ± 1.62 2.61 61.32 ± 1.72 2.82 62.77 ± 2.04 3.26 0.37 0.61 0.17 %of Swing 38.89 ± 1.14 2.94 39.31 ± 1.27 3.23 37.48 ± 2.00 5.35 0.62 0.08 0.03\* \*Not statistically significant after alpha error correction. ###### Comparison of the kinetic data of the forelimbs among three groups (G1, G2 and G3) of healthy Sheep   **G1** **G2** **G3** **P value** --------------------------- --------------------- -------- ---------------------- ------------- --------------------- ------- ------ ----------- ------ Peak Vertical Force (%BW) 54.67 ± 8.40^**a**^ 15.36 47.23 ± 4.06^**ab**^ 8.59 40.15 ± 3.83^**b**^ 9.53 0.09 **0.001** 0.12 Vertical Impulse (%BW\*s) 16.39 ± 2.53^**a**^ 15.43 14.53 ± 1.30^**ab**^ 8,94 12.46 ± 1.96^**b**^ 15.73 0.30 **0.005** 0.20 \% of Body distribution 31.38 ± 1.57 5.00 31.04 ± 1.20 3.86 31.60 ± 1.93 6.10 1.00 1.00 1.00 Values followed by different letters along each row are significantly different. ###### Comparison of the kinetic data of the hind limbs among three groups (G1, G2 and G3) of healthy Sheep   **G1** **G2** **G3** **P value** --------------------------- --------------------- -------- ---------------------- ------------- --------------------- ------- ------ ----------- ------ Peak Vertical Force (%BW) 33.14 ± 5.06^**a**^ 15.26 29.08 ± 2.63^**ab**^ 9.04 23.52 ± 3.35^**b**^ 14.24 0.25 **0.001** 0.06 Vertical Impulse (%BW\*s) 9.51 ± 1.96 20.6 8.92 ± 1.15 12.89 7.74 ± 1.89 24.41 1.00 0.22 0.67 \% of Body distribution 18.92 ± 1.12 5.91 19.05 ± 1.22 6.40 18.39 ± 1.81 9.84 1.00 1.00 1.00 Values followed by different letters along each row are significantly different. The inter-group lengths differed in both the forelimbs (G1 \< G3) and the hind limbs (G1 \< G2 and G3) (Table [5](#T5){ref-type="table"}). The velocity and the stride frequency did not differ significantly among the groups (Table [6](#T6){ref-type="table"}), which was in contrast to a significant difference in body mass (G1 \< G2 \< G3) (Table [7](#T7){ref-type="table"}). The relative velocity differed in both the forelimbs and hind limbs (G1 \> G3). The average Froude numbers were 0.27 (G1), 0.25 (G2) and 0.23 (G3) for the forelimbs and 0.25 (G1), 0.24 (G2), and 0.23 (G3) for the hind limbs. ###### Comparison of the forelimb and hind limb lengths (m) among three groups (G1, G2 and G3) of healthy sheep   **G1** **G2** **G3** **P value** ------------------ ---------------------- -------- ----------------------- ------------- ---------------------- ------ ----------- -------------- ------- Forelimb length 0.514 ± 0.022^**a**^ 4.31 0.586 ± 0.029^**ab**^ 5.03 0.607 ± 0.041^**b**^ 6.82 0.125 **0.008** 0.397 Hind limb length 0.545 ± 0.022^**a**^ 4.10 0.595 ± 0.019^**b**^ 1.46 0.628 ± 0.073^**b**^ 5.13 **0.001** **\< 0.001** 0.479 Values followed by different letters along each row are significantly different ###### Velocity (m/s) and Stride frequency (cycles/min) among three groups (G1, G2 and G3) of healthy sheep **Limb** **G1** **G2** **G3** **P value** ------------------------------- --------------- -------- --------------- ------------- --------------- ------ ------ ------ ------ Velocity (m/s) 1.175 ± 0.035 3.03 1.169 ± 0.034 2.91 1.185 ± 0.032 2.73 0.07 0.44 0.27 Stride frequency (cycles/min) 84.57 ± 5.47 6.48 83.98 ± 5.58 6.65 82.68 ± 5.80 7.02 0.58 0.06 0.16 ###### Values of body mass among three groups (G1, G2 and G3) of healthy sheep   **G1** **G2** **G3** **P value** ---------------- --------------------- ------------- --------------------- ------------- --------------------- ------------- ---------- -------------- -------------- Body mass (kg) 25.78 ± 4.75^**a**^ 19.5 - 33.0 31.52 ± 4.88^**b**^ 26.5 - 42.0 42.21 ± 3.11^**c**^ 37.3 - 45.0 **0.02** **\< 0.001** **\< 0.001** Values followed by different letters along each row are significantly different. The duty factors for the complete study population ranged from 0.55 to 0.62, with a mean of 0.58 (SD 0.01), for the forelimbs, and from 0.58 to 0.66, with a mean of 0.61 (SD 0.02), for the hind limbs. Discussion ========== In our study, characteristics that are considered unfavourable for performing gait analysis in sheep, such as flight zone and flocking behaviour \[[@B19]\], were reduced by training, as has been previously reported \[[@B20]\], and with the same halter when inducing the animals to walk on a pressure-sensing walkway. Another important strategy was using food to motivate the sheep to traverse the pressure-sensing walkway in a straight line. Moreover, the body mass (mean 33.10 kg, range 19.5-45.0 kg, comparable to that of a large-sized dog) of the sheep used in our study facilitated handling. Although sheep reach sexual maturity at approximately 7--12 months of age, depending on the breed, the closure of the physeal plates of the long bones may occur as late as 36 months \[[@B16]\]. Thus, it is important to evaluate sheep of different ages, as in the present study, when evaluating skeletal growth. Velocity measurements obtained by photoelectric cells and measurements from pressure-sensing walkway have been shown to be similar \[[@B7]\]. For this reason, only the latter measurement method was used. The designated software calculated the velocity by dividing the distance between consecutive foot strikes by the time between them \[[@B7],[@B10],[@B21]\]. The velocity used, 1.1-1.3 m/s, was considered comfortable for the sheep. The mean velocity used in a study assessing healthy sheep walking on a pressure-sensing walkway was 1.06 m/s \[[@B19]\]. In healthy dogs, the walking velocity on the same or similar type of pressure-sensing walkway has been reported to range from 0.5 to 1.14 m/s, depending on body sizes \[[@B10],[@B21]-[@B24]\]. For healthy cats, the mean velocity has been reported to range from 0.6 to 0.81 m/s \[[@B25]-[@B27]\]. The duty factor in this study was \>0.5, with a mean of 0.58 for the forelimbs and a mean of 0.61 for the hind limbs. In a study of healthy Suffolk-mix sheep, the duty factors were 0.66 and 0.69 for the forelimbs and hind limbs, respectively \[[@B19]\]. As has been reported previously \[[@B19]\], these values suggest a walking speed, as a duty factor \>0.5 in the hind limbs is indicative of walking and a factor \<0.5 is indicative of trotting or running \[[@B28]\]. Velocity and acceleration must be controlled because of their influence over the stance time, which is associated with the VI \[[@B2],[@B10],[@B29]\]. Studies of dogs have reported that velocity variations of 0.3 m/s may modify the ground reaction forces; as the velocity increased, the PVF increased and the VI decreased \[[@B4],[@B30]\]. In the present study, the velocity was similar among the groups and had a mean value of 1.17 m/s (SD 0.02). In another sheep study, the mean velocity was 1.06 m/s, but the velocity varied from 0.57 to 1.49 m/s in the forelimbs and from 0.57 to 1.76 m/s in the hind limbs \[[@B19]\]. The two directions in which the sheep walked on the pressure-sensing walkway did not show significant differences, suggesting that the side of the handler did not interfere with the values. The direction of locomotion relative to the camera did not influence any of the temporospatial or kinetic parameters evaluated in a prior sheep study \[[@B19]\]. In humans, the gait of children is different from that of adults. Characteristics such as the cycle time, stride length and velocity change with growth \[[@B2]\]. In this study, no differences were detected in the temporospatial parameters among the groups, but the sheep moved at the same velocity. Stride length is diminished in healthy elderly humans \[[@B2],[@B31]\]. This finding was not observed in Group 3, probably because the animals were younger adults. Sheep life expectancy has been reported to be approximately 10 to 12 years \[[@B32]\]. In addition, a relationship between stride length and height has been found in humans \[[@B2]\]. No stride length differences were found among the groups in our study, although the G1 animals had shorter limbs than those in other groups. The 8.2 cm and 6.7 cm disparities in the forelimbs and hind limbs, respectively, may not have been sufficient to cause detectable temporospatial differences. Additionally, a kinematic study of horses found that the duration of stance and stride in foals can be normalised using linear or dynamic scaling by the height at the withers and used to predict the values observed in adulthood. However, the height at the withers of the foals had increased by 30 cm when they reached adulthood \[[@B33]\]. In general, the stance and swing phases account for approximately 60% and 40%, respectively, of the gait cycle in healthy humans \[[@B2]\]. In the present study, the distribution was 58.86% stance phase and 41.49% swing phase for the forelimbs versus 61.88% stance phase and 38.85% swing phase for the hind limbs. Another study of sheep reported a mean of 66.31% for the forelimbs and 68.89% for the hind limbs in stance phase and 33.69% for the forelimbs and 31.11% for the hind limbs in swing phase \[[@B19]\]. The walking PVI (%BW) is generally higher for the forelimbs than the hind limbs when measured using a pressure-sensitive walkway \[[@B10],[@B19],[@B22],[@B26],[@B34]\]. Although this finding is consistent with the results from all of the groups in our study (mean values of 47.35% for the forelimbs and 28.58% for hind limbs), the values showed some variations compared to previous studies of healthy sheep (means of 52.52% and 38.52% for the forelimbs and hind limbs, respectively) \[[@B19]\], healthy dogs (means of 58.11% and 42.30% for the forelimbs and hind limbs, respectively) \[[@B10]\], and healthy cats (means of 48.2% and 38.3% for the forelimbs and hind limbs, respectively) \[[@B26]\]. The differences in velocity and calibration \[[@B26]\] and the animal species in question may have contributed to these differences in reported values. Differences in the calibration protocol may influence the number of activated sensors and result in differing readings \[[@B14],[@B26],[@B35]\]. In the present study, the calibration was performed according to the manufacturer's guidelines using a phantom (a short three-legged stool). Weight was added to the stool to match the weight of each animal, and hard rubber was attached to the bottom of the feet of the stool to mimic a sheep's hoof. Few studies have specified which methods were used to calibrate their pressure-sensitive walkways \[[@B7],[@B11],[@B14],[@B26]\], which may explain the different measurements. In a study that evaluated vertical limb forces in dogs at a trotting velocity, for example, a 65 kg human subject placed one foot on the mat for calibration \[[@B7]\]. In a kinetic evaluation of cats, the mat was calibrated by having a 50 kg human subject stand on it with both bare feet \[[@B26]\]. The body weight distribution among the limbs of healthy sheep and dogs during walking has been described as approximately 30% on each forelimb and 20% on each hind limb \[[@B19],[@B36]\]. Similar values were observed in our study, with 31.34% of the body weight placed on each forelimb and 18.79% placed on each hind limb. In a study of healthy dogs walking on a force plate, the peak vertical forces were inversely correlated with physical size \[[@B36]\]. A similar correlation was observed in the present study for the forelimbs (R = −0.57) and hind limbs (R = −0.44) Thus, the PFV values for the forelimbs and hind limbs were greater in Group 1 than in Group 3, with Group 3 being composed of larger sheep. Although body size of the Group 2 was similar to Group 3, no significant PFV differences were observed. Furthermore, the relative velocity was greater in Group 1 than in Group 3, which may have contributed to the differences in the PVF values. However, the VI of the forelimbs was higher in the G1 animals than in the G2 animals. The velocity may influence the stance time and consequently the VI \[[@B29]\], as the impulse reflects the association between force and the time that the foot is on the ground \[[@B4]\]. As a constant velocity was maintained in our study and as the stance time was statistically indistinguishable among the groups, the differences were associated with the different forces**.** These results are in contrast with those from another canine study in which the VI increased as the size of the dog increased \[[@B36]\], probably due to differences in velocity and individual dog size. Conclusions =========== The vertical forces created by young healthy sheep walking on a pressure-sensing walkway at the same velocity differ from those of older sheep under similar circumstances. This finding may be a source of variation and should be controlled in locomotion research studies. Authors' contributions ====================== FSA and SCR conceived and designed the study; FAPA and RTC helped collected the data; CAH and AOE assisted in data collection methods and FOBM helped draft the manuscript; all authors read, contributed to and approved the final manuscript. Acknowledgements ================ The authors would like to thank the FAPESP (The State of São Paulo Research Foundation) and CNPq (National Council for Scientific and Technological Development) for financial support, the Coordination for the Improvement of Higher Level Personnel (CAPES) - PROCAD-NF No. 21/2009, and Dr. Guy Beauchamp of the Faculty of Veterinary Medicine, University of Montreal -- Canada, for the statistical analysis. We would also like to thank the Universidade Estadual Paulista (UNESP) and the Universidade Federal Rural da Amazônia.

{#sp1 .429}

It can be confidently stated that global population aging is both the greatest success of global public health efforts of the past, as well as the greatest challenge for the further global public health efforts of the future. Over the past decades, life expectancy at birth has increased globally, reaching a world-wide average of about 71.5 in 2017 (7 years longer than in 1990) and around 80 in the developed countries (compared to about 50 years in the developed countries in the early 20th century). This achievement occurred primarily due to advances in basic public health such as sanitation, clean water, waste disposal, temperature controlled living and working environments, and advances in medical technology \-- all of which yielded large and rapid declines in communicable diseases. Beginning with the last quarter of the 20th century, progress made against chronic fatal diseases (especially heart disease) from new diagnostic tools, treatments, and improved behavioral risk factors (such as reductions in smoking), generated further increases in life expectancy, although at a slower pace. Currently, while the highest life expectancies are still found in the "developed" countries, the opportunity for rising longevity remains likely for "developing" world in the coming decades. Considering that both rising longevity and population aging are likely demographic events in the coming decades, by 2050 the proportion of people over 60 years is expected to double from about 13% to nearly 25%, which, in absolute terms, means an increase from 962 million to 2.1 billion people \[[@b1-ad-9-2-331]\]. Rising longevity during the last 150 years is a testament to human ingenuity, and there is reason to believe further advances are possible. Yet global population aging, and rising longevity are also accompanied by challenges and opportunities across all nations. The number of patients suffering from chronic non-communicable age-related diseases are estimated at tens of millions, and are expected to strongly increase worldwide due to the rapid population aging. Thus, 36 million people worldwide are living with old-age dementia, and their numbers are expected to double every 20 years, reaching 66 million by 2030, and 115 million by 2050 \[[@b2-ad-9-2-331]\]. According to World Health Organization's data, "Noncommunicable diseases (NCDs) kill 40 million people each year, equivalent to 70% of all deaths globally. Cardiovascular diseases account for most NCD deaths, or 17.7 million people annually, followed by cancers (8.8 million), respiratory diseases (3.9 million), and diabetes (1.6 million)." Also, according to WHO data, "Each year, 15 million people die from a NCD between the ages of 30 and 69 years; over 80% of these \"premature\" deaths occur in low- and middle-income countries" \[[@b3-ad-9-2-331]\]. In other words, of the 57 million deaths in the world each year, nearly 50% occur due to chronic non-communicable diseases in the world's oldest population (70+), and over 60% in the older population (60+), making the health of older persons the worst and most urgent global health problem. In view of the urgency of the problem, it seems highly surprising that in the forthcoming draft 13^th^ General Programme of Work of the World Health Organization for 2019-2023 - the issue of aging and aging-related ill health is excluded completely! Beside a cursory mention of the word "aging," this work program does not contain any specific objectives, deliverables and actions to improve the health of the aged \[[@b4-ad-9-2-331]\]. This means that, through 2023, according to this document, the World Health Organization is not obliged to provide any services to care for the health of older persons or to improve their health, not to mention conduct any research and development to create new therapies and technologies for improving the health of the aged. The issues of aged health are not in the WHO work program! This, in essence, appears to be a manifestation of "ageism" in health care and health research! The very cutoff at 70 years of age, discriminating "premature" from "mature" mortality, is questionable both scientifically and ethically. Does it imply that "mature" deaths should not be combated or reduced? Still, the complete exclusion of any actions for the improvement of aged health and reduction of aged mortality is not just unjustifiable, but inexplicable. It should be obvious, not just to any health professional but to any lay person, that the global population aging poses grave and urgent challenges for global health, in the "developed" as well as the "developing" world. In the developing world, the faster population aging also means a faster increase in the incidence of chronic age-related NCD's, while the geriatric and old-age NCD care and research capabilities in these countries may be under par and unprepared. Hence, in the "developing world" in particular, not addressing the problems of aging, not improving national geriatric care and research capabilities, will condemn millions of the world's poorest and most disadvantaged older people to misery that could be avoided. This is one of the reasons the explicit exclusion of the health needs of the global aging population in the draft work program of the WHO appears to be inadmissible, even incredible. Such an exclusion can have dire consequences for global health. First and foremost, it will signal to governments, in particular in developing countries, that aging health is not an important priority to be addressed with urgency. Practically, this attitude could result in many existing and future health care and health research programs on aging being eliminated. Even now, the issues related to aging health have not been strongly prioritized by WHO. Indicatively, as of 2016, the entire budget for the World Health Organization's "Ageing and Health" program was \$13.5M, out of about \$4.4 billion total WHO budget (0.3%). No budget portion was specified for anything indicative of "ageing research" \[[@b5-ad-9-2-331]\]. Now, with the explicit exclusion of aging health from the WHO work plan, will the "Ageing and Health" programs, and other ageing-related programs (both care and R&D) receive no funding and support whatsoever? How can this exclusion coexist with the mission of WHO's division on "Ageing and Life Course"? How can it coexist with the recently adopted WHO's Global Strategy and Action Plan on Ageing and Health (GSAP) for 2016-2020, endorsed by all the WHO member states? According to its goal statement, the GSAP must prepare for the "Decade of Healthy Ageing from 2020 to 2030" which was also announced by WHO \[[@b6-ad-9-2-331]\]. The coordination and consultation between various arms and branches of the WHO must improve. The developers of the WHO Work Program must avail of the world expertise on ageing health, within the WHO and externally, to develop an effective, strategically-minded and inclusive global health program. Most importantly, we urge the WHO to include and emphasize specific tasks and goals to maintain and improve the health of the global aging population in its work program. Among others, these tasks and goals must include enhancing scientific research and technological development aimed to provide new effective and safe therapies to meet the pressing health needs of the global aging population \[[@b7-ad-9-2-331]\]. We also urge the readers to make your voice heard, advocate and increase publicity about the need to include and implement concrete measures to improve aging health, including R&D for healthy longevity, as a priority in the WHO work program \[[@b8-ad-9-2-331]\].

1. Introduction {#sec1-ijerph-17-02822} =============== The enrichment of potentially harmful elements (PHEs) in soil has been widely concerned as a significant ecological and environmental issue \[[@B1-ijerph-17-02822]\]. The bedrock geochemistry and human activities are the main sources of PHEs in the environment \[[@B2-ijerph-17-02822],[@B3-ijerph-17-02822]\]. Lee et al. \[[@B4-ijerph-17-02822]\] reported that natural soils and rocks contain a high content of PHEs, such as Cd and Mo. Usually, the enrichment of soil PHEs are mainly caused by anthropogenic factors (industrialization, urbanization, agriculture, and transportation), and the enrichment of PHEs tend to aggravate with the increase of economic development and human activities \[[@B3-ijerph-17-02822],[@B5-ijerph-17-02822]\]. The accumulation of PHEs appeared in a typical rapid urbanization city of China, specifically, the concentration of PHEs gradually decreased with the increase of distance from the constructed land, which indicated the significant effect of the urbanization process \[[@B6-ijerph-17-02822]\]. The concentration of soil PHEs is also affected by land-use and rises along with a corresponding increase in land-use intensity \[[@B3-ijerph-17-02822],[@B7-ijerph-17-02822]\]. Li et al. \[[@B8-ijerph-17-02822]\] assessed the human health risk caused by soil PHEs pollution from mines in China and indicated that Cd, Pb, Cu, Zn, Hg, As, and Ni as the priority control elements. The high PHEs concentration in the industrial zone is mainly derived from the discharge of poorly treated "three wastes" (waste gas, water, and residues), which enters the soil through leaching, settlement, and run-off during rains \[[@B9-ijerph-17-02822]\]. It is reported that steel smelting is an important source to elevate Cd, Pb, and Zn \[[@B10-ijerph-17-02822]\]. The fossil fuel combustion from a coal-fired plant will cause the enrichment of Pb \[[@B10-ijerph-17-02822]\], while the concentration of Cd in the soil around the ferroalloy plant is significantly higher \[[@B11-ijerph-17-02822]\]. Zhang et al. \[[@B12-ijerph-17-02822]\] showed that the total pollution rate of farmland soil in China was 10.18%, mainly from Cd, Hg, Cu, and Ni. The external sources of farmland were agricultural activities (e.g., animal manure, inorganic fertilizers, and pesticides) in remote areas and industrial sewage irrigation in the suburbs \[[@B12-ijerph-17-02822]\]. Besides, Zn, Cd, Pb, Cr, Ni, and Cu are relatively high in the infrastructure area of the transportation (e.g., highway, railway) and its surrounding areas, which may originate from fuel combustion in vehicles and abrasion of tires and roads/tracks \[[@B8-ijerph-17-02822],[@B13-ijerph-17-02822]\]. In addition, land-use alters the soil texture, chemical composition, and microbial activity, thus leading to corresponding changes in the uptake and transfer capacity of soil PHEs \[[@B7-ijerph-17-02822],[@B14-ijerph-17-02822],[@B15-ijerph-17-02822]\]. The material flows and energy cycles between ecosystems have infinitely amplified the threat of soil PHEs to the survival and development of animals and humans \[[@B5-ijerph-17-02822]\]. Due to its non-biodegradability and persistence, PHEs damage the physicochemical properties of soil after long-term accumulation, leading to nutrient loss and degradation \[[@B16-ijerph-17-02822]\]. Meanwhile, soil PHEs will further threaten human health through the food chain with the pathways of ingestion, inhalation and dermal contact \[[@B3-ijerph-17-02822]\]. For example, Cd, Pb, and As are proven to be closely related to cancer, kidney dysfunction, nervous system dysfunction, and skin damage \[[@B1-ijerph-17-02822],[@B3-ijerph-17-02822],[@B16-ijerph-17-02822]\]. The coastal zone is an intersection between land and ocean where 25% of global primary productivity occurs, and it supplies nearly 70% of global fish catch \[[@B17-ijerph-17-02822],[@B18-ijerph-17-02822]\]. The coastal zone only occupies 12% of the Earth's surface, while supports about 50% of the world's population \[[@B18-ijerph-17-02822]\]. In many coastal countries and regions, sea reclamation (e.g., Netherlands, South Korea, and Japan), and tidal land reclamation (e.g., China, Vietnam) are the dominant ways to accommodate the increasing demand for space for living and development \[[@B19-ijerph-17-02822],[@B20-ijerph-17-02822]\]. Anthropogenic activities such as marine industry, tourism, fisheries, and shipping have extremely promoted the economic development of the coastal zone, however, they have also brought a series of ecological and environmental crises, including the decline of ecosystem services, destruction of biodiversity, metal and nutrient pollution \[[@B17-ijerph-17-02822],[@B18-ijerph-17-02822],[@B21-ijerph-17-02822]\]. It is widely reported that coastal fishery farming, agricultural cultivation, energy (wind, solar, etc.) base construction, industrial construction, and port construction lead to changes in the soil's PHEs concentration and the potential ecological threats \[[@B21-ijerph-17-02822],[@B22-ijerph-17-02822],[@B23-ijerph-17-02822]\]. Currently, a series of studies have been carried out on PHEs in the coastal zone, including the spatial pattern and source analysis \[[@B1-ijerph-17-02822]\], pollution assessment \[[@B24-ijerph-17-02822]\], ecological risk assessment \[[@B25-ijerph-17-02822]\], the influences of land use on distribution and morphology of PHEs \[[@B14-ijerph-17-02822],[@B26-ijerph-17-02822]\]. However, few studies have focused on the health risks of PHEs associated with different land use in the coastal tidelands reclamation area \[[@B22-ijerph-17-02822]\]. The coastal zone of Jiangsu Province is one of the most economically developed areas of China where the reclaimed tidal flats were up to 3350 km^2^ from 1950 to 2015, which provided important reserve land resources for regional urbanization and industrial development. In 2009, the State Council of China implemented a national strategy "*development planning in Jiangsu coastal areas*", which plans to reclaim 1805 km^2^ coastal tidal flats in Jiangsu from 2009 to 2020. The high-intensity, rapid reclamation of coastal zones within a short time period has affected the physicochemical properties of soil in tidelands and even accelerated the rapid enrichment of pollutants \[[@B14-ijerph-17-02822],[@B27-ijerph-17-02822]\]. There are over ten ports along the coast of Jiangsu Province, which undertake the transportation of numerous import and export commodities and energy (coal, petroleum and metal ore, etc.). Furthermore, port-centered industries such as metallurgy and petrochemicals, electricity, cement, and papermaking have also developed rapidly, which have posed huge threats to the soil ecosystem. Hence, the investigation of characteristics and trends of PHEs enrichment in the soil after coastal reclamation is important to clarify the influence of anthropogenic activities on tidelands and to understand and assess the ecological and environmental effects of coastal reclamation. In this study, multivariate statistical methods and a US EPA soil health risk assessment model were applied to explore the enrichment, source, and health risk of PHEs in a long-term reclaimed coastal area. The objectives of this study are (1) to explore the variation of soil PHEs under different land use; (2) to identify the source of soil PHEs in a coastal reclamation area; and (3) to assess the level of health risk status after long-term reclamation. Our study is designated to provide a scientific basis for the monitoring and pollution control of the tideland. 2. Materials and Methods {#sec2-ijerph-17-02822} ======================== 2.1. Study Area {#sec2dot1-ijerph-17-02822} --------------- The study area is located in the coast of the Dongtai City (120°78′86.7″--120°90′29.2″ E; 32°95′22.3″--33°00′38.9″ N) with an area of 200 km^2^, which closely adjoins Yancheng National Nature Reserve, Jiangsu Province ([Figure 1](#ijerph-17-02822-f001){ref-type="fig"}). The regional climate is characterized as a northern humid subtropical monsoon climate with four distinct seasons, and the annual average rainfall and temperature are 1035 mm and 15.0 °C, respectively. The soils in the study area are mainly derived from the modern marine and fluvial sediments, which belong to sandy loam and are classified as coastal saline-sodic soils. Actually, the study area is considered one of the fastest silting regions in eastern China, and the tidal land is currently expanding seaward at a rate of 200 m per year \[[@B20-ijerph-17-02822]\]. The reclamation of the study area mainly depended on different land uses to carry out the process of desalination after the construction of seawalls. The land-use/cover types varied with increasing reclamation years and occurred in three stages \[[@B28-ijerph-17-02822]\]. In the early period of reclamation, newly reclaimed tidal land changed to aquaculture ponds and salt-tolerant crops, such as *Sesbania cannabina*. Over 10--30 years, the land-use pattern gradually turned into cropland with an increase of soil fertility and decrease of soil salinity, and some crops were planted in this period, such as cotton, wheat, and corn. After 30 years of reclamation, soil quality basically met the cultivation for most crops with consistent cultivation and fertilization, such as paddy, wheat, corn, rapeseed, and broad bean. In addition, driven by the demands of economic development, some reclaimed land has been converted into industrial land to support industries such as power generation, textiles, chemicals, and automobile infrastructure. 2.2. Sampling Design and Chemical Analysis {#sec2dot2-ijerph-17-02822} ------------------------------------------ Soil sites were chosen in mid-August 2014 by using a uniform grid method in the study area. In total, 54 soil sampling sites were collected, including 5 sites respectively in tideland and halophyte land in coastal wetlands, and 44 sites (industrial land (10), forestland (6), residential land (7), aquaculture land (6), cropland (8), and vegetable land (7)) in the areas reclaimed for 30 years. Both surface (0--20 cm) and subsurface (20--40 cm) samples were collected from each sampling site and recorded using GPS ([Figure 2](#ijerph-17-02822-f002){ref-type="fig"}). Each soil sample consisted of 4--8 subsamples collected in a surrounding 10-m area. After air-drying at room temperature, the soil samples were grounded and sieved through a 2 mm nylon sieve to remove animal and plant residues, as well as other impurities, and then sieved through a 0.149 mm nylon sieve as described by Xie et al. \[[@B29-ijerph-17-02822]\]. From the sieved soil, 10--15 g preliminarily treated soil samples were used to form soil tablets. Eight PHEs (As, Co, Cr, Cu, Mn, Ni, Pb, and Zn) were analyzed according to X-ray fluorescence spectrometry methods, as described by Xia et al. \[[@B30-ijerph-17-02822]\]. 2.3. Descriptive and Multivariate Statistical Analysis {#sec2dot3-ijerph-17-02822} ------------------------------------------------------ Descriptive analysis (range, mean, standard deviation, coefficient of variation) was applied to understand the distribution and variation of PHEs in the study area. The differences of PHEs among different land-use types were compared using the analysis of variance (ANOVA) procedure and the mean comparisons were performed using the Fisher's least significant difference (LSD) test with a probability defined at 0.05 \[[@B24-ijerph-17-02822]\]. Pearson's correlation analysis and cluster analysis (CA) are effective to identify the source of PHEs in soil \[[@B15-ijerph-17-02822]\]. Pearson's correlation analysis could determine whether the sources of PHEs are the same through the correlation between PHEs, and CA could provide variable grouping results, which corroborates their correctness with correlation analysis results. Descriptive and multivariate statistical analysis was conducted in SPSS 22.0 for Windows (SPSS München, Germany). The enrichment factor (EF) is applied to determine the degree of PHEs enrichment and to differentiate between elements originating from human activities and those from natural provenance by using Equation (1) \[[@B1-ijerph-17-02822],[@B31-ijerph-17-02822]\]. $$EF = {(\frac{M_{x}}{N_{ref}})_{sample}}{\times /{(\frac{M_{x}}{N_{ref}})_{background}}}$$ where Mx and *Nref* are the concentration of PHEs and reference elements in soil, respectively. In this study, Fe was adopted as a reference because it is one of the abundant elements of soil \[[@B31-ijerph-17-02822]\]. The geochemical baseline of Fe (4.5 μg/g) in Jiangsu Province was chosen as reference background values \[[@B32-ijerph-17-02822]\]. Generally, EF is applied to differentiate whether the element originates from anthropogenic activities. The element is completely derived from natural processes under EF \< 1, while it might be from man-made sources under EF \> 1 and the proportion of human sources increased with the increasing of EF \[[@B33-ijerph-17-02822]\]. The EF is also applied to estimate polluted degree, which can be divided into smaller enrichment (EF \< 2), moderate enrichment (2 ≤ EF \< 5), significant enrichment (5 ≤ EF \< 20), highly enrichment (20 ≤ EF \< 40) and extremely high enrichment (EF ≥ 40) \[[@B31-ijerph-17-02822],[@B33-ijerph-17-02822]\]. 2.4. Health Risk Assessment {#sec2dot4-ijerph-17-02822} --------------------------- The long-term accumulation of PHEs in soil may cause a threat to the health of animals and human beings, and increased risk of cancer \[[@B31-ijerph-17-02822],[@B34-ijerph-17-02822]\]. The methodology used in this study to assess the exposure risks of PHEs to human health in the coastal reclaimed area based on risk assessment guidance for superfund (RAGS) health risk assessment as developed by the United States Environmental Protection Agency (US EPA) \[[@B35-ijerph-17-02822]\]. Direct exposure of an individual to PHEs in the soil through three main pathways: ingestion, inhalation, and dermal contact \[[@B36-ijerph-17-02822],[@B37-ijerph-17-02822]\]. The average daily dose through three pathways can be respectively estimated by the following Equation \[[@B8-ijerph-17-02822]\]. $$ADD_{ing} = C_{soil} \times \frac{IngR \times CF \times EF \times ED}{BW \times AT_{nc/ca}}$$ $$ADD_{inh} = C_{soil} \times \frac{InhR \times EF \times ED}{PEF \times BW \times AT_{nc/ca}}$$ $$ADD_{derm} = C_{soil} \times \frac{SA \times CF \times SL \times ABS \times EF \times ED}{BW \times AT_{nc/ca}}$$ Where *ADD~ing~*, *ADD~inh,~* and *ADD~derm~* are the average daily dose from ingestion, inhalation, and dermal contact, respectively (mg·kg^−1^·d^−1^). *C~soil~* is the concentration of PHEs in soil (mg·kg^−1^·d^−1^). Related definition and parameters used in the average daily dose of soil metals can be found in [Table 1](#ijerph-17-02822-t001){ref-type="table"}. **Non-carcinogenic risk assessment:** Hazard quotient (HQ) is calculated by the ratio of average daily dose to the corresponding reference dose, which is used to assess the potential non-carcinogenic risk of PHEs under certain exposure pathways. The Hazard Index (HI) is the sum of the HQ and represents the overall potential non-carcinogenic effects posed by all exposure pathways. The HQ of a single element is determined by Equation (5) \[[@B35-ijerph-17-02822]\]. $$HI = {\sum_{i = 1}^{n}{\sum_{j = 1}^{m}{HQ_{ij} =}}}{\sum_{i = 1}^{n}{\sum_{j = 1}^{m}\frac{ADD_{ij}}{RfD_{ij}}}}$$ $$CR = {\sum_{i = 1}^{n}{\sum_{j = 1}^{m}{CR_{ij} =}}}{\sum_{i = 1}^{n}{\sum_{j = 1}^{m}{SF_{ij}}}} \times ADD_{ij}$$ where *n* =8, *i* = As, Co, Cr, Cu, Mn, Ni, Pb, and Zn, *m* =3, *j* = ingestion, inhalation and dermal contact. *RfD~ij~* is the reference dose of *i* element in *j* exposure pathway (mg·kg^−1^·d^−1^). **Carcinogenic risk assessment:** Carcinogenic risk (CR) is the probability of cancer, which can be calculated by the product of daily average exposure multiplied by the corresponding slope factor. The CR of a single element is determined by Equation (6) \[[@B35-ijerph-17-02822]\]. *SF~ij~* is the slope factor *i* metal in *j* exposure pathway (mg·kg^−1^·d^−1^). The values of *RfD~ij~* and *SF~ij~* are shown in [Table 2](#ijerph-17-02822-t002){ref-type="table"} \[[@B41-ijerph-17-02822],[@B42-ijerph-17-02822],[@B43-ijerph-17-02822]\]. If *HQ* or *HI* \< 1, the non-carcinogenic risk did not exist to the exposed individual. However, if *HQ* or *HI* ≥ 1, the non-carcinogenic effect may occur with a probability which tends to increase with the increase of *HQ* or *HI* \[[@B8-ijerph-17-02822]\]. If *CR* \< 10^−6^, it will neglect the carcinogenic risk to human health. 10^−6^ \< *CR* \< 10^−4^ is considered an acceptable range and *CR* \> 10^−4^ is viewed as unacceptable \[[@B44-ijerph-17-02822]\]. 3. Results {#sec3-ijerph-17-02822} ========== 3.1. Descriptive Analysis {#sec3dot1-ijerph-17-02822} ------------------------- The average concentrations of soil PHEs (0--40 cm) showed that As, Co, Cr, Cu, Mn, Ni, Pb, and Zn were 3.31, 8.00, 52.50, 8.10, 458.76, 21.47, 15.90, and 45.45 mg·kg^−1^, respectively ([Table 3](#ijerph-17-02822-t003){ref-type="table"}). Compared with the Chinese Soil Environmental Quality Standard (*GB 15618-1995*), the average concentrations of PHEs were all within the range. In general, the concentrations of PHEs in the tidal flat reclamation area were relatively low, which is similar to the conclusions of other researches in Jiangsu coastal areas, and might be related to the local geochemical background \[[@B45-ijerph-17-02822],[@B46-ijerph-17-02822]\]. Lv et al. \[[@B47-ijerph-17-02822]\] demonstrated that the background concentrations of Cr, Cu, Ni, Pb, and Zn in marine sediments were significantly lower than that in river alluvial deposits, lagoonal facie sediments, and delta sediments. However, the highest concentrations of Pb and Cr exceeded the limit values of the first and second-grade environmental quality standard for soils, respectively. This showed that partial land use had the opportunity to cause soil pollution, which might pose a threat to the regional natural background or agricultural production. Specifically, the average values of eight elements (As, Co, Cr, Cu, Mn, Ni, Pb, and Zn) did not exceed regional soil background values and the maximum values were close to or exceeded that value. The variation coefficient identified that As has a high coefficient of variation (60.06%), which suggests that the distribution of As is extremely non-uniform. The distribution of Cr, Cu, and Pb have been relatively moderate and non-uniform, owing to their coefficients of variation ranging from 20--30%. Although the average values of PHEs were lower than the risk control standard and regional soil background values, the human health risk of PHEs needs to be of concern due to the strong variation of PHEs (As, Cr, Cu, and Pb) and the difference between the exposure parameters and chronic reference dose of human health risk. 3.2. PHEs Concentrations under Different Land Use {#sec3dot2-ijerph-17-02822} ------------------------------------------------- As shown in [Figure 3](#ijerph-17-02822-f003){ref-type="fig"}, the average concentrations of As, Cu, Mn, Ni and Zn in surface and subsurface soil were observed to be significantly different from different land-use types (*p* \< 0.05), which indicates that PHEs are greatly affected by the land use. However, soil depth had less effect on the concentrations of determined elements. The concentrations of Co and Cr did not change significantly under different land-use types, whatsoever, in surface or subsurface soil. However, the average concentration of Pb changed significantly in subsurface soil with the highest concentration observed in aquaculture land, and the concentration of As varied significantly under different land-use types with the highest concentration observed in industrial land. The average concentration of Cu in industrial land and aquaculture land was obviously higher than that of tideland and halophyte land, and the differences of other land-use types were not significant. Besides, no significant differences of Mn were observed in surface soil under different land-use patterns, whereas in subsurface soil, it was significantly higher in aquaculture land than that of other land-use types. Compared with the industrial land and aquaculture land, the concentrations of Ni and Zn in tideland, halophyte land, forestland, and vegetable land were relatively lower and changed mildly. The marine sediment sources and the degree of disturbance of human activities revealed that the lower concentration of PHEs was distributed in uncultivated soil (Tideland and Halophyte land) \[[@B49-ijerph-17-02822],[@B50-ijerph-17-02822]\]. As, Cu, Pb, and Zn of industrial land, residential land, aquaculture land, cropland, and vegetable land were relatively higher than that of forestland, which might be caused by the absorption of trees and the re-entry of toxic elements from industrial activities and agricultural practices \[[@B14-ijerph-17-02822]\]. 3.3. Source Analysis of PHEs in Soil {#sec3dot3-ijerph-17-02822} ------------------------------------ As presented in [Figure 4](#ijerph-17-02822-f004){ref-type="fig"}, the *EF* of measured elements were in the following order: Ni \> Mn \> Zn \> Cr \> Co \> Pb \> Cu \> As. The average *EF* of measured elements ranged from 1.0 to 1.5, except for Cu and As, which indicated that Co, Cr, Mn, Ni, Pb, and Zn were at minimal enrichment level. However, the highest *EF* of Cr, Mn, Ni, Pb, and Zn were above 1.5 and that of Cr was above 2.0, which was close to the medium enrichment level. This might be affected by anthropogenic activities (e.g., land use). As shown in [Figure 5](#ijerph-17-02822-f005){ref-type="fig"}, the halophyte land, tideland, and forestland belong to the low *EF* value group whereas the cropland and aquaculture land belong to the high EF value group. It was worth noting that the average value of Cu was less than 1.0 in any kind of land-use types, which indicates that it was mainly controlled by the parent material in the study area. Only in industrial land did the *EF* value of As exceed 1.0, indicating that the As might be mainly affected by the parent material and local industrial activities such as steel smelting might be another reason \[[@B48-ijerph-17-02822]\]. For Pb, the mean EF of halophyte land, tideland and forestland were less than 1.0, while that of other land-use types was more than 1.0, especially in aquaculture land and industrial land. Similarly, the mean *EF* of Co ranged between 1.0--1.2 and was relatively high in aquaculture land and industrial land. Therefore, there was a slight enrichment of Pb and Co in the study area, which might be partly due to the natural processes, and local aquaculture and industrial activities. In addition, the average *EF* of Cr, Mn, Ni and Zn under different land-use types were more than 1.0, and the cropland and aquaculture land was much higher than that of other land use types, indicating that they were mainly controlled by agricultural and aquaculture activities. Correlation analysis was performed to analyze the relationship between PHEs \[[@B36-ijerph-17-02822]\]. As depicted in [Figure 6](#ijerph-17-02822-f006){ref-type="fig"}, Cr was weakly or uncorrelated with other elements (r \< 0.4), which demonstrated the source of Cr was different from other measured elements. Most of the remaining PHEs were significantly correlated (r \> 0.5). Especially, As-Cu-Pb had significant correlations (r \> 0.5), Mn had significant correlation with Ni, Co (r = 0.62, 0.58), and Ni-Zn-Cu had extremely significant correlations (r \> 0.75). As, Cu, and Pb had homology; part of the source of Mn might be the same as Ni and the other part might be the same as Co; Ni, Zn, and Cu had homology. The results of cluster analysis ([Figure 7](#ijerph-17-02822-f007){ref-type="fig"}) indicated that all soil PHEs after reclamation were divided into four categories: (1) Co and Mn; (2) Cr; (3) As and Pb; (4) Cu, Zn and Ni. This was consistent with the results of the correlation analysis, indicating that PHEs in the study area were multi-sourced. 3.4. Health Risk Assessment {#sec3dot4-ijerph-17-02822} --------------------------- ### 3.4.1. Non-Carcinogenic Risk Assessment from Eight PHEs {#sec3dot4dot1-ijerph-17-02822} According to the risk assessment model, the non-carcinogenic risk exposure dose (*ADD*), hazard quotient (*HQ~i~*), and hazard index (*HI*) of PHEs for adults and children through three exposure pathways (ingestion, inhalation, and dermal contact) were calculated. It could be seen from [Table 4](#ijerph-17-02822-t004){ref-type="table"} that both the *HQ~i~* and the *HI* were less than 1, indicating that the PHEs in soil potentially had no non-carcinogenic risk to the public. Compared with the national average of non-carcinogenic risks, the non-carcinogenic risks of soil Cr, Cu, and Ni in the study area were lower than the national average, while the non-carcinogenic risks of As, Pb and Zn were slightly higher than the national average \[[@B31-ijerph-17-02822]\]. In general, the daily average exposure dose and non-carcinogenic risk for different people varied slightly, showing that children were higher than adults were. The average hazard quotients of PHEs decreased in the order of *HQ~Mn~* \> *HQ~Cr~* \> *HQ~As~* \> *HQ~Pb~* \> *HQ~Co~* \> *HQ~Ni~* \> *HQ~Cu~* \> *HQ~Zn~*, and the contribution rates to *HI* were 48.87%--73.30%, 13.65%--25.48%, 7.52%--16.09%, 1.29%--1.67%, 0.70%--1.49%, 0.13%--0.28%, 0.10%--0.21% and 0.15%--0.16%, respectively. For all PHEs, the average daily exposure dose of ingestion was almost 10^3^ times that of inhalation and dermal contact. Similarly, the non-carcinogenic risk of most PHEs through ingestion obviously exceeded the other pathways, and risk caused by dermal contact of Mn also accounted for a large proportion. Therefore, the non-carcinogenic risk was mainly contributed by the exposure pathway of ingestion in Cr and dermal contact in Mn. Although there was no non-carcinogenic risk in the study area, the HQ of children were nearly three times that of adults, indicating that children were more vulnerable to PHEs ([Figure 8](#ijerph-17-02822-f008){ref-type="fig"}). ### 3.4.2. Carcinogenic Risk Assessment from As, Co, Ni, Gr and Pb {#sec3dot4dot2-ijerph-17-02822} Due to only As, Co, Ni, Cr, and Pb have carcinogenic slope factor parameters, the *ADD~ing~*, *RI~i,~* and *RI* of adults and children through three exposure routes (oral intake, respiratory inhalation, and dermal contact) were evaluated. As shown in [Table 5](#ijerph-17-02822-t005){ref-type="table"}, the daily average exposure of children is higher than that of adults under different exposure routes, and the *RI~i~* and *RI* of children are higher than that of adults. The average *RI* of PHEs is following the order of *RI~Cr~* \> *RI~As~* \> *RI~Pb~* \> *RI~Co~* \> *RI~Ni~*, with the contribution rate of 83.04%--83.05%, 16.49%--16.52%, 0.42%--0.43%, 0.01%--0.04%, 0.0016%--0.008%, respectively. This indicated that the *RI* of PHEs (As, Co, Ni, Cr, and Pb) in the soil of the study area is at an acceptable risk level, which is mainly contributed by the oral exposure of Cr and As. From different exposure routes, the daily average exposure of oral intake is 2--3 orders of magnitude higher than that of inhalation and dermal contact. The *RI~ing~* of As, Cr, and Pb in oral intake was much higher than that of *RI~inh~* and *RI~derm~*. 4. Discussion {#sec4-ijerph-17-02822} ============= 4.1. The Influence of Different Land Use on PHEs Enrichment in Coastal Tideland Region {#sec4dot1-ijerph-17-02822} -------------------------------------------------------------------------------------- High-intensity reclamations have resulted in drastic changes in land use of tidal flats, which have made ecosystems more fragile and sensitive, including the increased risk of soil pollution \[[@B19-ijerph-17-02822]\]. [Table 3](#ijerph-17-02822-t003){ref-type="table"} indicated that the concentrations of PHEs in surface and subsurface soil layers were almost equal, but the maximum and minimum concentrations of whole sites differed greatly. Furthermore, soil depth was observed to have less influence on the concentration of PHEs than land use. Soil environmental status of the study area was generally in good condition due to the average concentrations of all PHEs lower than the environmental quality standard for soils in China ([Table 3](#ijerph-17-02822-t003){ref-type="table"}). However, the maximum concentration of some PHEs was close to or exceeded the limits of first grade (As and Pb) and second grade (Cr) of the environmental quality standard, which might be affected by reclamation activities, specifically manifested in the changes of land-use intensity and soil properties \[[@B24-ijerph-17-02822],[@B26-ijerph-17-02822],[@B45-ijerph-17-02822]\]. The traditional evolution pathway of land use in the reclaimed coastal areas in eastern China was mudflat→halophyte land→aquaculture land→arable land→construction land \[[@B51-ijerph-17-02822]\], which showed that the land-use intensity gradually increased with the increase of human activity interference. In this study, the concentrations of most PHEs (except Co) were in the following order: aquaculture land \> industrial land/cropland \> forestland/residential land \> vegetable land \> tideland/halophyte land ([Figure 3](#ijerph-17-02822-f003){ref-type="fig"}). Yao et al. \[[@B52-ijerph-17-02822]\] also reported similar conclusions after soil investigation on the reclaimed tideland of Dongtai City, Jiangsu Province, which found there was no significant excess of soil PHEs, while the total ecological risk and polluted level showed the trend following the order: industrial park \> residential area \> suburb farmland \> newly-reclaimed tidal flats. ANOVA ([Figure 3](#ijerph-17-02822-f003){ref-type="fig"}) illustrated that the concentrations of As, Cu, Pb and Zn in soil after 30 years of reclamation (industrial land, aquaculture land, farmland, residential land, and vegetable land) were significantly higher than those in uncultivated coastal wetlands (tideland and halophyte land), which might be caused by changes in soil physical and chemical properties under reclamation activities \[[@B19-ijerph-17-02822]\]. It is reported that the sand content, salinity, and pH gradually decreased, while the silt and clay content and organic matter content gradually increased following tidal land reclamation, which finally reached a stable state after 30 years reclamation \[[@B14-ijerph-17-02822],[@B19-ijerph-17-02822]\]. The dynamics of soil physicochemical properties have been proven as the key factors affecting the mobility of PHEs in soil \[[@B14-ijerph-17-02822]\]. Specifically, fine-grained sediments are easy to solidify PHEs in soil \[[@B53-ijerph-17-02822]\], the increase of SOM promotes PHEs accumulation \[[@B1-ijerph-17-02822],[@B24-ijerph-17-02822]\], and the soluble PHEs in pore water decreases with the decreasing of soil salinity after reclamation \[[@B27-ijerph-17-02822]\]. The sources of PHEs in the study area were complex based on the results of enrichment analysis, correlation analysis, and cluster analysis. The *EFs* were observed less than 2.0 in the study area, which indicated the sources of soil PHEs were largely contributed by parent materials and partly determined by human activities, such as land use. The enrichment analysis showed Co (0.86--1.39), Cr (0.96--2.01), Mn (1.21--1.75), Ni (1.11--1.84), Pb (0.78--1.85), and Zn (0.97--1.61) were generally in slight enrichment, while Cr, Ni, and Pb in some sampling sites were close to moderate enrichment ([Figure 4](#ijerph-17-02822-f004){ref-type="fig"}). Therefore, it is necessary to pay attention to the pollution of Cr, Ni, and Pb in the future reclamation and utilization of tidal land. The EFs of the determined PHEs in halophyte land, tideland, and forestland were generally classified as low concentrations group, and that in cropland and aquaculture land were classified as high concentrations group. Similarly, Sowan et al. \[[@B26-ijerph-17-02822]\] reported that the PHEs contamination in municipality areas, industrial zones, and dockyard were more serious, while shrimp farming and traditional land uses such as salt marshes, paddy fields, orchards, and mangrove forests showed low levels of metals in Pattani Bay. These conclusions proved the intensity of human activity further influenced the soil PHEs enrichment \[[@B7-ijerph-17-02822],[@B54-ijerph-17-02822]\]. It could be seen in [Table 3](#ijerph-17-02822-t003){ref-type="table"}, the highest concentration of Cu (14.10 mg·kg^−1^) was much smaller than the background of Jiangsu (22.3 mg·kg^−1^), and the EFs of Cu did not exceed 1.0 in all land-use types. This indicates that no significant Cu enrichment was observed in the study area. Actually, the average concentration of Cu (8.10 mg·kg^−1^) was close to the deep sediment of coastal mudflat in Jiangsu (13.0 mg·kg^−1^) \[[@B49-ijerph-17-02822]\], which indicated that Cu most likely originated in natural factors and was rarely disturbed by anthropogenic activities \[[@B55-ijerph-17-02822]\]. Co and Mn might have the same sources and be slightly accumulated by the influence of human activities, as their correlation coefficient was 0.58 and the average EF ranged from 1 to 1.5. [Table 3](#ijerph-17-02822-t003){ref-type="table"} showed that the average concentrations of Co and Mn were less than the background parameters, and their concentrations in the surface (8.00 mg·kg^−1^, 457.49 mg·kg^−1^) were lower than that in the subsurface layer (8.01 mg·kg^−1^, 460.03 mg·kg^−1^). This phenomenon was primarily attributed to the natural leaching, which indicated that Co and Mn were controlled by natural geological processes \[[@B15-ijerph-17-02822]\]. Li et al. \[[@B7-ijerph-17-02822]\] and Franco-Uría et al. \[[@B56-ijerph-17-02822]\] also presented that Co and Mn were originated from a lithogenic source \[[@B24-ijerph-17-02822]\]. The EF of Co and Mn in tideland (EF = 1.14, 1.43) and halophyte land (EF = 1.16, 1.39) were observed to be slightly lower than that in aquaculture land (EF = 1.17, 1.47), which also revealed that the Co and Mn background content in the tidal flats under the natural sedimentary process was accounted for high proportions \[[@B24-ijerph-17-02822]\]. Co and Mn highly accumulated in aquaculture land compared with other land-use types. The marine aquaculture fish feed usually contains Zn, Cu, Cd, Fe, Mn, Co, Ni, P \[[@B55-ijerph-17-02822]\], and it may be a major source of Co and Mn. In addition, the water submerged environment promotes metals and S^2-^ associate and precipitate to insoluble metal sulfides under redox status \[[@B57-ijerph-17-02822]\]. Thereby, natural sources mostly contribute to the distribution of Co and Mn, and aquaculture is another source. Cr seemed to be an isolated element, having low correlations with other elements (r \< 0.4) \[[@B58-ijerph-17-02822]\]. Even if Cr originates from a natural source which has been extensively mentioned in many pieces of literature \[[@B15-ijerph-17-02822],[@B59-ijerph-17-02822]\]. In this study, the distribution of Cr was mainly affected by agricultural input as the average EF of Cr in cropland (EF = 1.36) and aquaculture land (EF = 1.30) was significantly higher than that of others. The average concentration of determined Cr (52.50 mg·kg^−1^) in cropland was higher than that in phosphate fertilizers (47.3 mg·kg^−1^), whereas much less than that in sludge (86.73 mg·kg^−1^) \[[@B24-ijerph-17-02822]\]. Consequently, the enrichment of Cr in the study area may be the result of the application of fertilizer using sludge as the raw material. Field investigation found printing and dyeing mills and mining enterprises using Cr compounds around the reclamation area \[[@B60-ijerph-17-02822]\]. In summary, sewage irrigation and fish feed fish have become important sources of soil Cr in this area \[[@B55-ijerph-17-02822],[@B61-ijerph-17-02822]\]. The slight accumulation of soil As appeared locally rather than the whole area (*EF~average~* = 0.58, *EF~max~* = 1.19). As, Cu and Pb had homology as the correlation coefficients among them all exceeded 0.5. It had been concluded that Cu originated from natural factors, which indirectly revealed that As and Pb were partially derived from natural parent materials. The average concentration and EF of As in industrial land were significantly higher than those in other land use types, indicating that the industrial activities might be another source of As. Huang et al. \[[@B62-ijerph-17-02822]\] and Khairy et al. \[[@B63-ijerph-17-02822]\] reported that As was related to the emissions of industrial flue gas, sewage, and sludge. The average EF of Pb ranged from 0.93 to 1.21 under different land-use types and tended to accumulate in aquaculture land, industrial land, vegetable land, and cropland. The cropland and fishponds were adjacent to each other, which might cause mutual pollution. It is widely reported that pesticides and fertilizers (e.g., phosphate fertilizers, nitrate fertilizers, and organic fertilizers) applied in agricultural soils caused high concentrations of Cd, Pb, and Zn \[[@B15-ijerph-17-02822],[@B64-ijerph-17-02822]\]. Zn-Ni-Cu were significantly correlated (r \> 0.75), which revealed that partly, sources of Zn and Ni might derive from natural parent materials as well as Cu. This was consistent with the result of Sun et al. \[[@B59-ijerph-17-02822]\]. Slight enrichment of Zn and Ni occurred in aquaculture land, industrial land, and cropland (1.30 \< EF \< 1.60, [Figure 4](#ijerph-17-02822-f004){ref-type="fig"}). Zn accumulated in fishery sediments may be attributed to the input of fish feed or fish manure \[[@B55-ijerph-17-02822]\]. In addition, the application of poultry manure in agricultural land increased Zn accumulation \[[@B15-ijerph-17-02822],[@B56-ijerph-17-02822]\]. Furthermore, the emission of automobile exhaust and wear of the tires significantly affected the Zn and Ni concentrations in road dust, and then entered the surrounding soil through atmospheric sedimentation \[[@B15-ijerph-17-02822],[@B63-ijerph-17-02822]\]. Ni might be directly related to the "three wastes" (including some domestic waste) emissions during the production of batteries near the study area as the battery slag and wastewater contains high concentrations of Pb, Cd, Ni, etc. \[[@B49-ijerph-17-02822]\]. 4.2. The PHEs Health Risk in Coastal Tideland Reclamation Area {#sec4dot2-ijerph-17-02822} -------------------------------------------------------------- It is found that the total non-carcinogenic risk of soil PHEs in the study area is at the no non-carcinogenic risk level, while the total carcinogenic risk is at the acceptable risk level. This is consistent with the result that the average concentration of PHEs in [Table 3](#ijerph-17-02822-t003){ref-type="table"}, indicating that the concentration of soil PHEs in the study area has no obvious harm to human health. It is worth noting that the carcinogenic risk index of Cr (1.50 × 10^−5^--2.30 × 10^−5^) and As (2.99 × 10^−6^--4.58 × 10^−6^) exceeds the maximum acceptable carcinogenic risk of the human body. This might be attributed to the high slope coefficient of the carcinogenic risk of Cr and As \[[@B31-ijerph-17-02822],[@B43-ijerph-17-02822]\]. It is reported that As is a carcinogen for lung cancer and skin cancer, and excessive intake of Cr will cause respiratory and digestive diseases \[[@B3-ijerph-17-02822]\]. Accordingly, As and Cr should receive more attention in future soil environmental monitoring of coastal reclamation areas. Tideland reclamation accelerated the transformation from bare flat to cultivated land, aquaculture land and construction land, which had an important impact on the health risk of PHEs in soil. In this study, the highest non-carcinogenic risk and the carcinogenic risk appeared in industrial land, followed by farmland and vegetable land ([Figure 8](#ijerph-17-02822-f008){ref-type="fig"}). Yang et.al \[[@B60-ijerph-17-02822]\] found that the health risks of industrial areas were higher than that of agricultural areas in China. [Table 3](#ijerph-17-02822-t003){ref-type="table"} and [Table 4](#ijerph-17-02822-t004){ref-type="table"} showed that the ratio of average daily exposure dose between ingestion and inhalation, dermal contact was all close to 10^3^. However, the non-carcinogenic risk was mainly contributed by ingestion and dermal contact of Mn and ingestion of Cr ([Table 4](#ijerph-17-02822-t004){ref-type="table"}). Carcinogenic risk assessment ([Table 5](#ijerph-17-02822-t005){ref-type="table"}) showed that ingestion was the main exposure route for As, Cr, and Pb, while the inhalation was the unique exposure pathway for Co and Ni. This might be due to the increased risk exposure pathway by the reclamation of tideland. The smoke discharged by industry and transportation enters the human body through inhalation, while the rest enters the soil through natural sedimentation and rain. PHEs from industrial activities and agricultural fertilization management accumulate continuously after entering the soil and migrate in different systems (such as water systems and crops), thereby entering the human body through the food chain, resulting in harm to human health \[[@B3-ijerph-17-02822],[@B65-ijerph-17-02822]\]. Besides, compared with daily average exposure, non-carcinogenic risk index and carcinogenic risk index, it can be found that children are more vulnerable to PHEs in soil than adults ([Figure 8](#ijerph-17-02822-f008){ref-type="fig"}), which is similar to previous research results \[[@B44-ijerph-17-02822],[@B60-ijerph-17-02822],[@B65-ijerph-17-02822]\]. It is mainly because the children's immune resistance is much lower than that of adults, which is more sensitive to environmental pollution. In addition, children's exposure to soil due to outdoor activities is higher than that of adults, and children's physical behavior (such as finger sucking) makes their health risk higher than adults \[[@B3-ijerph-17-02822],[@B65-ijerph-17-02822]\]. 5. Conclusions {#sec5-ijerph-17-02822} ============== The enrichment, sources apportionment, and health risk assessment of eight PHEs (As, Co, Cr, Cu, Mn, Ni, Pb, and Zn) were identified in Jiangsu coastal reclamation area, eastern China. In general, none of the PHEs we concerned about exceeding the environmental quality standard for soils in China. Enrichment analysis showed that As and Cu were not enriched, while other elements were slightly enriched. Land use is observed to have an obvious enrichment effect on most of the PHEs, which resulted in a relatively high level of PHEs enrichment in aquaculture land, industrial land, and cropland compared with other land use types, such as tideland and halophyte land. The correlation analysis and hierarchically cluster divided PHEs into five categories: (1) Cu; (2) Co and Mn; (3) Cr; (4) As and Pb; (5) Zn and Ni. Additionally, it was observed that the Cu was completely derived from natural parent materials; the Mn and Co were mainly controlled by natural sources and aquaculture is another source, and; the Cr was an independent element and was mainly derived from agricultural fertilization and aquaculture activities. However, the As, Pb, Ni, and Zn are governed by both weathering of parent rock and human activities, including agricultural activities, industrial production, and transportation emissions. The health risk assessment showed that the soil PHEs potentially had no non-carcinogenic risk to the public, while there was an acceptable probability to have cancer due to Cr and As. Meanwhile, children were more susceptible to harm of soil PHEs than that of adults. Although the soil PHEs in the study area had not caused significant negative effects on the environment and individuals, the uncertainty of some PHEs (Pb, As, and Cr) was still high. That is to say, it may increase PHEs pollution and further increase the probability of human body damage if land management could not be rationally planned in the future. Conceptualization, X.C., X.X. and L.P.; Data curation, X.X. and M.Z.; Methodology, J.L. and X.W.; Project administration, L.P.; Writing---original draft, X.C. and S.H. All authors have read and agreed to the published version of the manuscript. This research was supported by the National Natural Science Foundation of China (41230751), the Open Fund of Key Laboratory of Coastal Zone Exploitation and Protection, Ministry of Natural Resource (2019CZEPK09), and the Scientific Research Foundation of Graduate School of Nanjing University (2014CL09). The authors declare no conflict of interest in this paper and study. {#ijerph-17-02822-f001} {#ijerph-17-02822-f002} {#ijerph-17-02822-f003} {#ijerph-17-02822-f004} {#ijerph-17-02822-f005} {#ijerph-17-02822-f006} {#ijerph-17-02822-f007} {#ijerph-17-02822-f008} ijerph-17-02822-t001_Table 1 ###### Parameters used to evaluate exposure risks of soil metals. Parameters Definitions Values References ------------ ------------------------------------------------------------------- ---------------------- ---------------------------- ---------------------------- *C~Soil~* Concentration of potentially harmful elements in soil/(mg·kg^−1^) Average *CF* Conversion coefficient 1 × 10^−6^ kg·mg^−1^ \[[@B38-ijerph-17-02822]\] *IngR* Ingestion rate of soil/(mg·d^−1^) 100 200 \[[@B38-ijerph-17-02822]\] *EF* Exposure frequency/(d·a^−1^) 365 \[[@B39-ijerph-17-02822]\] *ED* Exposure duration/year 24 6 \[[@B38-ijerph-17-02822]\] *BW* Weight/kg 60.6 19.6 \[[@B40-ijerph-17-02822]\] *AT~nc~* Averaged time of carcinogenic impact/day 70×365 \[[@B38-ijerph-17-02822]\] *AT~ca~* Averaged time of non-carcinogenic impact/day ED×365 \[[@B38-ijerph-17-02822]\] *InhR* inhalation rate of soil/(d·a^−1^) 20 7.65 \[[@B38-ijerph-17-02822]\] *PEF* Emission factor/(m^3^·kg^−1^) 1.36×10^9^ \[[@B38-ijerph-17-02822]\] *SA* Exposed skin surface area/cm^2^ 5 700 2 800 \[[@B40-ijerph-17-02822]\] *SL* Adherence factor/(mg·cm^−2^·d^−1^) 0.07 0.2 \[[@B38-ijerph-17-02822]\] *ABS* Dermal absorption factor 0.001 \[[@B38-ijerph-17-02822]\] ijerph-17-02822-t002_Table 2 ###### Reference dose and slope factor of potentially harmful elements (PHEs) under different exposure pathways. Parameter As Co Cr Cu Mn Ni Pb Zn ------------------------ ------------------------- ---------------- --------------- --------------- ---------------- --------------- --------------- --------------- --------------- Ingestion *RfD*/(mg·kg^−1^·d^−1^) 3.00 × 10^−4^ 2.00 × 10^−2^ 3.00 × 10^−3^ 4.00 × 10^−2^ 4.60 × 10^−2^ 2.00 × 10^−2^ 3.50 × 10^−3^ 3.00 × 10^−1^ *SF*/(mg·kg^−1^·d^−1^) 1.50 --- 5.00 × 10^−1^ --- --- --- 8.50 × 10^−3^ --- Inhalation *RfD*/(mg·kg^−1^·d^−1^) 3.00 × 10^−4^ 5.71 × 10^−6^ 2.86 × 10^−5^ 4.00 × 10^−02^ 1.43 × 10^−5^ 2.06 × 10^−2^ 3.52 × 10^−3^ 3.00 × 10^−1^ *SF*/(mg·kg^−1^·d^−1^) 1.51 × 10 ^1^ 9.80 4.20 × 10^1^ --- --- 8.40 × 10^−1^ --- --- D × 10rmal Contact *RfD*/(mg·kg^−1^·d^−1^) 1.23 × 10 ^−4^ 1.60 × 10^−5^ 6.00 × 10^−5^ 1.20 × 10^−2^ 1.84 × 10^−5^ 5.40 × 10^−3^ 5.25 × 10^−3^ 6.00 × 10^−2^ *SF*/(mg·kg^−1^·d^−1^) 3.66 --- --- --- --- --- --- --- PHEs with carcinogenic risk are As, Cd, Cr, Ni and Pb; others are non-carcinogenic metals. "---" in the table means not available. ijerph-17-02822-t003_Table 3 ###### Statistical description of PHEs contents in the study area (mg·kg^−1^). PHEs Min Max Mean Mean~1~ Mean~2~ S.D S.D~1~ S.D~2~ CV Background of Jiangsu \[[@B32-ijerph-17-02822]\] Background of China \[[@B48-ijerph-17-02822]\] Risk Control Standard China (1)/(2)/(3) ------ -------- -------- -------- --------- --------- ------- -------- -------- ------- -------------------------------------------------- ------------------------------------------------ ----------------------------------------- As 0.44 11.08 3.31 3.29 3.33 2.09 1.41 1.62 60.06 10.0 15.0 15/25/30 Co 5.04 10.81 8.00 8.00 8.01 1.13 1.15 1.04 14.09 12.6 − − Cr 38.94 116.53 52.50 51.84 53.16 11.60 7.42 14.52 22.10 77.8 90.0 90/200/300 Cu 4.13 14.10 8.10 8.36 7.85 2.21 2.35 1.97 27.22 22.3 35.0 30/100/400 Mn 374.19 643.80 458.76 457.49 460.03 35.32 39.10 30.71 7.70 585 − − Ni 16.49 28.25 21.47 21.35 21.58 2.09 2.31 1.81 9.73 26.7 40.0 40/50/200 Pb 9.18 38.76 15.90 16.31 15.49 3.52 3.39 3.57 22.14 26.2 35.0 35/300/500 Zn 33.37 62.04 45.45 46.82 44.07 5.76 6.22 4.84 12.67 62.6 100.0 100/250/500 *Risk control standard China* is Environmental quality standard for soils (GB 15618-1995) and has three levels of standards. (1) means the limit value of the soil environmental quality to protect the regional natural ecology and maintain the natural background. (2) means the limit value of the soil to ensure agricultural production and maintain human health; (3) means the limit value of the soil to ensure the agricultural and forestry production and normal plant growth. Mean~1~ and Mean~2~ represent the average concentration of PHEs in surface and subsurface. S.D~1~ and S.D~2~ represent standard deviations for the surface and subsurface concentrations. ijerph-17-02822-t004_Table 4 ###### Daily exposure and non-carcinogenic risk index for adults and children with PHEs and pathways in 0--40 cm soil layer. Metals Age Group *ADD~ing~* *ADD~inh~* *ADD~derm~* *HQ~ing~* *HQ~inh~* *HQ~derm~* *HQ~i~* ---------- --------------- ---------------- ---------------- --------------- --------------- --------------- --------------- --------------- As Adult 5.74 × 10^−6^ 8.44 × 10^−10^ 2.29 × 10^−8^ 1.91 × 10^−2^ 2.81 × 10^−6^ 1.86 × 10^−4^ 1.93 × 10^−2^ Childen 3.55 × 10^−5^ 9.99 × 10^−10^ 3.48 × 10^−8^ 1.18 × 10^−1^ 3.33 × 10^−6^ 2.83 × 10^−4^ 1.19 × 10^−1^ Co Adult 1.32 × 10^−5^ 1.95 × 10^−9^ 5.28 × 10^−8^ 6.62 × 10^−4^ 3.41 × 10^−4^ 3.30 × 10^−3^ 4.30 × 10^−3^ Children 8.18 × 10^−5^ 2.30 × 10^−9^ 8.02 × 10^−8^ 4.09 × 10^−3^ 4.03 × 10^−4^ 5.01 × 10^−3^ 9.51 × 10^−3^ Cr Adult 8.66 × 10^−5^ 1.27 × 10^−8^ 3.46 × 10^−7^ 2.89 × 10^−2^ 4.45 × 10^−4^ 5.76 × 10^−3^ 3.51 × 10^−2^ Children 5.36 × 10^−4^ 1.51 × 10^−8^ 5.25 × 10^−7^ 1.79 × 10^−1^ 5.27 × 10^−4^ 8.75 × 10^−3^ 1.88 × 10^−1^ Cu Adult 1.34 × 10^−5^ 1.97 × 10^−9^ 5.35 × 10^−8^ 3.35 × 10^−4^ 4.93 × 10^−8^ 4.46 × 10^−6^ 3.39 × 10^−4^ Children 8.29 × 10^−5^ 2.33 × 10^−9^ 8.12 × 10^−8^ 2.07 × 10^−3^ 5.83 × 10^−8^ 6.77 × 10^−6^ 2.08 × 10^−3^ Mn Adult 7.57 × 10^−4^ 1.11 × 10^−7^ 3.02 × 10^−6^ 1.65 × 10^−2^ 7.79 × 10^−3^ 1.64 × 10^−1^ 1.88 × 10^−1^ Children 4.68 × 10^−3^ 1.32 × 10^−7^ 4.59 × 10^−6^ 1.02 × 10^−1^ 9.21 × 10^−3^ 2.49 × 10^−1^ 3.60 × 10^−1^ Ni Adult 3.54 × 10^−5^ 2.53 × 10^−7^ 1.41 × 10^−7^ 1.77 × 10^−3^ 2.53 × 10^−7^ 2.62 × 10^−5^ 1.80 × 10^−3^ Children 2.19 × 10^−4^ 2.99 × 10^−7^ 2.15 × 10^−7^ 1.10 × 10^−2^ 2.99 × 10^−7^ 3.98 × 10^−5^ 1.10 × 10^−2^ Pb Adult 2.62 × 10^−5^ 3.86 × 10^−9^ 1.05 × 10^−7^ 7.50 × 10^−3^ 1.10 × 10^−6^ 1.99 × 10^−5^ 7.52 × 10^−3^ Children 1.62 × 10^−4^ 4.56 × 10^−9^ 1.59 × 10^−7^ 4.64 × 10^−2^ 1.30 × 10^−6^ 3.03 × 10^−5^ 4.64 × 10^−2^ Zn Adult 7.50 × 10^−5^ 1.10 × 10^−8^ 2.99 × 10^−7^ 2.50 × 10^−4^ 3.68 × 10^−8^ 4.99 × 10^−6^ 2.55 × 10^−4^ Children 4.64 × 10^−4^ 1.30 × 10^−8^ 4.55 × 10^−7^ 1.55 × 10^−3^ 4.35 × 10^−8^ 7.58 × 10^−6^ 1.55 × 10^−3^ *HI* Adult --- ^1^ --- --- --- --- --- 2.57 × 10^−1^ Children --- --- --- --- --- --- 7.37 × 10^−1^ ^1^ "---" in the table means not available. ijerph-17-02822-t005_Table 5 ###### Daily exposure and carcinogenic risk index for adults and children with PHEs and pathways in 0--40 cm soil layer. Metals Age Group *ADD~ing~* *ADD~inh~* *ADD~derm~* *CR~ing~* *CR~inh~* *CR~derm~* *CR~i~* ---------- --------------- ---------------- ---------------- --------------- ---------------- --------------- ---------------- --------------- As Adult 1.97 × 10^−6^ 2.90× 10^−10^ 7.86 × 10^−9^ 2.95 × 10^−6^ 4.37 × 10^−9^ 2.88 × 10^−8^ 2.99 × 10^−6^ Children 3.04 × 10^−6^ 8.56 × 10^−11^ 2.98 × 10^−9^ 4.57 × 10^−6^ 1.29 × 10^−9^ 1.09 × 10^−8^ 4.58 × 10^−6^ Co Adult 4.54 × 10^−6^ 6.67 × 10^−10^ 1.81 × 10^−8^ --- ^1^ 6.54 × 10^−9^ --- 6.54 × 10^−9^ Children 7.01 × 10^−6^ 1.97 × 10^−10^ 6.87 × 10^−9^ --- 1.93 × 10^−9^ --- 1.93 × 10^−9^ Cr Adult 2.97 × 10^−5^ 4.37 × 10^−9^ 1.19 × 10^−7^ 1.49 × 10^−5^ 1.83 × 10^−7^ --- 1.50 × 10^−5^ Children 4.59 × 10^−5^ 1.29 × 10^−9^ 4.50 × 10^−8^ 2.30 × 10^−5^ 5.42 × 10^−8^ --- 2.30 × 10^−5^ Ni Adult 1.21 × 10^−5^ 1.79 × 10^−9^ 4.85 × 10^−8^ --- 1.50 × 10^−9^ --- 1.50 × 10^−9^ Children 1.88 × 10^−5^ 5.28 × 10^−10^ 1.84 × 10^−8^ --- 4.44 × 10^−10^ --- 4.44 × 10^−10^ Pb Adult 9.00 × 10^−6^ 1.32 × 10^−9^ 3.59 × 10^−8^ 7.65 × 10^−8^ --- --- 7.65 × 10^−8^ Children 1.39 × 10^−5^ 3.91 × 10^−10^ 1.36 × 10^−8^ 1.18 × 10^−7^ --- --- 1.18 × 10^−7^ *CR* Adult --- --- --- --- --- --- 1.81 × 10^−5^ Children --- --- --- --- --- --- 2.77 × 10^−5^ ^1^ "---" in the table means not available. [^1]: Co-first Author.

All relevant data are within the paper and its Supporting Information files. Introduction {#sec005} ============ Febrile neonates (aged \<30 days) are a high-risk group for serious bacterial infections \[[@pone.0120959.ref001],[@pone.0120959.ref002]\]. Reasons for this include susceptibility to infections, difficulty with discriminatory clinical examination, and poor outcomes if not treated or diagnosed promptly \[[@pone.0120959.ref001]\]. According to a recent World Health Organization update, 1/3 of 4 million annual neonatal deaths worldwide were the result of severe infections \[[@pone.0120959.ref003],[@pone.0120959.ref004]\]. Due to the under-recognition of the severity of illness and lack of access to services, most sick neonates in developing countries die before even reaching medical care \[[@pone.0120959.ref005],[@pone.0120959.ref006]\]. This is not a problem unique to poor and developing countries. In 2007, among 17 peer developed nations, Canada's infant mortality rate was second worst \[[@pone.0120959.ref007]\]. In 2009, the neonatal mortality rate in Canada was 4 per 1000 live births, higher than the 3.6 per 1000 average among 75 high income countries \[[@pone.0120959.ref008]\]. The provision of high level neonatal and infant care faces specific challenges created by Canada's geography, low population density and a large proportion of new immigrants from developing countries \[[@pone.0120959.ref009],[@pone.0120959.ref010]\]. As a result, prevention of infant and neonatal mortality is now a national policy priority for this country \[[@pone.0120959.ref010],[@pone.0120959.ref011]\]. In Canada, a contributing cause of death of neonates is sepsis. Sepsis is defined as the presence (probable or documented) of infection together with systematic manifestations of infection \[[@pone.0120959.ref012]\]. Fever (defined in this population as a rectal temperature of 38 degrees Celsius or greater \[[@pone.0120959.ref013],[@pone.0120959.ref014]\]) can often be the only sign of a serious bacterial infection in the neonate \[[@pone.0120959.ref001],[@pone.0120959.ref013],[@pone.0120959.ref015]\], but may not always be present \[[@pone.0120959.ref013],[@pone.0120959.ref014]\]. In the last two decades, clinical guidelines about how to recognize and treat fever in the neonate have been studied, reported, and scrutinized in many journals \[[@pone.0120959.ref016]--[@pone.0120959.ref018]\]. As well, variations in practice patterns in the evaluation of the febrile newborn have also been observed \[[@pone.0120959.ref019]\]. Despite ongoing discussions about optimal medical treatment, care cannot commence in this population if the parent does not recognize the fever and seek out medical attention. Consequently, parental understanding of fever is critical. Remarkably in developed nations such as Canada, very little is known about expectant parents' knowledge of fever, a potential sign of modifiable and treatable illness in this age group. Recent studies from developed countries have reported that many parents do not know the precise definition of "fever" or believe fever itself is dangerous \[[@pone.0120959.ref020],[@pone.0120959.ref021]\]. Expectant parents that are new immigrants, and of lower socioeconomic and educational status, have been shown to possess less knowledge on pediatric fever, and may be less likely to seek care for a febrile neonate \[[@pone.0120959.ref021]--[@pone.0120959.ref025]\]. To the best of our knowledge, no study has been undertaken to assess Canadian expectant parents' understanding of fever in the neonate. Thus, the objectives of this study were to estimate the extent to which Canadian expectant parents would seek medical care in a febrile neonate (age 30 days or less), evaluate expectant parents' knowledge of signs and symptoms of fever in a neonate, and describe the actions Canadian expectant parents would take if they recognized fever in their child. Methods {#sec006} ======= Setting {#sec007} ------- In 2011, the population of Canada was 33.4 million persons, with an estimated 378,762 new babies born that year. At this time, Canada had an immigrant population of 6.1 million persons \[[@pone.0120959.ref007]\], and over 4.7 million persons (14.2% of the population) spoke a language other than English or French most often at home \[[@pone.0120959.ref007]\]. Montreal is Canada's second largest city with a population of 1.6 million in 2011 and an area of 4,259 square kilometers. All health care in Canada is delivered through a publically funded health care system, which is mostly free at the point of use. Ethics Consideration {#sec008} -------------------- This study was reviewed and approved by the institutional research ethics boards at the Jewish General Hospital in Montreal, Canada. Study Design, Sample, and Data Collection {#sec009} ----------------------------------------- We conducted a cross-sectional survey of expectant parents from February to April 2011 in a large Canadian urban setting. We invited English or French speaking expectant parents, aged 18 years and over, to participate in the study. We recruited participants from an obstetrics clinic at the Jewish General Hospital, a tertiary care center serving a diverse patient population in a largely immigrant-based area of Montreal, Quebec, Canada. We invited expectant parents to our study while they waited for their scheduled obstetrics appointments. If an individual agreed to participate, we asked him/her to provide informed written consent. Once consent was obtained, we asked participants to complete an anonymous survey in either English or French and drop the completed survey in an opaque box. After survey completion, or in the case of participants who did not consent to participate, we gave all individuals an educational pamphlet about the definition and significance of fever in the neonate (see [S1 Appendix](#pone.0120959.s001){ref-type="supplementary-material"} and [S2 Appendix](#pone.0120959.s002){ref-type="supplementary-material"}). Variables and Measurement {#sec010} ------------------------- ### Background information {#sec011} Gender was coded as man or woman. Age was dichotomized into "≤27 years" or "\>27 years". First language spoken at home was coded as "English", or "French/other". Stage of pregnancy was coded as "0--26 weeks", and "27--40 weeks". Immigration status was coded as "born in Canada" or "moved to Canada". Education was coded as "high school or less" or "more than high school". Parent of a second child was coded as "yes" or "no". Income was coded as "\$0-\$49,999" or "\$50,000+". All continuous variables were coded based on median values from found in the initial analysis of the data. ### Knowledge of fever {#sec012} In order to assess knowledge of fever, we asked expectant parents nine questions (see [S3 Appendix](#pone.0120959.s003){ref-type="supplementary-material"} and [S4 Appendix](#pone.0120959.s004){ref-type="supplementary-material"}). Questions were developed based on a review of the literature and consultation with experts in the field of pediatrics, pediatric emergency medicine, and internal medicine. We measured knowledge in three ways. First, we asked one question about whether parents correctly identified the need to obtain medical care in a febrile neonate (scored correct/incorrect). Second, we asked five questions about whether expectant parents could identify potential signs of fever (scored correct/partially correct/incorrect). Finally, we asked three questions to explore the actions the expectant parent would take if they suspected fever in their neonate (see [S3 Appendix](#pone.0120959.s003){ref-type="supplementary-material"} and [S4 Appendix](#pone.0120959.s004){ref-type="supplementary-material"} for list of questions). Statistical Methods {#sec013} ------------------- As mentioned above, we assessed expectant parental knowledge in three ways. First, in order to estimate what proportion of expectant parents would seek medical care in the case they suspect fever in the neonate, we summarized the frequency that parents endorsed one item "*If your baby less than 30 days old has a fever*, *you should bring them to seek medical care immediately*". We calculated crude odds ratios (OR) and 95% confidence intervals (CI) to ascertain the effects of age, gender, immigration status, education, language spoken, income, stage of pregnancy, and number of other children. Second, for the five questions asked about potential signs of fever, we summarized the number of parents who answered each question correctly or incorrectly. As we are in the exploratory phase of this research, and do not yet have a psychometrically sound screening tool, we did not add these items to produce a total score. Instead, we counted and displayed the individual responses and displayed the response to each item using frequency tables and cross tabulation. Finally, for the last three questions about the actions expectant parents would take if they suspected fever in their neonate, we counted the frequency that parents endorsed pre-selected categories for each question. For all questions in the survey, we also recorded the number of participants who completed each question. In order to test if missing data was missing at random, we performed chi-square tests to test whether there were differences amongst the individuals who completed the items, and those who did not. All calculations and analyses were performed using SAS software, Version 9.2 (SAS Institute, Inc, Cary, NC). Results {#sec014} ======= We approached a total of 428 expectant parents for this study. As shown in [Fig 1](#pone.0120959.g001){ref-type="fig"}, 371 (87%) agreed to fill in the survey. A total of 377 surveys were partially completed. Twenty three participants (4%) only completed baseline demographic questions, whereas 355 respondents (96%) completed both the demographic and survey parts of the questionnaire. The majority of participants who completed the survey were female (94%), over age 27 (84%), highly educated (60% university or post-graduate degree), and in their second or third trimester of pregnancy (94%). Forty four percent of respondents reported English as their first language, followed by 39% reporting a language other than English or French as their first language, and 17% reporting French as their first language (see [Table 1](#pone.0120959.t001){ref-type="table"}). Individuals who did not fully complete the survey were not statistically different, in terms of demographics, compared to those who fully completed the survey. {#pone.0120959.g001} 10.1371/journal.pone.0120959.t001 ###### Demographic characteristics of the sample[\*](#t001fn001){ref-type="table-fn"} of people who completed survey questions (n = 355). {#pone.0120959.t001g} Characteristic N (%) ------------------------------ ---------- Age  \<18 years 1 (1)  18--26 years 42 (12)  27--35 years 214 (60)  35+ years 96 (26)  Missing 2 (1) Female 332(94) Spoken First Language  English 156 (44)  French 62 (17)  Other 137 (39) Education  High School or less 34 (10)  Community College 76 (21)  University or post-graduate 213 (60)  Missing 32(9) Ethnicity/Religion  African American 16 (5)  White 64 (18)  Asian 38 (11)  First Nation 0 (0)  Jewish 60 (17)  Christian 92 (26)  Muslim 57 (16)  Hindu 8 (2)  Other 20 (5) Immigrant Status  Born in Canada 158 (45)  Born outside of Canada 197 (55) Stage of Pregnancy  1st trimester 15 (4)  2nd trimester 175 (49)  3rd trimester 160 (45)  Missing 5(2) \* In 2011, the population of Canada was 33.4 million persons, with an estimated 378,762 new babies born that year. At this time, Canada had an immigrant population of 6.1 million persons\[[@pone.0120959.ref003]\], and over 4.7 million persons (14.2% of the population) spoke a language other than English or French most often at home\[[@pone.0120959.ref009]\]. Montreal is Canada's second largest city with a population of 1.6 million in 2011 and an area of 4,259 square kilometers. All health care in Canada is delivered through a publically funded health care system, which is mostly free at the point of use. The first objective was to estimate the extent to which Canadian expectant parents would seek medical care in a febrile neonate (age 30 days or less). [Fig 2](#pone.0120959.g002){ref-type="fig"} summarizes expectant parents' knowledge of fever. We found that 75% of expectant parents in this study would bring their child to seek medical care if they suspected fever. Nearly one fifth of parents (17%) reported that they would not take their neonate to seek medical care if they suspected a fever. We found that 28 individuals (8%) did not answer this question. As shown in [Table 2](#pone.0120959.t002){ref-type="table"}, we found no significant associations between the choice to seek medical care and any socio-demographic variables. {#pone.0120959.g002} 10.1371/journal.pone.0120959.t002 ###### Associations between socio-demographics and knowledge of whether to take febrile neonate to seek medical care. {#pone.0120959.t002g} Characteristics Would seek medical care n = 287/355 (%) OR (95% CI) --------------------------------------------- ----------------------------------------- ------------------- Gender  Male 21/23 (91.3) 0.42 (0.10--1.75)  Female 266/289 (92.0) 1.0 (ref) Age  \< 27 years 37/40 (92.5) 0.42 (0.12--1.49)  ≥ 27 years 248/310 (80.0) 1.0 (ref) Education level  ≤ high school 88/103 (85.4) 0.87 (0.56--1.78)  \> high school 199/249 (79.9) 1.0 (ref) Immigration Status  Born outside of Canada 158/194 (81.4) 1.0 (ref)  Born in Canada 122/149 (81.9) 0.92 (0.51--1.65) First language  English 127/152 (83.6) 1.0 (ref)  French or Other 159/198 (80.3) 0.84 (0.60--1.19) Income[\*](#t002fn001){ref-type="table-fn"}  ≤ \$49,999/year 129/154 (83.8) 0.92 (0.69--1.23)  ≥ \$50,000/year 139/175 (79.4) 1.0 (ref) Stage of Pregnancy  ≤ 26 weeks 122/149 (81.9) 0.89 (0.67--1.23)  ≥ 27 weeks 158/194 (81.4) 1.0 (ref) Parent of other children  No 121/149 (81.2) 0.72 (0.81--1.93)  Yes 162/198 (81.8) 1.0 (ref) \* Dollars reported in Canadian Dollars; OR = Odds Ratio, CI- Confidence Interval, Ref = referent group. The second objective was to evaluate expectant parents' knowledge of signs and symptoms of fever in a neonate. As shown in [Fig 2](#pone.0120959.g002){ref-type="fig"}, we found that most parents were knowledgeable about the signs of illness in the neonate, description of fever, and reasons why fever is dangerous. However, few parents correctly identified the correct rectal temperature (38%) for fever in the neonate. The third objective was to describe the actions Canadian expectant parents would take if they recognized fever in their child. [Table 3](#pone.0120959.t003){ref-type="table"} outlines the steps that participants reported they would take if they suspected a fever. Most commonly, parents said that they would call a medical doctor (n = 211, 59.4%), and/or would go to a medical doctor (n = 206, 58%). Approximately one third of the sample reported that they would give their child acetaminophen (n = 118, 33.2%). Amongst participants who reported they would not seek medical care if they suspected a fever (n = 65), 34% (n = 22) reported that they would "relax", 8% (n = 5) said they do not know what they would do, and 3% said they would rely on personal experience to help their child. 10.1371/journal.pone.0120959.t003 ###### Actions expectant parents reported that they would take if they suspected fever in their neonate (n = 355). {#pone.0120959.t003g} What is your main source of information about fever in babies less than 30 days old? If your baby is less than 30 days old and has a fever, what would you do? Where would you take your baby, less than 30 days old, to seek medical attention when they are ill? -------------------------------------------------------------------------------------- --------------------------------------------------------------------------- ----------------------------------------------------------------------------------------------------- ------- -------------------------------------------------------- ------- **Response Categories** **n** **Response Categories** **n** **Response Categories** **n** book given by MD/nurse 102 relax 22 I take care of baby at home 21 book I bought 46 give baby sponge bath 61 I call Info-Sante[\*](#t003fn001){ref-type="table-fn"} 91 magazine or newspaper 13 acetaminophen 118 Clinic 46 internet 19 call MD 211 My pediatrician or family MD 230 nurse/physician 150 go to MD 206 Hospital emergency room 138 family 65 ask friend for help family 42 Other practitioner healer, acupuncturist, etc) 10 Ask family 393 other 3 \*Info-Santé is a Quebec health information phone hotline; MD = medical doctor. Participants reported accessing varying sources about fever in the neonate. Primarily, expectant parents would seek information from health care professionals, or from books or pamphlets given to them by a nurse or doctor (n = 252, 71%). Participants reported that the Internet (n = 91, 25.6%) and a family member (n = 65, 18.3%) would be other sources of information to learn about how to manage fever in the neonate. Discussion {#sec015} ========== Our study evaluated Canadian expectant parents' knowledge of fever in the neonate. The main finding of this research was that nearly a fifth of expectant parents in the study reported that they would not take their neonate to seek medical care if they suspected a fever. No demographic variables predicted if expectant parents would seek care for a febrile neonate. Though we found that although most parents were able to identify the signs of a fever and why it may be dangerous, few parents were able to correctly identify the temperature at which a neonate is considered febrile. Although neonatal mortality is less common in Canada compared to developing countries, our findings suggest that further investigation is needed to understand the barriers to knowledge about fever in the neonate in this population. Most studies to date have examined this topic in populations of parents living in developing countries, summarizing a common need to enhance education about fever in mothers in antenatal care as well as those discharged from health facilities after delivery \[[@pone.0120959.ref006],[@pone.0120959.ref026],[@pone.0120959.ref027]\]. Despite living in a country that offers free universal health care, our study shows that a significant portion of Canadian expectant parents would not consult a physician if they suspected fever in their neonate. This raises questions about pre and postnatal education about fever (or neonatal health in general) in this country. Understanding reasons for these deviations is critical and could shed light how to optimize health outcomes in young infants. In our study, although immigration status was not a predictor of seeking medical care, we found a significant portion of participants were foreign born immigrants (55%). This was three times greater than the national average of 18% \[[@pone.0120959.ref007]\]. Efforts to develop any pre/postnatal education materials should carefully consider the cultural backgrounds and norms for the varying groups of individuals who receive health services in countries with a diverse immigrant population such as Canada \[[@pone.0120959.ref028]--[@pone.0120959.ref030]\]. This is supported by previous studies on this topic \[[@pone.0120959.ref001],[@pone.0120959.ref002],[@pone.0120959.ref006],[@pone.0120959.ref017]\]. Based on the results of our study, in order to refine current health promotion programs targeted at Canadian expectant parents, future qualitative work is needed to understand the potentially diverse educational needs around this topic \[[@pone.0120959.ref031],[@pone.0120959.ref032]\]. As well, prior studies on this topic have been conducted in countries with high population density and smaller geographical challenges than Canada, and less cultural diversity \[[@pone.0120959.ref001],[@pone.0120959.ref002],[@pone.0120959.ref006]\]. Thus, future qualitative efforts will need to consider the unique geographical challenges inherent in Canadian health care. Finally, we found that most expectant parents reported that they would seek information about fever in the neonate from a medical doctor or a nurse (see [Table 3](#pone.0120959.t003){ref-type="table"}). Even in today's technological era, only 25% of the sample identified that they would rely on the Internet to query how to treat fever in a neonate. Our study suggest that expectant parents feel that the role of the health care professional in educating around this topic is very important. If not already part of practice, health care professionals involved in prenatal care should provide education about this topic in appointments before and after the delivery. Further, collaboration between the professional societies of family physicians, paediatricians, obstetricians, and nursing is essential to promote early, frequent, and accurate education to expectant parents about this information \[[@pone.0120959.ref033]\]. Limitations of Study {#sec016} -------------------- Our study was not without limitations. First, given that the sample came from only one Canadian city, it may not be fully representative of the Canadian population. Specifically, our study lacked younger parents (aged 26 years or less), First Nations (aboriginal peoples of Canada) representation, and parents with a low education. It is hard to know exactly how this would have influenced the results; however, we postulate that a population of younger, less experienced, less educated, marginalized parents may have had less knowledge around fever in the neonate. Second, our study was conducted in a single-center. Although the site chosen was quite ethnically and economically diverse, we recognize that a similar study conducted in other large Canadian urban hospitals may give different results. Third, we used a newly developed survey in this study. Further development and refinement of the survey is needed to justify content validity, reliability, and addition of items to produce a total score. Specifically, we acknowledge that future studies that survey expectant parental knowledge of fever in the neonate should include questions that capture other danger signs of severe illness such as those outlined by the World Health Organization in the Integrated Management of Newborn and Childhood Illness guidelines \[[@pone.0120959.ref003]\]: 1) history of difficulty feeding, 2) movement only when stimulated, 3) temperature below 35.5 degrees Celsius, 5) temperature above 37.5 degrees Celsius, 5) respiratory rate of over 60 breaths per minute, 6) severe chest in drawings, and 7) history of convulsions. In addition, the statement, "If your baby seems unwell, check a rectal temperature" should be used in future assessments of expectant parents' knowledge of fever. Accurate and timely parental assessment of fever in the neonate, and continued heightened awareness around the topic, may optimize health outcomes of neonates globally. Supporting Information {#sec017} ====================== ###### Information for participants page 1. (TIFF) ###### Click here for additional data file. ###### Information for participants page 2. (TIFF) ###### Click here for additional data file. ###### Study survey page 1. (TIFF) ###### Click here for additional data file. ###### Study survey page 2. (TIF) ###### Click here for additional data file. [^1]: **Competing Interests:**The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: SA DM JN. Performed the experiments: SK KLR PB. Analyzed the data: SA SB. Contributed reagents/materials/analysis tools: SB DB. Wrote the paper: SA SB JN DM DB.

Background ========== Chinese hamster ovary (CHO) cells are the most widely used host for large scale production of recombinant therapeutic proteins. A combination of several gene editing approaches applied to Novartis proprietary CHO cell line resulted in a superior cell line with a significant increase of titer and improved product quality. Inter alia we have surprisingly identified a key protease responsible for proteolytic degradation of mainly non-antibody format therapeutic proteins. The recently published CHO genome in combination with screening methods and cell line engineering tools has enabled the development of this novel CHO cell line. Materials and methods ===================== CHO cell lines growing in suspension were cultivated at 36.5 C° and 10% CO2 in shake flasks using a proprietary, chemically defined culture medium. Gene editing was performed according to manufacturer\'s protocols, cells were single cell sorted using FACS device and subsequently screened for the desired phenotype. Cell viabilities and growth rates were monitored using an automated system (ViCell, Beckman Coulter). Cells were stably transfected by electroporation (Amaxa Nucleofection system, Lonza, Germany) with expression plasmids encoding human monoclonal antibodies. This and all other kits were used according to the manufacturer\'s instructions. Spike-in experiments were performed in conditioned medium. Conditioned medium was collected by centrifugation of the cells grown for 7/8 days for 15 min at 90 g. After centrifugation, the supernatant was transferred and passed through a 0.22 μm filter to remove remaining cell particles from the conditioned medium. Under these conditions and at this stage of cell growth, the maximum amounts of secreted proteases are expected to be active in the cell culture medium without release of intracellular proteases due to cell death. The polypeptide of interest was added to the conditioned medium with a final concentration of 0.7 μM and incubated at 37°C with continuous shaking at 500 rpm. After incubation, samples were analyzed by SDS-PAGE followed by Western Blot analysis to determine the amount of clipping. Results ======= Elimination of telomeric region of chromosome 8 ----------------------------------------------- Gene expression profiles of high versus low producing CHO clones were compared to identify genes correlated with increased titer. Improved production rates were surprisingly correlating with loss of the telomeric region of chromosome 8 (figure [1A](#F1){ref-type="fig"}). Based on this finding three new parental CHO cell lines lacking this region were generated and their capability for protein production was assessed. After transfection and two consecutive selection steps (geneticin followed by methotrexate selection) a massive increase of productivity could be detected for all three new cell lines compared to the original cell line (35 fold increase for G418 selection and 7 fold increase for methotrexate selection, respectively) (figure [1B](#F1){ref-type="fig"}). The increased productivity obtained with the new cell lines is facilitating the supply of early drug substance material. In addition, significantly more cell clones with a higher average productivity were obtained after single cell cloning. This results in reduced efforts in single cell sorting and the need to screen fewer clones. {#F1} Identification of major clipping protease and gene knockout using gene editing technologies ------------------------------------------------------------------------------------------- Proteolytic activity in cell cultures is derived from endogenously expressed proteases. However of the major challenge is that more than 700 endogenous proteases are known. We applied different approaches e.g. a variety of protease inhibitors, siRNAs and next generation sequencing approaches to identify the main protease(s) involved in clipping of therapeutic proteins produced in CHOK1 derived cell lines (figure [1C](#F1){ref-type="fig"}).We have surprisingly identified one protease which is mainly responsible for proteolytic degradation. To eliminate the proteolytic activity, the protease gene was knocked out using gene editing technology. To evaluate the effect of the knockout cell line, protein candidates of diverse formats which were degraded in CHO wildtype cells were co-incubated in conditioned medium derived from knockout and wildtype CHO cell lines. No or significant reduced degradation of the proteins was detected using the protease knockout cell line (figure [1D](#F1){ref-type="fig"}). Conclusions =========== The combination of the recently published CHO genome \[[@B1]-[@B3]\]with screening methods and cell line engineering tools has enabled the development of a superior CHO cell line. Gene expression profile analysis of high vs. low antibody producing CHO cell lines revealed that high production is correlated with loss of the telomeric region of chromosome 8. New parental cell lines lacking this telomeric region showed increased productivities after transfection. Additionally the knockout of a surprisingly identified protease eliminates/significantly reduces proteolytic degradation evaluated with a variety of therapeutic proteins. In conclusion novel cell line engineering methods are powerful tools to improve productivity and can solve issues in production of therapeutic proteins in biopharmaceutical industry in a short timeframe with minimal screening effort.

INTRODUCTION ============ Arf-like (Arl) proteins are structurally similar to members of the Arf protein family, which belong to the Ras superfamily of small GTPases and regulate vesicular transport, membrane trafficking, organelle structure, and cytoskeletal remodeling via cyclic regulation between their GTP-bound active form and their GDP-bound inactive form ([@B7]; [@B4]). Like other GTP-binding proteins, the GTP-GDP cycle is regulated by guanine nucleotide exchange factors (GEFs) to exchange GDP for a triphosphate nucleotide and GTP­ase-activating proteins (GAPs) to stimulate GTP hydrolysis. Arl4 proteins (Arl4A, Arl4C, and Arl4D) are distinct from other Arf/Arl proteins due to their unique structures, which include a nuclear localization signal peptide at the carboxy terminus and a long interswitch region between two switch domains ([@B25]). Constituting the subfamily, Arl4A, Arl4C, and Arl4D have different expression patterns during fetal development ([@B33]; [@B11]; [@B22], [@B23]). Our previous studies showed that Arl4C and Arl4D modulate actin cytoskeleton remodeling and cell migration by affecting their interacting partners FLNa and ARNO, respectively ([@B20]; [@B5]). Arl4A shares a 90% amino acid sequence consensus with Arl4C and Arl4D and is mainly expressed in the forebrain--midbrain and midbrain--hindbrain junctions ([@B22]). Recently, Arl4A was found to play a role in actin cytoskeleton rearrangement via a pathway that stimulates ELMO/DOCK180-induced Rac signaling ([@B26]). However, the mechanism underlying how Arl4A affects cell motility remains unclear. Cell migration is a highly regulated event that is initiated by protrusion of the cell membrane ([@B18]). Among the Rho GTPase family, Cdc42 reportedly plays a major role in regulating cell polarity, cell migration, and actin reorganization ([@B8]; [@B29]; [@B30]). The secreted proteins Slits and their transmembrane receptor Roundabout (Robo) are highly expressed in the neuronal system and are crucial for neuronal guidance and cell migration ([@B2]). Recent studies have shown that a pathway mediated by Slit2 and Robo1 also plays important roles in other physiological and pathological processes outside of the nervous system, including the vascular system and tumorigenesis ([@B38]; [@B9]; [@B19]). Many studies have indicated that Slit2/Robo1 signaling regulates multiple types of signaling responses, such as cell proliferation, adhesion, and migration ([@B1]; [@B19]). Interestingly, emerging evidence postulates that Slit2/Robo1 often function in both the promotion and prevention of cell migration in various cell types in the same tissue ([@B32]; [@B28]). Slit2/Robo1 showed an ability to prevent cell migration and promote cell--cell adhesion via E-cadherin and β-catenin in lung cancer and breast cancer cells, respectively, as well as to attenuate Cdc42 activity in epithelial cell lines ([@B28]; [@B40]; [@B36]). By contrast, Slit2/Robo1 also reportedly promotes cell migration by chemokines or by regulating Rho GTPases in breast cancer and epithelial cells ([@B32]; [@B17]). In this study, we first identified Robo1 as a novel effector of Arl4A and found that Robo1 is required for Arl4A-induced cell migration. We showed that decreased cell migration resulting from Robo1 knockdown can be rescued by the expression of wild-type Robo1 but not by a Robo1 mutant deficient in Arl4A binding. Furthermore, the Arl4A-Robo1 interaction promotes Cdc42 activation by decreasing the association of a Cdc42-GAP, srGAP1, with Robo1. The binding of Slit2 to Robo1 decreases the Arl4A-Robo1 interaction and increases the srGAP1-Robo1 association, resulting in decreased Cdc42 activation and prohibited cell migration. Our results demonstrate that Arl4A functions together with Robo1 to modulate Cdc42 activation and cell migration via regulating the Robo1-srGAP1 association. RESULTS ======= Identification of Robo1 as an Arl4A GTPase interaction partner -------------------------------------------------------------- To identify Arl4A interaction proteins, a yeast two-hybrid screen with a human fetal brain cDNA library was performed using Arl4A-QL, a GTP-bound form of Arl4A, as the bait; one Arl4A-interacting protein was identified, the cytoplasmic domain of Robo1 (residues 921--1651). To identify the region within Robo1 required for its interaction with Arl4A, we constructed fragments of the intracellular region of Robo1 as illustrated in [Figure 1A](#F1){ref-type="fig"}, and their interactions with Arl4A were tested using a yeast two-hybrid system. Only the CC2+CC3 domain of Robo1 was found to interact with Arl4A-WT ([Figure 1B](#F1){ref-type="fig"}). To determine the specific region of Robo1 that interacts with Arl4A, the CC2+CC3 domain was divided into CC2 and CC3 fragments. Only Robo1-CC3 was accountable for its interaction with Arl4A ([Figure 1C](#F1){ref-type="fig"}). To further investigate the specific regions within Robo1-CC3 responsible for its interaction with Arl4A, we generated three truncated forms of the Robo1-CC3 domain. Both Robo1-CC3-1 and Robo1-CC3-2 positively interacted with Arl4A in a yeast two-hybrid system, indicating that the Robo1-CC3-1 region (residues 1342--1475) is necessary for the Robo1-Arl4A interaction ([Figure 1D](#F1){ref-type="fig"}). In addition, in vivo coimmunoprecipitation assays also suggested that the interaction between Arl4A and Robo1 is GTP-dependent ([Figure 1E](#F1){ref-type="fig"}). Furthermore, yeast two-hybrid and pull-down assays showed that an interaction was not detected between the Robo1-CC3 region and the other two Arl4 family members, Arl4C and Arl4D ([Figure 1, F and G](#F1){ref-type="fig"}). These results suggest that Arl4A specifically and directly interacts with Robo1 in a GTP-dependent manner. {#F1} Residual Robo1 amino acids 1394--1398 are necessary for the Arl4A-Robo1 interaction ----------------------------------------------------------------------------------- We further narrowed the amino acids responsible for the rlA4A-Robo1 interaction to Robo1 fragments 1342--1475 (Robo1 CC3-1) and 1370--1475 (Robo1-CC3′) using the yeast two-hybrid system (unpublished data). To identify which residues are important for its interaction with Arl4A, we generated Arl4A-binding defective Robo1 mutants by alanine scanning and assayed these mutants using the yeast two-hybrid system. Among these 13 mutants, only the Robo1-A1 (1394--1398 amino acids) mutant was unable to interact with Arl4A-WT in the yeast two-hybrid assay ([Figure 2A](#F2){ref-type="fig"}). As shown in [Figure 2B](#F2){ref-type="fig"}, Arl4A directly interacted with Robo1-CC3-1-WT fused with glutathione *S*-transferase (GST) beads but not with beads containing Robo1-CC3-1-A1, Robo1-CC0+CC1, or GST alone. Notably, Robo1-CC3-1-A2 fused with GST beads did not lose much interaction ability with the purified Arl4A protein, suggesting that residues 1394--1398 of Robo1 play a more critical role in the Arl4A-Robo1 interaction than other regions in this fragment. The interaction was also verified by an in vivo coimmunoprecipitation assay. HeLa cells cotransfected with Arl4A and the Robo1-Flag-WT or Robo1-Flag-A1 mutant were immunoprecipitated using magnetic beads conjugated to an anti-Flag M2 antibody. Similarly, the interaction ability of the Robo1-A1 mutant with Arl4A was decreased significantly compared with that of Robo1-WT ([Figure 2C](#F2){ref-type="fig"}). To elucidate the effects of the Arl4A-binding defective Robo1-A1 mutant on other Robo1 partner proteins, interactions between Robo1-srGAP1 and Robo1-Nck1 were assayed by the yeast two-hybrid system. Like wild-type Robo1, the Robo1-A1 mutant was capable of interacting with both srGAP1 and Nck1, indicating that the interaction defect of Robo1-A1 was specific to Arl4A but not to the other Robo1-interacting proteins srGAP1 and Nck1 (Supplemental Figure 1). {#F2} Arl4A induces Robo1 localization at the plasma membrane ------------------------------------------------------- Several studies have shown that the expression of Robo1 on the cell surface is regulated by factors involved in exocytosis and the endosomal system ([@B16]; [@B24]; [@B27]; [@B13]). We next examined whether the Arl4A-Robo1 interaction corresponds with their localization in HeLa cells by immunofluorescence staining. As previously reported, the typical pattern of Arl4A signals includes Golgi and plasma membrane signals ([@B21]). Robo1, which had a C-terminal myc tag in our experiments, localized primarily in the cytosol in diffuse or vesicle-like punctate forms in HeLa cells ([Figure 3A](#F3){ref-type="fig"}). When Robo1 and Arl4A were coexpressed, Robo1, which appeared to be sequestered in the cytosol when expressed alone, localized partially at the plasma membrane ([Figure 3B](#F3){ref-type="fig"}). To carefully determine the extent of Robo1 potentiation at the plasma membrane upon the expression of Arl4A, we costained HeLa cells with the plasma membrane marker CD44 to clearly demarcate the margins of the cells. With the boundaries of the HeLa cells defined by the CD44 signal, we calculated the plasma membrane-to-cytosol (PM/C) ratio (see *Materials and Methods*) for Robo1 and found that this ratio significantly increased upon Arl4A expression ([Figure 3C](#F3){ref-type="fig"}). In addition, Arl4A expression significantly promoted plasma membrane localization of the Arl4A-binding defective Robo1 A1 mutant ([Figure 3, D and E](#F3){ref-type="fig"}), although the PM/C ratios for Robo1 A1 were slightly lower than those for Robo1 WT. We further monitored the change in localization of Arl4A or Robo1 by knocking down either Robo1 or Arl4A and showed that the cellular localization of Arl4A or Robo1 did not change ([Figure 3, F and G](#F3){ref-type="fig"}). These results suggest that although Arl4A facilitates the plasma membrane localization of Robo1, the Arl4A-Robo1 interaction is not the major factor contributing to the Arl4A-induced localization of Robo1 at the plasma membrane. {#F3} Arl4A-induced cell migration requires interaction with Robo1 ------------------------------------------------------------ Although Arl4A induces cellular protrusion and plays a role in the regulation of actin dynamics ([@B26]), its function in modulating cellular mobility remains to be established. The wound healing assay showed that HeLa cells expressing Arl4A-WT and Arl4A-Q79L had higher migration abilities than those expressing either Arl4A-T34N or Arl4-T51N ([Figure 4, A and B](#F4){ref-type="fig"}). Knocking down Arl4A or Robo1 decreased the motility of HeLa cells (Supplemental Figure 2), although the endogenous Arl4A protein level was extremely low. Importantly, the coexpression of Arl4A and Robo1 had an additive effect on promoting cell mobility, as the increased migration abilities of Arl4A- or Robo1-expressing cells were enhanced by the coexpression of these two proteins in HeLa cells ([Figure 4, C and D](#F4){ref-type="fig"}). The increased migration ability of Arl4A-expressing cells was abolished by Robo-1 knockdown ([Figure 4, E and F](#F4){ref-type="fig"}). Our preliminary screening showed that HEK293T cells possess abundant endogenous Robo1; testing whether Robo1 is required for Arl4A-induced cell migration in a different cell model is thus ideal. Owing to the tendency of HEK293T cells to detach easily from the plate after the wound is made, it therefore compromises the suitability of the wound closure assay for these cells ([@B14]). Transwell migration inserts were used instead to assess the migration ability of HEK293T cells. According to the Trans­well assay results, the overexpression of Arl4A did not restore the reduced motility of Robo1-knockdown cells, supporting a critical role for Robo1 in Arl4A-induced cell migration. This assumption was further supported by the observation that Arl4A and Robo1-WT coexpression but not Arl4A and Robo1-A1 coexpression rescued the cell migration ability of Robo1-knockdown cells ([Figure 4, G and H](#F4){ref-type="fig"}). These results collectively demonstrate that the Arl4A-Robo1 interaction is necessary for Arl4A-induced cell migration. {#F4} The Arl4A-Robo1 interaction promotes cell migration by activating Cdc42 ----------------------------------------------------------------------- Because Cdc42 is reportedly important for regulating cell motility, we examined its role in affecting the migration of HEK293T and HeLa cells expressing Arl4A and Robo1. We tested whether the Arl4A-Robo1 interaction promotes Cdc42 activation using an activity pull-down assay with PAK1-PBD beads. No active Cdc42 was found in mock-transfected HEK293T cells, while only a low level of active Cdc42 was detected in cells expressing exogenous Cdc42. The amount of active Cdc42 increased in cells cotransfected with Cdc42, Arl4A, and Robo1-WT, suggesting that the coexpression of Arl4A and Robo1 induces Cdc42 activation. By contrast, the amount of active Cdc42 decreased when Robo1-WT was replaced with the Arl4A-binding defective Robo1-A1 mutant ([Figure 5A](#F5){ref-type="fig"}). Similar results were also found in HeLa cells (Supplemental Figure 3). Moreover, the level of active Cdc42 decreased when Robo1 was knocked down in Arl4A-expressing HEK293T cells ([Figure 5B](#F5){ref-type="fig"}). These results indicate that the Arl4A-Robo1 interaction is critical for promoting Cdc42 activation. {#F5} Robo1 and Arl4A binding decreases the association of Robo1 with srGAP1 ---------------------------------------------------------------------- Previous studies reported that the association between Robo1 and srGAP1, a Cdc42 GTPase-activating protein, regulates Cdc42 activity ([@B37]). We examined whether srGAP1 also plays a role in Arl4A-induced Cdc42 activation. The Robo1-srGAP1 association was verified by an in vivo coimmunoprecipitation assay in which HEK293T cells were transfected with Arl4A, srGAP1, Robo1-Flag-WT, and Robo1-Flag-A1. A higher level of srGAP1 coimmunoprecipitated with Robo1-A1 than with Robo1-WT ([Figure 6A](#F6){ref-type="fig"}). To further investigate whether Arl4A affects the association between Robo1 and srGAP1, a coimmunoprecipitation assay was performed in HEK293T cells coexpressing srGAP1, Robo1-Flag, and different amounts of Arl4A. The level of immunoprecipitated srGAP1 decreased with increasing Arl4A concentrations in a dose-dependent manner ([Figure 6B](#F6){ref-type="fig"}). Together, the Arl4A-Robo1 interaction promotes Cdc42 activation by down-regulating the Robo1-srGAP1 association. {#F6} Slit2 down-regulates Cdc42 activation and cell motility by affecting the Robo1-srGAP1 and Robo1-Arl4A associations ------------------------------------------------------------------------------------------------------------------ A previous study showed that upon Slit2 stimulation, Robo1 recruits srGAP1 in order to inhibit Cdc42 activity, thus controlling neuronal migration ([@B37]). Slit2, a secreted extracellular matrix protein, binds the Robo1 receptor and controls many morphogenesis and cellular functions, including migration, proliferation, and adhesion ([@B41]). We herein showed that Slit2 treatment reduced cell migration in a dose-dependent manner; however, this treatment did not abolish Arl4A-induced cell migration ([Figure 7, A and B](#F7){ref-type="fig"} and Supplemental Figure 4). To test the activity of Cdc42 under Slit2 treatment, HEK293T cells were transiently transfected with Cdc42, Robo1, and Arl4A. Consistently, Slit2 treatment reduced the amount of active Cdc42 in a dose-dependent manner ([Figure 7C](#F7){ref-type="fig"}). Notably, the quantitative level of active Cdc42 in Arl4A-overexpressing cells treated with the highest amount of Slit2 was reduced to nearly its basal level, but this reduction did not occur in cells not overexpressing Arl4A ([Figure 7C](#F7){ref-type="fig"}), thus indicating that Slit2 fully suppresses Arl4A-induced Cdc42 activation. Next, we examined whether Slit2 modulates cellular migration by regulating the Robo1-srGAP1 and Robo1-Arl4A associations. As shown in [Figure 7D](#F7){ref-type="fig"}, Slit2 treatment significantly reduced the association between Robo-1 and Arl4A in a dose-dependent manner. By contrast, the association between Robo1 and srGAP1 was increased when cells were treated with Slit2 ([Figure 7E](#F7){ref-type="fig"}). These results collectively demonstrate that Slit2 plays a role in modulating the association between Robo1 and its two interacting proteins Arl4A and srGAP1 and thus contributes to the dynamic regulation of cell migration. {#F7} DISCUSSION ========== In this study, we identified Robo1 as a novel effector of Arl4A and demonstrated that the Arl4A-Robo1 interaction modulates cell migration by potentiating Cdc42 activation. Arl4A-induced Cdc42 activation was impaired in Robo1-knockdown cells ([Figure 5B](#F5){ref-type="fig"}). Moreover, the Robo1-A1 mutant, which can no longer interact with Arl4A, failed to activate Cdc42 in either HeLa or HEK293T cells ([Figure 5A](#F5){ref-type="fig"} and Supplemental Figure S3). The Arl4A-Robo1 interaction reduces protein--protein associations between Robo1 and the Cdc42 GAP srGAP1 ([Figure 6](#F6){ref-type="fig"}). We further showed that the binding of Robo1 to the neuronal repulsive factor Slit2 inhibits HEK293T cell motility and Cdc42 activity by decreasing the Arl4A-Robo1 interaction and increasing the Robo1-srGAP1 association ([Figure 7](#F7){ref-type="fig"}). Our previous study showed that Arl4A plays a role in actin cytoskeleton rearrangement via a pathway that stimulates ELMO/DOCK180-induced Rac signaling ([@B26]). Studies have demonstrated that Arl4A and its close relatives Arl4C and Arl4D promote actin restructuring via the recruitment of ARNO, an Arf6 GEF, to the plasma membrane. Interestingly, Arf6 is positioned upstream of Rac activation in various biological processes. Arf6 activation has been proposed to recruit the DOCK180--ELMO complex to the leading edge of a cell to promote lamellipodia formation ([@B31]). However, our previous studies demonstrated that Arl4A-induced cytoskeletal remodeling occurs via an Arf6-independent pathway. This finding may signify that Arl4A can act as a central signaling node for two divergent GTPase pathways. RhoG is known to promote migration and phagocytosis when coexpressed with ELMO ([@B15]). In contrast to the RhoG-ELMO interaction, Arl4A was incapable of synergizing with ELMO/DOCK180 in both cell migration and engulfment assays. These data suggest that Arl4A and RhoG are not located in the same membrane microdomain because their interaction with ELMO led to different biological outputs. Robo1, originally observed in the nervous system, plays a vital role in axon guidance and cell motility during the forebrain development process. Previous studies have shown that Robo1 regulates cell proliferation and motility in both neuronal and nonneuronal cells. Apart from the central nervous system, several studies have shown that Robo1 plays dual roles in cancer cell migration, as it acts as both an oncogene and a suppressor in different types of cancers ([@B2]). Robo1 has been found to either inhibit or promote cell migration depending on its downstream interacting partners ([@B34]; [@B35]; [@B40]). The different directions of Robo1\'s regulation of cell motility have something in common, as both phenomena have been reported to affect the activities of Rho GTPases, such as Cdc42, in several cell types ([@B38]; [@B34]; [@B35]; [@B40]). Our results demonstrated that Arl4A-expressing Robo1-knockdown cells exhibited significantly decreased migration. These data are consistent with those of previous studies showing that Robo1 knockdown by small interfering RNA (siRNA) in human retinal pigmented epithelium cells significantly reduced their adhesion, proliferation, and migration capacities ([@B10]). However, cells coexpressing Arl4A and Robo1 had higher migration abilities than those expressing Arl4A or Robo1 alone. Based on our findings, we postulated that Arl4A might participate in the Robo1-regulated migration pathway, as HeLa and HEK293T cells expressing Arl4A in which Robo1 was knocked down exhibited down-regulated cell motility. The expression of Robo1 on the cell surface is regulated by factors involved in exocytosis and the endosomal system ([@B16]; [@B24]; [@B27]; [@B13]). The Commissureless protein (Comm) in *Drosophila* can bind Robo1 in the Golgi and directly traffic Robo1 to endosomes, thus preventing it from reaching the cell surface ([@B16]; [@B24]). In vertebrates, Robo1 was shown to be predominantly stored in Rab11-positive vesicles, and RabGDI controls axonal midline crossing by up-regulating Robo1 surface expression ([@B27]). Consistent with this report, we observed that the majority of the Robo1 protein appeared to be located in vesicle-like puncta in HeLa cells and that Arl4A can induce Robo1 localization at the plasma membrane. Our previous study showed that Arl4A has a role in the maintenance of the Golgi structure and endosome-to-Golgi transport ([@B21]). ARL4A was partially localized at the plasma membrane and presented a punctate pattern in intracellular areas, where it partially colocalized with the *trans*-Golgi, early endosomes, late endosomes, and recycling endosomes. Although the Arl4A-Robo1 interaction is not required for Arl4A-induced Robo1 localization at the plasma membrane, whether Arl4A-dependent vesicle trafficking plays a role in axonal behavior remains to be investigated. Robo1 is a transmembrane receptor with a Slit2 binding site at its amino terminus to control axonal guidance, cell motility, and active downstream signaling pathways. The intracellular region of Robo1, located at the carboxy terminus, is responsible for its association with interacting proteins, including Nck1, Dock, and srGAP1. These Robo1 interacting partners are involved in Slit2-Robo1 signaling in different ways, which are coordinated with a variety of cellular processes and result in various functions and phenotypes. Slit2 promotes Robo1 to inhibit glioma cell migration and invasion by down-regulating Cdc42 activation ([@B34]; [@B40]). Stimulation of Robo1 by Slit2 leads to the recruitment and activation of srGAP1, consequently inhibiting Cdc42 activity. Conversely, our results showed that Arl4A binding to Robo1 decreased the srGAP1-Robo1 association in vivo following Cdc42 activation. Consistent with most Robo1 interacting proteins, Arl4A also bound to the C-terminal region of Robo1. Although the Arl4A-binding defective Robo1-A1 mutant did not lose its ability to interact with srGAP1 or Nck1 in the yeast two-hybrid assay, this mutant showed a stronger interaction ability with srGAP1 in vivo. Conversely, Arl4A binding to Robo1 reduced the Robo1-srGAP1 association ([Figure 5](#F5){ref-type="fig"}). The Robo1-A1 mutant, which could not interact with Arl4A, failed to rescue cell migration in Robo1-knockdown HEK293T cells expressing Arl4A, suggesting that Arl4A and Robo1 function in collaboration to regulate cell motility outside the nervous system. Our study suggests that the plasma membrane association of GTP-bound Arl4A can induce Robo1 to localize at the plasma membrane, and its binding to Robo1 may lead to a conformational change in Robo1, decreasing the recruitment of srGAP1 to Robo1. In addition, we speculate that an intermediate Robo1 structure associates with an unknown Cdc42 GEF, which is much more susceptible to Cdc42 activation by Arl4A stimulation. Our studies suggest that Arl4A plays a role in regulating the conformation of Robo1, which can also be modulated by other stimulators. Slit2 binds Robo1 and decreases the Arl4A-Robo1 association, leading to the recruitment and activation of srGAP1 ([Figure 8](#F8){ref-type="fig"}). {#F8} Our previous study showed that the expression of mouse Arl4A mRNA is developmentally regulated and is consistent with involvement of the protein in early events of embryogenesis of the central nervous system, somitogenesis, and spermatogenesis ([@B22]). At mouse embryonic day 14.5 (E14.5), Arl4A mRNA is expressed strongly in the cortical plate in an overlapping pattern with that of Robo1 and Slit1/Slit2 mRNA ([@B22]; [@B39]), suggesting possible comodulation of these molecules during the development of the dorsal telencephalon. Given that Slit-Robo signaling plays a crucial role in the axon guidance of developing cortical neurons ([@B3]; [@B6]) and that rsGAP1 is functionally involved in neuronal migration in response to Slit ([@B37]), our current study suggests that Arl4A might functionally participate in axonal pathfinding or neuronal migration via modulation of the Slit-Robo/rsGAP1 signal axis. Because Arl4A binds with Robo1 in a competitive manner with rsGAP1, a possible scenario for the developing migratory neurons is that Arl4A keeps the migratory-brake function of Robo1 inactivated by constitutively binding with Robo1, thus preventing Robo1 from association with the downstream signaling rsGAP1 for execution of migratory inhibition. When migratory neurons navigate through the developing cortex, contact with the guidance cue Slit releases Robo1 from Arl4A\'s binding, allowing for Robo1\'s association with rsGAP1 and subsequent Cdc42 down-regulation concomitant with inhibition of neuronal migration. It is also possible that other environmental cues or intrinsic factors, such as changes in intracellular calcium concentration or neuronal activity, could regulate the expression level or activity of Arl4A to fine-tune the Slit-Robo/rsGAP1/Cdc42 signal axis in the developing migratory neurons. Our preliminary data in cultured nonneuronal cells showed that Arl4A activity was reduced when cells were grown in serum starvation conditions. They suggest that Arl4A activity requires certain extrinsic factors, which remain to be further investigated. The proceeding dissection of possible promigratory or environmental cues will cast light on the regulation of Arl4 activation for cell migration. In conclusion, we demonstrate that the activated GTP-bound form of Arl4A functions as a novel interacting partner of Robo1 to modulate cell motility. The Arl4A-Robo1 interaction regulates cell migration via Cdc42 activation. The cell migration enhanced by the Arl4A-Robo1 interaction is caused by the down-regulation of the srGAP1 association with Robo1. Our work addresses a long-standing knowledge gap in the effects of regulating Robo1 signaling on cell migration by defining a role for dynamic protein interactions governing Cdc42 activation. MATERIALS AND METHODS ===================== Antibodies ---------- Rabbit anti-Robo1 and anti-Arl4A/C/D antibodies were generated as described previously ([@B20]). The following antibodies were used: anti-Myc (1:1000, catalogue no. MMS-150R; Covance, Princeton, NJ), anti-HA (1:1000, catalogue no. SC-7392; Santa Cruz Biotechnology), anti-LexA (1:1000, catalogue no. 5397-1; Clontech, Mountain View, CA), anti-CD44 (1:200 for IF, catalogue no. MA4400; Invitrogen), anti-Cdc42 (1:500, catalogue no. 2462; Cell Signaling, Danvers, MA), anti-srGAP1 (1:1000, catalogue no. 76926; Abcam), and anti-α-tubulin (1:5000, catalogue no. T5168; Sigma-Aldrich, St. Louis, MO). A blue-fluorescent DNA stain 4′,6-diamidino-2-­phenylindole (DAPI) solution was purchased from Millipore (1:5000, catalogue no. S7113). Horseradish peroxidase (HRP)--conjugated goat anti-rabbit and anti-mouse antibodies were purchased from GE Healthcare (1:5000, catalogue nos. NA934V and NA931V, respectively; Waukesha, WI). Alexa Fluor--conjugated secondary antibodies were purchased from Invitrogen (Grand Island, NY). Alexa Fluor 594/488--conjugated anti-rabbit/anti-mouse and Alexa Fluor 647--conjugated anti-rat secondary antibodies were from Invitrogen (1:1000, catalogue nos. A-11012 for Alexa Fluor 594-rabbit, A-11034 for Alexa Fluor 488-rabbit, A-11001 for Alexa Fluor 488-mouse, A-11032 for Alexa Fluor 594-mouse, and A-21247 for Alexa Fluor 647-rat). Cell culture and transfection ----------------------------- HEK293T, COS-7, and HeLa cells were purchased from the American Type Culture Collection (ATCC) and cultured in high-glucose DMEM (HG-DMEM; HyClone, Logan, UT) containing 10% fetal bovine serum (HyClone). All cells were maintained in a humidified incubator supplemented with 5% CO~2~ at 37°C. Cells were transiently transfected with the indicated short hairpin RNA (shRNA), siRNA, and plasmids using the Lipofectamine 2000 transfection reagent (Life Technologies, Grand Island, NY) following the manufacturer\'s procedure. All siRNAs were purchased from Dharmacon (GE Healthcare Life Sciences). The specific siRNA and shRNA used were as follows: Arl4A: GGAACUCAUUGUCACUUUCUU and Robo1: GCAGGUACUUG-GAGGAUAU. After transfection, the transfected cells were harvested at 24 or 48 h for overexpression and knockdown, respectively. Plasmid construction -------------------- cDNA sequences of Arl4A/C/D and its mutants were cloned as previously described ([@B20]). Briefly, Arl4A was cloned into the mammalian expression vector pSG5 (Stratagene, La Jolla, CA) to generate untagged Arl4A. The cDNA sequence was subcloned into pET32a (Novagen, Madison, WI) for His-tagged Arl4A expressed in *Escherichia coli* or pBTM116 for LexA-tagged Arl4A expressed in yeast. Full-length Robo1 and srGAP1 cDNA sequences were amplified from a human fetal brain cDNA library (Clontech) using PCR. Robo1 and srGAP1 were cloned into the mammalian expression vectors pCMV-Tag4A and pcDNA3.1A (Invitrogen), the *E. coli* expression vector pGEX4T-1, or the yeast expression vector pACT2 (Clontech). Replacement of alanine was accomplished using a two-step PCR technique. All constructs were confirmed by DNA sequencing. Total protein extraction and immunoblotting analysis ---------------------------------------------------- Total proteins were extracted as described elsewhere ([@B20]). The proteins in the lysates were separated by 7.5% or 12.5% SDS--PAGE and transferred to Immobilon-P membranes (Millipore). The membranes were blocked in PBST buffer (0.1% Tween 20 in phosphate-buffered saline \[PBS\]) containing 5% skim milk powder or bovine serum albumin (BSA) and then incubated with the indicated primary antibodies. After washing, the membranes were incubated in PBST containing the appropriate HRP-conjugated secondary antibody (1:5000). α-Tubulin was used as the internal control for protein loading. Total RNA isolation, reverse transcription, and quantitative-PCR ---------------------------------------------------------------- Total RNA isolation, reverse transcription, and quantitative-PCR (Q-PCR) were performed as previously described ([@B5]). RNA (2 μg) was reverse transcribed into cDNA and then used for Q-PCR with the TaqMan system (Applied Biosystems \[ABI\], Foster City, CA). The Arl4A primer/probe (Hs01932504-s1; ABI) and the housekeeping gene glyceraldehyde-3-phosphate dehydrogenase (GAPDH, Hs00266705-g1; ABI) were used. Yeast two-hybrid interaction assay ---------------------------------- The yeast two-hybrid screen was performed as previously described ([@B20]). In brief, the *Saccharomyces cerevisiae* strain L40 (MATα trp1 leu2 his3 LYS::\[lexAop\]4-HIS3 URA3::\[lexAop\]8-lacZ) was used. A human fetal brain cDNA library in pACT2 (Clontech) was screened using Arl4A-Q72L as the bait. Colonies exhibiting histidine auxotrophy were patched onto selective plates and assayed for β-galactosidase activity using a colony-lift filter assay. Yeast protein extraction ------------------------ Yeast cells were harvested after 5 d of culture. Yeast protein extraction was performed as described previously ([@B20]), and the extracted proteins were subjected to immunoblotting analysis. Immunofluorescence staining and quantification ---------------------------------------------- To stain Robo1 and Arl4A, HeLa cells were seeded on coverslips and transfected with the indicated plasmids. After transfection, cells were fixed with 4% paraformaldehyde supplemented with 1 mM CaCl~2~ and 1 mM MgCl~2~ for 15 min. Cells were then permeabilized with PBS supplemented with 1 mM CaCl~2~ and 1 mM MgCl~2~ containing 0.1% saponin for 10 min followed by blocking in 0.02% Tween-20 PBST containing 1% BSA for 1 h at room temperature. Cells were incubated with either rabbit anti-Arl4A (1:500) or mouse anti-Myc (1:200) and anti-CD44 (1:200) antibodies. After extensive washing, cells were incubated with the corresponding Alexa Fluor--conjugated secondary antibody (1:1000) for at least 1 h. Cells were then mounted in 90% glycerol in PBS containing 1 mg/ml *p*-phenylenediamine and were visualized using a Carl Zeiss LSM780 system (Carl Zeiss MicroImaging, Jena, Germany). LSM780 confocal microscope analyses of HeLa cells singly expressing or coexpressing Robo-myc and Arl4A were performed with a 63× oil immersion objective. Midplane optical sections were captured for presentation and quantification. ImageJ2 software was used for intensity and PM/C ratio calculations. We determined the border of the plasma membrane by CD44 signals, which clearly outlined the cell boundary. A 10-pixel-wide freehand *Line* was drawn along the boundary demarcated by the CD44 signal as a cell margin denoting the plasma membrane. The recorded coordinates were transferred to the Robo-myc detection channel to obtain the intensity of the plasma membrane signal. The freehand selection tool was used to enclose the cytosolic *Region* along the inner boundary of the CD44 signal inside the cell outline. Similarly, the coordinates were transferred to the Robo-myc detection channel to measure the intensity of the cytosolic signal. The PM/C ratio was calculated using the following equation: where intensity on the line and intensity in the region are defined in the CD44 detection channel but measured in the Robo-myc detection channel. In vitro binding assay ---------------------- His-tagged Arl4A and GST-tagged Robo1 were purified using Ni-NTA resins (Qiagen, Valencia, CA) and glutathione resins (GE Healthcare), respectively. In total, 4 µg of purified His-Arl4A was incubated with 4 µg of various GST-Robo1 fragments in binding assay buffer (20 mM Tris-HCl \[pH 8.0\], 100 mM NaCl, 5 mM MgCl~2~, 1 mM EDTA, 1 mM dithiothreitol \[DTT\], 10% glycerol, 1% Triton X-100, 1 mM NaN~3~, 1× protease inhibitor cocktail) for at least 2 h at 4°C. Unbound proteins were washed away with assay buffer, and proteins bound to GST beads were then eluted with protein loading buffer for immunoblotting analysis. Coimmunoprecipitation --------------------- Immunoprecipitation was performed according to a previously described protocol with slight modifications ([@B5]). Cells were lysed with IP buffer for 30 min on a shaker and cleared by centrifugation at 15,000 × *g* for 10 min at 4°C. The IP buffer for immunoprecipitation of the Robo1-srGAP1 complex comprised 50 mM Tris-HCl (pH 7.5), 50 mM NaCl, 0.1% NP40, 10 mM NaF, 1 mM EDTA, 1 mM DTT, and 1× protease inhibitor cocktail. The IP buffer for immunoprecipitation of the Robo1-Arl4A complex comprised 50 mM Tris-HCl (pH 7.5), 150 mM NaCl, 1% NP40, and 1× protease inhibitor cocktail. The cell supernatants were incubated with 20 µl of anti-Flag M2 magnetic beads (Sigma-Aldrich) for 2 h at 4°C and then extensively washed with IP buffer at least five times. The immunoprecipitated proteins were analyzed by immunoblotting. Preparation and purification of Slit2 ------------------------------------- COS-7 cells were transiently transfected with the empty pSecTag2 vector or the pSecTag 2-human Slit2 plasmid. Purification was performed as previously described ([@B12]). Finally, ∼500 µl of the concentrated Slit2 solution was collected ([@B12]). Concentrated Slit2 was analyzed by Coomassie Blue staining and stored at 4°C for up to 1 wk. In addition, we prepared the empty pSecTag2 vector as the "mock" and used it as the negative control in all the experiments. Cdc42 activity pull-down assay ------------------------------ Cells were transfected with the indicated plasmids and then lysed in Cdc42 activity pull-down assay buffer (50 mM Tris-Cl \[pH 7.5\], 150 mM NaCl, 1 mM DTT, 1% NP-40, 5 mM MgCl~2~, 5% glycerol, and 1× protease inhibitor cocktail). Then, the cells were incubated with purified PAK-PBD-GST beads for 2 h at 4°C. The beads were washed with assay buffer, and bound proteins were eluted with protein loading buffer for immunoblotting analysis. Migration assay --------------- Wound healing migration assays were performed as described previously ([@B5]). Cells were observed using a time-lapse microscope (Axiovert 200M; Carl Zeiss MicroImaging) equipped with a temperature and CO~2~ controller. The cell migration capacity was determined as described previously using ImageJ software ([@B5]). In addition, Transwell cell migration assays were performed using the Boyden chamber assay method as previously described ([@B5]). Cells were serum-starved for 16 h before being seeded into the upper chamber. For Slit2 treatment experiments, the indicated concentrations of purified Slit2 and the vector control were added to the lower chamber. After the assay, the migrated cells were stained with 1% crystal violet and images were captured using a phase-contrast microscope (Eclipse TS-100; Nikon, Melville, NY) equipped with a digital camera (DS-5M; Nikon). Statistical analysis -------------------- All data are presented as the mean ± SD, and *P* values were calculated by one-way analysis of variance (ANOVA) followed by Dunnett\'s post hoc multiple comparison test or two-tailed Student\'s *t* test. Significant statistical differences (\*, *P* \< 0.05; \*\*, *P* \< 0.01; \*\*\*, *P* \< 0.001) are indicated. For each independent in vitro experiment, three biological replicates were used. Supplementary Material ====================== ###### Click here for additional data file. This article was published online ahead of print in MBoC in Press ([http://www.molbiolcell.org/cgi/doi/10.1091/mbc.E18-01-0001](http://www.molbiolcell.org/cgi/doi/10.1091/mbc.E18-09-0550)) on November 14, 2018. We thank Randy Haun, Chia-Jung Yu, Ya-Wen Liu, and Pei-Hsin Huang for their critical review of this article. This work was supported by Grant no. NHRI-EX106-10601B1 from the National Health Research Institutes (NHRI) in Taiwan to F.-J.S.L. Arf : ADP-ribosylation factor Arl : ADP-ribosylation factor--like GDP : guanosine-5'-diphosphate GTP : guanosine-5'-triphosphate [^1]: The authors declare no competing financial interests. [^2]: T.C. and F.S.L. designed the study and interpreted the results; T.C., M.L., and M.T. performed the majority of the experiments and analyzed the data; C.C. and L.J. conducted and supported the biological experiments; T.C., M.L., and F.S.L. wrote and edited the manuscript; F.S.L. provided supervision, funding acquisition, and project administration.

**Funding information** Magliozzi was supported by Italian MS Foundation grant (FISM 16/17/F14). Calabrese and Rossi were supported by the GR‐2013‐02‐355322 grant from Italian Ministry of Health. Introduction {#acn350893-sec-0005} ============ Accumulating neuropathological evidences indicate that multiple sclerosis (MS) is not only characterized by the presence of white matter (WM) demyelinated lesions, readily detectable by conventional magnetic resonance imaging (MRI), but also by focal and diffuse cortical grey matter (GM) demyelination.[1](#acn350893-bib-0001){ref-type="ref"}, [2](#acn350893-bib-0002){ref-type="ref"} Advanced MRI methodology has been developed to visualize GM demyelination,[3](#acn350893-bib-0003){ref-type="ref"}, [4](#acn350893-bib-0004){ref-type="ref"} which likely plays a key role in the accumulation of motor and cognitive disability in MS, especially during the progressive stages of the disease. Compartmentalized inflammation within the meninges of MS patients is spatially related to subpial demyelination, which accounts for about 70% of GM demyelination in progressive MS,[5](#acn350893-bib-0005){ref-type="ref"} and neuronal/glial alterations in the adjacent GM, following a "surface‐in\" gradient[6](#acn350893-bib-0006){ref-type="ref"} from the pial surface toward the WM. In a recent combined ex vivo and in vivo study, compartmentalized inflammation in meningeal infiltrates and cerebrospinal fluid (CSF) was found significantly associated with increased subpial demyelination.[7](#acn350893-bib-0007){ref-type="ref"} In particular, a distinctive CSF pro‐inflammatory pattern, including increased levels of CXCL13, TNF, IFN*γ*, CXCL12, IL6, IL8, and IL10, has been associated with a higher GM lesion load both at the time of diagnosis and at death.[7](#acn350893-bib-0007){ref-type="ref"}, [8](#acn350893-bib-0008){ref-type="ref"} Several lines of evidence support the notion that regional differences in CSF flow might facilitate focal or diffuse intrathecal trapping of immune cells and stasis of inflammatory molecules, which may diffuse through the pial surface toward the inner cortical layers, in turn mediating a detrimental effect in the adjacent GM.[9](#acn350893-bib-0009){ref-type="ref"}, [10](#acn350893-bib-0010){ref-type="ref"} Taken together, these studies suggest a crucial role of CSF/meningeal compartmentalized, intrathecal immune response in the pathogenesis of cortical lesions (CLs). Spurred by the above evidences, using advanced proteomic analysis of CSF combined with advanced MRI techniques, we have attempted to identify more specific molecular pathways involved in early MS‐specific tissue damage at the time of the diagnosis. Methods {#acn350893-sec-0006} ======= Patients {#acn350893-sec-0007} -------- Sixty‐four consecutive treatment‐naïve relapsing‐remitting MS patients from the MS Centre of Verona University Hospital (Italy) were enrolled at diagnosis between March 2015 and February 2016. All the MS patients had a confirmed diagnosis of MS according to Revised McDonald criteria and underwent neurological evaluation, including Expanded Disability Status Scale (EDSS)[11](#acn350893-bib-0011){ref-type="ref"} assessment, a 3T MRI, and a CSF examination at the time of diagnosis (Table [1](#acn350893-tbl-0001){ref-type="table"}). MS patients were stratified in two groups according to the presence of low (n° 28 MSlow, \<10 CLs) or high (n° 36 MShigh,\>10 CLs) levels of CL number (median value: 10 CLs, range: 0--34) detected using 3D Double Inversion Recovery (DIR) sequences. In addition, we performed CSF analysis in a control group of 26 patients (age, mean, SD:46.2 ± 10.1; 16 female/10 male) with other neurological disorders, including 12 subjects with noninflammatory neurological diseases (NIND) and 14 subjects with other inflammatory neurological diseases (OIND) (Table [S1](#acn350893-sup-0001){ref-type="supplementary-material"}). ###### Demographic, clinical, MRI and CSF (protein concentration) details of the two examined MS populations with low and high degree of cortical demyelination. MS damage Low (28) High (36) ----------------------------------------- -------------------------- ------------------------- Age studied (mean ± SD) 41.6 ± 11.9 39.7 ± 13.8 Gender (f:m) 16:12 23:13 EDSS at recruitment (median; range) 2 (1.0--4.0) 2 (0.0--5.0) OCBs positive/negative 16/12 34/2 IgG index (mean ± SD) 0.66 ± 0.2 0.92 ± 0.46 CLs volume (mm^3^‐ range) 132 ± 186 (0--575) 1620 ± 728 (365--3910) Intracortical CLs volume (mm^3^‐ range) 62 ± 109 (0--364) 757 ± 512 (142--2311) CLs number (mean ± SD) 1.2 ± 1.7 (0--5) 15.7 ± 6.7 (10--34) Intracortical CLs number (mean ± SD) 0.53 ± 1.03 (0--4) 6.6 ± 4.01 (2--16) CTh (mm; range) 2.78 ± 0.24 (2.39--3.17) 2.8 ± 0.27 (2.15--3.23) T2WMLV (cm^3^; range) 6.6 ± 4.2 (2.4--23.3) 5.4 ± 2.2 (1.4--11.7) Brain volume (cm^3^; mean ± SD) 1513.3 ± 119.7 1506.4 ± 163.7 GM Volume (cm^3^; mean ± SD) 496.8 ± 46.8 494.6 ± 65.3 WM Volume (cm^3^; mean ± SD) 415.1 ± 38.7 402.8 ± 57.6 MIG/CXCL9 (pg/mL; mean ± SD) 24.28 ± 16.21 47.11 ± 45.87 CXCL10/IP10 (pg/mL; mean ± SD) 253.04 ± 264.54 475.84 ± 369.24 SDF1*αβ*/CXCL12 (pg/mL; mean ± SD) 1205.47 ± 842.97 2639.02 ± 1375.13 CXCL13/BCA1 (pg/mL; mean ± SD) 3.11 ± 4.78 29.13 ± 2.968 6Ckine/CCL21 (pg/mL; mean ± SD) 619.08 ± 527.00 1085.29 ± 697.26 GM‐CSF (pg/mL; mean ± SD) 30.40 ± 20.05 59.32 ± 64.78 MIF (pg/mL; mean ± SD) 1534.75 ± 1808.30 2624.57 ± 2805.85 TNF (pg/mL; mean ± SD) 14.23 ± 12.26 43.70 ± 35.40 sTNF‐R1 (pg/mL; mean ± SD) 4357.82 ± 1835.17 11860.49 ± 18986.05 sTNF‐R2 (pg/mL; mean ± SD) 1642.13 ± 902.52 2301.40 ± 1559.97 TWEAK/TNFSF12 (pg/mL; mean ± SD) 1506.11 ± 1158.68 3157.59 ± 2386.68 APRIL/TNFSF13 (pg/mL; mean ± SD) 25564.85 ± 18266.06 59598.08 ± 44038.57 BAFF/TNFSF13B (pg/mL; mean ± SD) 5314.47 ± 2135.23 10001.39 ± 6901.69 LIGHT/TNFSF14 (pg/mL; mean ± SD) 528.62 ± 395.08 1174.72 ± 1001.65 IFN‐G (pg/mL; mean ± SD) 2.51 ± 1.76 8.93 ± 7.63 INF‐alfa 2\* (pg/mL; mean ± SD) 16.13 ± 15.86 11.59 ± 8.74 INF‐beta (pg/mL; mean ± SD) 57.37 ± 27.13 68.17 ± 39.66 IFNlambda2\* (pg/mL; mean ± SD) 1304.70 ± 1475.29 528.94 ± 655.79 IL‐6 (pg/mL; mean ± SD) 7.50 ± 7.53 11.78 ± 8.07 IL‐8 (pg/mL; mean ± SD) 17.11 ± 8.57 31.66 ± 21.41 IL‐10 (pg/mL; mean ± SD) 4.07 ± 1.67 7.99 ± 3.76 IL‐16 (pg/mL; mean ± SD) 35.11 ± 15.64 64.57 ± 53.41 IL‐12(p40) (pg/mL; mean ± SD) 3.29 ± 6.47 4.24 ± 5.37 IL‐12(p70) (pg/mL; mean ± SD) 0.80 ± 1.00 1.01 ± 1.45 MMP‐2 (pg/mL; mean ± SD) 192.83 ± 111.78 715.47 ± 317.47 Osteopontin (pg/mL; mean ± SD) 114611.26 ± 163433.93 151589.51 ± 157961.25 Pentraxin ‐3 (pg/mL; mean ± SD) 417.62 ± 592.90 818.47 ± 799.79 Chitinase 3 ‐like 1 (pg/mL; mean ± SD) 31930.36 ± 41535.25 59571.12 ± 63352.69 sCD163 (pg/mL; mean ± SD) 35199.65 ± 9514.72 65939.36 ± 13129.70 Abbreviations: EDSS, Expanded Disability Status Scale; OCB, oligoclonal bands; IgG index, immunoglobulin‐G index; CL, cortical lesion; CTh, cortical thickness; T2WMLV, T2 white matter lesion volume; MIG/CXCL9, monokine induced by gamma interferon or chemokine (C‐X‐C motif) ligand 9; CXCL10/IP10, C‐X‐C motif chemokine 10 or interferon gamma‐induced protein 10; SDF1αβ/CXCL12, stromal cell‐derived factor or C‐X‐C motif chemokine 12; CXCL13/BCA, chemokine (C‐X‐C motif) ligand 13 or B lymphocyte chemoattractant; 6Ckine/CCL2, Chemokine (C‐C motif) ligand 21; GM‐CSF, Granulocyte‐macrophage colony‐stimulating factor ; MIF, Macrophage migration inhibitor factor; TNF, tumor necrosis factor; sTNF‐R1, soluble tumour necrosis factor‐receptor 1; sTNF‐R2, soluble tumour necrosis factor‐receptor 2; TWEAK/TNFSF12, TNF‐related weak inducer of apoptosis or tumour necrosis factor ligand superfamily member 12; APRIL/TNFSF13, A proliferation‐inducing ligand, or tumour necrosis factor ligand superfamily member 13; BAFF/TNFSF13B, B cell activating factor or tumour necrosis factor ligand superfamily member 13B; LIGHT/TNFSF14, tumor necrosis factor ligand superfamily member 14 or tumour necrosis factor superfamily member 14; IFN‐G, interferon gamma; IFN‐alfa2, interferon alfa2; IFN‐lambda2, interferon lambda2; IFN‐beta, interferon beta; IL‐6, interleukin 6; IL‐8/CXCL8, interleukin‐8 or (C‐X‐C motif) chemokine ligand 8; IL‐10, interleukin10; IL‐16, interleukin16; IL‐12(p40), interleukin‐12 subunit p40; IL‐12(p70), interleukin‐12 subunit p40; MMP2, matrix metallopeptidase 2; sCD163, soluble‐CD163 (cluster of differentiation 163). John Wiley & Sons, Ltd MRI acquisition protocol and analysis {#acn350893-sec-0008} ------------------------------------- In each MS patient, MRI was performed at least 2 months after the last relapse using a Philips Achieva 3T MRI Scanner and the following image sets were acquired: (1) 3D T1‐weighted Turbo Field Echo (TFE); (2) 3D Double Inversion Recovery (DIR); (3) 3D Fluid‐Attenuated Inversion Recovery (FLAIR). Optimized parameters of each sequence were set as previously reported.[7](#acn350893-bib-0007){ref-type="ref"} WM and GM lesion number and load and estimation of cortical thickness were assessed by consensus of experienced observers, as previously described.[7](#acn350893-bib-0007){ref-type="ref"}, [12](#acn350893-bib-0012){ref-type="ref"} More specifically, details on MRI analysis are reported in Supplementary Materials (Data [S1](#acn350893-sup-0005){ref-type="supplementary-material"}). CSF analysis {#acn350893-sec-0009} ------------ ### Immunoassay protein analysis {#acn350893-sec-0010} Cerebrospinal fluid samples of MS patients were obtained at least 2 months after the last relapse and within 1 week of the MRI (ethical approval n° 35315), according to Consensus Guidelines for CSF and Blood Biobanking.[13](#acn350893-bib-0013){ref-type="ref"} The IgG index and the presence/absence of oligoclonal bands (OCB) for each MS patient are reported in Table [1](#acn350893-tbl-0001){ref-type="table"}. The levels of 69 inflammatory mediators (Table [1](#acn350893-tbl-0001){ref-type="table"}) were assessed using a combination of immune‐assay multiplex techniques (Bio‐Plex X200 System, BioRad, Hercules, CA) as previously optimized[7](#acn350893-bib-0007){ref-type="ref"}, [8](#acn350893-bib-0008){ref-type="ref"} and described in Data [S1](#acn350893-sup-0005){ref-type="supplementary-material"}. The levels of neurofilament light chain (NF‐L) in CSF were measured using Human NF‐light enzyme‐linked immunosorbent assays (ELISA) kit (MyBioSource, San Diego, CA) according to procedures previously optimized.[7](#acn350893-bib-0007){ref-type="ref"}, [8](#acn350893-bib-0008){ref-type="ref"} The levels of CSF sCD14 (Quantikine Human sCD14 Immunoassay, R&D Systems), haptoglobin (Quantikine Human Haptoglobin, R&D Systems), free‐hemoglobin (Hb) (Abcam ab157707), and fibrinogen total antigen (\#MBS135523, MyBiosource) were measured in duplicate by ELISA assays according to manufacturer\'s instructions (Table [2](#acn350893-tbl-0002){ref-type="table"}). ### Proteomic analysis {#acn350893-sec-0011} Six CSF samples obtained from three MS patients representative of the MSlow group and three MS patients representative of the MShigh patients, according to the immunoassay (BioPlex) CSF profiling as described above (Table [S2](#acn350893-sup-0002){ref-type="supplementary-material"}), were selected and analyzed using TRIDENT analysis[14](#acn350893-bib-0014){ref-type="ref"} with denaturation by three different protocols as previously described[14](#acn350893-bib-0014){ref-type="ref"} and fully explained in Data [S1](#acn350893-sup-0005){ref-type="supplementary-material"}. The obtained protein datasets for each experiment were analyzed by Database for Annotation, Visualization and Integrated Discovery (DAVID, v6.8) software (<https://david.ncifcrf.gov>), and gene ontology analyses and protein--protein interactions using STRING software as described in detail in Data [S1](#acn350893-sup-0005){ref-type="supplementary-material"}. Statistics {#acn350893-sec-0012} ---------- Nonparametric Mann--Whitney *U* tests were used to test differences in MRI, EDSS, and proteomic data between MS and control groups or between MSlow and MShigh groups. Pearson correlation coefficients were calculated to analyze the strength of correlation between clinical, MRI, and CSF proteomic data. False discovery rate (FDR), with a significance level of 0.05, was adopted to correct for multiple testing problem. Statistical analysis was performed using GraphPad PRISM‐8‐Software. Results {#acn350893-sec-0013} ======= MS stratification according to cortical lesion load {#acn350893-sec-0014} --------------------------------------------------- Upon stratification of MS population according to median value of CL number, the MShigh group (56%, mean CL number = 15.7 ± 6.7; CL volume = 1620 ± 728 mm^3^) was characterized by about 12‐fold higher numbers and volume of CLs compared to MSlow group (44%, mean CL number = 1.2 ± 1.7; CL volume = 132 ± 186 mm^3^) (Table [1](#acn350893-tbl-0001){ref-type="table"}). Similar 12‐fold increase was found when type I leukocortical lesions were excluded (Table [1](#acn350893-tbl-0001){ref-type="table"}). No significant differences were found between the two MS groups regarding age (mean MSlow = 41.6 years; mean MShigh = 39.7 years) and disease duration (mean MSlow = 5.6 years; mean MShigh = 5.6 years). As expected, the female/male ratio showed higher number of females in both groups (MSlow = 16:12; MShigh = 23:13). The median EDSS was 2.0 in both MSlow group (range = 1.0--4.0) and MShigh group (range = 0.0--5.0), (Table [1](#acn350893-tbl-0001){ref-type="table"}). No significant differences were found regarding cortical thickness, T2WMLV, brain volume, GM volume, and WM volume between the two MS cohorts (Table [1](#acn350893-tbl-0001){ref-type="table"}). CSF immune‐assay analysis {#acn350893-sec-0015} ------------------------- By examining the presence and levels of 69 inflammatory mediators in the CSF of all the examined MS patients versus controls, 29 proteins were found significantly (*P* \< 0.01) elevated in MS compared to controls (Table [1](#acn350893-tbl-0001){ref-type="table"}). When comparing the two subgroups of MS patients, the levels of 27 of the 29 molecules were found significantly higher (*P* \< 0.01) in the CSF of MShigh compared to MSlow group, whereas only interferon α 2 (IFN*α*2) and INF*λ*2 were significantly higher in MSlow than MShigh (*P* \< 0.01) (asterisks in Table [1](#acn350893-tbl-0001){ref-type="table"}). CSF proteomic analysis {#acn350893-sec-0016} ---------------------- Using TRIDENT approach and mass spectrometry analyses of CSF obtained from three MSlow and three MShigh patients, representative of the two groups identified as above described, a total of 227 proteins were identified to be differently expressed in the CSF of the two examined groups of patients (Table [S3](#acn350893-sup-0003){ref-type="supplementary-material"}). Functional classification analysis (DAVID) identified 10 key pathways differentiating the MShigh and MSlow patients (Table [2](#acn350893-tbl-0002){ref-type="table"}; Table [S4](#acn350893-sup-0004){ref-type="supplementary-material"}): complement activation and its regulation (30% of total identified 227 proteins); receptor‐mediated endocytosis (25%); innate immune response (14%); positive regulation of B‐cell activation and B‐cell receptor signaling pathways (7%); cellular iron ion homeostasis (6%); Fc‐gamma receptor signaling pathway involved in phagocytosis (5%); negative regulation of blood coagulation (4%); fibrinolysis (3%); acute phase response (3%); inflammatory response (3%), (Table [2](#acn350893-tbl-0002){ref-type="table"}). By applying protein--protein interaction network to further investigate the connection between the 227 total protein identified, 201 nodes and 338 edges were identified in the network. STRING predictive network analysis depicted strong connections among proteins involved in complement and coagulation cascade, cellular iron homeostasis, and protein binding (Fig. [1](#acn350893-fig-0001){ref-type="fig"}, Table [S4](#acn350893-sup-0004){ref-type="supplementary-material"}), including, in particular, CD14, Hb, haptoglobin, and fibrinogen. ###### List of the 10 main protein pathways differentially expressed in MShigh respect to MSlow patients. ---------------------------------------------------------------------------------------- \(1\) Complement activation, regulation of complement activation (30%)^\*^ \(2\) Receptor‐mediated endocytosis (25%) \(3\) Innate immune response (14%)^\*^ \(4\) Positive regulation of B cell activation/B cell receptor signaling pathways (7%) \(5\) Cellular iron ion homeostasis (6%)^\*^ \(6\) Fc‐gamma receptor signaling pathway involved in phagocytosis (5%)^\*^ \(7\) Phagocytosis, recognition (4%)^\*^ \(8\) Fibrinolysis (3%)^\*^ \(9\) Acute‐phase response (3%) \(10\) Negative regulation of blood coagulation (3%)^\*^ ---------------------------------------------------------------------------------------- The asterisks indicate pathways characterizing by common proteins. John Wiley & Sons, Ltd {#acn350893-fig-0001} Differential expression of macrophage CSF biomarkers correlates with cortical lesion load and neurodegeneration {#acn350893-sec-0017} --------------------------------------------------------------------------------------------------------------- In order to validate the nonquantitative proteomic analysis on all the 64 examined MS patients and 26 controls, using more quantitative assays, ELISA, we found increased levels of soluble CD14 protein (sCD14) in the CSF of MSlow compared to MShigh patients (*P* \< 0.01). Conversely, increased sCD163 was found in the CSF of MShigh in comparison with MSlow patients (*P* \< 0.0001; Fig. [2](#acn350893-fig-0002){ref-type="fig"}). Free‐Hb levels were significantly higher in the CSF of both MSlow (*P* \< 0.05) and MShigh (*P* \< 0.01), compared to controls. Haptoglobin levels in the CSF were significantly elevated in the MShigh group (*P* \< 0.001), compared to the MSlow and control groups; lastly, CSF fibrinogen levels were significantly higher in MSlow (*P* \< 0.05) and MShigh (*P* \< 0.001) compared to controls, and in MShigh compared to MSlow (*P* \< 0.01) (Fig. [2](#acn350893-fig-0002){ref-type="fig"}). {#acn350893-fig-0002} Differential CSF levels of coagulation cascade and iron homeostasis molecules correlate with cortical lesion load {#acn350893-sec-0018} ----------------------------------------------------------------------------------------------------------------- CSF levels of sCD163 positively correlated with CL volumes (*r* = 0.54; *P* \< 0.0001) and numbers (*r* = 0.54; *P* \< 0.0001), while an opposite correlation was observed for sCD14 (CL volume: *r* = −0.33; *P* = 0.035; and CL numbers: *r* = −0.34; *P* = 0.030) (Fig. [3](#acn350893-fig-0003){ref-type="fig"}). Interestingly, NF‐L correlated positively with sCD163 (*r* = 0.53; *P* \< 0.001) and negatively with sCD14 (*r* = −0.51; *P* \< 0.001) (Fig. [3](#acn350893-fig-0003){ref-type="fig"}). Furthermore, sCD163 correlated positively with several inflammatory mediators of the MShigh CSF profile, including matrix metalloproteinase 2 (MMP2) (*r* = 0.52; *P* \< 0.001), IL10 (*r* = 0.47, *P* \< 0.001), CXCL13 (*r* = 0.42; *P* \< 0.001), and CXCL12 (*r* = 0.42, *P* \< 0.001) (Fig. [3](#acn350893-fig-0003){ref-type="fig"}). In addition, CSF levels of sCD163 positively correlated with CSF levels of free‐Hb (*r* = 0.44; *P* \< 0.001), haptoglobin (*r* = 0.63; *P* \< 0.001), and fibrinogen (*r* = 0.492; *P* \< 0.01) (Figs. [3](#acn350893-fig-0003){ref-type="fig"} and [4](#acn350893-fig-0004){ref-type="fig"}). Interestingly, the two cohorts of examined MS patients were grouped separately, as reflected by the two color clusters in Figure [4](#acn350893-fig-0004){ref-type="fig"}. {#acn350893-fig-0003} {#acn350893-fig-0004} B‐cell and macrophage CSF correlations {#acn350893-sec-0019} -------------------------------------- Cerebrospinal fluid protein levels of sCD163 positively correlated (*P* \< 0.001) with the levels of BAFF (*r* = 0.4), IL10 (*r* = 0.5), CXCL13 (*r* = 0.5), CXCL12 (*r* = 0.5), and TNF (*r* = 0.5) (Figs. [3](#acn350893-fig-0003){ref-type="fig"} and [4](#acn350893-fig-0004){ref-type="fig"}). In contrast, CSF levels of sCD14 only correlated positively (*r* = 0.50; *P* = 0.0006) with high protein levels of IFN*λ*2 (Figs. [3](#acn350893-fig-0003){ref-type="fig"} and [4](#acn350893-fig-0004){ref-type="fig"}). Albumin CSF/serum ratio correlates with CSF protein and fibrinogen concentration {#acn350893-sec-0020} -------------------------------------------------------------------------------- Using CSF/serum albumin ratio as one of the possible indicator of BBB permeability alteration, no significant differences have been detected between MShigh and MSlow patients at the time of diagnosis. Conversely, CSF/serum albumin ratio significantly correlated only with CSF protein concentration (*r* = 0.59; *P* \< 0.01) and with CSF fibrinogen levels (*r* = 0.522; *P* \< 0.01) (Fig. [5](#acn350893-fig-0005){ref-type="fig"}), but not with the other examined biomarkers. {#acn350893-fig-0005} Discussion {#acn350893-sec-0021} ========== The present study, besides confirming our previous findings of a strong association between severe cortical pathology and a distinctive CSF inflammatory profile[7](#acn350893-bib-0007){ref-type="ref"} in an independent MS subgroup, enabled us to detect previously unexpected or unknown candidate molecules involved in altered CSF profiles and associated cortical pathology since early disease stages. In particular, we identified, as strictly associated with severe cortical pathology, a group of molecules involved in the uptake and metabolism of iron, including free‐Hb, haptoglobin, and sCD163. Previous studies have indicated disturbed iron metabolism as a potential amplification factor for demyelination and neurodegeneration in MS patients, particularly in those with progressive disease.[15](#acn350893-bib-0015){ref-type="ref"} Iron accumulates with aging in human brain mainly in oligodendrocytes and myelin and may be liberated into the extracellular space in the course of MS‐related demyelination, potentially amplifying oxidative injury.[15](#acn350893-bib-0015){ref-type="ref"} Our current study additionally suggests iron‐containing free‐Hb as an another important pathogenetic factor,we found significantly increased CSF free‐Hb concentrations particularly in MS patients with high CL load. So far, few studies have investigated the role of Hb in MS pathogenesis. In a recent study of a well‐characterized cohort of 140 secondary progressive MS (SPMS) patients,[16](#acn350893-bib-0016){ref-type="ref"} the rate of brain atrophy significantly correlated with increased serum concentration of alpha‐Hb and beta‐Hb together with enhanced serum lactate dehydrogenase activity in SPMS patients. Our results provide robust evidence for accumulation of free‐Hb in the CSF of MS patients, particularly in those patients with severe cortical pathology. Free‐Hb can be oxidized to methemoglobin (containing ferric iron), ferryl heme intermediate (Fe4+), hemichromes, and free heme or iron, and can cause oxidative injury to lipids, nucleic acids, and proteins.[17](#acn350893-bib-0017){ref-type="ref"} Uncomplexed heme is instable and either decomposes or is enzymatically oxidized to biliverdin, which is rapidly metabolized to bilirubin, ferrous iron (Fe2+), and carbon monoxide (CO).[18](#acn350893-bib-0018){ref-type="ref"} Thus, free‐Hb may directly lead to oxidative damage of oligodendrocytes, maltose‐binding protein (MBP), and myelin lipids, and involve in the formation of globin radicals and heme transfer.[19](#acn350893-bib-0019){ref-type="ref"} Free‐Hb also influences endothelial tight junction proteins, such as zonula occludens‐1 (ZO‐1) and claudin‐5, leading to BBB dysfunction due to loosening of tight junctions,[20](#acn350893-bib-0020){ref-type="ref"} which may lead to the entry of leukocytes into the cortical parenchyma. For all these reasons, free‐Hb in the CSF of MS patients may be deleterious by increasing oxidative injury in underlying cortex. Our study has shown a parallel increase in CSF levels of haptoglobin and soluble CD163 (sCD163) molecules involved in transport and macrophage uptake of free‐Hb. Haptoglobin binds Hb with very high affinity, thus representing an efficient buffering mechanism for free‐Hb liberated from damaged erythrocytes.[21](#acn350893-bib-0021){ref-type="ref"} Binding of Hb to haptoglobin and formation of haptoglobin‐Hb complexes are essential for cellular uptake of Hb via CD163.[21](#acn350893-bib-0021){ref-type="ref"} CD163 is the haptoglobin‐Hb complex receptor, expressed by monocytes, macrophages, and dendritic cells, which mediates endocytosis of haptoglobin‐Hb complexes in a Ca^2+^‐dependent manner.[22](#acn350893-bib-0022){ref-type="ref"} The soluble form of CD163 (sCD163), which we detected in the CSF of MS patients with high CL load, may be formed via ectodomain shedding, that is, proteolytic cleavage of the extracellular domain from macrophage membrane‐bound CD163.[23](#acn350893-bib-0023){ref-type="ref"} Shedding of CD163 may be induced by inflammatory stimuli such as IL10, IL6, glucocorticoids,[24](#acn350893-bib-0024){ref-type="ref"}, [25](#acn350893-bib-0025){ref-type="ref"} and by inflammation‐responsive protease ADAM17, which also cleaves pro‐TNF to bioactive TNF.[23](#acn350893-bib-0023){ref-type="ref"} sCD163 has been proposed as a potential predictor of MS activity[26](#acn350893-bib-0026){ref-type="ref"} as it was found expressed on macrophages and microglia in MS plaques,[27](#acn350893-bib-0027){ref-type="ref"} and increased sCD163 levels have been shown in patients with different stages of MS both in serum[28](#acn350893-bib-0028){ref-type="ref"} and CSF.[26](#acn350893-bib-0026){ref-type="ref"} As CD163 shedding likely reduces the detoxification capacity of macrophages for free‐Hb from the extracellular space,[23](#acn350893-bib-0023){ref-type="ref"} we suppose that the presence of sCD163 might be related to reduced Hb detoxification via macrophages and, as a consequence, increased toxic free‐Hb in CSF. On the contrary, we found elevated CSF levels of sCD14, an innate immune receptor expressed on monocyte, in MS patients with low CL load. The dissimilar sCD14/sCD163 levels detected in the CSF of patients with different degree of GM damage at the time of diagnosis suggest that an altered balance of monocyte profile and activation that may possibly be linked with different amounts of GM injury. To date, discordant data are available on the correlation between blood CD14 expression and disease activity.[25](#acn350893-bib-0025){ref-type="ref"}, [29](#acn350893-bib-0029){ref-type="ref"} In particular, an inverse correlation was found between serum levels of sCD14 and disease activity, showing that CD14 may be a good indicator of stable MS.[30](#acn350893-bib-0030){ref-type="ref"} Thus, a more comprehensive analysis of the CD14/CD163 CSF balance at the time of diagnosis is mandatory as it may indicate a new potential tool to better evaluate the intrathecal stage of macrophage activation to predict and monitor MS evolution. The correlation found between CSF levels of sCD163 and NF‐L may reflect the activity of CD163^+^ cells in destruction or phagocytosis of neurons and axons.[31](#acn350893-bib-0031){ref-type="ref"} This finding suggests that combined analysis of CSF sCD163 and NF‐L levels at diagnosis may represent a possible early indicator of CL activity. In addition, we found a pronounced increase of fibrinogen, normally absent in healthy brain, in the CSF of MS patients at early stage of the disease with high GM lesion load. Fibrinogen deposition was previously found associated with blood brain barrier disruption, neuroinflammation, and neurodegeneration in MS and several other neurological conditions.[32](#acn350893-bib-0032){ref-type="ref"}, [33](#acn350893-bib-0033){ref-type="ref"} In addition, fibrinogen was found associated with reduced neuronal density in progressive MS CLs[33](#acn350893-bib-0033){ref-type="ref"} suggesting that it can have a key role in disability accumulation, in particular, in the progressive phase of the disease. Fibrinogen activates the bone morphogenetic protein (BMP) signaling pathway in OPCs, thus suppressing remyelination.[34](#acn350893-bib-0034){ref-type="ref"} Fibrinogen was recently detected at the edge of chronic active but not inactive lesions of MS cases[35](#acn350893-bib-0035){ref-type="ref"} and was demonstrated to stimulate a unique transcriptional signature in CD11b+/CD18+ microglia that activate a cascade of pro‐inflammatory events, including the recruitment and central nervous system (CNS) activation of myelin antigen‐specific Th1 cells and reactive oxygen species (ROS), TNF and IL1*β* release.[32](#acn350893-bib-0032){ref-type="ref"}, [34](#acn350893-bib-0034){ref-type="ref"}, [36](#acn350893-bib-0036){ref-type="ref"}, [37](#acn350893-bib-0037){ref-type="ref"} Chronification of these pathological events may contribute not only to tissue damage and neurodegeneration, but also to inhibition of potential repair mechanisms. Targeting the coagulation pathway might therefore have therapeutic potential.[38](#acn350893-bib-0038){ref-type="ref"} It still remains unclear how the whole fibrinogen molecule or some of the protein subunits may diffuse in CSF and brain tissue. The positive correlation that we found between albumin CSF/serum ratio and CSF levels of fibrinogen at the time of diagnosis suggests that blood/CSF/brain barrier alterations and/or BBB leakage may somehow influence intrathecal accumulation of molecules, potentially linked to periphery circulation, such as fibrinogen. On the contrary, the fact that similar correlations have not been detected between inflammatory mediators detected in the CSF and albumin CSF/serum ratio may suggest that CSF inflammatory markers are mainly associated with intrathecal inflammation, mediated by both CNS cells or intracerebral compartmentalized inflammation, as demonstrated by the strong correlation with CSF levels of sCD163. The significant correlation between fibrinogen and sCD163 supports the hypothesis that the CSF levels of the BBB disruption marker fibrinogen may also reflect to the degree of monocyte/macrophage activation and possibly to the presence of active pathological processes. In particular, it has been shown that the binding of the 377--395 amino acid portion of the *γ* chain fibrinogen molecule with the CD11b/CD18 receptor induces the activation of numerous intracellular pathways, including the activation of microglia with the production of a pro‐inflammatory milieu (cytokines and chemokines), able to recruit within the CNS monocytes and myelin antigen‐specific Th1 cells, and mediate axonal and neuronal damage[36](#acn350893-bib-0036){ref-type="ref"}. The combined assessment of CSF levels of fibrinogen and sCD163 may, therefore, represent a potential tool to early investigate the activity stage of the lesions. Studies on experimental autoimmune encephalomyelitis models suggested that upon BBB disruption, fibrinogen together with prothrombin may enter the CNS and local activation of thrombin may induce fibrin deposition since early stage of the disease[39](#acn350893-bib-0039){ref-type="ref"} (Davalos et al. Ann Neurol 2014). Therefore, complement and coagulation pathway possibly represents one of the earliest signs of inflammatory activity in lesions and its CSF examination may help to detect early pathological CNS injuries and neuroinflammation. This study corroborates our previous suggestions that high CL load is associated with profound CSF inflammatory changes. In particular, CSF abundance of B cell‐linked inflammatory mediators, such as CXCL13, CXCL12, CXCL10, BAFF, IL6, IL10, GM‐CSF, and TNF, supports the hypothesis of a prominent role of B cells in the pathogenesis of CLs.[5](#acn350893-bib-0005){ref-type="ref"} These inflammatory mediators as well as one or more cyto‐ and/or myelinotoxic soluble CSF factors could then diffuse through the cortex and mediate a "surface‐inward" pattern of subpial cortical damage.[5](#acn350893-bib-0005){ref-type="ref"}, [7](#acn350893-bib-0007){ref-type="ref"}, [8](#acn350893-bib-0008){ref-type="ref"}, [9](#acn350893-bib-0009){ref-type="ref"}, [40](#acn350893-bib-0040){ref-type="ref"} A number of in vitro studies further support this hypothesis, showing that cultured neurons exposed to CSF of MS patients, but not of controls, undergo oxidative stress and axonal damage and that ceramides (C16:0 and C24:0) enriched in MS CSF patients may represent possible mediators of this injury.[41](#acn350893-bib-0041){ref-type="ref"}, [42](#acn350893-bib-0042){ref-type="ref"} In other works, mixed CNS glial cells exposed to supernatants of B cells isolated from MS patients but not from controls exhibited increased death of both oligodendrocytes[43](#acn350893-bib-0043){ref-type="ref"} and neurons.[44](#acn350893-bib-0044){ref-type="ref"} Moreover, monoclonal recombinant antibodies derived from expanded B‐cell clones isolated from CSF of MS patients induced demyelination and astrocyte activation on spinal cord explants.[45](#acn350893-bib-0045){ref-type="ref"} In addition, the CSF correlations found between B cell‐related molecules and macrophage markers may imply complex immunity interactions evocative of immunologic reactions occurring in the subcapsular sinus of lymph nodes or in the marginal zone of the spleen. These data suggest that CSF and meningeal infiltrating B cells might intrathecally interact with macrophages recognizing, internalizing, and retaining antigens through a variety of receptors including toll like receptors, C‐type lectin receptors but also scavenger receptors such as CD163.[46](#acn350893-bib-0046){ref-type="ref"} It remains to be clarified whether potential link between B‐cell immunity and complement‐coagulation cascade may indicate mechanisms of cortical damage  directly mediated by complement and/or opsonization . This work may have some limitations: the proteomic analysis was performed only on a representative subset of MS patients. Considering that the applied mass spectrometry analysis was not quantitative, we performed the data validation with quantitative techniques (ELISA or Luminex experiments) involving a larger number of patients. We suggest for the first time that intrathecal altered balance of complement/coagulation cascade, iron uptake, and innate response, in combination with dysregulated B‐cell immunity, may have a direct role in cortical damage since earliest MS stages and, possibly, in disease progression. Combined evaluation of CSF sCD14/sCD163 balance and of Hb/haptoglobin/fibrinogen levels, as well as of B‐cell activity biomarkers at the time of diagnosis may therefore represent a useful tool that, together with clinical and MRI assessment, may help to predict specific disease immunophenotypes. Study approval {#acn350893-sec-0022} -------------- The local Ethic Committee approved the study (Protocol number 35315, 31/07/2014). Written informed consent was obtained from each patient. Author Contributions {#acn350893-sec-0024} ==================== RM and MC contributed to the conception and design of the study; RM, SH, FF, DM, SR, and MC and MC contributed to the acquisition and analysis of data; RM, SH, DM, SM, HL, and MC contributed to interpretation of the results; RM, SH, DM, FF, SR, VM, FL, AV, RN, RR, SM, HL, and MC contributed to drafting and revision of the text. Conflict of Interest {#acn350893-sec-0025} ==================== The author(s) declare(s) that there is no conflict of interest. Supporting information ====================== ###### **Table S1.** Clinical parameters at the diagnosis of the control group. ###### Click here for additional data file. ###### **Table S2.** Clinical and MRI parameters at the diagnosis of the MS patients examined for CSF TRIDENT proteomic analysis. ###### Click here for additional data file. ###### **Table S3.** List of proteins and respective accession numbers detected by TRIDENT methodology followed by LC--MS/MS analysis. ###### Click here for additional data file. ###### **Table S4.** List of the 10 main pathways as key players differentiating the MShigh and MSlow patients ###### Click here for additional data file. ###### **Data S1.** Supplementary materials and methods. ###### Click here for additional data file. We thank the technical support by C. Senatore and M. Cordella at the Complex Protein Mixture (CPM) Facility at Istituto Superiore di Sanità, Rome, Italy. Magliozzi was supported by Italian MS Foundation grant (FISM 16/17/F14). Calabrese and Rossi were supported by the GR‐2013‐02‐355322 grant from Italian Ministry of Health.

Introduction ============ Peatlands are important to the global carbon cycle as they are large repositories of atmospheric carbon, with estimates ranging between 270 and 455 Pg (10^15^) carbon, despite only covering 3% of the Earth's surface (Gorham, [@B32]; Turunen et al., [@B77]). This is due to a relative imbalance between rates of net primary production by plants and microbial decomposition (Clymo, [@B18]), resulting from persistently water saturated environments and consequent anoxia (Moore and Basiliko, [@B52]). Such conditions also favor microbial CH~4~ production, thus peatlands are also significant sources of atmospheric CH~4~ (Moore et al., [@B55]; Laine et al., [@B46]; Roulet et al., [@B65]), which has a global warming potential 23 times higher than CO~2~ under the IPCC's 100-year timeframe (Forster et al., [@B26]). The James Bay Lowlands (JBL) are a large peatland-complex that form the southeast part of the Hudson Bay Lowlands, Canada, the second largest peatland area in the world (325,000 km^2^) after the Western Siberian Lowlands (Riley, [@B62]; Gorham, [@B32]). Few studies have been conducted on carbon biogeochemistry in the JBL (Roulet, [@B64]; Yu et al., [@B86]), which represents a massive atmospheric carbon store characterized by a mosaic of peatland types that were formed under varying environmental conditions and may differ in their sensitivity to environmental change (Martini, [@B50]). The JBL therefore represents both an important frontier in peatland ecology and a significant factor for the atmospheric carbon balance. Concern has been raised over the stability of the JBL carbon store as climate models predict an increase in mean annual air temperature of 3--4°C by 2020 and 5--10°C by 2050, with only minor increases in precipitation of 10--20% in this region by the end of the twenty-first century (Hengeveld, [@B35]; Gagnon and Gough, [@B29]). Warmer temperatures are expected to increase evapotranspiration resulting in lower water table levels (Gorham, [@B32]; Roulet et al., [@B66]). Short-term water table level fluctuations may increase decomposition as the presence of bimolecular oxygen (O~2~) activates phenol oxidases responsible for the degradation of phenolic compounds (Sinsabaugh, [@B70]). This change can release hydrolase enzymes, responsible for labile carbon degradation, from inhibition and has been referred to as the "enzymic latch theory" (Freeman et al., [@B28], [@B27]). Continued lowering of the water table height may lead to a complete shift in plant community composition (Weltzin et al., [@B83]; Laiho, [@B45]; Robroek et al., [@B63]), potentially altering litter decomposability (Dorrepaal et al., [@B21]), heterotrophic respiration (Moore and Basiliko, [@B52]), and increased CH~4~ production (Hines et al., [@B37]). Many studies have demonstrated that temperature and hydrology-related redox state directly control carbon mineralization in peatlands (e.g., Moore and Dalva, [@B53]; Updegraff et al., [@B78]; Keller et al., [@B41]). However, microorganisms ultimately control peat decomposition and evidence indicates that the soil microbial community composition is important to ecosystem processes (Balser and Firestone, [@B8]; Reed and Martiny, [@B61]); but its effects on carbon mineralization are unresolved (Waldrop and Firestone, [@B81]; Bardgett et al., [@B9]; McGuire and Treseder, [@B51]). Soil pH has been identified as one of the most influential factors controlling microbial diversity and community composition across soils types (Fierer and Jackson, [@B24]; Allison and Treseder, [@B2]; Rousk et al., [@B67]). However, peatland microbial diversity has not been extensively categorized (Kraigher et al., [@B44]) but differences in microbial community structure have been identified between peatland types (i.e., bogs and fens; Jaatinen et al., [@B40]; Ausec et al., [@B6]; Peltoniemi et al., [@B59]) as well as among depths down the peat profile (Morales et al., [@B56]; Jaatinen et al., [@B40]), potentially due to changes in pH. Differences in microbial community composition among peatlands should be expected as vegetation varies from poorly decomposable mosses and shrubs in ombrotrophic bogs that only receive nutrients from dry and wet deposition (Aerts et al., [@B1]) to more easily decomposable sedges, herbs, and shrubs in minerotrophic fens that receive nutrients from groundwater (Bragazza et al., [@B15]). Although ecosystem models assume that microbial communities are functionally redundant (Reed and Martiny, [@B61]) differences in microbial activity have been observed among peatland habitats and depths (Fisk et al., [@B25]). Moreover, plant litter decomposition rates have been shown to vary depending on both the structure of the microbial community and the past resource inputs a community has experienced (Strickland et al., [@B73]). Given the contrasting botanical composition between bog and fen peatlands, microbial communities may preferentially utilize their "native" organic compounds to a greater extent than microbes that have been exposed to peat with a different chemical composition. In this study we used multiple approaches to characterize depth-dependent microbial community structure and function in two bogs and a fen within the JBL. This included molecular fingerprinting techniques, two independent estimates of microbial activity and *in vitro* assays of microbial responses to changes in aeration and temperature. We hypothesized that (i) pH is the dominant control of microbial community composition and activity among sites and depths, with more diverse communities observed at higher pH, (ii) substrate utilization patterns of carbon compounds will differ between the bogs and the fen, with the fen preferentially mineralizing labile compounds and bogs mineralizing more chemically complex compounds, (iii) measures of microbial activity and community composition correlate with carbon mineralization, and (iv) higher peat temperature and aeration results in greater rates of microbial activity, with the largest response occurring in surface peat. Materials and Methods ===================== Sample collection and preparation --------------------------------- The climate of the JBL is described as being humid microthermal-subarctic (Dcf), according to the Köppen Climate Classification System (Christopherson and Byrne, [@B17]). Annual total precipitation ranges from 610 to 660 mm and mean temperatures range from −29 to −15°C in January and 12 to 18°C in July between 50° and 52°N (Martini, [@B50]). The region is essentially roadless, thus it is difficult to access and required the use of helicopters. Three sampling locations were chosen based on their nutrient status and distance from one another. Kinoje (51°35′N 82°12′W) and Victor (52°42′N 85°49′W) Bogs are separated by 206 km and have similar characteristics. Both are ombrotrophic, characterized by *Sphagnum* mosses and lichen (\>50%), ericaceous shrubs (primarily members of the genera *Vaccinium* and *Kalmia*; \>50% cover), and stunted black spruce (*Picea mariana*; ∼10% cover). Victor Fen is located near Victor Bog (∼1.1 km), but is minerotrophic and dominated by *Carex* spp. (\>50%), with patches of *Sphagnum* and brown mosses (*Drepanocladus*, *Scorpidium*, *Aulacomnium*; \<50% cover), ericaceous shrubs (\<25% cover), and larch (*Larix laricina*; ∼10%). Due to logistical constraints, we were able to collect one peat core (Russian pear corer) per site in the summer of 2009 to a depth of 2 m. All cores were collected from hollows and immediately frozen at −18°C at a nearby research station prior to transport back to Toronto. Subsamples from each core were taken at 0--10 cm (Surface, above the water table), 50--60 cm (Middle, in a zone of fluctuating water table), and 100--110 cm (Deep, always below the water table). Surface samples from both bog cores were weakly decomposed and classified as H3 using the von Post system of humification (Rydin and Jeglum, [@B68]), while the fen surface was classed as H4. All middle and deep samples were highly decomposed and categorized as H8 and H9, respectively. To preserve the *in situ* microbial community from being altered by thaw--refreeze cycles, each sub-section was cut vertically in half while still frozen. The microbial community composition and activity was assessed on one of the sections while the second vertical section was subjected to a temperature and aeration manipulation experiment (below). When required for analysis the peat was thawed at 21°C and gently homogenized to reduce the variation within the sub-sample. Peat moisture content was determined gravimetrically by drying 1 g of wet peat at 105°C for 24 h and peat pH was measured in reverse osmosis purified water after 1 h (soil-to-water ratio of 1:4). Profiling the microbial community --------------------------------- Archaeal, bacterial, and fungal community structure within the peat samples was profiled using terminal restriction fragment length polymorphism analysis (T-RFLP). DNA was extracted from each peat sample using the MoBio PowerSoil DNA Isolation Kit (MoBio Laboratories Inc., Carlsbad, CA, USA) according to the manufacturer's instructions. To minimize DNA extraction bias (e.g., Feinstein et al., [@B23]), three successive extractions were pooled per peat sample. Archaeal rDNA genes were amplified from total DNA using an unlabeled forward primer Ar109 (5′-ACK GCT CAG TAA CAC GT-3′) and the fluorescently labeled reverse primer Ar912r (5′-\[6FAM\]CTC CCC CGC CAA TTC CTT TA-3′; Lueders and Friedrich, [@B47]). Bacterial rDNA genes were amplified using a fluorescently labeled forward primer Eu27f (5′ \[6FAM\]AGA GTT TGA TCM TGG CTC AG-3′) and an unlabeled reverse primer Eu1492r (5′-ACG GYT ACC TTG TTA CGA CTT-3′; Bräuer et al., [@B16]). Fungal rDNA genes were amplified with an unlabeled forward primer Fu-817f (5′-TTA GCA TGG AAT AAT RRA ATA GGA-3′) and a fluorescently labeled reverse primer Fu-1536r (5′-\[6FAM\]ATT AGC AAT GCY CTA TCC CCA-3′; Edel-Hermann et al., [@B22]). The reaction mixture (50 μl) consisted of 20 ng soil DNA, 36.5 μl H~2~O, 5 μl 10× buffer, 4 μl MgCl~2~, 1.25 μl dNTPs, 2 μl primers (forward and reverse), and 0.25 μl Taq (Sigma-Aldrich, USA). DNA was amplified by PCR using a Primus 96 plus thermal cycler (MWG Biotech, Germany) with the following reaction conditions: 5 min 94°C, followed by 30 cycles of 94°C for 1 min, annealing temperature 1 min, 72°C for 2 min and a final extension at 72°C for 10 min. The annealing temperature for the bacterial primers was 50 and 52°C for archaeal and fungal primers. For each sample, the PCR product of three reactions was pooled to minimize amplification bias. The PCR products were purified in order to remove residual primers using a GenElute PCR clean-up kit (Sigma-Aldrich, USA) according to the manufacturer's instructions and quantified using a Nanodrop spectrophotometer (Thermo Scientific, USA). Archaeal T-RFLP analysis followed the method outlined by Lueders and Friedrich ([@B47]). Aliquots of purified archaeal amplicons (50 ng) were digested using *Taq*I (3 U) and 1 μl of appropriate buffer supplied by the manufacturer (New England BioLabs, USA) and 1 μl of bovine serum albumin were combined to a total volume of 10 μl and digested at 65°C for 2 h. Bacterial T-RFLP analysis followed the method outlined by Lukow et al. ([@B49]). Aliquots of purified bacterial amplicons (50 ng) were digested with the restriction endonuclease *Msp*I (10 U) and 1 μl of appropriate buffer supplied by the manufacturer to a total volume of 10 μl and digested at 37°C for 3 h. Fungal T-RFLP analysis followed the method outlined by Edel-Hermann et al. ([@B22]). Aliquots of purified fungal amplicons (50 ng) were double digested with *Alu*I and *Mbo*I (5 U) and 1 μl of the appropriate buffer supplied by the manufacturer to a total volume of 10 μl and digested at 37°C for 3 h. Terminal restriction fragments were measured on an ABI3730 DNA Analyzer (Applied Biosystems, Foster City, CA, USA) with GeneScan1000-ROX™ fragment size standards (Applied Biosystems) at the University of Guelph Lab Services (ON, Canada). Terminal restriction fragment (T-RFs) lengths between 50 and 1200 bp were determined relative to the standards. T-RFs differing by less that 2 bp were considered to be the same fragment and their data combined and only T-RFs that represented ≥ 1% of the total sample were included for statistical analysis. The abundance (A) of each T-RF was calculated as: A = n i / N × 10 , 000 where *n*~i~ is the peak height of one T-RF, N is the sum of all peak heights in that sample, and multiplied by 10,000 so the data could be analyzed using traditional ecological methods. Peak heights were used in preference to peak area due to overlapping peaks (Blackwood et al., [@B12]) and this method is more accurate at establishing relative gene abundances (Lueders and Friedrich, [@B48]). Microbial biomass carbon (MB-C) and nitrogen (MB-N) was determined using slight modifications to the chloroform (CHCl~3~) fumigation extraction method by Vance et al. ([@B79]). Three replicate 10 g (wet weight) samples of peat from the sub-sections were each split into two 5 g samples One sample was placed inside a vacuum desiccator and fumigated with ethanol-free CHCl~3~ in the dark for 24 h. Due to the high moisture content of the peat soil, 0.5 ml of CHCl~3~ was added directly to the surface to aid cell lysis (Ocio and Brookes, [@B57]) and a beaker of 1 N NaOH was placed inside the desiccator to remove any excess CO~2~. CHCl~3~ vapor and residue was removed through repeated evacuation. All peat samples were sealed in containers with 40 ml of 0.5 M K~2~SO~4~, shaken for 1 h at 200 rpm on an oscillating shaker, and filtered through 0.45 μm glass fiber filters. The dissolved organic carbon (DOC) and N (DON) concentration of the peat soil extracts was measured with a TOC/TN analyzer (IL550, Lachat Instruments, USA). MB-C and MB-N was calculated by subtracting the non-fumigated extractable DOC/DON from the fumigated extractable DOC/DON after correcting for differences in initial peat moisture content. Estimating the activity of the microbial community -------------------------------------------------- Microbial metabolic potential within each JBL sample was estimated by measuring the mineralization rates of four natural and six synthetic substrates of varying chemical complexity and molecular weight. Briefly, synthetic substrates were selected based on their prevalence in natural plant tissues and ranged from simple, low molecular weight compounds to large chemically complex substances. These substrates included: alkali lignin and methylcellulose, potentially representative of major, complex, high carbon concentration, structural compounds in many plants and *p*-coumaric acid (a lignin derivative from non-woody plant tissues), sodium benzoate (an aromatic organic acid and potential lignin derivative), glucose (a cellulose derivative), and a commercially available mixture of amino acids (protein derivatives, Sigma-Aldrich catalog \# R7131). The pH of the synthetic solutions was adjusted to that of reverse osmosis water to avoid a pH effect. Dissolved organic matter extracts from four plant types (two sedge and two *Sphagnum*) were made by autoclaving chopped plant material in reverse osmosis water and filtered through a 0.45-μm glass fiber filer. Each substrate was adjusted to a standard concentration of 1 mg carbon ml^−1^. Substrate induced respiration (SIR) assays were conducted by adding 10 ml of substrate (10 mg carbon) to 5 g of peat (wet mass) in 30 ml serum vials each capped with a rubber stopper and crimped. Reverse osmosis water acted as the control (basal microbial activity). Vials were vigorously shaken by hand for 30 s to ensure mixing and incubated under oxic conditions for 48 h and anoxic conditions for 72 h. Anoxic conditions were created through repeated evacuation with a vacuum pump and back-flushing with nitrogen (N~2~) gas and brought to atmospheric pressure. Gas samples were taken using a 1-ml syringe following mixing of the gas in the vial and CO~2~ concentrations were measured using an infrared gas analyzer (Qubit Systems, Kingston, ON, Canada). CO~2~ production was monitored at five times periods (0, 4, 16, 24, 48 h) for the oxic assay and six time periods (additional measurement at 72 h) for the anoxic assay and was calculated and expressed as total mass of CO~2~ produced (using the ideal gas law) per gram of dry peat. Preliminary analysis revealed that CO~2~ production rates increased exponentially after 16 (oxic) and 48 (anoxic) h. Thus to represent the initial responses of *in situ* microbial community to substrate addition, comparisons of CO~2~ production were made after 16 and 48 h for oxic and anoxic conditions, respectively. SIR data are expressed as the ratios of CO~2~ production divided by basal (control) respiration, allowing inter-site comparisons. Potential extracellular enzyme activities (EEA) were determined following the method outlined by Saiya-Cork et al. ([@B69]). Briefly, the activity of 1,4-β-glucosidase, cellobiohydrolase, 1,4,-β-*N*-acetyl-glucosaminidase, phosphatase, and sulfatase were determined fluorometrically using 200 μM methylumbelliferone (MUB)-linked substrates. Peroxidase and phenol oxidase activities were determined with a colorimetric assay using 25 mM [l]{.smallcaps}-3,4,-dihydroxy-phenylalanine ([l]{.smallcaps}-DOPA) as a substrate; 10 μl of 0.3% H~2~O~2~ was added to the peroxidase assay. A soil slurry was created by combining 0.5 g of peat (wet mass) to 125 ml of acetate buffer (50 mM, pH 5) and homogenized for 60 s with a hand blender. A preliminary study showed that estimated enzyme activity was not influenced by the mass of peat used (max. 2.5 g). For each enzyme assay, 200 μl of soil slurry and 50 μl of specific substrate were loaded onto a 96-well microplate with eight analytical replicates. Each plate contained a positive and negative control. All assays were incubated at 21°C for 2 h, except for phosphatase and 1,4,-β-*N*-acetyl-glucosaminidase assays which were incubated for 0.5 h. To stop the reaction in the fluorometric assays 10 μl of 0.5 N NaOH was added at the end of the incubation period. Fluorescence (excitation energy 365 nm, emission 460 nm) and absorbance (460 nm) was measured using a multi-plate spectrophotometer (Synergy HT, Biotek, USA). Temperature and aeration control of microbial activity ------------------------------------------------------ To test how future environmental change might influence microbial activity, peat microcosms for each sub-sample were established by adding 5 g of homogenized wet peat to 20 ml serum vials. Each vial was capped with a rubber stopper and crimped. Vials were randomly allocated to one of four treatments with two temperatures and either oxic or anoxic conditions: 4°C oxic, 4°C anoxic, 14°C oxic, and 14°C anoxic. Each treatment had three replicates. Anoxic conditions were established by repeated evacuation and back-flushing with N~2~ gas and brought to atmospheric pressure. Gas samples were taken at 90, 144, and 270 h and transferred to 10 ml pre-evacuated glass 10 ml crimped glass vials with butyl rubber stoppers (Geo-Microbial Technologies Inc., Ochelata, OK, USA), where they were later measured for CO~2~ and CH~4~ concentration using a SRI 8610-0040 gas chromatograph with a FID and methanizer (SRI Instruments, Torrance, CA, USA). As the enzyme assays required destructive sampling (0.5 g was removed), one replicate per treatment was analyzed for each enzyme activity per time period. The mass of the remaining peat was recorded and the treatment conditions re-established. The fractional rate of increase in microbial activity in the 14°C over 4°C temperature (the Q~10~) was calculated for each site and depth. Statistics ---------- Enzyme activity and SIR ratios were analyzed with non-metric multidimensional scaling (NMDS) ordination using Bray--Curtis distance, while T-RF proportional abundance data (square-root transformed) was analyzed with Sørensen's distance. All the NMDS ordinations were run using a random starting configuration, a maximum of 250 iterations and an instability criterion of 0.00001; two dimensions were used for all plots and Monte Carlo tests (1,000 randomized runs) was used to determine significance. The Shannon--Wiener index (Hill et al., [@B36]) was used to estimate T-RF diversity. Analysis of variance (ANOVA) with Tukey's *post hoc* test was used to determine differences in basal respiration. Classification and regression tree analysis (CART) was conducted to partition the relative effects of peat sample and the experimental variables on CO~2~ and CH~4~ production in the microcosm experiment. Relationships between CO~2~ and CH~4~ concentration were determined using Pearson's correlation analysis. Exploratory analysis of the measured variables was analyzed in two steps, as there are more variables than peat samples, following a similar method to Hudson et al. ([@B39]). Firstly, variables that correlated with basal CO~2~ respiration, pH, enzyme activity, and SIR data to a probability level of \< 0.1 were identified. Then, backward stepwise linear multiple regression was used to identify the most important variables. The residuals met the parametric assumptions of the model. All statistics were performed following tests for normality and heteroscedasticity. Data were natural log transformed when parametric assumptions were not met and pH data were log~10~ transformed prior to analysis. All statistical analyses were performed using R v. 2.13. (R Development Core Team, [@B60]) with the package vegan for NMDS ordination (Oksanen et al., [@B58]) and mvpart for CART analysis (De'ath, [@B20]). Results ======= Peat pH ------- Victor Fen surface peat had the highest pH of 5.8 and there was no difference in pH between middle and deep samples (pH 5.0). Kinoje and Victor bogs were more acidic and had the same pH value of 3.4 at surface and similar pH values of 3.2 and 3.4 were observed in the middle peat samples and 3.8 and 3.5 in the deep peat, respectively. Microbial community characterization ------------------------------------ NMDS ordination showed that the bacterial T-RFs among the peat samples appeared to form a single cluster (Figure [1](#F1){ref-type="fig"}A), as many abundant T-RFs were found among the peat samples (Table [A1](#TA1){ref-type="table"} in Appendix). However, bacterial richness was greatest in Kinoje and Victor bogs and many of the same taxa were present in the two peatlands although more unique T-RFs were found at Victor Bog primarily at 0--10 cm and 50--60 cm (Table [A1](#TA1){ref-type="table"} in Appendix). NMDS ordination of the archaeal T-RFs showed clustering among the peatland types (Figure [1](#F1){ref-type="fig"}B), however many of the same T-RFs were identified at each site and clustering appears to be driven by a few rare T-RFs (Table [A1](#TA1){ref-type="table"} in Appendix). There was not enough dissimilarity among the fungal T-RFs to separate by peat sample with NMDS ordination. {#F1} Kinoje Bog MB-C at 0--10 cm was approximately six times greater than at both Victor Bog and Fen (Table [1](#T1){ref-type="table"}). MB-C generally decreased with depth except at Victor Bog 50--60 cm, which had a biomass 1.7 times greater than the surface. The carbon:nitrogen ratio typically decreased from the surface to deep samples, but the ratio was similar among all depth samples at Victor Bog. ###### **Microbial biomass C and N within peat soil at three depths from cores collected from three James Bay peatlands (*n* = 1)**. Peatland Sample depth (cm) MB-C (μg g^−1^) MB-N (μg g^−1^) C:N ------------ ------------------- ----------------- ----------------- ------ Kinoje Bog 0--10 1463 75.5 19.7 50--60 211 14.9 14.7 100--110 61 6.5 10.7 Victor Bog 0--10 254 9.9 26.8 50--60 431 22.6 19.5 100--110 92 4.6 22.8 Victor Fen 0--10 210 17.7 11.8 50--60 9 5.9 13.5 100--110 80 14.8 5.5 Microbial activity ------------------ Aerobic basal CO~2~ respiration was significantly higher in the Victor Fen surface and middle depth peat samples compared with the deep fen peat and peat from all depths at both Kinoje and Victor Bogs (*F*~4,18~ = 84.0, *p* \< 0.001; Figure [2](#F2){ref-type="fig"}). There was little difference in CO~2~ production among the bog peat samples, except for Victor Bog middle depth where CO~2~ production was approximately two times greater than surface and deep peat. Patterns of anaerobic basal respiration were similar to aerobic basal respiration and were strongly correlated (data not shown). {#F2} The overall magnitude of substrate utilization in the oxic SIR assay generally increased with depth as indicated by the mean ratio, except for the Victor Bog middle depth sample (Table [2](#T2){ref-type="table"}). Deep peat had the highest SIR ratios on average and the NMDS ordination of the oxic SIR ratios revealed that most of the deep peat samples clustered together (Figure [3](#F3){ref-type="fig"}A). Kinoje Bog and Victor Bog had similar responses to substrate addition in both magnitude and relative ability to mineralize different substrates, while Victor Fen formed a separate cluster on the NMDS ordination, particularly as the surface and middle samples had higher average ratios than the two bogs. In contrast, substrate utilization in the anoxic SIR assay was greatest in the surface samples, except at Victor Bog where the middle peat sample was highest on average. NMDS ordination of the anoxic SIR ratios showed little clustering among the peat depth samples within the sites, but Kinoje and Victor Bog appear to have a similar response. There was little clustering among the Victor Fen samples (Figure [3](#F3){ref-type="fig"}B). ###### **Substrate induced respiration ratios (substrate/basal respiration; mean ± 1 SE) under oxic (16 h) and anoxic conditions (48 h) of peat soil at three depths from peat cores collected from the James Bay Lowlands**. Study site Depth (cm) Synthetic substrates ---------------- ---------------- ------------------------ -------------------- ------------------------- ------------------------- ---------------- ------------- **OXIC** Kinoje Bog 0--10 2.08 (0.24) 1.08 (0.10) 0.89 (0.06) 1.25 (0.12) 0.92 (0.09) 0.66 (0.05) 50--60 2.03 (0.08) 2.06 (0.13) 0.99 (0.08) 1.27 (0.09) 0.94 (0.04) 0.86 (0.06) 100--110 1.98 (0.25) 1.49 (0.17) 1.04 (0.12) 0.94 (0.03) 0.84 (0.09) 0.99 (0.07) Victor Bog 0--10 1.98 (0.20) 1.67 (0.12) 1.66 (0.02) 1.13 (0.07) 0.78 (0.01) 0.93 (0.11) 50--60 1.24 (0.12) 1.28 (0.10) 0.72 (0.05) 0.78 (0.07) 0.66 (0.06) 0.62 (0.04) 100--110 3.48 (0.04) 2.37 (0.41) 1.69 (0.17) 1.74 (0.14) 1.48 (0.13) 0.97 (0.03) Victor Fen 0--10 1.62 (0.17) 2.28 (0.13) 3.27 (0.25) 0.88 (0.04) 1.47 (0.13) 2.06 (0.08) 50--60 1.96 (0.08) 2.16 (0.14) 2.47 (0.06) 1.09 (0.12) 1.23 (0.14) 0.78 (0.09) 100--110 1.85 (0.06) 2.71 (0.15) 1.49 (0.12) 1.02 (0.08) 0.96 (0.05) 0.81 (0.06) **ANOXIC** Kinoje Bog 0--10 2.04 (0.74) 2.65 (0.24) 2.48 (0.92) 1.99 (0.46) 1.40 (0.19) 0.65 (0.09) 50--60 1.30 (0.33) 1.36 (0.22) 1.44 (0.34) 1.06 (0.06) 0.89 (0.20) 0.40 (0.13) 100--110 1.97 (0.26) 1.26 (0.30) 1.30 (0.08) 1.03 (0.11) 0.95 (0.08) 0.51 (0.09) Victor Bog 0--10 0.91 (0.24) 0.93 (0.05) 1.26 (0.27) 1.08 (0.04) 0.52 (0.04) 0.47 (0.05) 50--60 2.13 (0.45) 1.82 (0.08) 1.71 (0.16) 1.39 (0.22) 1.11 (0.21) 0.89 (0.04) 100--110 1.50 (0.30) 1.26 (0.44) 1.09 (0.25) 1.10 (0.41) 0.86 (0.18) 0.53 (0.15) Victor Fen 0--10 2.58 (0.61) 3.02 (0.50) 5.24 (1.52) 1.02 (0.12) 1.51 (0.05) 2.81 (0.49) 50--60 1.57 (0.18) 2.96 (0.35) 2.91 (1.03) 0.94 (0.24) 0.25 (0.06) 0.76 (0.08) 100--110 1.17 (0.04) 1.96 (0.59) 1.52 (0.05) 0.85 (0.11) 0.81 (0.15) 0.47 (0.15) **Study site** **Depth (cm)** **Organic substrates** **Rich fen sedge** **Int. fen sedge** **Int. fen *Sphagnum*** **Poor fen *Sphagnum*** **Mean ratio** **OXIC** Kinoje Bog 0--10 2.97 (0.65) 1.75 (0.03) 3.72 (0.94) 4.65 (0.48) 2.00 50--60 3.80 (0.48) 2.36 (0.25) 4.36 (0.27) 4.37 (0.59) 2.30 100--110 6.01 (0.58) 2.26 (0.38) 4.36 (0.52) 4.92 (0.51) 2.48 Victor Bog 0--10 3.12 (0.30) 2.26 (0.29) 5.46 (0.13) 6.68 (0.53) 2.57 50--60 3.11 (0.40) 1.78 (0.22) 2.83 (0.13) 4.53 (0.32) 1.75 100--110 9.53 (0.55) 4.85 (1.33) 6.55 (1.23) 6.60 (0.64) 3.93 Victor Fen 0--10 3.63 (0.39) 3.28 (0.38) 2.51 (0.15) 3.42 (0.33) 2.44 50--60 4.88 (0.30) 5.72 (0.54) 3.09 (0.29) 4.48 (0.42) 2.79 100--110 7.25 (0.88) 4.85 (0.42) 4.58 (0.02) 6.24 (0.25) 3.18 **ANOXIC** Kinoje Bog 0--10 2.06 (0.50) 1.89 (0.63) 2.69 (0.08) 3.91 (1.06) 2.18 50--60 1.65 (0.30) 1.50 (0.42) 2.49 (0.31) 2.16 (0.61) 1.43 100--110 2.26 (0.49) 3.08 (0.26) 2.16 (0.57) 2.62 (0.27) 1.71 Victor Bog 0--10 1.31 (0.21) 1.15 (0.04) 1.82 (0.42) 1.43 (0.53) 1.09 50--60 2.22 (0.36) 1.91 (0.19) 3.01 (0.42) 2.94 (0.78) 1.91 100--110 1.99 (0.46) 2.27 (0.44) 2.75 (0.59) 2.80 (0.50) 1.62 Victor Fen 0--10 4.3 (0.94) 5.79 (1.31) 3.55 (0.58) 4.99 (0.30) 3.48 50--60 5.25 (1.56) 6.32 (0.65) 4.15 (0.32) 6.56 (1.93) 3.17 100--110 5.35 (0.82) 5.95 (0.80) 4.55 (0.91) 6.36 (1.31) 2.90 Int. represents Intermediate {#F3} Enzyme activity generally decreased with depth, but the Kinoje Bog middle depth sample had higher *B*-xylosidase, cellobiohydrolase, sulfatase, and peroxidase activities (Table [3](#T3){ref-type="table"}). NMDS ordination of the initial EEAs showed that Victor Bog and Fen surface samples had similar activities (Figure [3](#F3){ref-type="fig"}C). While Kinoje Bog surface peat had much lower enzyme activities (between 2.5 and 6 times; Table [3](#T3){ref-type="table"}) the samples did cluster close to the Victor peatland surface samples. Similarly, Kinoje and Victor bog middle and deep samples clustered together, whereas Victor Fen middle and deep peat formed their own group as enzyme activities were much higher (range 0.5--12.7 times). ###### **Activity of enzymes (mean ± 1 SE nmol h^−1^ g^−1^) within peat soil collected at three depths (*n* = 3) from peat cores collected in the James Bay Lowlands**. Sample location Sample depth (cm) *N*-acetyl-beta-[d]{.smallcaps}-glucosaminide Phosphatase β-glucosidase Sulfatase *B*-xylosidase Cello-biohydrolase Per-oxidase Phenol oxidase ----------------- ------------------- ----------------------------------------------- ------------------- ---------------- -------------- ---------------- -------------------- ------------- ---------------- Kinoje Bog 0--10 360.6 (67.5) 8261.1 (1058.4) 1096.8 (101.5) 4.8 (0.9) 36.6 (0.6) 88.1 (5.8) 1.9 (0.7) 0.3 (0.1) 50--60 16.1 (9.6) 735.9 (291.5) 339.2 (80) 23.6 (3.4) 78.2 (12) 21.7 (5.5) 5.6 (0.8) 2.1 (0.8) 100--110 48.7 (9.3) 1845 (244.3) 338.5 (48) 0 (0) 45.5 (14.3) 21.8 (1.7) 2.4 (1.3) 0 (0) Victor Bog 0--10 900.2 (333) 45151.8 (16612.7) 2611.9 (500.3) 32.3 (20.3) 196.3 (94.4) 248.5 (34.2) 9.6 (1.3) 1.6 (0.9) 50--60 139.3 (76) 2164.9 (1264) 1047.2 (313.6) 20.6 (15.7) 119.1 (5.1) 64.6 (36.7) 2.1 (1.1) 3.4 (2.4) 100--110 41.2 (3.5) 1126.9 (334.7) 387.2 (81.3) 0 (0) 89.4 (14.3) 4.7 (0.5) 4.2 (2.3) 1 (1) Victor Fen 0--10 1667.6 (112.5) 39231.2 (4371.8) 3913.3 (200.2) 145.2 (52.2) 362.5 (52.6) 415.5 (16.6) 29.4 (6.5) 3.5 (1.9) 50--60 1777.5 (266.8) 3227.4 (60.1) 2686.2 (150.8) 85.3 (15) 145 (5.7) 344.3 (24) 17.3 (2.6) 2 (0.7) 100--110 647.8 (69.9) 1072.6 (280.5) 2199.1 (217.8) 0 (0) 79.6 (2.1) 275 (22.6) 17.2 (3.9) 1.1 (0.6) Influence of temperature and aeration on microbial activities ------------------------------------------------------------- Temperature only affected potential CO~2~ production rates in the surface peat samples at Victor Bog and Victor Fen (Figure [4](#F4){ref-type="fig"}), where rates were highest in the oxic treatment at 14°C and lowest in the anoxic treatment at 4°C. Although potential CO~2~ production rates in an oxic environment were greatest at Victor Fen, the largest effect of temperature was observed at Victor Bog in both aeration conditions. Mean Q~10~ values at Victor Bog were 1.57 ± 0.05 and 2.34 ± 0.07, and 1.46 ± 0.03 and 1.52 ± 0.16 at Victor Fen in an oxic and anoxic environment, respectively. In contrast, similar potential CO~2~ production rates were observed in the middle and deep samples at both Victor Bog and Victor Fen. Interestingly, temperature and aeration states did not have an affect on potential CO~2~ production rates at Kinoje Bog, where higher rates were found in deeper peat. Overall, anoxic potential CO~2~ production rates followed a similar pattern to oxic production rates, and there was a strong correlation between the two (*t*~52~ = 15.5, *p* \< 0.001, *r*^2^ = 0.82). {#F4} Similarly, temperature had no effect on potential CH~4~ production rates in any peat sample and there was no difference in potential CH~4~ production rates among the middle and deep peat samples at all three peatlands (Figure [5](#F5){ref-type="fig"}). Compared to middle and deep peat samples the potential CH~4~ production rates in surface peat at Victor Fen were 2 times greater and 3.7 times greater at Victor Bog, but were approximately 50% lower at Kinoje Bog. Moreover, anoxic CH~4~ production rates correlated with oxic (*t*~52~ = 5.2, *p* \< 0.001, *r*^2^ = 0.34) and anoxic (*t*~52~ = 4.7, *p* \< 0.001, *r*^2^ = 0.29) CO~2~ production rates. {ref-type="fig"} for CART explanation.](fmicb-03-00070-g005){#F5} Measured EEA in the microcosm experiment at 90, 180, and 270 h did not correlate with CO~2~ production in either the oxic or anoxic environments, except for a weak correlation between oxic CO~2~ production and sulfatase (*t*~106~ = 4.3, *p* \< 0.001, *r*^2^ = 0.15), but this was primarily driven by the absence of enzyme activity in deep peat. Interactions among measured variables ------------------------------------- Although basal CO~2~ respiration correlated well for several variables with *r* values ranging from 0.6 to 0.85 (Table [4](#T4){ref-type="table"}), only pH was included in the final regression model (basal CO~2~ mg g^−1^ = 0.07 + 0.48(pH); *F*~1,7~ = 8.5, *p* \< 0.05, adjusted *r*^2^ = 0.49). Similarly, pH could also explain anoxic SIR values (Anoxic NMDS SIR = 1.41--0.35(pH); *F*~1,7~ = 32.3, *p* \< 0.001, adjusted *r*^2^ = 0.77), but not oxic SIR values. Despite pH correlating well with several enzyme activities, it was not significant in the final regression model. Interestingly, bacterial and fungal diversity did not correlate with pH, but good correlations were observed for measures of archaeal diversity and pH. Overall enzyme activity (NMDS values) correlated with bacterial diversity, but the regression was not significant, but there was no relationship with archaeal or fungal diversity. ###### **Pearson's product moment correlations between measures of microbial activity/pH and potential explanatory variables**. Dependent variable Independent variable Test statistic *r* Consistent with hypothesis -------------------- ------------------------------ ---------------------------- ------- ---------------------------- Basal CO~2~ mg pH *t*~7~ = 2.9, *p* \< 0.05 0.74 Yes Archaea Shannon *t*~7~ = 2.0, *p* \< 0.1 0.6 Yes NMDS enzyme *t*~7~ = 2.2, *p* \< 0.1 0.65 Yes β-1-4-glucosidase *t*~7~ = 3.9, *p* \< 0.05 0.83 Yes Cellobiohydrolase *t*~7~ = 3.4, *p* \< 0.05 0.79 Yes β-*N*-acetyl-glucosaminidase *t*~7~ = 4.3, *p* \< 0.05 0.85 Yes Peroxidiase *t*~7~ = 3.1, *p* \< 0.05 0.76 Yes Phenol oxidase *t*~7~ = 2.5, *p* \< 0.05 0.69 Yes Sulfatase *t*~7~ = 5.3, *p* \< 0.001 0.9 Yes *B*-xylosidase *t*~7~ = 4.1, *p* \< 0.01 0.84 Yes NMDS initial EEA NMDS bacteria *t*~7~ = 2.2, *p* \< 0.01 −0.64 Yes NMDS oxic SIR MB-C *t*~7~ = 2.4, *p* \< 0.05 −0.68 Yes MB-N *t*~7~ = 2.2, *p* \< 0.01 −0.65 Yes NMDS archaea *t*~7~ = 2.2, *p* \< 0.01 −0.63 Yes NMDS anoxic SIR pH *t*~7~ = 5.7, *p* \< 0.001 −0.91 Yes β-1-4-glucosidase *t*~7~ = 1.9, *p* \< 0.1 −0.59 Yes Cellobiohydrolase *t*~7~ = 2.5, *p* \< 0.05 −0.68 Yes β-*N*-acetyl-glucosaminidase *t*~7~ = 2.5, *p* \< 0.05 −0.67 Yes Peroxidase *t*~7~ = 2.9, *p* \< 0.05 −0.73 Yes Sulfatase *t*~7~ = 1.9, *p* \< 0.1 −0.58 Yes pH Archaea Shannon *t*~7~ = 2.5, *p* \< 0.05 0.68 Yes NMDS Archaea *t*~7~ = 2.2, *p* \< 0.01 −0.63 Yes β-1-4-glucosidase *t*~7~ = 3.4, *p* \< 0.05 0.79 Yes Cellobiohydrolase *t*~7~ = 4.2, *p* \< 0.01 0.85 Yes β-*N*-acetyl-glucosaminidase *t*~7~ = 3.6, *p* \< 0.01 0.8 Yes Peroxidase *t*~7~ = 6.5, *p* \< 0.001 0.93 Yes Sulfatase *t*~7~ = 2.9, *p* \< 0.05 0.73 Yes Xylase *t*~7~ = 2.1, *p* \< 0.01 0.63 Yes *NMDS represents the first axis from the respective NMDS plot*. Discussion ========== Microbial community characterization ------------------------------------ The same dominant microbial taxa identified using T-RFLP were found among all three peatlands despite differences in geographic location, nutrient status, and plant community composition. This contradicts our initial hypothesis and previous studies where differences in microbial community composition have been identified among peatlands with different vegetation characteristics (Ausec et al., [@B6]; Peltoniemi et al., [@B59]; Andersen et al., [@B3]). However, this finding is consistent with previous peatland studies comparing bacterial communities in two similar fens in Slovenia (Kraigher et al., [@B44]) and archaea and bacteria in *Sphagnum* dominated bogs across the northeastern USA (Basiliko et al., [@B11]; Morales et al., [@B56]). In contrast to previous studies (Bååth and Anderson, [@B7]; Fierer and Jackson, [@B24]; Rousk et al., [@B67]), where higher bacterial diversity is associated with more neutral pH, a greater number of taxa were actually detected in the more acidic bogs rather than the fen. Moreover, there was no correlation between pH and the microbial community structure or diversity indices. However, the range of peat pH values in the JBL was relatively small, which may explain why diversity was not explained by pH. Furthermore, pH also appeared not to influence microbial community composition in other peatland studies (Kraigher et al., [@B44]; Ausec et al., [@B6]), suggesting that the relationship between pH and microbial diversity in general might be less important in peatland ecosystems. Richness was also similar among the depths with few taxa confined to a single depth profile, in contrast to Morales et al. ([@B56]) who found a greater number of bacterial T-RFs in surface peat (0--15 cm) than in deep peat (1 m). Although this may have been due to the relatively low detected richness level relative to Morales et al. ([@B56]), it suggests that at least the dominant members of the microbial community are always present, despite presumed differences in oxygen content with depth. Little is known about the role of fungi in peatlands, but diverse communities have been found beneath the *Sphagnum* mat and in close association with ericaceous shrubs (Thormann, [@B75]; Artz et al., [@B4]) and different fungal communities have been linked to the vegetation and nutrient status of peatlands (Artz et al., [@B4]; Trinder et al., [@B76]). Fungi are generally thought to predominate in peatland lawns and hummocks (Jaatinen et al., [@B40]) as they cannot utilize alternative electron acceptors in anoxic environments and successfully compete with bacteria for carbon substrates (Killham and Prosser, [@B42]). Thus it was interesting that most fungal taxa were identified at all three depths among the three JBL peatlands. Fungi are typically most competitive under an oxic environment, however they can carry out fermentation. So their abundance at depth may in part be due to a lack of alternative inorganic electron acceptors (e.g., sulfate) that might otherwise have facilitated anaerobic bacterial respiration in these deep anoxic peat samples. This might also imply that facultative fungi supply fermentation products that ultimately fuel methanogenesis (Conrad, [@B19]). Consistent with prior studies, microbial biomass decreased with depth probably due to the presence of more labile organic matter and higher redox states typically found near the surface of peat profiles (Blodau and Moore, [@B14]; Blodau et al., [@B13]; Basiliko et al., [@B10]). Although microbial biomass across sites was generally quite small compared to other bogs and fens (Moore and Basiliko, [@B52]), it was similar to that in mined peatlands in more southern regions of Canada that have relatively low nutrient and carbon availability (Basiliko et al., [@B10]); which might indicate that substrate or nutrient availability for heterotrophic microbial communities is low even in the fen site. Decreasing MB-C:MB-N ratios with depth has been observed elsewhere (e.g., Basiliko et al., [@B10]) and is probably due to decreasing fungal:bacterial ratios with depth and the fact that bacteria typically have higher anabolic N requirements (Killham and Prosser, [@B42]). Bacteria often predominate in deeper more anoxic peat layers as they can utilize alternative electron acceptors (Killham and Prosser, [@B42]). Microbial biomass did not correlate with potential CO~2~ production rates in contrast to previous studies (Blodau et al., [@B13]; Basiliko et al., [@B10]). However, this may simply reflect the low microbial biomass and low amount of labile carbon in our sites. Microbial activities -------------------- Despite the similarity of the microbial communities among the three peatlands, there were differences in basal respiration, initial enzyme activities and substrate utilization patterns. Greater microbial activity was observed in the fen compared to the bogs, possibly due to the higher *in situ* nutrient concentrations and greater carbon substrate availability typically found in fens (Moore and Basiliko, [@B52]; Knorr and Blodau, [@B43]). Overall the average oxic SIR ratio response was very similar among deep peat samples in all three peatlands and greater than those in both the surface and middle peat samples. Deeper peat is older and typically has lower levels of labile carbon (Glatzel et al., [@B31]; Artz et al., [@B5]). Thus, the microbial community at these depths was probably substrate limited relative to the surface peat. However, this trend disappeared in the anoxic SIR assay, as there was little difference in the response among depths between Kinoje and Victor bogs. In contrast, Victor Fen had greater average SIR values in the surface peat sample, likely as a result of higher redox substrate concentrations (Knorr and Blodau, [@B43]; Webster and McLaughlin, [@B82]). This apparent contradiction between microbial activity and community composition may be explained by differences in peat organic matter quality. Plant litter from bogs and fens have different nutrient concentrations (Bragazza et al., [@B15]) and therefore microorganisms in these contrasting habitats will have different nutrient requirements. Recently, Straková et al. ([@B72]) identified litter type as the main factor affecting microbial EEA in bog peat soil. Similar to our study the authors reported higher enzyme activities in peat composed of vascular plants, typical in fens, rather than *Sphagnum* dominated peat and also failed to identify a strong relationship between microbial community composition and EEA. Moreover, the differences in SIR activities, identified by the NMDS ordination, among sites was primarily due to the magnitude of the response, but the relative substrate utilization patterns (i.e., among the 10 substrates) were actually similar among the three peatlands despite our hypothesis that the two bogs and fen would preferentially mineralize different substrates, probably because the microbial communities were similar. The subdued response to substrate addition in the bogs may be due to inhibitory effects of *Sphagnum* leachates on microbial activity (Verhoeven and Toth, [@B80]; Hättenschwiler and Vitousek, [@B34]; Hájek et al., [@B33]). Furthermore, vegetation type has been shown to influence microbial activity more than water table drawdown (Trinder et al., [@B76]; Straková et al., [@B72]) and small changes in the plant community composition can have profound affects on microbial processes, such as methanogenesis (Hines et al., [@B37]). The relatively high *N*-acetyl-beta-[d]{.smallcaps}-glucosaminidase activity throughout the peat profile suggests a high metabolic demand for N released during chitin turnover (Sinsabaugh et al., [@B71]) and corresponds with presence and identification of fungal taxa at all peat depths. This contradicts our previous finding that carbon:nitrogen ratio decreases with depth and might represent bacterial dominance of decomposition due to higher N demand. However, it is not certain whether fungi or bacteria produced these enzymes or what the specific activity of the fungi is in deep peat. Interestingly, a recent meta-analysis found that shifts in bacterial to fungal dominance across environmental gradients often do not correlate with changes in element cycling and may be due to niche overlap between fungal and bacterial communities (Strickland and Rousk, [@B74]). In an antibiotic inhibition assay, Winsborough and Basiliko ([@B84]) showed that bacteria dominated peatland microbial activity; while fungal activity did increase in acidic dry bogs it was still lower than bacterial activity. Although we do not have direct measurements of fungal biomass, these results suggest that our understanding of the relationship between fungi and ecosystem function in peat soil is incomplete. We did not find a relationship between microbial community composition and activities in contrast to our hypothesis. Although this could be due to the issues of the taxa-level resolution using T-RFLP small-subunit rDNA, this finding highlights the importance of carbon substrate as a proximate control of microbial community dynamics because similar communities can perform different functions given different resources and conditions. Environmental change -------------------- As predicted and consistent with previous studies (Moore and Dalva, [@B53]; Updegraff et al., [@B78]; Keller et al., [@B41]), an increase in aeration of the peat soil stimulated CO~2~ production rates, but only in Victor Bog and Fen, and temperature only affected rates in the surface peat samples. This suggests that oxygen availability is more important than temperature in increasing mineralization rates, possibly brought about by an increase in enzyme activity in response to a release from inhibition through the "enzymic latch theory" (Freeman et al., [@B28], [@B27]). However, CO~2~ production did not correlate with enzyme activity probably because the enzymes measured in this study may only be a subset of those actually involved in peat decomposition (Horwath, [@B38]). Potential CO~2~ production rates among the depths were lower in anoxic conditions but there was a strong correlation between oxic and anoxic CO~2~ production consistent with previous investigations (Moore and Dalva, [@B54]; Yavitt et al., [@B85]; Glatzel et al., [@B30]). These findings suggest that similar peat properties control the activities of both aerobic and anaerobic microbial metabolism (Glatzel et al., [@B30]; Basiliko et al., [@B10]). Similarly, CH~4~ production rates were low throughout the peat profile but still within the range of previously reported values (Moore and Dalva, [@B53]; Glatzel et al., [@B30]; Basiliko et al., [@B10]). Conclusion ========== Microbial community composition was very similar among peatlands and at depths within the JBL despite differences in geographic location and nutrient status. In contrast, microbial activity appears to be determined by the quality of the peat substrate and the presence of potential microbial inhibitors. As climate change is expected to cause a shift in JBL plant community composition, this study suggests that the microbial community will respond quickly to changes in plant litter and root exudate quality but the existing peat substrate will probably have a large influence on future microbial activity. For example, a shift from *Sphagnum* to sedge-dominated peatlands may not necessarily result in the expected increase in carbon mineralization due to the antimicrobial properties of the *Sphagnum*-peat. Interestingly, we identified fungal taxa in deep peat but it is unclear what anaerobic processes are occurring and how these organisms influence carbon cycling. Thus, more in depth profiling of the JBL microbial community and identification of the potential constraints on microbial activity is required in order to better predict future peatland carbon dynamics. Conflict of Interest Statement ============================== The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The authors gratefully acknowledge Benoit Hamel, Mark Crofts, and Adam Kinnunen for collection of peat samples, Varun Gupta and Charlotte Hewins for assistance with laboratory analyses and the two reviewers for comments and suggestions to improve the manuscript. Funding was provided by Ontario Ministry for Natural Resources, The National Science Foundation Enzymes in the Environment Research Coordination Network, The Corning Institute for Education and Research, and The Reinberger Foundation. ###### **Operational taxonomic units (OTU) of archaea, bacteria, and fungi identified at three depths in peatlands across the James Bay Lowlands, Canada using terminal restriction fragment length polymorphism analysis (T-RFLP) of rDNA**. Site T-RF length Proportion within each depth -------------- ------------- ------------------------------ ------ ------ **ARCHAEA** Kinoje bog 51 -- -- 0.73 178 1.00 0.25 -- 384 -- -- 0.02 483 -- 0.75 0.25 Victor bog 51 0.03 -- -- 178 0.03 0.03 -- 193 0.07 0.07 -- 384 0.43 0.45 0.60 483 0.41 0.42 0.40 729 0.03 0.03 -- Victor fen 51 -- 0.07 -- 67 -- -- 0.04 75 -- 0.05 0.11 83 0.11 0.10 0.42 249 -- 0.04 -- 277 0.27 0.26 0.11 384 0.28 0.20 0.14 483 0.34 0.23 0.13 729 -- 0.04 0.05 **BACTERIA** Kinoje bog 59 -- -- 0.02 61 0.07 -- -- 66 0.88 0.15 0.04 82 -- 0.08 0.02 89 -- 0.41 0.01 136 -- -- 0.02 140 0.02 0.13 -- 144 0.03 0.03 0.02 160 -- 0.02 -- 165 -- -- 0.12 260 -- 0.10 0.04 429 -- 0.02 -- 447 -- 0.03 -- 476 -- -- 0.17 481 -- -- 0.26 484 -- -- 0.24 491 -- -- 0.03 509 -- 0.02 -- Victor bog 54 0.06 -- -- 59 0.04 -- -- 61 0.01 0.07 0.04 66 0.37 0.13 0.08 89 0.03 0.08 0.10 110 0.02 -- -- 123 0.02 -- -- 133 -- 0.03 -- 136 0.09 0.16 0.15 140 0.07 -- -- 144 0.08 0.14 0.20 147 0.01 -- -- 160 0.01 -- -- 165 0.01 0.07 0.08 260 0.04 0.06 -- 280 -- 0.03 -- 284 0.04 -- -- 301 0.02 -- -- 429 0.03 -- -- 432 0.01 -- -- 447 -- 0.03 -- 450 0.01 -- -- 476 -- 0.15 0.14 484 -- -- 0.22 496 0.02 -- -- 519 -- 0.04 -- Victor fen 66 0.30 0.60 0.30 79 -- -- 0.07 89 -- 0.04 -- 97 -- 0.04 -- 144 -- -- 0.19 176 0.04 -- -- 484 0.21 0.08 0.30 496 -- -- 0.13 513 0.21 0.12 -- 519 0.10 0.06 -- 521 0.15 0.06 -- 193 0.34 -- 0.05 **FUNGI** Kinoje bog 150 0.32 0.83 0.85 178 -- 0.03 -- 193 0.34 -- 0.05 294 -- 0.08 -- 306 0.15 0.06 0.10 328 0.20 -- -- Victor bog 150 0.64 0.82 0.18 153 -- -- 0.82 193 0.29 0.04 -- 294 -- 0.06 -- 306 0.07 0.03 -- 799 -- 0.05 -- Victor fen 150 0.09 0.61 0.24 178 -- -- 0.06 193 0.72 0.17 0.16 306 0.19 0.22 0.53 *Each OTU is represented by a terminal restriction fragment (T-RF) expressed as a proportion of the total sample (e.g., Kinoje Bog 0--10 cm)*. [^1]: Edited by: Svetlana N. Dedysh, Russian Academy of Sciences, Russia [^2]: Reviewed by: Neil Duncan Gray, University of Newcastle, UK; Peter Frenzel, MPI for Terrestrial Microbiology, Germany [^3]: This article was submitted to Frontiers in Terrestrial Microbiology, a specialty of Frontiers in Microbiology.

Apoptosis has been widely recognized as a well controlled and conserved process which is crucial for the normal development and function of multiple organisms[@b1]. Deregulations of apoptosis lead to either inappropriate killing of vital cells or survival of unwanted cells, which has been considered to be a hallmark of most cancers[@b2][@b3]. B-cell lymphoma-2 (BCL-2) family proteins are essential regulators of apoptosis and consists of both pro- and anti-apoptotic members, which all share sequence homology in their BCL-2 homology domains[@b4]. These proteins can promote cell survival (BCL-2 and BCL-xL), initiate cell killing (BIM, PUMA and BID) or activate the effecter pathways of apoptosis (BAX and BAK)[@b4]. *BCL-2* was firstly identified during the investigation of t(11;14) chromosome translocation in B-cell lymphoma[@b5]. BCL-2 protein locating on intracellular membranes prevents apoptosis in response to various death inducing stimuli and its overexpression could promote cancer cell survival[@b5]. The discovery of BCL-2 established a new paradigm in cancer biology, namely that apoptosis defects give cells selective survival superiority[@b5]. As one of the most common and fatal malignancies worldwide, esophageal squamous cell carcinoma (ESCC) shows a relatively high incidence in Asian including China[@b6]. Cigarette smoking, heavy ethanol consumption, micronutrient deficiency as well as dietary carcinogen exposure have been identified as main environmental etiological factors of ESCC[@b7][@b8]. Accumulated evidences indicate that genetic makeup may also contribute to ESCC susceptibility. For example, ESCC genome-wide association studies (GWAS) highlight the involvement of single nucleotide polymorphisms (SNP) in cancer development, alone and in combination with environmental risk factors[@b9][@b10][@b11][@b12][@b13][@b14]. In ESCC, BCL-2 plays its role in regulating cancer cell growth, especially in the early stage. Additionally, BCL-2 expression has been positively associated with cancer cell differentiation and inversely with disease progression[@b15]. There are multiple functional genetic polymorphisms have been identified in the *BCL-2* gene locus which is located on chromosome 18q21.3 and consists of three exons and two promoters. These two promoters show different functional properties. That is, *BCL-2* mRNA transcription is driven by the P1 promoter, while the P2 promoter acts as a negative regulatory element[@b16][@b17]. There is a functional rs2279115 (−938 C \> A) promoter SNP in the inhibitory P2 promoter[@b18]. Interestingly, the rs2279115 A allele may render a better interaction with TP53, leading to a decrease in the BCL2 expression, an up-regulated programmed cell death or reduced longevity of transformed cells, and thus a subsequent decrease in the risk of malignances, such as squamous cell carcinoma of the head and neck (SCCHN)[@b19]. There is only one study with a relatively small sample size investigated the role of this SNP in the etiology of ESCC without genotype-phenotype association investigations[@b20]. Considering the importance of BCL-2 in tumorigenesis, we hypothesized that the *BCL-2* functional polymorphisms (rs2279115, rs1801018 and rs1564483) might be also involved in ESCC development through deregulating BCL-2 expression. To test this hypothesis, we conducted a two-stage case-control study of ESCC. To validate the biological function of *BCL-2* rs2279115 genetic variant *in vivo*, we detected the association between its genotypes and *BCL-2* mRNA expression levels in normal and cancerous esophagus tissues. Materials and Methods ===================== Study subjects -------------- Two case-control sets were included in the current study. Huaian case-control set consists of 588 ESCC cases from Huaian No. 2 Hospital (Huaian, Jiangsu Province, China) and sex- and age-matched 600 healthy controls. Jinan case-control set contains 540 patients with ESCC from Shandong Cancer Hospital, Shandong Academy of Medical Sciences (Jinan, Shandong Province, China) and sex- and age-matched (±5 years) 550 controls. We used the group match considering sex- and age-match between cases and controls. The detailed information about the two case-control sets has been reported previously[@b21][@b22][@b23]. Twenty-nine ESCC tissues and twenty nine paired esophagus normal tissues adjacent to the tumors were obtained from surgically removed specimens of patients in Huaian No. 2 Hospital. The normal tissues sampled at least 2 cm away from the margin of the tumor. All subjects were ethnic Han Chinese. This study was approved by the Institutional Review Boards of Huaian No. 2 Hospital and Shandong Cancer Hospital, Shandong Academy of Medical Sciences. At recruitment, the written informed consent was obtained from each subject. The methods were carried out in accordance with the approved guidelines. Genotyping of BCL-2 polymorphisms --------------------------------- Three *BCL-2* candidate SNPs were analyzed by the MassArray system (Sequenom Inc., San Diego, California, USA). A 5% blind, random sample of study subjects was genotyped in duplicates and the reproducibility was 98.8%. To reduce the costs of the study, we genotyped the *BCL-2* rs2279115 SNP in the validation set using PCR-based restriction fragment length polymorphism (RFLP). The genotyping primers used for amplifying DNA segments with the SNP site were 5′-GCATTTGCTGTTCGGAGTTT-3′ and 5′-TTCGCAGAAGTCCTGTGATG-3′. The 25 μL PCR reaction mixture contains 0.2 mmol/L of deoxynucleoside triphosphate, 0.1 mmol/L of each primer, 100 ng of DNA, 1.0 U of rTaq DNA polymerase (TaKaRa), 1.5 mmol/L MgCl~2~, and 1 × reaction buffer. The PCR profile included an initial 2 minutes melting step at 95 °C, followed by 35 cycles of 30 seconds at 94 °C, 30 seconds at 60 °C, 30 seconds at 72 °C, and a final 10 minutes elongation step at 72 °C. Restriction enzyme *Bcc*I (New England Biolabs) was used to distinguish the rs2279115 C \> A genotypes. A 15% random sample was reciprocally tested by different person, and the reproducibility was 99.0%. Real-time Analysis of BCL-2 mRNA -------------------------------- Total RNA was extracted from ESCC tissue samples using TRIzol Reagent (Invitrogen) and converted to cDNA using the ReverTra Ace qPCR RT Kit (TOYOBO). *BCL-2* mRNA expression in cancerous and normal esophagus tissues was examined using the SYBR-Green real-time quantity PCR (qPCR) method as described previously[@b23][@b24][@b25]. Gene expression for *BCL-2* and *β-actin* as an internal reference gene was carried out using the ABI 7500 real-time PCR system in triplicates. The primers used for *BCL-2* were 5′-TCGCCCTGTGGATGACTGA-3′ and 5′-CAGAGACAGCCAGGAGAAATCA-3′; and for *β-actin* were 5′-GGCGGCACCACCATGTACCCT-3′ and 5′-AGGGGCCGGACTCGTCATACT-3′. The samples size of each qPCR assay is 10 μL. Relative gene quantitation for *BCL-2* was calculated by -delta delta ct methods. To control quality of the qPCR data, we repeated the qPCR assays for some specific sample if the ct values of the triplicates for this sample showed variations more than 0.5 in very few cases. Paired *t*-test was used to calculate the differences between individuals with different genotypes. Statistics ---------- The differences in demographic variables and genotype distributions of *BCL-2* SNPs between ESCC cases and controls were examined using Pearson's χ^2^ test. Unconditional logistic regression model was utilized to estimate associations between *BCL-2* genotypes and ESCC risk by *odds ratio* (*OR*) and their 95% *confidence intervals* (*CIs*). During calculating associations between functional SNP candidates in *BCL-2* and ESCC risk in Huaian case-control set, we used the common genotypes of rs2279115 (CC), rs1801018 (AA) and rs1564483 (GG) as the reference genotype. All *ORs* were adjusted for age, sex, drinking and smoking status, where it was appropriate. A *P* value of less than 0.05 was used as the criterion of statistical significance, and all statistical tests were two-sided. All analyses were performed with SPSS software package (Version 16.0, SPSS Inc., Chicago, IL). Results ======= There were no statistically significant differences between cases and controls for both case-control sets in terms of median age and sex distribution (all *P* \> 0.05), which indicated that the frequency matching of age and sex was adequate ([Table 1](#t1){ref-type="table"}). More smokers were observed among ESCC cases compared with controls in both case-control sets (Huaian set: 74.3% vs. 33.8%, *P* \< 0.001; Jinan set: 75.2% vs. 39.6%, *P* \< 0.001). Similarly, there were more alcohol drinkers among patients than among control subjects in these two sets (Huaian set: 56.8% vs. 40.3%, *P* \< 0.001; Jinan set 55.3% vs. 40.1%, *P* \< 0.001). The genotype frequencies of *BCL-2* candidate SNPs (rs2279115 C \> A, rs1801018 A \> G and rs1564483 G \> A) are summarized in [Table 2](#t2){ref-type="table"}. The allele frequencies for rs2279115 C, rs1801018 G and rs1564483 A were 0.362, 0.084, and 0.375 in ESCC cases and 0.410, 0.098, and 0.338 in control subjects in Huaian training case-control set. All observed genotype frequencies in either controls or cases conform to Hardy-Weinberg equilibrium. Distributions of the rs2279115, rs1801018 and rs1564483 genotypes were then compared among patients and controls. Frequencies of rs2279115 CC, CA, and AA genotypes among ESCC cases differed significantly from those among controls (χ^2^ = 8.68, *P* = 0.013, *df* = 2), with the frequency of AA homozygote being significantly lower among patients than among controls (13.7% vs. 15.0%). However, no statistically significant differences of rs1801018 and rs1564483 genotypes were observed between cases and control subjects (both *P* \> 0.05) ([Table 2](#t2){ref-type="table"}). Therefore, we did no other analyses of these two polymorphisms in the next studies. Associations between genotypes of *BCL-2* rs2279115 C \> A SNP and ESCC risk were calculated using unconditional logistic regression analyses ([Table 3](#t3){ref-type="table"}). The *BCL-2* rs2279115 A allele was shown to be a protective allele. Individuals with the rs2279115 CA genotype had an *OR* of 0.66 (95% *CI* = 0.50--0.88, *P* = 0.004) for developing ESCC in Huaian Set, compared with individual having the rs2279115 CC genotype. However, the rs2279115 AA genotypes had a marginally decreased risk for ESCC compared with the rs2279115 CC genotype (*OR* = 0.85, 95% *CI* = 0.70--1.03, *P* = 0.095). In Jinan set, carriers of the rs2279115 CA or AA genotypes were significantly associated with decreased ESCC risk (*OR* = 0.62, 95% *CI* = 0.50--0.77, *P* = 0.002, or *OR* = 0.49, 95% *CI* = 0.36--0.66, *P* = 4.2 × 10^−4^) ([Table 3](#t3){ref-type="table"}). In the pooled analyses, we observed that only individuals with the rs2279115 AA genotype had a 0.72-fold decreased risk to develop ESCC compared to the CC genotype carriers (95% *CI* = 0.57--0.90, *P* = 0.005) ([Table 3](#t3){ref-type="table"}). All *ORs* were calculated with adjustments of sex, age, smoking and alcohol drinking status. Associations between genotypes of *BCL-2* rs2279115 genetic variant and ESCC risk was further examined by stratifying for age, sex, smoking and alcohol drinking status using the pooled data of two Chinese case-control sets ([Table 4](#t4){ref-type="table"}). Compared with the *BCL-2* rs2279115 CC genotype, a significantly decreased risk of ESCC was associated with AA genotypes only among males (*OR* = 0.70, 95% *CI* = 0.54--0.91, *P* = 0.008), but not among females (*OR* = 0.74, 95% *CI* = 0.42--1.32, *P* = 0.309). However, the *BCL-2* rs2279115 CC genotype was not significantly associated with ESCC susceptibility in males or females (*OR* = 0.86, 95% *CI* = 0.71--1.04, *P* = 0.123, or *OR* = 0.97, 95% *CI* = 0.66--1.42, *P* = 0.868). In age-stratified analyses, either rs2279115 CA or AA genotype was significantly associated with decreased risk in subjects aged older than 57 years (*OR* = 0.76, 95% *CI* = 0.60--0.96, *P* = 0.021, or, *OR* = 0.56, 95% *CI* = 0.40--0.78, *P* = 0.001). However, among subjects aged 57 years or younger, neither rs2279115 CA nor AA genotype showed impacts on ESCC risk (*OR* = 0.56, 95% *CI* = 0.40--0.78, *P* = 0.001, or, *OR* = 0.92, 95% *CI* = 0.65--1.29, *P* = 0.619). Because tobacco smoking and alcohol drinking are both risk factors for ESCC, we then examined whether the *BCL-2* rs2279115 genetic variant influence ESCC susceptibility in combination with these pathogenic factors ([Table 4](#t4){ref-type="table"}). In nonsmokers, compared with the rs2279115 CC carriers, individuals with CA or AA genotype had a 0.70-fold or 0.42-fold decreased risk to develop ESCC (95% *CI* = 0.54--0.92, *P* = 0.010, or 95% *CI* = 0.29--0.59, *P* = 0.001). There was no significantly decreased risk for smokers with CA or AA genotype compared with CC smokers (both *P* \> 0.05). Nondrinkers carrying rs2279115 CA or AA genotype showed significantly decreased risk to develop ESCC compared with CC carriers who did not drink (*OR* = 0.78, 95% *CI* = 0.61--0.99, *P* = 0.041, or *OR* = 0.44, 95% *CI* = 0.32--0.62, *P* = 0.002). However, there were no association between rs2279115 CA and AA genotypes and ESCC risk in drinkers (both *P* \> 0.05) ([Table 4](#t4){ref-type="table"}). Due to rs2279115 C-to-A change could influence *BCL-2* P2 promoter activity and gene expression in cancer cells, we investigated whether there is an allele-specific effect of rs2279115 SNP on *BCL-2* expression in esophagus tissues. We found that there were significantly lower *BCL-2* mRNA levels (mean ± SE) among carriers of the rs2279115 CA and AA genotypes compared to carriers of the CC genotype in normal esophagus tissues (0.083 ± 0.012 \[n = 17\] vs. 0.109 ± 0.021 \[n = 12\], *P* = 0.031) ([Table 5](#t5){ref-type="table"}). Similar results have also been observed in ESCC tissues (the rs2279115 CA and AA genotypes: 0.153 ± 0.022 \[n = 17\] vs. the CC genotype: 0.184 ± 0.045 \[n = 12\], *P* = 0.040) ([Table 5](#t5){ref-type="table"}). Discussion ========== In the current study, we investigated the association between three *BCL-2* functional candidate SNPs and ESCC susceptibility through a case-control approach. We found that only *BCL-2* rs2279115 polymorphism is significantly associated with decreased ESCC susceptibility in Chinese populations, with the rs2279115 AA genotype as the protective genotype. Genotype-phenotype correlation studies demonstrated that subjects with the rs2279115 CA and AA genotypes had a statistically significant decrease of *BCL-2* mRNA expression compared to the CC genotype in both normal and cancerous esophagus tissues. Our data support the hypothesis that SNPs in gene expression regulatory elements of tumor suppressor genes or oncogenes might impact genetic susceptibility of cancers. The *BCL-2* rs2279115 polymorphism has been extensively studied in multiple cancer types, including ESCC. Liu *et al.* reported in a case-control study conducted in western China including 205 esophageal cancer patients and 224 controls[@b20]. They found that the *BCL-2* rs2279115 AA genotype was significantly associated with increased risk of developing esophageal cancer. In contrast, we found that subjects with the rs2279115 AA genotype have significantly decreased risk to develop ESCC in both Northern and Southern Chinese populations. There are several possible explanations for the insistent results. Firstly, Liu *et al.* only recruited 205 esophageal cancer patients to evaluate association between rs2279115 and esophageal cancer risk. The relative large sample size of the present study (1588 ESCC patients and 1600 controls) may provide more statistic power to the moderate effect of this genetic polymorphism on ESCC susceptibility. Secondly, we only included ESCC, but not esophageal adenocarcinoma in this study. Considering the distinct etiology and clinical behaviors of these two subtypes of esophageal cancer, we believe that mixed study subjects might also lead to the discrepancy. Our results are consistent to functional relevance of rs2279115 polymorphism in SCCHN[@b19]. That is, the rs2279115 AA genotype may result in decreased BCL-2 expression, elevated apoptosis rates of cancer cells, and thus decreased risk of malignances[@b19]. Song *et al.* recently reported that hierarchical clustering analyses of whole-genome sequencing in 17 ESCC cases and whole-exome sequencing in 71 cases indicate that ESCC and HNSCC mutation spectra were intermingled, whereas esophageal adenocarcinomas were clearly distinctive from ESCC[@b26]. Therefore, it is biologically plausible that the functional *BCL-2* rs2279115 polymorphism influence ESCC genetics thoroughly through regulating BCL-2 expression and apoptosis *in vivo*. The *BCL-2* rs2279115 polymorphism showed a consistent association with ESCC risk in two independent case-control cohorts. Additionally, our results are unlikely to be attributable to unknown confounding factors due to having relatively large sample sizes, significantly increased odd ratios with small *P* values. More importantly, our results on the genotype-phenotype relationship between the rs2279115 polymorphism and gene expression supports our conclusion. However, there might be several limitations in the current case-control study. For instance, since all ESCC cases were recruited from the hospital, inherent selection bias may exist. Therefore, the findings of our study warrant to be validated in a population-based prospective study in the future. In addition, relatively small sample size for non-smokers and non-drinkers should be further analyzed in a larger population. In conclusion, we demonstrated that functional *BCL-2* rs2279115 SNP was associated with a significantly decreased risk of ESCC in Chinese populations, especially in nonsmokers or nondrinkers. Our data may support the hypothesis that genetic variants can influence gene regulation might be important modifiers of ESCC susceptibility. These results may lead to better understanding of ESCC etiology in different populations. Additional Information ====================== **How to cite this article**: Pan, W. *et al.* Functional *BCL-2* regulatory genetic variants contribute to susceptibility of esophageal squamous cell carcinoma. *Sci. Rep.* **5**, 11833; doi: 10.1038/srep11833 (2015). This study was financially supported by National Natural Science Foundation of China (31271382, 81201586 & 91229126); the Fundamental Research Funds for the Central Universities (YS1407); the National High-Tech Research and Development Program of China (2015AA020950); the open project of State Key Laboratory of Molecular Oncology (SKL-KF-2015-05); Beijing Higher Education Young Elite Teacher Project (YETP0521); Program for Changjiang Scholars and Innovative Research Team in University (IRT13045). **Author Contributions** M.Y. and L.Z. conceived and designed the experiments; W.P. performed the experiments; W.P. and J.Y. analyzed the data; J.W., H.C., Y.G., J.Z., Z.W., C.Z. and Q.Y. contributed materials/analysis tools; M.Y. and L.Z. wrote the manuscript. All authors reviewed and approved the manuscript prior to submission. ###### Distribution of selected characteristics among patients with esophageal squamous cell carcinoma and controls. Variable Huaian set Jinan set --------------------------------------- ------------ ------------ --------- ----------- ----------- ---------   588 600   1000 1000   Sex     0.678     0.426  Male 413(70.2) 428(71.3)   776(77.6) 761(76.1)    Female 175(29.8) 172(28.7)   224(22.4) 239(23.9)   Age (year)[2](#t1-fn2){ref-type="fn"}     0.725     0.474  ≤59(or 56) 288(49.0) 300(50.0)   516(51.6) 500(50.0)    \>59(or 56) 300(51.0) 300(50.0)   484(48.4) 500(50.0)   Smoking status     \<0.001     \<0.001  No 151(25.7) 397(66.2)   248(24.8) 604(60.4)    Yes 437(74.3) 203(33.8)   752(75.2) 396(39.6)   Drinking status     \<0.001     \<0.001  No 254(43.2) 358 (59.7)   447(44.7) 599(59.9)    Yes 334(56.8) 242(40.3)   553(55.3) 401(40.1)   ^1^Two-sided χ^2^ test. ^2^Median ages of patients for Huaian set and Jinan set are 59 and 56 years. ###### Associations between functional SNP candidates in *BCL-2* and ESCC risk in Huaian case-control set (Training set). \# Identity Location Position[1](#t2-fn1){ref-type="fn"} Case Common genotype No. (%) Heterozygous genotype No. (%) Rare genotype No. (%) Common genotype *OR*[2](#t2-fn2){ref-type="fn"} (95% *CI*) for heterozygote *P* *OR*[2](#t2-fn2){ref-type="fn"} (95% *CI*) for rare genotype *P* ---- ----------- -------------- ------------------------------------- --------- ------------------------- ------------------------------- ----------------------- ----------------- ------------------------------------------------------------- ------- -------------------------------------------------------------- ------- 1 rs2279115 5\' promoter 63319604 ESCC 242(41.2) 265(45.1) 80(13.7) Reference 0.66(0.50--0.88) 0.004 0.85(0.70--1.03) 0.095   (C \> A)     Control 198(33.0) 312(52.0) 90(15.0)           2 rs1801018 Exon 2 63318646 ESCC 493(83.8) 91(15.5) 4(0.7) Reference 1.01(0.72--1.41) 0.972 0.77(0.39--1.52) 0.451   (A \> G)     Control 487(81.2) 108(18.0) 5(0.8)           3 rs1564483 3'-UTR 63127421 ESCC 229(38.9) 278(47.2) 82(13.9) Reference 1.30(0.98--1.70) 0.062 1.21(0.98--1.47) 0.068   (G \> A)     Control 271(45.1) 253(42.2) 76(12.7)           Abbreviations: ESCC, esophageal squamous cell carcinoma; *OR, odds ratio; CI, confidence interval*; 3\'-UTR, 3\'-untranslated region. ^1^Position in NCBI build 38. ^2^Data were calculated by unconditional logistic regression, adjusted for sex, age, drinking and smoking status. ###### Genotype frequencies of *BCL-2* rs2279115 genetic variant among patients and controls and their association with ESCC risk. Genotypes *BCL-2* rs2279115 C \> A ------------ -------------------------- ------------ ------------------ -------------- --- Huaian set   *n* = 587 *n* = 600     CC 242(41.2) 198(33.0) Reference   CA 265(45.1) 312(52.0) 0.66(0.50--0.88) 0.004 AA 80(13.7) 90(15.0) 0.85(0.70--1.03) 0.095 Jinan set   *n* = 1000 *n* = 1000     CC 416(41.6) 312(31.2) Reference   CA 453(45.3) 516(51.6) 0.62(0.50--0.77) 0.002 AA 131(13.1) 172(17.2) 0.49(0.36--0.66) 4.2 × 10^−4^ Total   *n* = 1587 *n* = 1600     CC 658(41.5) 510(31.9) Reference   CA 718(45.2) 828(51.7) 0.86(0.73--1.02) 0.079 AA 211(13.3) 262(16.4) 0.72(0.57--0.90) 0.005 Abbreviations: ESCC, esophageal squamous cell carcinoma; *OR, odds ratio; CI, confidence interval.* ^1^Data were calculated by logistic regression with adjustment for age, sex, smoking and drinking status. ###### Risk of ESCC associated with*BCL-2* rs2279115 C \> A genotypes by age, sex, smoking status and drinking status.   *BCL-2* rs2279115 C\>A *BCL-2* rs2279115 C\>A ----------------- ------------------------ ------------------------ ------------------ ------- --------- --------- ------------------ ------- Sex  Male 485/380 530/617 0.86(0.71--1.04) 0.123 485/380 159/207 0.70(0.54--0.91) 0.008  Female 173/130 188/211 0.97(0.66--1.42) 0.868 173/130 52/55 0.74(0.42--1.32) 0.309 Age (year)  ≤57 327/273 364/412 1.02(0.81--1.30) 0.843 327/273 96/131 0.92(0.65--1.29) 0.619  \>57 331/237 354/416 0.76(0.60--0.96) 0.021 331/237 115/131 0.56(0.40--0.78) 0.001 Smoking status  No 414/202 341/334 0.70(0.54--0.92) 0.010 414/202 108/109 0.42(0.29--0.59) 0.001  Yes 352/308 377/494 1.03(0.83--1.28) 0.808 352/308 103/153 1.01(0.81--1.50) 0.532 Drinking status  No 306/202 341/334 0.78(0.61--0.99) 0.041 306/202 108/109 0.44(0.32--0.62) 0.002  Yes 352/308 377/494 0.98(0.78--1.23) 0.855 352/308 103/153 1.12(0.80--1.56) 0.514 Abbreviations: ESCC, esophageal squamous cell carcinoma; *OR, odds ratio; CI, confidence interval.* ^1^Number of case patients with genotype/number of control subjects with genotype. ^2^Data were calculated by logistic regression, adjusted for sex, age, smoking, and drinking status, where it was appropriate. ###### An allele-specific effect of rs2279115 genetic variant on *BCL-2* mRNA expression in esophagus tissues. *BCL-2* rs2279115 C\>A -------------------------- --------------- --------------- ------- Normal esophagus tissues 0.109 ± 0.021 0.083 ± 0.012 0.031 ESCC tissues 0.184 ± 0.045 0.153 ± 0.022 0.040 Abbreviations: ESCC, esophageal squamous cell carcinoma. [^1]: These authors contributed equally to this work.

###### Strengths and limitations of this study - The first large-scale study quantifying experiences of Cancer of Unknown Primary (CUP) patients compared with those with known primary cancers; - The first qualitative analysis of free-text comments drawn from a national sample of CUP patients; - The profile of CUP patients who responded to the Cancer Patient Experience Survey were found not to be representative of this patient population; - Reasons for sample limitations include: time between patient identification and survey participation; inconsistent administration of International Classification of Diseases 10th Revision codes; neglect of CUP typology; - These limitations highlight the need for prospective, observational cohort studies to investigate the experiential, informational and psychosocial issues faced by CUP patients. Introduction {#s1} ============ Patients with cancer of unknown primary (CUP) have metastatic malignant disease, for which an identifiable primary site has not been identified after extensive clinical evaluation. CUP is not a single disease, but rather a heterogeneous collection of cancer types with a wide variety of clinical presentations, which are hypothesised to share a common tendency to metastasise early.[@R1] CUP was the fifth most common cancer in the UK in 2014, accounting for 3% of new cancers and 6% of cancer deaths.[@R2] Worldwide it has been identified as between the sixth to eighth most common cancer, accounting for 2.3%--5% of all new cancer cases and is the third to fourth most common cause of death.[@R3] The UK National Institute of Clinical Excellence (NICE) published guidelines for the management of CUP patients in 2010,[@R5] which for the first time developed a taxonomy of definitions that reflected different phases of investigations for CUP: malignancy of undefined primary origin (MUO); provisional CUP (pCUP) and confirmed CUP ([table 1](#T1){ref-type="table"}). The guidelines also recommended the establishment of specialist CUP multidisciplinary teams (MDTs) in each National Health Service (NHS) cancer centre to include an oncologist, palliative care physician and clinical nurse specialist.[@R5] ###### Terms used in NICE guideline to define CUP ------------------------ ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- MUO Metastatic malignancy identified on the basis of a limited number of tests, without an obvious primary site, before comprehensive investigation Provisional CUP origin Metastatic epithelial or neuroendocrine malignancy identified on the basis of the histology or cytology, with no primary site detected despite a selected initial screen of investigations, before specialist review and possible further specialised investigations Confirmed CUP origin Metastatic epithelial or neuroendocrine malignancy identified on the basis of final histology, with no primary site detected despite a selected initial screen of investigations, specialist review and further specialised investigations as appropriate ------------------------ ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- CUP, cancer of unknown primary; MUO, malignancy of undefined primary origin; NICE, National Institute of Clinical Excellence. However, with the exception of three previous studies,[@R6] there is very little published research on quality of life (QoL) and psychosocial aspects of CUP. Patients with CUP struggle with uncertainty and distress, especially regarding prognosis,[@R6] possible recurrence and the primary's hereditary potential.[@R7] Problems with care continuity commonly faced by cancer patients were amplified for those with CUP, particularly in relation to coordination, accountability and timeliness of care.[@R7] A recent cross-sectional study from Greece, using matched-sample analysis, found patients with CUP experienced higher depression and anxiety and poorer QoL compared with those who have metastatic disease of either breast or colorectal cancer.[@R8] These findings suggest CUP patients may have unique psychosocial and supportive care needs, which require development of targeted supportive care interventions. Internationally, routine assessments of patient experiences of care are used increasingly to drive service quality improvements.[@R9] In the UK, the NHS Cancer Reform Strategy,[@R11] Outcomes Strategy for Cancer[@R12] and recent Cancer Taskforce[@R13] documents highlight the important role of patient experiences in measuring and improving clinical quality. The national Cancer Patient Experience Survey (CPES) is an extensive, UK-wide programme of research about cancer patients' experiences of care while undergoing inpatient or day-case treatment.[@R14] The CPES has been administered annually since 2010 and is the biggest survey programme of its kind in the world. The data are a public resource and available from the survey provider, Quality Health. This represented an opportunity to interrogate these data to understand CUP patients' experiences of care to underpin the development of targeted care interventions for this population. Drawing from the 2013 CPES, this paper reports findings from two related studies: qualitative and quantitative. The aim of both studies was to better understand the experiences of care among CUP patients. The quantitative study described perceived experiences of care of patients diagnosed with CUP compared with patients with metastatic cancer of a comparable known primary (non-CUP). The second, qualitative study analysed free-text responses of CUP patients in the CPES to identify emerging themes and insights regarding their experiences of cancer care. Methods {#s2} ======= Survey design {#s2a} ------------- The two studies reported here consist of secondary analyses of data collected as part of the English 2013 CPES, to assess differences in responses between CUP and non-CUP patients and to explore freetext responses of patients with CUP. The 2013 CPES was administered by the survey provider, Quality Health. The CPES questionnaire was mailed to all adult patients (aged ≥16 years) in England with a diagnosis of cancer, who had been admitted to an NHS hospital as an inpatient or day case patient over a 3-month period (1 September 2013--30 November 2013). Non-responders were sent one reminder letter and a further reminder letter with questionnaire if necessary. The overall response rate for the 2013 CPES was 64% (n=68 737).[@R14] Survey instrument {#s2b} ----------------- The CPES instrument contained 63 closed question items, measuring eight key areas of patient experience across the care trajectory from diagnosis to leaving hospital: access to care; respect for patients' preferences; information and education; physical comfort; emotional support; involvement of family and friends; continuity and transition and coordination of care. Three freetext comment boxes at the end of the questionnaire asked the following questions: Was there anything particularly good about your NHS care? Was there anything that could be improved? Any other comments? Identification of respondents {#s2c} ----------------------------- All NHS hospitals treating adult patients with cancer in England were included. Patients were identified from data provided by these organisation, selected from local patient administration systems. Patients were identified as CUP using the 10 revision of the International Statistical Classification of Diseases and Related Problems (ICD-10) codes: C77 (secondary and unspecified malignant neoplasm of lymph nodes), C78 (secondary malignant neoplasm of respiratory and digestive organs), C79 (secondary malignant neoplasm of other and unspecified sites) and C80 (malignant neoplasm, without specification of site).[@R15] Analysis {#s3} ======== Frequency matching analysis: matching procedure {#s3a} ----------------------------------------------- A frequency matching analysis was conducted to compare responses to the closed survey questions between CUP and non-CUP respondents. CPES (2013) respondents of 4535 were identified as CUP patients. The original plan was to match CUP and non-CUP samples for analysis on three variables: age group in deciles, sex and type of admission (ordinary admission, day case admission or regular day case admission). After exploration of the data, two further variables needed to be accounted for: tumour type and time since treatment start. Prevalence of breast and prostate cancer were high among non-CUP respondents to the CPES questionnaire, and it seemed inappropriate to match the CUP sample against a high proportion of such patients given estimates of likely site of origin in CUP cases from autopsy and biomarker studies.[@R16] For this reason, the dataset was restricted so these two tumour types represented only 5% each in the known primary sample. Preliminary analysis also found a marked difference between the CUP and non-CUP respondents regarding time since commencing cancer treatment ([figure 1](#F1){ref-type="fig"}). Patients with CUP were found disproportionally to have commenced treatment \>1 year prior to the survey (34% of CUP vs 73% of non-CUP); many more than would normally be expected among a random CUP sample and suggests a high proportion of favourable CUP subtypes.[@R4] Due to the small proportion of non-CUP patients who responded that it had been longer than 1 year since they began treatment, it was impossible to match on this variable, so the sample was restricted to only those patients who responded 'less than 1 year'. {#F1} The final sample included 1496 CUP patients who began treatment in the past year, with no missing data on the matching variables (age, sex and administration type). All patients were assigned identification and CUP patients were matched with randomly selected non-CUP patients, who had no missing data on the matching variables, began treatment in the past year and were diagnosed with metastatic disease in specific tumour types (colorectal, breast, head and neck, kidney/adrenal, prostate, pancreas and upper and lower gastrointestinal). These sites were selected as the primary sites from which CUP is most commonly thought to arise.[@R16] The final combined dataset contained 2992 respondents ([table 2](#T2){ref-type="table"}). ###### Demographic and clinical characteristics of the sample are shown in [table 2](#T2){ref-type="table"} Observation Free-text sample (n=3055) Marched-pair samples (2992) --------------------- --------------------------- ----------------------------- ------------- ------ ----- ------ Age  Mean (SD) 65.3 (11.3) 66.5 (11.7) 66.6 (11.5)  Median (IQR) 66 (58, 73) 67 (59, 75) 67 (59, 75)  Range 16, 95 20, 98 23, 94  Sex n \% n \% n \%  Male 1121 36.7 583 39.0 583 39.0  Female 1934 63.3 913 61.0 913 61.0 Diagnosis  CUP 3055 100 1496 100 --  Breast -- 111 7.4  Head and neck -- 185 12.4  Lung -- 271 18.1  Pancreatic -- 44 2.9  Prostate -- 52 3.5  Renal -- 89 5.9  Upper and lower GI -- 744 49.7 ICD-10 codes  C77 514 16.8 399 27 -- --  C78 1172 38.4 496 33 -- --  C79 1209 39.6 435 29 -- --  C80 160 5.2 166 11 -- -- CUP, cancer of unknown primary; ICD-10, International Classification of Diseases 10th Revision. Frequency matching analysis: analysis {#s3b} ------------------------------------- χ^2^ tests assessed associations between diagnosis type (CUP vs non-CUP) and responses to each item. Given the very large sample, the likelihood of finding statistically significant associations at p\<0.05 was high. It was therefore determined a priori that a 'small' or greater effect would be classified as meaningful; this corresponds to a Cramer's V of greater than 0.1 for comparisons where the degrees of freedom (df)=1, 0.07, where df=2, 0.06, where df=3,0.05 and where df=4 or 5.[@R17] Free-text analysis {#s3c} ------------------ Free-text responses provided by patients diagnosed with CUP were analysed to provide insights into their experiences of care. Comments were extracted from the CPES data set as individual text files and loaded into the NVivo qualitative data analysis software package. A coding framework for sorting free-text data from a previous study of responses to the Welsh CPES 2013 (WCPES)[@R18] was used. A sample of 200 randomly selected comments were double coded by two researchers (MB, RW) to ensure compatibility of the framework. Coding agreement between the two researchers was 80% (Cohen's Kappa). Any conflicts were resolved through discussion between coders. The framework was then used to categorise the free-text comments. Search criteria were developed for each category, using terms gleaned from term frequency and unique terms analyses of the coded data from the WCPES study. The search strategy was designed to identify relevant comments for each theme. During this process, all comments were read and coded. Following categorisation of comments, a second level of thematic analysis was conducted on specific free-text categories to provide greater depth of insight.[@R19] Ethics and data management {#s3d} -------------------------- CPES data is a public resource and available from the survey provider, Quality Health. Data were anonymised and secondary analysis of closed questions permitted with agreement from NHS England. Because free-text data may contain personal identifiable data, approval from NHS England to analyse CPES free-text data was separately secured. Free-text analysis of CPES data was approved by the University of Southampton Ethics Committee on 12 November 2014 (UoS Ethics ID:12313). Findings {#s4} ======== The demographic and clinical characteristics of respondents in both the matched analysis and free-text samples are shown in [table 2](#T2){ref-type="table"}. The mean age of samples were similar; 65.3 for the CUP free-text sample, 66.5 and 66.6, respectively, for the matched CUP and non-CUP samples. Matched samples contained equal numbers of men (39%, n=583) and women (61%, n=913), proportions were broadly similar for the free-text sample (36.7% (n=1121) and 63.3% (n=1934), respectively). Finally, the ICD-10 codes of respondents indicated c78 (33%, n=496) as the largest categorisation in the CUP matched sample and c79 (39.6%, n=1209) the largest in the free-text sample, while c80 was the smallest categorisation for both samples; 11% (n=166) and 5.2% (n=160), respectively. Patient experiences: closed responses {#s4a} ------------------------------------- Sixty comparisons were conducted. Of these, nine items showed differences between the CUP and non-CUP respondents that were classified as meaningfully different ([table 3](#T3){ref-type="table"}). In response to question 8 ('Beforehand, were you given written information about your (diagnostic) tests?'), CUP patients were more likely to respond 'No, but I would have liked written information about the test/s' or 'I did not need written information' and less likely to respond 'Yes, and it was easy to understand'. Similarly, Q9 'Were the results of your tests explained to you in a way you could understand?', CUP patients were less likely to respond 'Yes, completely'. In response to Q10 ('Who first told you that you had cancer?'), CUP participants were more likely to respond 'A GP' and less likely to respond 'A hospital nurse', with similar proportions indicating 'a hospital doctor'. In relation to Q13 ('Did you understand the explanation of what was wrong with you?'), CUP patients were less likely to answer 'Yes, I completely understood it' and more likely to respond 'Yes, I understood some of it' or 'No, I did not understand it'. For Q14 (When you were told you had cancer, were you given written information about the type of cancer you had?), CUP patients were less likely to respond 'Yes, and it was easy to understand' and more likely to respond 'No, I was not given written information about the type of cancer I had'. In response to Q16 'Do you think your views were taken into account when the team of doctors and nurses caring for you were discussing which treatment you should have?', CUP patients were less likely to answer 'Yes, definitely'. CUP patients were less likely to report having had surgery in the past 12 months (Q32), and in response to Q34 'Beforehand, were you given written information about your operation?', were less likely to respond 'Yes, and it was easy to understand' and more likely to respond 'No, I was not given written information about my operation'. Finally, CUP patients were more likely to report having received treatment from a lymphoedema specialist (Q66). ###### Matched paired analysis: CUP and non-CUP patients ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Item CUP\ Known primary n (%) Cramer's V n (%) ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ ----------- --------------------- ------------ Q8. Beforehand, were you given written information about your test(s)?  Yes, and it was easy to understand 808 (83) 917 (90) 0.102  Yes, but it was difficult to understand 37 (4) 32 (3)  No, but I would have liked written information about the test(s) 127 (13) 73 (7)  I did not need written information/Don't know/can't remember 392 334 Q9. Were the results of the test(s) explained to you in a way you could understand?  Yes, completely 961 (72) 1042 (78) 0.076  Yes, to some extent 322 (24) 259 (20)  No, but I would have liked an explanation 48 (4) 28 (2)  I did not need an explanation/Don't know/can't remember/Missing *33* *27* Q10. Who first told you that you had cancer?  A hospital doctor 1191 (81) 1199 (82) 0.104  A hospital nurse 44 (3) 91 (6)  A GP 155 (11) 104 (7)  Another health professional 40 (3) 51 (4)  A friend or relative 5 (0.3) 3 (0.2)  Nobody---I worked it out for myself 30 (2) 17 (1)  Missing 31 31 Q13. Did you understand the explanation of what was wrong with you?  Yes, I completely understood it 1006 (68) 1162 (78) 0.123  Yes, I understood some of it 438 (30) 313 (21)  No, I did not understand it 32 (2) 9 (1)  Can't remember/Missing 20 12 Q14. When you were told you had cancer, were you given written information about the type of cancer you had?  Yes, and it was easy to understand 670 (54) 834 (67) 0.132  Yes, but it was difficult to understand 104 (8) 91 (7)  No, I was not given written information about the type of cancer I had 466 (38) 326 (26)  I did not need written information/Don't know/can't remember/Missing *256* *245* Q16. Do you think your views were taken into account when the team of doctors and nurses caring for you were discussing which treatment you should have?  Yes, definitely 848 (61) 951 (68) 0.078  Yes, to some extent 343 (25) 284 (20)  No, my views were not taken into account 112 (8) 82 (6)  I didn't know my treatment was being discussed by a team of doctors/nurses 91 (7) 80 (6)  Not sure/can't remember/Missing *102* *99* Q32. During the last 12 months, have you had an operation (such as removal of a tumour or lump) at one of the hospitals in the covering letter?  Yes 824 (55) 1004 (67) 0.124  No 635 (42) 469 (31)  Missing 37 23 Q34. Beforehand, were you given written information about your operation?  Yes, and it was easy to understand 472 (64) 649 (71) 0.077  Yes, but it was difficult to understand 36 (5) 30 (3)  No, I was not given written information about my operation 229 (31) 234 (26)  Don't know/can't remember/Missing 87 91 Q66. Have you had treatment from any of the following (cancer specialists) for your cancer (patients were asked to tick as many as apply from the following list: physiotherapist; occupational therapist; dietician; speech and language therapist; lymphoedema specialist)  Yes 132 (9) 37 (3) 0.138 ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- CUP, cancer of unknown primary. Patient experiences: free text {#s4b} ------------------------------ 3055 CUP patients provided comments to one or more of the three free-text questions. The mean length of comments was 64.2 words, with female respondents providing longer comments than male respondents (54.2 words and 69.9 words, respectively). Comments were retrieved from the dataset for 17 categories ([table 4](#T4){ref-type="table"}). Positive comments were predominant in eight categories ('communication with patients', 'consultants', 'nursing', 'clinical nurse specialists', 'chemotherapy', 'radiotherapy', 'surgery' and 'palliative care'). Predominantly, negative comments were provided by participants for the remaining nine categories ('interagency communication', 'waiting for appointments/investigations', 'waiting on the day', 'receiving results of investigations', 'GPs', 'accident and emergency', 'emotional, social and psychological needs', 'financial concerns' and post-treatment care'). Ratios of negative to positive comments varied widely between categories, with the greatest proportion of negative comments reported for 'waiting time on the day', while the greatest proportion of positive comments was for 'palliative care' (though numbers were small). A trend existed within all themes that positive comments tended to be of a more general quality and scope than negative comments. Essentially, if participants were reporting a negative experience they provided more detail. ###### Comment categories with counts and ratios of positive and negative comments ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ Comment category Negative comments (n=) Positive comments (n=) Negative to positive ratio (n: 1) Overall ratio\ CUP dataset coverage (%) (+ve/--ve) -------------------------------------------- ------------------------ ------------------------ ----------------------------------- ---------------- -------------------------- 1\. Cross cutting themes  Interagency communication 351 139 2.38 -ve 15.3  Patient information 89 221 0.40 +ve 10.4  Waiting for appointments/investigations 88 72 1.24 -ve 5.2  Waiting on the day 299 12 24.9 -ve 10.2  Investigations---receiving results 165 37 4.46 -ve 6.32 2\. Healthcare professionals  GPs 220 91 2.41 -ve 6.32  Consultants 49 98 0.50 +ve 4.8  Nursing 284 409 0.69 +ve 22.7  Clinical nurse specialists 28 72 0.39 +ve 3.3 3\. Treatment specialisms  Accident and emergency 28 12 2.33 -ve 1.3  Chemotherapy 58 282 0.21 +ve 11.1  Radiotherapy 32 81 0.39 +ve 3.7  Surgery 170 350 0.49 +ve 17.0  Palliative care 2 40 0.05 +ve 1.3  Post-treatment care 38 32 1.19 -ve 2.3 4\. Other quality of life concerns  Emotional, social and psychological needs 39 17 2.29 -ve 1.3  Financial concern 75 7 10.71 -ve 2.7 ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ CUP, cancer of unknown primary; GPs, general practitioners. Three of the nine closed questions for which responses between CUP and non-CUP patients showed significant differences (Q10, Q13 and Q14) broadly mapped against particular free-text categories. Comments in these categories were subsequently thematically analysed in greater depth. Patient information support from health professionals {#s4c} ----------------------------------------------------- When investigated further, comments pertaining to --patient information (n=310) provided insights on patient responses to the closed questions 9, 13, 14 and 34; whether patients understood explanations given of their condition or their test results and whether they were provided with sufficient written information about their cancer or tests received. This free-text category had a net ratio of positive comments (0.4:1), but several themes were identified among negative comments (n=89) that indicated why patients found explanations of their condition difficult to understand and why they needed more information. For example, one CUP patient related her difficulty understanding why clinicians were unable to locate the primary tumour site, despite several investigations: > I had two liver biopsies (the first one did not have sufficient to discern whether benign or malignant). I had a scan while in hospital, also a CT scan and an MRI scan. On my discharge, I later received an appointment for an endoscopy when two small nodules were discovered in the gullet, a biopsy was taken -- they were benign, the tumour on the liver is a secondary and they still do not know where the primary is. (Female, 64 years) Patients were therefore sometimes uncertain or confused with regards the type of cancer they had and alluded to the frustration of not knowing and fears of progression: > Further detailed information of my cancer needed. Exactly where secondaries are and what kind of problems they could cause. Not knowing can be most distressing as you try to second guess things too much. (Female, 72 years) > They cannot find the primary source so it is rather upsetting as to how the disease is progressing. (Male, 69 years) Some patients wished to receive more information about what to expect from the diagnosis, but moreover, wanted an opportunity to discuss what this meant for them: > I would have liked to have had information about how my type of cancer usually progresses, plus a care plan for my own case. I would have liked to have had a chat, as a matter of course, to someone about the practical side of having cancer. (Male, 71 years) Nevertheless, despite receiving explanations of their conditions from health professionals, patients sometimes reported difficulties with understanding the terminology used: > The surgeon I see cannot explain things without using clinical terms I do not understand. He could if he tried! And so often I do not understand the detail. In addition he is relatively emotionless. (Female, 57 years) In addition, patients sometimes reported receiving conflicting information from the many doctors involved in their care: > Sometimes difficult to get answers as too many doctors and nurses dealing with me and not passing on the relevant information and then having to explain it all over again. (This happened many times.) (Male, 68 years) Several clinicians could be involved with a CUP patient's care, leading to conflicting information, as they were sometimes passed between clinical teams, as this comment suggests: > The only thing that could be improved was because it was a process of elimination I went from chest specialist, to organ oncologist and after my scans and biopsies finally back to my breast oncologist of 11 years ago. So my treatment didn't' start until end of March---6 months later. (Female, 62 years) GPs' diagnosis of CUP {#s4d} --------------------- Analysis of closed items indicated CUP patients were more likely than non-CUP patients to be given their diagnosis by a GP. Analysis of comments regarding interactions with GPs (n-311) provided by participants found CUP patients were more likely to report negative rather than positive experiences with GPs (ratio of 2.4:1). The theme containing most comments concerning GPs related to speed of diagnosis/referral for further investigations. Negative comments often indicated months and sometimes years of consulting GP services with cancer symptoms before diagnosis and/or specialist referral was made: > I wish my original GP had listened properly over the months I complained about weight loss (over two stone). Instead I had to change my GP who fast tracked me into hospital where a scan showed metastasized tumours. (Female, 81 years) Comments often related delayed referrals as a consequence of GPs attributing symptoms to conditions other than cancer or giving insufficient regard to previous cancer diagnoses. Some comments described 'misdiagnosis' of symptoms were later found to be inaccurate. > Very upset that I have to call the GP, a couple of times a week for 5 weeks, my pain getting worse by the day. My wife asked on a couple of visits if I could have x-ray, blood tests, GP said nothing wrong with me, it was just back ache. In the end my wife took me to A&E. She had to get me into wheelchair from the car. I was in terrible pain. (Male, 79 years) Poor communication between GP and secondary care services was also a prevalent topic within the 'GP' category and revealed a lack of continuity between GP and secondary care. It is important to note that these comments were often not critical of GP services but of the information provided to them by secondary services. > I should like my GP to be kept informed more quickly of my treatment at hospital. At the moment, information does not get to her quickly enough, so if I want to discuss something with her, she does not have up to date information, sometimes 3--4 weeks behind. (Female, 65 years) Discussion {#s5} ========== To our knowledge, these two investigations represent the first, large-scale studies quantifying the experiences of people with CUP compared with people with metastatic known primary cancers and the first qualitative analysis of free-text comments from participants drawn from a national sample of CUP patients. Experiences of patients with CUP were broadly similar to those with advanced metastatic cancer, with the exception of the nine areas reported. This might suggest the success of recommendations advocated by NICE in the UK that relate to the introduction of MDTs and specialist nurses dedicated to patients with CUP,[@R5] and that equivalent standards of care now exist irrespective of whether the primary site was known or unknown. However, it is not yet clear how consistently CUP MDTs have been established across the UK, and it is doubtful they were well established by 2013. Patients in the CUP sample were significantly more likely to answer they would have liked more information than the non-CUP sample about the type of cancer they had and the investigations they underwent and significantly less likely to understand the explanations of what was wrong with them. Providing accurate and helpful information and preparing CUP patients for their treatment trajectory is especially difficult given the uncertainty that pervades this diagnosis. The location of the primary tumour is the main reference point for treatment decisions and prognostic information,[@R20] treatment regimens may change several times during a patient's illness trajectory and patients often receive conflicting information from clinicians.[@R7] Doctors face challenges when communicating with CUP patients in the face of such uncertainty,[@R20] and consequently doctors often experience discomfort.[@R22] CUP patients also frequently undergo a greater number of investigations than patients with a known primary, as clinicians sometimes inappropriately 'chase the primary'.[@R23] Many doctors prefer to estimate a primary site on the basis of clinical signs but there is little consistency in the language used to describe the diagnosis to patients.[@R21] The number and diversity of health professionals involved with their care is often greater for CUP than non-CUP patients as they are often reviewed and moved between multidisciplinary clinical teams.[@R7] This can result in confusion and anxiety for patients,[@R6] and may partly explain why CUP patients reportedly felt their views were less often taken into account during treatment decision-making. CUP patients were significantly more likely to be first told by their GP that they had cancer, compared with the non-CUP sample. It may be that, as CUP patients are often not adopted by a specific MDT or are 'bounced' between MDTs,[@R7] GPs may take responsibility for relaying the diagnosis to the patient on the basis of accumulating evidence, possibly prior to referring them to specialist consultation. Nevertheless, free-text responses also found CUP patients were more likely to report negative rather than positive experiences of interactions with GPs (ratio of 2.4:1), which often related to referrals being delayed as a consequence of GPs attributing symptoms to conditions other than cancer or not sufficiently investigating patients' health concerns. That referral systems for MUO from primary to secondary care is variable across the UK may also contribute to such delays. CUP patients were significantly less likely to have surgery; by definition, patients with CUP are diagnosed after a primary has metastasised and hence surgery is not usually an option. In contrast, those patients with an advanced known primary may well have had surgery prior to the discovery of metastases. Finally, the finding that patients with CUP are more likely to have seen a lymphoedema specialist is difficult to explain and may be an artefact of the limitations of the sample as discussed below. A major limitation of this study is that the CUP sample is not representative of the profile of patients with CUP. During 2006--2010, the 1-year relative survival for CUP patients in the UK was 16%,[@R25] yet in this sample 62% of respondents with CUP had begun treatment for their cancer more than 1 year previous to their diagnosis. Potentially, many of the patients in the CUP sample may have a favourable subtype of CUP: those with likely primary sites that have greater survival rates and reliable treatments,[@R4] who may be still receiving follow-up care at hospital in the years that followed their original diagnosis. We attempted to control for this potential bias by restricting the CUP sample to those who were first treated for their cancer less than 1 year previously. Nevertheless, these people were still well enough to complete the CPES several months postdischarge. Speculatively, the CUP sample in this analysis may be more like patients with advanced known cancer than a CUP sample recruited prospectively after diagnosis of confirmed CUP, which would have included many more individuals with poorer prognoses. Indeed, while almost half (49%) of patients diagnosed with CUP in 2009 were coded to the ICD-10 code C80, which also accounted for 93% of deaths from CUP in 2010,[@R26] a much smaller proportion of patients coded C80 were included in the CPES CUP population sample. Therefore, a prospectively recruited sample of CUP patients might have less positive experiences than the CUP sample in this analysis. Moreover, as we selected patients with a known primary for the non-CUP sample to approximate proportions of likely site of origin in CUP cases from autopsy and biomarker studies,[@R16] we have enriched the non-CUP sample with patients who have poorer prognoses, such as those with pancreatic and lung cancers. These patients may have less positive experiences than a more omnibus sample of patients with advanced known primaries. Compounding these issues, the diagnoses used in this study were not self-reported or clinician reported but based on administrative data. The ICD-10 classification system does not differentiate between MUO, pCUP and cCUP.[@R5] For CPES, the ICD code was extracted directly from each Trust's administrative data records, and two errors may subsequently occur with this method. First, as ICD codes are generated by MDT administrators and not by doctors, there may be errors in coding. The second, more serious issue is that because no distinction is made in the data capture between MUO, pCUP and cCUP, it is possible that if a primary site is later confirmed the coding will not be updated in the administrative file. As the survey maybe completed by patients some months postdiagnosis, some patients within our CUP sample may have already received a site-specific diagnosis, and the sample is therefore not fully representative of the profile of patients with MUO/CUP. Errors in classifying CUP are likely to be more pronounced for the CUP sample due to the uncertainty of whether or when a primary might be found and because patients with MUO can pass among several multidisciplinary teams managing different cancer types.[@R7] Finally, using administrative data to define the CUP sample also fails to account for the patient's perception of their disease. Indeed, the limited research that exists indicates that patients are often very confused about a diagnosis of CUP and for many patients clinicians convey a 'best guess' primary site to direct treatment or access subsidised medical therapy.[@R21] What and how they are told of their diagnosis may influence their questionnaire responses. Conclusion {#s6} ========== These findings, when combined, suggest CUP patients may experience delays in diagnosis and to specialist referral and greater uncertainty with regards understanding their diagnosis and be less prepared for what to expect regarding diagnostic investigations than patients with a known primary site. These represent specific areas where targeted psychoeducational interventions might be developed. But, given we identified significant limitations with the CPES CUP sample data, possibly as a consequence of inconsistent or erroneous coding and neglect of the NICE taxonomy of CUP, it is crucial to conduct prospective, observational cohort studies to develop a more complete understanding of the issues faced by CUP patients. Supplementary Material ====================== ###### Reviewer comments ###### Author\'s manuscript We would like to thank Dr Karla Gough, Senior Research Fellow, Department of Cancer Experiences Research, Peter MacCallum Cancer Centre, who assisted with analysis. **Contributors:** PS, RW and AR conceived the idea of the study and acquisition of the data. PS was responsible for the design of the quantitative study, RW for design of the qualitative study. AD was responsible for undertaking quantitative data analysis. MB and RW were responsible for undertaking analysis of qualitative data. PS, LM, AR, JS, AD, MB and RW were all involved with interpretation of the results from both studies. The initial draft of the paper was produced by RW and MB and circulated repeatedly between all authors for critical revision. All authors read and approved the final manuscript. **Funding:** This work was supported by the Cancer of Unknown Primary (CUP) Foundation and Cancer Australia. Additional funding was provided by the 'Adventures in Research' funding stream at the University of Southampton. **Competing interests:** None declared. **Provenance and peer review:** Not commissioned; externally peer reviewed. **Data sharing statement:** Source data for the study are closed question responses and free text responses to the Cancer Patient Experience Survey in England for 2013. These data are available from the survey provider (Quality Health, UK) <https://www.quality-health.co.uk/> ---email: info\@quality-health.co.uk.

1. Introduction {#s0005} =============== The genus *Leontopodium* (R.Br. ex Cassini) belongs to the Asteraceae family and comprises about 30 species [@bb0005]. The well-known alpine Edelweiss (*Leontopodium nivale* ssp. *alpinum* (Cass.) Greuter or *L. alpinum* Cass.) is found in the mountain regions of Europe, mostly in the Alps, Carpathians, Pyrenees, the Tatra and the Balkan Peninsula [@bb0010]. Edelweiss has been used in traditional medicine since centuries against ailments like diarrhea, abdominal aches, bronchitis or fever [@bb0015; @bb0020]. Recent studies revealed that extracts from aerial parts and roots possess anti-inflammatory [@bb0025; @bb0030] and analgesic [@bb0030] activities. Phytochemical research has revealed the occurrence of various secondary compounds like diterpenes [@bb0035], sesquiterpenes [@bb0040; @bb0045], benzofuranoids [@bb0050] and lignans [@bb0035; @bb0050]. In the latter class the metabolite, leoligin, has recently been isolated [@bb0050]. This lariciresinol-type lignan (see [Fig. 1](#f0005){ref-type="fig"}) has been shown to inhibit in vitro leukotriene biosynthesis and intimal hyperplasia of venous bypass grafts, seemingly without toxic side effects [@bb0055]. It inhibits in vivo neointima formation without causing endothelial damage, and it is not thrombogenic [@bb0055]. Vein graft disease, i.e. the progressive degeneration of veins used in surgical bypass operations, is characterized by endothelial damage, smooth muscle cell proliferation and pro-inflammatory signaling [@bb0055; @bb0060]. Drug eluting stents are considered to be the most likely approach in the prevention of graft failure, but currently there is a lack of drugs with specific activity. Leoligin is considered to have a great potential for the treatment of vein graft disease [@bb0055]. The structurally related compound 5-methoxy-leoligin (see [Fig. 1](#f0005){ref-type="fig"}) has recently been shown to be a very promising candidate for the development of the first low molecular weight pro-angiogenic and pro-arteriogenic drug for the treatment of myocardial infarction [@bb0065]. Myocardial infarction is a major cause of mortality worldwide. Recent treatment strategies in the therapy of myocardial infarction aim at the improvement of ventricular function, among others by stimulating angiogenesis and arteriogenesis, namely the induction of artery growth to bypass occluded arteries [@bb0065]. In this, there is an urgent need to find new drugs, and 5-methoxy-leoligin has shown to possess corresponding pro-angiogenic and pro-arteriogenic activities. Edelweiss is a protected plant in many countries. While it is cultivated in large quantities in Switzerland [@bb0070], the isolation of relevant amounts from Edelweiss roots remains a laborious task due to the low content [@bb0050; @bb0075] and the thin and fibrous nature of the roots of cultivated plants [@bb0080]. As the chemical synthesis has not yet been described the biotechnological production might be an alternative approach to the procurement of relevant amounts of these lignans. Hairy root cultures, i.e. in vitro cultured roots which result from the infection (transformation) of higher plants with the soil-born bacterium *Agrobacterium rhizogenes*, have been investigated for a few decades as biological systems for the production of secondary compounds from medicinal plants [@bb0085]. Hairy roots can in many cases produce the same compounds found in normal roots of the parent plant, but while in callus or cell suspension cultures this productivity frequently diminishes over time it remains stable in transformed roots [@bb0090]. Furthermore, in the recent past hairy root technology has been significantly improved concerning accumulation and excretion of secondary metabolites after elicitation, and scale-up of the culture process [@bb0085]. With regard to the production of lignans from hairy roots a number of studies have focused on podophyllotoxin and its derivatives, an important lead for anticancer drugs. For example, hairy root lines of *Linum album* produced 105 μg/L [@bb0095] or 5.12 mg/L [@bb0100] of the lignan while a content of 14.11 mg/L was measured in the roots of wild growing plants [@bb0095]. In *Linum flavum*, hairy roots were reported to contain up to 3.5% 5-methoxypodophyllotoxin [@bb0105] which is comparable to the amount of 3.68% described for roots of greenhouse grown plants [@bb0110]. Another arylnaphthalene lignan, justicin B, is found in normal root cultures (12.5 mg/L) and hairy root cultures (16.9 mg/L) of *Linum austriacum*[@bb0115]. Silymarin is a flavonolignan complex with hepatoprotective properties isolated from the fruits of the milk thistle plant, *Silybum marianum*. While hairy roots contained more isosilybin A and B than untransformed root cultures, the content of four other flavonolignans was higher in the latter culture type [@bb0120]. It has been demonstrated previously that *L. nivale* ssp. *alpinum* can be transformed with *A. rhizogenes*, and resulting hairy root lines were shown to produce anthocyanins, hydroxycinnamic acid esters, and essential oil [@bb0125]. In the present study a hairy root clone of Edelweiss was investigated specifically in respect of its leoligin and 5-methoxy-leoligin content and the influence of the treatment with elicitors, molecules that stimulate defense or stress-induced responses in plants [@bb0130], on product formation. 2. Experimental {#s0010} =============== 2.1. Plant material {#s0015} ------------------- Seeds of *L. nivale* ssp. *alpinum* were purchased from Austrosaat AG (Vienna, Austria) and were surface sterilized for 30 min with an aqueous NaOCl solution (3.4% active chlorine). They were then aseptically germinated on modified semisolid half-strength Murashige and Skoog (MS) medium [@bb0135] supplemented with 1% sucrose and 0.3% Gelrite® (Carl Roth, Karlsruhe, Germany). Shoots of 4 weeks old seedlings were subsequently transferred to modified semisolid MS medium with 3% sucrose, 0.4 mg/L thiamine hydrochloride, 80 mg/L myo-inositol, 100 mg/L caseine hydrolysate, 0.7% agar (Merck, Darmstadt, Germany), 0.55 μM 1-naphthaleneacetic acid and 0.25 μM kinetin [@bb0140] for further multiplication. Every 4 weeks shoot clusters were divided into single shoots which were transferred to fresh medium. All cultures were kept at 25 ± 1 °C under a 16 hour photoperiod with a light intensity of 40 μM · m^− 2^ · s^− 1^ (Sylvania Gro-Lux® fluorescent tubes). A number of seedlings of this seed batch were also potted into gardening soil and grown to the flowering stage. Voucher specimens and living plants are kept at the Department of Pharmacognosy, University of Vienna. Three normal root samples of cultivated *L. nivale* ssp. *alpinum* plants were obtained from the Mediplant Swiss Research Centre on Medicinal and Aromatic Plants, Conthey, Switzerland (two samples) and from W. Faulhammer, Innsbruck, Austria (one sample). Vouchers are stored at the herbarium of the Institute of Pharmacy/Pharmacognosy, University of Innsbruck. 2.2. Hairy root culture {#s0020} ----------------------- *A. rhizogenes* strain ATCC15834 was grown in liquid YMB medium [@bb0145]. The middle veins of leaves of in vitro grown Edelweiss shoots were gently scratched with a scalpel dipped in the bacterial suspension and the shoots where then cultivated on modified semisolid half-strength MS medium with 3% sucrose in the light. Single roots which emerged from the wounded sites after 4 weeks in average were dissected from the leaves and treated twice for 10 days each with liquid modified MS medium containing 3% sucrose and 500 mg/L of the antibiotic, cefotaxim. Subsequently the hairy roots were routinely transferred to liquid modified MS medium with 3% sucrose every 4 weeks: An inoculum of ca. 0.5 g was transferred to 50 mL of medium in a 250 mL Erlenmeyer flask and cultivated at 25 ± 1 °C in the dark on a rotary shaker at 100 RPM. The root line K8A was used for the elicitation experiments. 2.3. Elicitor treatments {#s0025} ------------------------ Elicitors were added to 3 weeks old hairy root cultures except for the treatments with elevated sucrose concentrations which occurred for the whole culture period. Silver nitrate (Merck, Darmstadt, Germany, analytical grade) was dissolved in distilled water and the filter sterilized (0.22 μm) solution was added to hairy root cultures at final concentrations of 15, 30 and 60 μM AgNO~3~, respectively. Similarly, filter sterilized solutions of methyl jasmonate (MeJa; Sigma--Aldrich, St. Louis, USA; 95% purity) in 96% ethanol were added to achieve final concentrations of 50, 100, 200 and 300 μM. Yeast extract (YE; Sigma--Aldrich, St. Louis, USA) was purified by dual precipitation with ethanol as described by Hahn and Albersheim [@bb0150]. After autoclaving (20 min at 121 °C) of the resulting aqueous solution aliquots were added to hairy root cultures at final concentrations of 1, 2 or 5 g/L. For osmotic treatment, hairy roots were inoculated in liquid modified MS media supplemented with 5, 6 or 7% sucrose. Root materials were always harvested after a total cultivation period of 4 weeks and were dried at room temperature. 2.4. HPLC-UV analysis {#s0030} --------------------- ### 2.4.1. Sample preparation {#s0035} A small amount of ground root sample was weighted (100.0 mg). The plant material was sonicated with 20 mL of dichloromethane for 10 min in the ultrasonic bath before being filtrated over cotton wool. This manipulation was repeated two times with 10 mL of dichloromethane each. After evaporation of the solvent of the combined extracts at reduced pressure the dry residue was dissolved in 1.00 mL of methanol, again filtered over cotton wool and analyzed by HPLC-UV. The efficacy of the extraction protocol was proved by a fourth extraction with 10 mL dichloromethane. The extract was evaporated separately, dissolved in 1.00 mL MeOH and analyzed by HPLC-UV. Since no signals of leoligin and 5-methoxy-leoligin were found, extraction was found to be exhaustive. Two extracts of each hairy root culture sample were prepared. Both were analyzed three times, and the six values of the peak area were used to perform the quantification of leoligin and 5-methoxy-leoligin in the samples. ### 2.4.2. HPLC-UV method and quantification parameters {#s0040} An HP 1050 system (Agilent, Waldbronn, Germany) equipped with auto sampler, DAD and column thermostat was used. The stationary phase was a Phenomenex Kinetex 2.6 μ C18 100 A (100 mm × 2.1 mm) with 2.6 μm particles equipped with the corresponding guard column. The temperature was set to 40 °C. The mobile phase consisted of two solvents: A = water, B = acetonitrile. The composition during run was set as following: 0 min: 65% A, 35% B; 20 min: 50% A, 50% B; 25 min: 1% A, 99% B; stop: 35 min; post time: 15 min. The flow rate was 0.250 mL/min and the detection wavelength was 205 nm. As standard for quantification pure leoligin and 5-methoxy-leoligin were dissolved in 1.00 mL of methanol and diluted to obtain five reference solutions (73.1 μg/mL to 4.6 μg/mL; 54.4 μg/mL to 3.4 μg/mL), which were analyzed to afford a calibration curve of y = 230818.83405x + 496.66528; R^2^ = 0.99819 for leoligin as well as y = 190969.00754x − 47.98333; R^2^ = 0.9995219 for 5-methoxy-leoligin. ### 2.4.3. Chemicals and reagents {#s0045} All solvents (dichloromethane, methanol and acetonitrile) were of analytical grade (99.9%) provided from Merck (Darmstadt, Germany). The used water was produced via reverse-osmosis. Leoligin and 5-methoxy-leoligin had a purity level higher than 98%, according to LC-DAD/MS and NMR examination. These references were prepared at the Institute of Pharmacy/Pharmacognosy of the University of Innsbruck [@bb0035]. 3. Results and discussion {#s0050} ========================= The hairy roots line K8A which was investigated in this study was chosen due to its fast growth. When cultivated under standard conditions, i.e. in liquid modified MS medium with 3% sucrose, it yielded 0.0062% LG and 0.0049 MLG ([Table 1](#t0005){ref-type="table"}). In many cases, secondary metabolite accumulation in hairy root cultures can be enhanced by elicitation, i.e. treatment of the culture with biotic and abiotic elicitors [@bb0155]. We therefore treated our hairy root line with the two abiotic elicitors silver nitrate and sucrose, and the two biotic elicitors yeast extract and methyl jasmonate. As elicitation can, despite an improved product yield, also result in decrease of biomass accumulation [@bb0160], hairy root growth was assessed in terms of final dry weight, too. Silver nitrate at a concentration of 15 μM led to a ca. 5-fold increase of the levels of both LG and MLG, but on the other hand root growth was negatively influenced (about 30% less biomass). Elicitation with 30 or 60 μM AgNO~3~ had a less pronounced and not significant impact on lignan biosynthesis. Similarly, enhanced production of silymarin in hairy root cultures of *S. marianum* has also been achieved through the use of silver nitrate [@bb0165]. A treatment with 30 or 60 μM of this elicitor lowered the total phenolics content in hairy roots of *Salvia miltiorrhiza*, but slightly stimulated the formation of these metabolites at a concentration of 15 μM [@bb0155]. But, in the same culture type Ge and Wu [@bb0170] found Ag^+^ to stimulate the production of tanshinones, and silver nitrate significantly increased the accumulation of the tropane alkaloids, scopolamine and hyoscyamine in hairy roots of *Brugmansia candida*[@bb0175]. Beside its role as carbon source in in vitro cultures, sucrose at elevated concentrations can act as abiotic elicitor due to increased osmotic pressure. We therefore investigated the effect of this sugar at concentrations of 5, 6 or 7%. While lignan formation was not significantly enhanced when 5 or 7% sucrose was used, at 6% the content of LG and MLG increased to 0.0678% and 0.0372%, respectively, which was 10.9-fold and 7.6-fold higher than that in the control roots with the standard sucrose concentration of 3%. Hairy root growth was not significantly impaired at any of the tested sucrose concentrations. An elevated sucrose concentration of 6% also stimulated glycyrrhizin production in hairy roots of *Glycyrrhiza inflata*[@bb0180]. In hairy roots of *Withania somnifera*, accumulation of the steroidal lactone, withaferin A was enhanced by treatment with 4% but not with 6% sucrose [@bb0185]. Plant cell suspension cultures of *Melastoma malabathricum* produced more anthocyanins with 4.5--7.5% sucrose than with 3% [@bb0190], and using 5% sucrose the formation of rosmarinic acid in suspension cultures of *Ocimum sanctum* was higher than with 4 or 7% of the sugar [@bb0195]. Yeast extract (YE) and preparations thereof have been widely used as elicitors in plant tissue culture [@bb0200]. When purified YE [@bb0150] was added to cultures of the Edelweiss hairy root clone K8A, 1 or 5 g/L did not significantly influence neither lignan formation nor root growth. At a concentration of 2 g/L a 5-fold (LG) and 4-fold (MLG) promotion of product formation was observed, while biomass increase was not affected. In cell cultures of *S. miltiorrhiza* 100 mg/L but not 50 or 200 mg/L YE led to the highest tanshinone yield [@bb0205]. In hairy root cultures of the same plant species rosmarinic acid formation was enhanced similarly irrespective of the concentration of YE elicitor, while cryptotanshinone biosynthesis was improved in a dose dependent manner [@bb0210]. As a second biotic elicitor we chose methyl jasmonate (MeJa) which has been shown to stimulate secondary metabolite formation in a wide variety of plant in vitro cultures. In particular, the production of lignans was shown to rise in cell suspension cultures of *Forsythia* x *intermedia*[@bb0215] and *L. album*[@bb0220] upon treatment with MeJa. Elicitation of Edelweiss hairy root clone K8A with 50 or 100 μM MeJa resulted in accumulation of up to 0.0498% LG and 0.0188% MLG (8-fold and 3.8-fold higher than in the control, respectively), although at 100 μM MeJa root growth was significantly suppressed. Higher concentrations (200 or 300 μM MeJa) showed no influence on biomass production but enhanced lignan formation only to a moderate non significant extent. The leoligin content of wild growing Edelweiss has been described to account for 0.005--0.010% [@bb0075]. Within the present study, analysis of two samples of cultivated *L. nivale* ssp. *alpinum* from Switzerland (Mediplant Swiss Research Centre on Medicinal and Aromatic Plants) revealed a content of 0.0155 and 0.0547% LG and 0.1293 and 0.0672% MLG, respectively. Another sample from field culture in Innsbruck, Austria yielded 0.0632% LG and 0.0421% MLG. Although no further data concerning the lignan content in Edelweiss are available so far, a variability regarding both the content and the ration of LG to MLG seem to be likely. Furthermore, it is not clear whether the growing conditions (wild vs. controlled cultivation) have an influence on lignan production. Anyhow, the basic LG content of the hairy root clone K8A used in the present study was found to be of the order of wild growing Edelweiss [@bb0075]. Similarly, untransformed roots and hairy roots of *L. flavum* accumulated comparable amounts of the lignan, 5-methoxypodophyllotoxin [@bb0105; @bb0110]. As delineated above, lignan contents in hairy roots have been described to be higher, lower or equal to that in normal roots depending on factors like e.g. plant genotype or *Agrobacterium* strain used for transformation. The production of LG could be significantly increased nearly 11-fold through elicitation with 6% sucrose. Similarly, the MLG content of clone K8A could be enhanced 7.6-fold, and elicitation had no significant impact on biomass accumulation. 4. Conclusions {#s0055} ============== Only recently the highly active lignan compounds leoligin and 5-methoxy-leoligin have been isolated from the roots of the alpine Edelweiss, casting new light on this tradition-steeped plant. As an alternative to the laborious extraction of the low amounts found in the plant, we could show that a transformed hairy root line of Edelweiss accumulates the valuable lignans in concentrations which, upon elicitation with 6% sucrose, resembles the content in normal roots. Analyses of three samples of normal roots revealed a degree of variation in both total lignan content and ratio of the two compounds. Future work will therefore focus on the establishment of further hairy root clones starting from high yielding Edelweiss genotypes, and using additional *A. rhizogenes* strains. In any case, as a continuous, sustainable and renewable production system independent of climatic or environmental effects [@bb0225] the biotechnological production of leoligin and 5-methoxy-leoligin would be of advantage over the extraction from field grown plants. The authors wish to thank F. Rayé, K. Ondratschek, S. Prisching and K. Schmidtbauer for their technical support as well as C. Carlen and W. Faulhammer for providing plant material. This study was supported by the Austrian Science Fund (FWF), NFN Projects S10703 and S10706. {#f0005} ###### Final biomass (g dry wt.) and contents (% dry wt.) of leoligin and 5-methoxy-leoligin in hairy root clone K8A of *Leontopodium nivale* ssp. *alpinum* treated with various elicitors. Sample/treatment Final biomass[⁎](#tf0005){ref-type="table-fn"} (g dry wt.) Leoligin (w%)[⁎⁎](#tf0010){ref-type="table-fn"} 5-methoxy-leoligin (w%)[⁎⁎](#tf0010){ref-type="table-fn"} ------------------ ------------------------------------------------------------ ------------------------------------------------- ----------------------------------------------------------- K8A control 0.62 ± 0.03^ade^ 0.0062 ± 0.0021^a^ 0.0049 ± 0.0014^ab^ 15 μM AgNO~3~ 0.44 ± 0.03^bc^ 0.0321 ± 0.0094^bc^ 0.0260 ± 0.0075^e^ 30 μM AgNO~3~ 0.43 ± 0.02^bc^ 0.0183 ± 0.0021^ab^ 0.0153 ± 0.0019^bcd^ 60 μM AgNO~3~ 0.36 ± 0.03^b^ 0.0217 ± 0.0015^ab^ 0.0177 ± 0.0014^cde^ 5% Sucrose 0.64 ± 0.04^ae^ 0.0142 ± 0.0032^ab^ 0.0083 ± 0.0018^abc^ 6% Sucrose 0.59 ± 0.03^ade^ 0.0678 ± 0.0042^d^ 0.0372 ± 0.0025^f^ 7% Sucrose 0.70 ± 0.05^a^ 0.0221 ± 0.0062^ab^ 0.0101 ± 0.0027^abcd^ 1 g/L YE 0.69 ± 0.04^a^ 0.0192 ± 0.0054^ab^ 0.0118 ± 0.0028^abcd^ 2 g/L YE 0.61 ± 0.05^ade^ 0.0337 ± 0.0094^bc^ 0.0204 ± 0.0050^de^ 5 g/L YE 0.59 ± 0.05^ade^ 0.0035 ± 0.0016^a^ 0.0019 ± 0.0008^a^ 50 μM MeJa 0.50 ± 0.05^cd^ 0.0316 ± 0.0097^bc^ 0.0116 ± 0.0028^abcd^ 100 μM MeJa 0.45 ± 0.05^bc^ 0.0498 ± 0.0152^cd^ 0.0188 ± 0.0050^cde^ 200 μM MeJa 0.54 ± 0.04^cde^ 0.0249 ± 0.0037^ab^ 0.0098 ± 0.0013^abcd^ 300 μM MeJa 0.59 ± 0.03^ade^ 0.0173 ± 0.0035^ab^ 0.0113 ± 0.0021^abcd^ Values are mean ± S.E. (n = 5) per culture flask with 50 mL nutrient medium after 4 weeks of culture; means followed by the same letter are not significantly different (p = 0.05) according to Duncan\'s multiple range test. Values are mean ± S.E. (n = 2, each measured 3 times); means followed by the same letter are not significantly different (p = 0.05) according to Duncan\'s multiple range test.

INTRODUCTION {#s1} ============ Cancer initiation and development are associated with the accumulation of numerous genetic alterations in the cancer genome. These alterations include both small-size mutations and large-scale genomic alterations consisting of copy number variants (CNVs - deletions, duplications or amplifications), as well as copy-number-neutral genomic rearrangements (inversions or translocations). Interactions between these alterations (in certain situations, in addition to germline mutations) allow cancer to clonally evolve due to deactivation of tumor suppressor genes (loss-of-function mutations) and activation of oncogenes (gain-of-function mutations). Lung cancer is the leading cause of cancer-related death (<http://www.who.int/mediacentre/factsheets/fs297/en/>; \[[@R1]\]). There are several subtypes of lung cancer, the most common of which is non-small-cell lung cancer (NSCLC). NSCLC can be further divided into adenocarcinoma, squamous-cell carcinoma, and large-cell carcinoma. Lung cancer occurs predominantly in smokers (\>60%). Regardless of histological and risk-factor divisions, lung cancer is one of the most genomically heterogeneous type of cancer. Recently, several whole-genome sequencing projects utilizing next-generation sequencing technologies revealed the presence of thousands of small-size mutations in the individual lung cancer genome \[[@R2]--[@R5]\], with an almost 10 times higher frequency of mutations in smoker than in non-smoker samples \[[@R6]\]. An even higher level of variation seems to be attributed to copy number alterations. It was shown with the use of SNP-array-based analysis that approximately 50% of the lung cancer genome undergoes recurrent copy number alterations \[[@R7]\]. On average, over 40% of the genome undergoes copy number alteration in individual lung cancers \[[@R8]\]. However, only a small fraction of alterations occurring in cancer genomes are functional ("driver") mutations; others are "passenger" mutations that occur as a consequence of the general cancer genome destabilization. Although "passenger" mutations are not critical for cancer genome evolution, they are often selected in parallel with closely located or commonly regulated targets of "driver" mutations. The role of "passenger" mutations for particular cancers is mostly unknown (it is not necessarily neutral). A substantial progress in lung cancer treatment (especially adenocarcinomas) has been made recently due to personalized therapy based on genomic biomarkers. The distinctive biomarkers in lung cancer are mutations in the *epidermal growth factor receptor* (*EGFR*) \[[@R9]\] or gain-of-function translocations and inversions involving the *anaplastic lymphoma receptor tyrosine kinase* (*ALK*) \[[@R10]\]. However, the general prognosis of lung cancer is still poor and its 5-year survival is one of the lowest among cancer patients at approximately 10%. Therefore, many lung cancer studies are currently focused on understanding the impact of genetic alterations on cancer biology and development and on the identification of new prognostic biomarkers. Among the most intensively studied candidate biomarkers are microRNAs (miRNAs), a class of short (∼21 nt long), single-stranded, noncoding RNAs. MiRNAs are primarily involved in the post-transcriptional regulation of gene expression, either by mRNA degradation or inhibition of translation efficiency \[[@R11], [@R12]\]. Mature miRNAs are generated in two subsequent steps from long primary precursors (pri-miRNAs). Pri-miRNAs are encoded either by independent transcriptional units or by protein-coding genes. In the first step of miRNA biogenesis that takes place in the nucleus, the secondary precursor (∼60 nt long pre-miRNA), which adopts a hairpin structure, is cleaved out from pri-miRNA by the nuclease DROSHA. Upon export to the cytoplasm, the pre-miRNA is further processed into a miRNA-duplex by the nuclease DICER. One of the miRNA-duplex strands is released, and the other becomes the mature miRNA that, as a key element of the miRNA-induced silencing complex (miRISC) recognizes complementary target sequences usually located within the 3′ untranslated regions of mRNAs. The biological functions of most miRNAs identified so far (miRBase; <http://www.mirbase.org>; \[[@R13], [@R14]\] remain unknown. However, it has been well documented that miRNAs downregulate numerous genes and either stimulate or inhibit many important biological processes and diseases, including cell proliferation and differentiation, apoptosis, development and cancer \[[@R15]--[@R18]\]. The role of miRNAs in the development of cancer was first identified in chronic lymphocytic leukemia in 2002 \[[@R19]\]. Since then, it has been shown that overexpression or downregulation of certain miRNAs contributes to the development, progression and metastasis of many types of cancer. Such miRNAs can therefore be classified as either oncogenes (oncomirs) or tumor suppressors \[[@R20]\]. It has also been shown that some miRNAs, such as *miR-21*, *miR-205* or *miR-155*, seem to be universal for different cancers \[[@R12]\]. There have been numerous studies of miRNA expression in lung cancer, and many miRNAs that are specifically over- or underexpressed in lung cancer or in particular lung cancer subtypes were identified. For example, it was shown that 6 miRNAs constituting the polycistronic miRNA cluster, *miR-17/92*, are overexpressed in lung cancer and enhance cell proliferation \[[@R21]\]. It was later shown that an elevated level of these miRNAs may be detected in the plasma of lung cancer patients \[[@R22], [@R23]\] and is associated with poor disease prognosis \[[@R24]\]. Other miRNAs consistently found to be either overexpressed or underexpressed in lung cancer are *miR-21*, *miR-210* and *miR-126*. However, it should be noted that substantial discordances between miRNA profiling results also exist. Although the functional relevance of some of the miRNAs that are differentially expressed in lung cancer has been demonstrated (e.g., \[[@R25]--[@R27]\]), the roles of most of these miRNAs in cancer are unknown or poorly recognized. One factor that may shed more light on the role of particular miRNAs in cancer is the mechanism underlying their aberrant expression. Among the most pronounced mechanisms underlying aberrant expression in cancer are point mutations, epigenetic modifications and copy number alterations. However, it has been suggested that point mutations and epigenetic modifications are not important factors in the global miRNA regulation in lung cancer \[[@R24], [@R28]\]. It has also been shown that miRNA genes are overrepresented and cluster in genomically fragile sites and other regions that undergo frequent copy number changes in cancer genomes. Thus, it has been suggested that somatic copy number variation may lead to the activation/deactivation of miRNAs in cancer \[[@R29], [@R30]\]. For example, systematic analysis of three cancer types (ovarian, breast, and melanoma) with the use of comparative genome hybridization microarrays showed that 37% (ovarian) to 89% (melanoma) of analyzed miRNA genes undergo copy number changes \[[@R30]\]. There are known examples of both miRNA- and protein-coding genes whose expression in cancer is either increased or decreased by their copy number variation. These high-copy-number amplifications and recurrent deletions (loss of heterozygosity) are often used as a confirmation of oncogenic and tumor-suppressive function of the analyzed gene, respectively. The role of copy number variations in the regulation of miRNAs in cancer and the potential cancer-related implications have been reviewed before \[[@R31]--[@R33]\]. The most recent review provides an excellent summary and discusses this notion using ovarian cancer as an example \[[@R33]\]. In this study, with the use of homemade multiplex ligation-dependent probe amplification (MLPA) assays, we analyzed the somatic copy number variation of 14 miRNA genes consistently found to be either over or underexpressed in lung cancer. Additionally, we analyzed the copy number variation of *DICER1* and *DROSHA*, two main miRNA biogenesis genes. We analyzed these genes in 254 NSCLC samples and observed high copy number variation in most of the analyzed genes. Among the frequently amplified miRNA genes were *miR-21*, *miR-17/92* and *miR-155*, which are commonly recognized as oncomirs, as well as *miR-30a* and *miR-30d* which were downregulated in lung cancer. Surprisingly, a high average copy number value and frequent amplifications were present in both miRNA biogenesis genes. We also showed that amplification of *DROSHA* is not driven by other closely located oncogenes. The most frequently deleted miRNA gene turned out to be *miR-126*, which is commonly found to be downregulated in lung cancer. Our analysis showed that a substantial fraction of differentially expressed miRNAs may be explained by and are consistent with the copy number variation of their genes. RESULTS {#s2} ======= Selection of miRNA genes for copy number analysis in lung cancer {#s2_1} ---------------------------------------------------------------- To select miRNA genes for our analysis, we took advantage of two recently published meta-analysis studies \[[@R34], [@R35]\] summarizing the results of dozens of whole-genome miRNA expression studies in lung cancer (references within \[[@R34], [@R35]\]). Although these two studies utilized completely different strategies of meta-analyses, the top significantly up- and downregulated miRNAs identified in both studies overlap perfectly (with minor differences in the order of identified miRNAs). Based on these meta-analyses, we selected 6 genes/genomic regions (*miR-21*, *miR-210*, *miR-182*, *mir-31*, *mir-200b*, *mir-205*) encoding miRNAs most consistently identified as upregulated, and 6 genes (*miR-126*, *miR-30a*, *miR-30d*, *miR-486*, *miR-451a*, *miR-143*) encoding miRNAs most consistently identified as downregulated in lung cancer. Additionally, for our analysis we selected the genomic regions of *miR-155* and *miR-17* (identified in one meta-analysis), which were consistently associated with poor prognosis of lung cancer, as well as two genes (*DICER1* and *DROSHA*) encoding miRNA processing enzymes. The genomic positions of all selected genes are indicated in Figure [1](#F1){ref-type="fig"}, and the criteria for their selection are summarized in Table [1](#T1){ref-type="table"}. Note that some of the selected miRNA genes encompass miRNA clusters (e.g., *miR-17/92* and *miR-143/145*). ![The positions of selected miRNA and miRNA biogenesis genes in human genome\ The positions of miRNA and miRNA biogenesis genes are indicated on chromosome ideograms (left-hand side). Arrowheads on the right-hand side of the ideograms indicate lung cancer relevant genes catalogued in COSMIC: the Cancer Gene Census (associated with one of the following terms: "lung", "NSCLC" or "multiple tumor types"), the most reliable list of cancer-related genes annotated and curated by the Wellcome Trust Sanger Institute (originally published in \[[@R78]\]). Additionally, the position of *GOLPH3*, which is discussed in this study, is indicated. Red and blue arrowheads indicate oncogenes and tumor suppressor genes, respectively. IDs of the most relevant genes are indicated next to the arrowheads. The figure was prepared with the use of the "Ensembl karyotypes" tool available on the Ensembl portal.](oncotarget-06-23399-g001){#F1} ###### List of miRNA and miRNA biogenesis genes selected for analysis expression change top-ranked in meta-analysis other lung cancer relevant features ------------------------------- -------------------------------------------------------------------------------------------- ------------------------------------------------------------------------------ ------------------------------------------------------------- -------------------------- -------------------------------------------- ----------------------------------- -------------------------- miRNA and miRNA-cluster genes *miR-21* ↑ 2E-14 ↑ 4.4 \+ ^\[[@R24],\ [@R80],\ [@R81]\]^ \+ ^\[[@R23],\ [@R82],\ [@R83]\]^ \+ ^\[[@R12],\ [@R84]\]^ *miR-210* ↑ 6E-11 ↑ 2.7 \+ ^\[[@R23]\]^ *miR-182* *182*[^E^](#tfn_001){ref-type="table-fn"}, *183*[^E^](#tfn_001){ref-type="table-fn"}, *96* ↑ 3E-8, 4E-2 ↑ 6.3, 5.9 \+ ^\[[@R23],\ [@R83],\ [@R85],\ [@R86]\]^ \+ ^\[[@R84]\]^ *miR-31* ↑ 1E-4 ↑ 2.89 \+ ^\[[@R84]\]^ *miR-200b* *200b*[^E^](#tfn_001){ref-type="table-fn"}, *200a*, *429* ↑ 1E-3 ↑ 3.7 \+ ^\[[@R82],\ [@R83]\]^ \+ ^\[[@R84]\]^ *miR-205* ↑ 7E-3 ↑ 23.2 \+ ^\[[@R83]\]^ *miR-126* ↓ 7E-12 ↓ .33 \+ ^\[[@R23],\ [@R86]\]^ \+ ^\[[@R56]\]^ *miR-30a* *30a*[^E^](#tfn_001){ref-type="table-fn"}, *30b* ↓ 1E-9 ↓ .36 *miR-30d* ↓ 2E-8 ↓ .34 *miR-486* ↓ 4E-7 ↓ .39 \+ ^\[[@R23],\ [@R82]\]^ *miR-451a* *451a*[^E^](#tfn_001){ref-type="table-fn"}, *4732*, *144* ↓ 7E-5 ↓ .37 \+ ^\[[@R83]\]^ *miR-143* *143*[^E^](#tfn_001){ref-type="table-fn"}, *145*[^E^](#tfn_001){ref-type="table-fn"} ↓ 7E-4, 1E-3 ↓ .33, .23 \+ ^\[[@R24]\]^ \+ ^\[[@R83]\]^ *miR-155* ↑[\*](#tfn_002){ref-type="table-fn"} ^\[[@R12],\ [@R24],\ [@R80],\ [@R85]\]^ \+ ^\[[@R24],\ [@R81]\]^ \+ ^\[[@R85]\]^ \+ ^\[[@R12],\ [@R84]\]^ *miR-17* *17*, *18a*, *19a*, *20a*, *19b-1*, *92a-1* ↑[\*](#tfn_002){ref-type="table-fn"} ^\[[@R12],\ [@R51],\ [@R80]\]^ \+ ^\[[@R24]\]^ \+ ^\[[@R12],\ [@R84]\]^ miRNA biogenesis genes *DICER1* ↑/↓[\*](#tfn_002){ref-type="table-fn"} ^\[[@R87]--[@R89]\]^ \+ ^\[[@R87],\ [@R88]\]^ \+ ^\[[@R90]--[@R93]\]^ *DROSHA* ↑[\*](#tfn_002){ref-type="table-fn"} ^\[[@R87]\]^ \+ ^\[[@R87]\]^ \+ ^\[[@R73],\ [@R91]--[@R93]\]^ miRNAs reported as top-ranked in both meta-analyses; expression changes non top-ranked or not analyzed in meta-analyses; altered in plasma/serum/blood/sputum of lung cancer patients and/or associated with early stage NSCLC ^\[[@R23],\ [@R82],\ [@R85],\ [@R86]\]^ MLPA assays design {#s2_2} ------------------ To analyze the somatic copy number variation of selected genomic regions, we designed two MLPA assays, each covering 7 miRNA genes and 1 miRNA biogenesis gene. Each miRNA or miRNA cluster region was covered by two MLPA probes located in close proximity (mostly within 1 kb) to an annotated pre-miRNA sequence, preferentially on both sides of the pre-miRNA sequence. Each of the miRNA biogenesis genes (*DICER1* and *DROSHA*) was covered by 3 MLPA probes located in exons distributed about equally across the genes. Additionally, each MLPA probe-set contained 4 control probes specific for different chromosomes. The exact genomic location and sequence of each probe is indicated in [Supplementary Table S1](#SD2){ref-type="supplementary-material"}. MLPA assays were designed and generated according to a strategy developed and have been described in detail previously \[[@R36], [@R37]\]. We validated the performance of the assays with the panel of reference non-cancer DNA samples and showed that all covered genomic regions are genetically stable and always occur in 2 copies. Analysis of the somatic copy number variation of selected miRNA genes {#s2_3} --------------------------------------------------------------------- With the use of the developed MLPA assay, we analyzed 254 NSCLC samples and determined the relative copy number value of all analyzed regions in these samples. As shown in Figure [2](#F2){ref-type="fig"}, the signals of probes representing particular regions in most cases are strongly synchronized. If one probe in a particular region indicates a copy number increase, the other probe or probes in these regions also show similar levels of copy number increase. As each MLPA probe recognizes different target sequence, such a correlation provides independent validation of the obtained results. The copy number value of a particular region was calculated as the average of the copy number values of the respective probes. The regions for which inter-probe variation was too high were considered uninterpretable and were excluded from further analysis. The relative copy number values of all analyzed regions are shown in [Supplementary Table S2](#SD3){ref-type="supplementary-material"} and graphically summarized in Figure [3](#F3){ref-type="fig"}. As analyzed NSCLC samples are contaminated with different amounts of normal DNA (in most samples percentage of tumor cells (PTC) is \>50%, and an average PTC is approximately 70%) the estimated copy number changes are generally diluted and lower than in actual cancer cells. For comparison, copy number values corrected for PTC (dilution) factor are shown in [Supplementary Figure S1](#SD1){ref-type="supplementary-material"}. As shown in Figure [3](#F3){ref-type="fig"} and [Supplementary Figure S1](#SD1){ref-type="supplementary-material"}, the average copy number of analyzed regions differs substantially and is highest for *DROSHA*, *miR-30d*, *miR-30a*, *miR-21*, *DICER1*, *miR-205*, *miR-17*, and *miR-155* and lowest for the *miR-126* region (Table [2](#T2){ref-type="table"}). {#F2} {ref-type="fig"}. Dots in brackets (above) indicate samples with a copy number value \>8. Color lines represent threshold values of homozygous deletions, losses, gains and amplifications. The outlined Tukey box-and-whisker plots indicate 1st quartile, median and 3rd quartile, and summarize the distribution of the presented copy number values. **B.** The heatmap graph showing the distribution of copy number categories of analyzed genes (columns) in 254 lung cancer samples (rows). The genes (from the left) and samples (from the top) were ordered from the lowest to highest average copy number value. Copy number categories are indicated by colors as shown in the legend on the right.](oncotarget-06-23399-g003){#F3} ###### Summary of copy number changes observed in analyzed miRNA and miRNA biogenesis genes in NSCLC samples expression copy number: median (average) gains: number (%) amplifications: number (%) losses:number (%) hom. deletions: number (%) informative samples \# ---------------- ------------ ------------------------------- ------------------- ---------------------------- ------------------- ---------------------------- ------------------------ ***miR-126*** ↓ 1.73 (1.76) 1 (0.4) 0 (0) 26 (10.8) 3 (1.2) 241 ***miR-200b*** ↑ 1.76 (1.84) 2 (0.9) 1 (0.4) 18 (7.7) 0 (0) 235 ***miR-182*** ↑ 1.78 (1.81) 1 (0.4) 0 (0) 9 (3.8) 0 (0) 240 ***miR-451a*** ↓ 1.84 (1.89) 6 (2.5) 0 (0) 12 (5.0) 4 (1.7) 239 ***miR-210*** ↑ 1.85 (1.87) 1 (0.4) 0 (0) 9 (4.0) 0 (0) 227 ***miR-31*** ↑ 1.99 (2.11) 22 (10.0) 6 (2.7) 20 (9.1) 8 (3.7) 219 ***miR-486*** ↓ 2.06 (2.11) 7 (3.0) 2 (0.9) 3 (1.3) 2 (0.9) 233 ***miR-143*** ↓ 2.14 (2.16) 12 (5.9) 2 (1.0) 8 (3.9) 3 (1.5) 205 ***miR-155*** ↑ 2.33 (2.50) 33 (13.5) 23 (9.4) 21 (8.6) 5 (2.0) 245 ***miR-17*** ↑ 2.42 (2.62) 39 (16.0) 28 (11.5) 15 (6.1) 5 (2.0) 244 ***miR-205*** ↑ 2.59 (2.61) 45 (19.0) 8 (3.4) 1 (0.4) 0 (0) 237 ***DICER1*** ↓/↑ 2.60 (2.69) 51 (21.8) 23 (9.8) 8 (3.4) 3 (1.3) 234 ***miR-21*** ↑ 2.63 (2.90) 48 (22.7) 25 (11.8) 6 (2.8) 0 (0) 211 ***miR-30a*** ↓ 2.67 (2.90) 59 (23.8) 40 (16.1) 14 (5.6) 6 (2.4) 248 ***miR-30d*** ↓ 2.77 (3.02) 63 (26.6) 35 (14.8) 3 (1.3) 1 (0.4) 237 ***DROSHA*** ↑ 2.79 (3.00) 67 (30.7) 23 (10.6) 0 (0) 0 (0) 218 ***MET*** ↑ 2.45 (2.50) 23 (9.9) 5 (2.2) 1 (0.4) 0 (0) 232 ***EGFR*** ↑ 2.41 (2.55) 13 (5.3) 5 (2.0) 1 (0.4) 1 (0.4) 246 Pronounced copy number changes may be more indicative of the role of a particular region in cancer. Therefore, based on criteria similar to those applied before \[[@R38], [@R39]\], we classified the identified copy number changes to the following categories (from highest to lowest copy number): amplifications (≥4 copies; ≥2x increase), gains (≥3 copies; ≥1.5x increase), losses (≤1.33 copies; ≤1.5x decrease) and homozygous deletions (≤1 copy; ≤2x decrease). The categorized copy number changes of all analyzed samples are visualized in a heatmap graph (Figure [3B](#F3){ref-type="fig"}) and are summarized in Table [2](#T2){ref-type="table"}. The results indicate that the number of amplifications detected in particular genes generally correlates with an increase in the average copy number value of these genes. The highest frequency of amplifications was observed in *miR-30a*, *miR-30d*, *miR-21*, *miR-17*, *DROSHA*, *DICER1*, and *miR-155*. In some of these genes both amplifications and isolated cases of deletions were detected. The genes for *miR-182*, *miR-200b* and *miR-210* turned out to be relatively stable, showing neither amplifications nor homozygous deletions. Only a few homozygous deletions but no amplifications were detected in *miR-126* and *miR-451a*. For comparison, the results of copy number changes of genes analyzed in this study are presented along with corresponding results of two oncogenes, *EGFR* and *MET*, obtained previously with the use of a similar methodology \[[@R40]\]. Extended analysis of the *DROSHA* locus on chromosome 5 {#s2_4} ------------------------------------------------------- One of the genes with the highest average copy number and the highest frequency of amplifications was *DROSHA*. To investigate whether copy number increases in *DROSHA* result from amplification of other nearby genes or regions, we designed an additional MLPA assay (LC-5p) covering the short arm of chromosome 5 (5p-arm). Except for the 4 control probes that were used before, the assay was composed of (i) 6 probes more or less evenly distributed along the entire chromosome arm, (ii) 5 probes covering the *DROSHA* gene (3 probes used before and 2 new probes), and 3 probes covering the *GOLPH3* gene, recently identified as oncogene \[[@R41]\] located in close proximity (∼0.5 Mb upstream) to *DROSHA*. The locations of the probes are indicated in Figure [4A](#F4){ref-type="fig"} and [Supplementary Table S1](#SD2){ref-type="supplementary-material"}. With the use of the developed assay, we analyzed 20 samples selected based on the increased signal of *DROSHA* observed in the first experiment (18 amplifications and 2 gains). The copy number values of *DROSHA* determined by two independent experiments (with the use of LC-miR_1 and LC-5p assays) showed a very strong correlation (*R* = 0.92, *p* \< 0.0001, data not shown). As shown in Figure [4](#F4){ref-type="fig"}, increased copy number is observed along almost the entire 5p-arm and no specific region shows sign of focal amplification. The region of amplifications observed in particular samples extends from the probe 5p_10, 2M to the probes covering *DROSHA*, and usually does not encompass *GOLPH3* (Figure [4](#F4){ref-type="fig"}). The above experiment clearly demonstrates that amplification of *DROSHA* is part of a chromosome-level amplification of the 5p-arm and is not a "passenger" effect of focal amplification of some other oncogene. {#F4} Survival analysis of patients stratified by copy number categories of miRNA and miRNA biogenesis genes {#s2_5} ------------------------------------------------------------------------------------------------------ The overall survival data were available for 120 of the analyzed patient samples. Median overall survival of these patients was 416 days (14 months). Kaplan-Meier survival analysis of patients grouped based on copy number categories showed significant decreases in the survival of patients with the *miR-200b* deletion (log-rank test, *p* = 0.022) and patients with gain or amplification of *miR-30d* (*p* = 0.013) (Figure [5](#F5){ref-type="fig"}). This corresponds to a lower 5-year survival rate (0%) of patients with the above mentioned copy number aberrations compared to patients without the aberrations in *miR-200b* (6%) and *miR-30d* (10%). {#F5} Similar analyses performed for *DICER1* and *DROSHA* showed that samples with an increased copy number of *DROSHA* have significantly decreased survival and that the survival rate corresponds to the degree of copy number increase (log-rank test for trend, *p* = 0.032) (Figure [5](#F5){ref-type="fig"}). Association of clinical data with copy number categories of miRNA and miRNA biogenesis genes {#s2_6} -------------------------------------------------------------------------------------------- The copy number categories of any of the analyzed regions showed substantial association with the sex or age of the analyzed patients ([Supplementary Table S3](#SD4){ref-type="supplementary-material"}). Somewhat higher average age of diagnosis showed samples with *miR-126* deletion (with del/without del; 64.9/61.2 years; *p* = 0.046), *miR-451a* deletion (with del/without del; 66.8/61.2 years; *p* = 0.041), and with *miR-31* deletion (with del/without del; 65.7/61.0 years; *p* = 0.017). It has to be noted, however, that these associations are only marginally significant on the nominal level but not after adjustment for multiple comparisons. We also did not find any significant association of copy number categories with clinical data, such as stage of lung cancer at time of sample collection and metastasis/progression/remission status during the last examination ([Supplementary Table S3](#SD4){ref-type="supplementary-material"}). It has to be noted, however, that clinical data were available only for part of the analyzed samples (*N* = 120) and therefore, the lack of association may result from relatively low statistical power of our analysis. Computational analysis of the association of *DICER1* and *DROSHA* copy number categories with their expression and cancer patient survival {#s2_7} ------------------------------------------------------------------------------------------------------------------------------------------- Because we do not have access to mRNA/cDNA material or the expression data for our samples to determine whether copy number changes in *DICER1* and *DROSHA* correlate with their expression, we used data deposited in the cBioPortal for Cancer Genomics \[[@R42], [@R43]\]. As shown in Figure [6](#F6){ref-type="fig"}, there is a dose-dependent correlation between the copy number categories and the expression of *DICER1* and *DROSHA* in lung cancer (based on TCGA Cancer Genome ATLAS data \[[@R44]\]). A similar correlation can be observed in other cancers analyzed in different studies ([Supplementary Figure S2](#SD1){ref-type="supplementary-material"}). Further analysis with the use of another oncogenomic tool, PPISURV \[[@R45]\], showed that the increased expression of *DROSHA* generally (across cancers/datasets) correlates with decreased survival (Figure [6B](#F6){ref-type="fig"} and [Supplementary Figure S2](#SD1){ref-type="supplementary-material"}). In most deposited datasets/cancer types, including lung cancer, correlations show the same negative direction (in 6 of 36 datasets association show significance at *p* ≤ 0.05). Similar analysis performed for *DICER1* shows the opposite effect of increased expression. In most deposited datasets, increased expression of *DICER1* shows the association (positive correlation) with increased survival (14 of 42 datasets show association at *p*-≤ 0.05; Figure [6B](#F6){ref-type="fig"} and [Supplementary Figure S2](#SD1){ref-type="supplementary-material"}). ![Computational analysis of clinical (survival) and oncogenomic data of *DROSHA* and *DICER1*\ Mutual relation between copy number and expression (oncogenomic data) of *DROSHA* **A and B.** and *DICER1* **C and D.** and the relation of their expression to survival of cancer patients. A) and C) Correlation analysis of copy number categories and expression level performed with the use of a dataset (lung adenocarcinoma TCGA \[[@R79]\]) deposited and tools available in cBioPortal for Cancer Genomics. B) and D) Survival analysis performed with the use of a dataset (stage i-ii lung adenocarcinoma; GEO: GSE31210) deposited in and tools available from the PPISURV web portal.](oncotarget-06-23399-g006){#F6} DISCUSSION {#s3} ========== With the use of two homemade MLPA assays, we analyzed the copy number variation of 14 miRNA genes reported as either over- or underexpressed in lung cancer. Additionally, we analyzed two critical miRNA biogenesis genes, *DROSHA* and *DICER1*. Each analyzed gene was tested by at least two independent MLPA probes, providing additional internal validation for the obtained results. To avoid any potential false results, the substantially discordant signals of matched probes were excluded from analysis. A similar strategy of somatic copy number variation analysis may be applied to almost any genomic region of interest in cancer samples. It should be noted, however, that the obtained copy number values are relative and to some extent may depend on the copy number variation of selected control regions (probes). The analysis showed a substantial somatic copy number variation (both gains and losses) of all selected regions in cancer samples (compared variation in cancer vs. control, non-cancer samples; Figure [3](#F3){ref-type="fig"}). However, the observed copy number alterations are not random, and some regions show a substantial increase (frequent amplifications), while the others show decrease in the average copy number value. The genes showing the highest average level of copy number include *miR-30d*, *miR-30a*, *miR-21*, *miR-205*, *miR-17*, *miR-155* as well as *DROSHA* and *DICER1*. Surprisingly, the average copy number and the frequency of amplifications of some of these genes (e.g., *DROSHA, miR-30d, miR-30a* and *miR-21)* are substantially higher than the corresponding values of well-known lung cancer-related oncogenes, *EGFR* and *MET*, analyzed in the same set of samples. In contrast, *miR-126* showed the lowest average copy number and a relatively high frequency of deletions and homozygous deletions. It should be noted, however, that due to the contamination of cancer samples with normal DNA and the inherent lower amplitude of copy number losses than copy number gains, the power of our analysis to detect deletions was substantially lower than the power to detect copy number gains/amplifications. Some genes, such as *miR-31*, show a relatively high frequency of both gains/amplifications and deletions. As expected, the copy number variation of analyzed miRNAs does not correlate perfectly with the global expression changes of these miRNAs observed in lung cancer. However, our results indicate that copy number gains/amplifications may contribute substantially and may be an important mechanism underlying overexpression of miRNAs such as *miR-21*, *miR-17*, *miR-205* or *miR-155*. In short, these miRNAs are the best known oncomirs implicated not only in lung cancer but also in many other types of cancer (reviewed in \[[@R20], [@R26], [@R46], [@R47]\]). *MiR-21* was originally recognized as an antiapoptotic miRNA \[[@R48]\] that was strongly overexpressed in most types of cancer. Later, it was shown that *miR-21* promotes growth, metastasis and invasiveness, as well as chemo- and radioresistance of NSCLC, most likely by targeting tumor suppressor *PTEN* \[[@R49], [@R50]\]. In our experiment, *miR-17* represents 6 miRNAs coded in the *miR-17/92* cluster located within intron 3 of the *C13orf25* on chromosome 13. It was shown that the *miR-17/92* cluster may be upregulated by gene amplification, which is consistent with our results, or by *MYC* overexpression. It was also shown that upregulation of the *miR-17/92* cluster promotes cell proliferation and inhibits lung cell differentiation (the role of *miR-17/92* cluster was reviewed in \[[@R51]\]). *MiR-205* acts either as a tumor suppressor or as an oncogene. As an oncogene, it promotes tumor initiation, progression, resistance to therapies and inhibits apoptosis. It was shown that the oncogenic role of *miR-205* is expressed mostly by downregulation of tumor suppressors such as *PTEN* and *SHIP2* (references within \[[@R46]\]). *MiR-155* is encoded by the non-protein-coding gene *BIC*, originally identified as B-cell integration cluster for the avian leukosis virus, inducing lymphomas \[[@R52]\]. It was shown that *miR-155* targets several tumor suppressors such as *SOCS1*, *FOXO3*, and *VHL* and is involved in the regulation of cell survival, growth, chemosensitivity and tumor angiogenesis \[[@R53]--[@R55]\]. On the other hand, *miR-126* showed the lowest average copy number and frequent deletions in our study and is also recurrently found as downregulated in lung cancer. *MiR-126* was recognized as a tumor suppressor in most of the cancers studied. It was shown that *miR-126* may negatively control and inhibit cell proliferation, migration, invasion, and cancer cell survival. Among the validated targets of *miR-126* are such oncogenes as *ADAM9*, *CRK*, *EGFL7*, *HOXA9*, *IRS1*, *KRAS*, *PI3K*, *SLC7A5*, *SOX2*, and *VEGF* (reviewed and references within \[[@R56]\]). The example of miRNAs which show discordant directions of expression and copy number changes are *miR-30a* and *miR-30d*, both belonging to *miR-30* family. *MiR-30a* and *miR-30d* belong to the miRNAs most frequently reported to be downregulated in lung cancer. On the other hand, these two miRNAs exhibit average copy number values and amplification frequencies that are among the highest of the genes analyzed in our study. It should be noted, however, that the copy number increases in *miR-30d* observed in our study correspond well to the results obtained previously by Li et al.. They showed that *miR-30d* is frequently amplified in different types of cancer (∼30%) including lung cancer (27%), and that amplification of *miR-30d* correlates with its overexpression \[[@R57]\]. It was also shown that *miR-30d* downregulates many cancer-related genes, including apoptotic caspase *CASP3*, and is involved in the upregulation of such processes as cell proliferation, apoptosis, and migration \[[@R57]\]. The above facts strongly suggest the oncogenic character of *miR-30d*. Additionally, our results suggest that increased copy number of *miR-30d* (gains or amplifications vs. others) correlate with significantly reduced survival (Figure [5](#F5){ref-type="fig"}). On the other hand, *miR-30a* has been frequently implicated as a tumor suppressor. It was shown that *miR-30a* targets and downregulates the transcription factor Snai1 and consequently inhibits the epithelial-to-mesenchymal transition (EMT), invasion, mobility and metastasis of NSCLC cells \[[@R25]\]. The opposite characteristics of these two miRNAs may be reflected by the different frequency of deletions of these two miRNAs observed in our study. Although *miR-30a* showed a substantially increased average copy number, it was also one of the most frequently deleted in our analysis. Of our analyzed samples, 20 (8%) showed deletion of *miR-30a*, including 6 samples (2.4%) with homozygous deletions. For comparison, only 5 samples showed deletion of *miR-30d*. Another example of miRNA with opposite trends in global expression and copy number changes is *miR-200b*. Although upregulation of *miR-200b* was recurrently identified in lung cancer, its character suggests it is likely a tumor suppressor. *MiR-200b* belongs to the *miR-200* family that maintains the general characteristics of the epithelia and inhibits EMT, tumor cell motility, and invasiveness (\[[@R58]\] and references within). Among the experimentally identified and validated targets of *miR-200b* are numerous genes involved in the regulation of cytoskeletal organization and cell morphology in addition to *EGFR* \[[@R58]\]. Additionally, our analysis showed significantly decreased survival of patients with either deletion or homozygous deletion of *miR-200b*. In addition to miRNA genes, we analyzed also two key miRNA biogenesis genes, *DICER1* and *DROSHA*. Both of these genes, but especially *DROSHA*, show substantial copy number increases and frequent high-copy number amplifications in analyzed samples. Review of the Cancer Gene Census (COSMIC database) reveals no proto-oncogene in close proximity of either *DROSHA* or *DICER1* that might drive their amplification. However, meticulous review of the literature allowed us to identify *GOLPH3* located in direct proximity (∼600 kb upstream) of *DROSHA*. *GOLPH3* encodes a Golgi-localizing protein that was recently identified as a candidate oncogene driving the amplification of the 5p13 region. This amplification has frequently been observed in multiple solid tumors, including lung cancer \[[@R41]\]. It was shown that Golph3 enlarges cell size, enhances growth-factor-induced mTOR signaling in human cancer cells, and increases the sensitivity to an mTOR inhibitor \[[@R41]\]. The detailed analysis showed that the region of amplification comprising *GOLPH3* is very narrow and does not extend to *DROSHA*. However, the frequency of *GOLPH3* amplification in lung cancer observed previously (56%) corresponded well to the frequency of gains/amplifications of *DROSHA* observed in our study (42%). To verify whether the *DROSHA* amplifications observed in our study might be driven by the closely located *GOLPH3*, we reanalyzed this region with the use of the new 5p-arm-specific MLPA assay. This experiment confirmed *DROSHA* amplifications in analyzed samples and showed that amplification of *DROSHA* results mostly from the chromosome-level amplification of almost the entire 5p-arm. This experiment clearly demonstrated that amplification of *DROSHA* does not depend on the focal amplification of closely located *GOLPH3* or any other specific oncogene on the 5p-arm. Regardless of whether *DROSHA* and *DICER1* are drivers of their amplifications, the amplifications of these two key miRNA biogenesis genes may increase their expression and, as a consequence, may contribute to the global destabilization of miRNA expression observed in many types of cancer. The computational analysis of publically available oncogenomic data showed that the copy number variation of *DROSHA* correlates well with its expression and that increased expression of *DROSHA* is associated with worse survival. The above analyses of oncogenomic data are in line with our experimental results suggesting decreased survival of patients with gain or amplification of *DROSHA* (Figure [5](#F5){ref-type="fig"}). A similar computational analysis of *DICER1* also showed a good correlation between its copy number categories and expression. However, in contrast to *DROSHA*, increased expression of *DICER1* was associated with longer survival in various cancers including lung cancer. Although such results must be interpreted with caution, the opposite effects of increased expression of *DROSHA* and *DICER1* on survival (positive and negative, respectively) may suggest the oncogenic role of intermediate products of these two enzymes, that is, pre-miRNAs (either specific or as a class). It should be noted that the advantage of the computational results discussed above is that they are based on independent (objectified) whole genome datasets generated in projects not focused specifically on *DICER1*, *DROSHA* or any other miRNA biogenesis gene. Our results add to the complex picture of the role of *DICER1* and *DROSHA* in cancer. The miRNA biogenesis genes were primarily considered as haploinsufficient tumor suppressors \[[@R59]\]. This notion results mostly from the observation that the overall level of miRNAs is often reduced in cancer \[[@R60]--[@R62]\] and from the fact that germline loss-of-function mutations in *DICER1* are causative variants in the so called DICER1 syndrome, which is associated with increased risk of numerous, mostly early, childhood malignancies and benign tumors \[[@R63]\]. The representative (most common) malignancy for this syndrome is pleuropulmonary blastoma, which occurs in the lungs. More recently, analysis of cancers associated with DICER1 syndrome as well as other early childhood cancers (e.g., Wilms tumor) led to the identification of a peculiar pattern of somatic second-hit mutations in *DICER1* and *DROSHA*. These mostly missense mutations are not randomly distributed over the genes but form clear hotspots, mostly affecting few amino acid residues located in or adjacent to metal-ion-binding residues in the RNaseIIIb domain of either *DICER1* (D1709, E1813) or *DROSHA* (E1147, D1151) \[[@R63]--[@R68]\]. Functional analyses suggest that these mutations are not deleterious (as expected for typical second-hit mutations) but rather modify the function of DICER1 or DROSHA, making it favorable for cancer (oncogenic) (recently discussed in \[[@R69], [@R70]\]). It was shown that modified enzymes selectively reduce the processing of miRNAs generated from the 5′ arm of pre-miRNA hairpins and as a consequence modify the miRNA expression profile in cancer \[[@R65], [@R66], [@R71], [@R72]\]. Our results and the notion about the oncogenic role of *DROSHA* are very much in line with previous results suggesting that *DROSHA* is a key gene driving frequent gains of the 5p-arm in cervical squamous cell carcinoma (SCC) \[[@R73], [@R74]\]. Analysis of primary cervical SCC samples and cell lines showed that the frequent copy number gains and overexpression of *DROSHA* led to an altered profile of miRNA expression, including the expression of many cancer-related miRNAs. Among the miRNAs showing the highest overexpression was *miR-31*. Functional *in vitro* analyses (including wound healing test) showed that upregulation of *DROSHA* increases motility and invasiveness of squamous SCC cell lines \[[@R73], [@R74]\]. It was also shown that overexpression of *DROSHA* is associated with metastasis and decreased survival in esophageal cancer patients \[[@R75]\]. It should be noted that other genes of miRNA biogenesis enzymes may contribute to the regulation of global or individual miRNA expression in cancer. Therefore, to better understand and evaluate the impact of somatic copy number variation of miRNA biogenesis genes on miRNA expression in cancer, a more complex analysis is needed. In conclusion, our results show a substantial somatic copy number variation in genomic regions comprising miRNA genes. Among these regions were those showing a substantial increase in the average copy number (frequently amplified), and regions with decreased average copy number. Concordance of copy number and expression changes of some miRNAs suggest that copy number variation may be an important mechanism responsible for regulation of these miRNAs in lung cancer. Therefore our observations support the proposed earlier notion, implying the high genomic instability of miRNA gene regions and contribution of copy number variation in the regulation of miRNA expression in cancer \[[@R29], [@R30]\]. It should be emphasized however that the amplitude and recurrence of copy number changes cannot be simply interpreted as the oncogenic role of a variable region/gene in cancer. Our results also indicate the important role of miRNA biogenesis genes, especially *DROSHA*, in lung cancer. Even if these genes are not drivers of their copy number changes, they may affect global regulation of miRNA expression in cancer. Finally, somatic copy number changes of some of the analyzed genes including *DROSHA* correlate with survival of cancer patients. Although the results of our survival analyses are only marginally significant (relatively low number of samples) and must be replicated in an independent group of samples, the copy number changes would be attractive biomarkers due to (i) the relatively high stability of genomic DNA, even extracted from formalin-fixed paraffin embedded (FFPE) samples; (ii) the small amount of DNA necessary for analysis; (iii) relatively low cost; (iv) simplicity; and (v) the reliability of copy number analysis. The drawback of such analysis is, however, contamination of the cancer samples with a difficult to estimate amount of normal DNA. MATERIALS AND METHODS {#s4} ===================== Selection and processing of NSCLC samples for molecular analysis {#s4_1} ---------------------------------------------------------------- We retrospectively reviewed a cohort of 254 patients with histopathologically confirmed NSCLC diagnosed at the Franciszek Lukaszczyk Oncology Center in Bydgoszcz (central Poland). The age of the patients ranged from 35 to 81. A total of 254 specimens that passed the quality control steps (microscopic analysis and tumor content qualification as well as qualitative and quantitative DNA analysis) were obtained following surgeries, fine-needle aspirations (FNAs), endobronchial ultrasound with guided transbronchial needle aspiration (EBUS-TBNA) procedures or pleural fluid sampling. The samples were stained with hematoxylin and eosin for the qualitative and quantitative analysis of tumor cells in the analyzed material (including macrodissection in marked out samples) as described previously \[[@R76]\]. The study was approved by the Committee of Ethics of Scientific Research of Collegium Medicum of Nicolaus Copernicus University, Poland (KB 265/2012). The data were analyzed anonymously. DNA extraction was performed after the microdissection of a region indicated by the pathomorphologist, and the quality and quantity of DNA samples were evaluated as described previously \[[@R40]\]. Copy number analysis by MLPA {#s4_2} ---------------------------- MLPA analysis was performed with the use of three in-house designed and generated assays, LC-miR_1, LC-miR_2 and LC-5p. Both LC-miR_1 and LC-miR_2 assays contained 14 probes specific for 7 miRNA genes (two probes for each miRNA or miRNA-cluster gene), 3 probes specific for one of miRNA biogenesis gene, and 4 control probes (located on different chromosomes outside of chromosome 5 and regions of known cancer-related genes). The LC-5p assay contained 6 probes more or less evenly covering the short arm of chromosome 5 (5p-arm), 5 probes specific for *DROSHA*, 3 probes specific for *GOLPH3*, and 4 control probes. The detailed characteristics, genomic positions and sequences of all probes used in this study are presented in [Supplementary Table S1](#SD2){ref-type="supplementary-material"}. The MLPA probes and the general layout of the probe sets were designed according to a previously proposed strategy \[[@R36], [@R37]\]. This strategy utilizes only short oligonucleotide probes that can easily be generated via standard chemical synthesis. Briefly, each probe was composed of two half-probes of equal size, and the total probe length ranged from 93 to 164 nt. The target sequences for the probes were selected to avoid SNPs, repeat elements and sequences of extremely high or low GC content. The MLPA probes were synthesized by IDT (Skokie, IL, USA). The MLPA reactions were run according to the manufacturer\'s general recommendations (MRC-Holland, Amsterdam, the Netherlands), as described earlier in \[[@R37], [@R77]\]. All reagents except the probe mixes were purchased form MRC-Holland (<http://www.mlpa.com>). The products of the MLPA reaction were subsequently diluted 20x in HiDi formamide containing GS Liz600, which was used as a DNA sizing standard, and separated via capillary electrophoresis (POP7 polymer) in an ABI Prism 3130XL apparatus (Applied Biosystems, Carlsbad, CA, USA). The obtained electropherograms were analyzed using GeneMarker software v2.4.0 (SoftGenetics, State College, PA, USA). The signal intensities (peak heights) were retrieved and transferred to prepared Excel sheets (available upon request). For each individual sample, the signal intensity of each probe was divided by the average signal intensity of the control probes to normalize the obtained values and to equalize run-to-run variation. Due to high signal variation, the control probe 3 (ctrl_3) was excluded from analysis. To calculate relative copy number value of particular probe, the normalized signal of this probe was divided by a corresponding value of this probe in the reference (non-cancer) sample and multiplied by 2. The relative copy number of a particular gene was calculated as an average of the normalized copy number value of 2 or 3 probes specific to this gene. If the difference between the maximum and minimum signal of the averaged probes was higher than one-third of an average copy number value or if the coefficient of variation of the averaged probes was higher than 0.3, the result was excluded from further analyses. Databases and statistical analysis {#s4_3} ---------------------------------- All statistical analyses were performed using Statistica (StatSoft, Tulsa, OK) or Prism v. 4.0 (GraphPad, San Diego, CA). All *p*-values were provided for two-sided tests. All human genome positions indicated in this report refer to the February 2009 (GRCh37/hg19) human reference sequence. The datasets for analysis and visualization of the relationship between copy number category and expression level of *DROSHA* and *DICER1* were obtained from cBioPortal for Cancer Genomics (MemorialSloan-Kettering Cancer Center, New York, NY, USA; <http://www.cbioportal.org/>) \[[@R42], [@R43]\] and were analyzed with the use of the cBioPortal Plots tool. The survival analyses of the cancer patients with high and low levels of either *DICER1* or *DROSHA* expression were performed with the use of datasets and tools available in the PPISURV portal (<http://www.bioprofiling.de/GEO/PPISURV/ppisurv.html>) \[[@R45]\]. SUPPLEMENTARY FIGURES AND TABLES {#s5} ================================ This work was supported by National Science Centre 2011/01/B/NZ5/02773, and KNOW program of the Polish Ministry of Science and Higher Education. **CONFLICTS OF INTEREST** All authors declare no conflict of interest.

The use of topical drops, corneal irrigation, and mechanical maneuvers during cataract surgery disturbs the balance of the corneal surface.^[@R1]--[@R4]^ Intraoperatively, a lid speculum is placed to prevent blinking of the eyelids, and an eye lubricant is required as a tear film substitute to ensure epithelial hydration and provide sufficient optical clarity. For anterior segment surgery, balanced salt solution (BSS) is usually used as an irrigating agent; however, the hydrating effect is short lasting, leading to frequent application by the surgeon or nurse. Repeated irrigations with BSS present a number of disadvantages. First, they may disturb and prolong the surgical procedure and be unpleasant for the patient, especially if the surgery is performed under topical anesthesia. Second, they may increase the risk of corneal epithelial damage, with possible discomfort caused by epithelial alterations and potentially prolonged postoperative recovery.^[@R5],[@R6]^ As an alternative solution to BSS irrigation, coating using viscoelastic agents to ensure prolonged corneal hydration and optical clarity during cataract surgery has been studied since the late 1990s.^[@R5]--[@R8]^ Reports suggested that fewer rewetting events were required when a viscoelastic fluid was applied over the corneal surface, instead of BSS.^[@R7]--[@R9]^ However, most of the currently available viscoelastic agents were developed primarily for intraocular use to maintain anterior chamber stability during surgical maneuvers and protect the corneal endothelium.^[@R6],[@R10]^ Therefore, it has been suggested that refinements in viscoelastic formulations may improve their efficacy and ease of use during cataract surgery.^[@R6]^ eyeDRO (AL.CHI.MI.A. S.R.L, Italy) is a commercially available medical device made of an advanced tripolymeric gel containing hydroxypropyl methylcellulose (HPMC), xanthan gum, and carrageenan and is intended to protect and hydrate the corneal surface during ophthalmic surgery and eye examination and to maintain maximum clarity of the operating field during surgery. In this study, we retrospectively analyzed 51 cases in which patients received either eyeDRO or BSS during cataract surgery. Our aim was to compare both treatments for intraoperative clarity of the operative field, ease of manipulation, epithelial integrity, and patient comfort. MATERIALS AND METHODS {#s1} ===================== Study Population {#s1-1} ---------------- This study was conducted in compliance with the Declaration of Helsinki. Institutional review board approval was not required for the retrospective analysis of deidentified data, including quality management questionnaire data of 51 consecutive patients who underwent routine cataract surgery at Casa di Cura S. Camillo in Brescia, Italy, within a 20-day period in 2016. Patients having evidence of eye dryness (corresponding approximately to ≥ grade II of the Oxford test), any other ocular pathologies revealed by slit-lamp preoperative screening, and/or previous corneal or refractive surgery were excluded from the analysis. Patients with any systemic disease, such as diabetes, that could influence the ocular surface were also excluded. In addition, patients were excluded if complications occurred during or after surgery, which affected the outcomes. Surgical Procedures {#s1-2} ------------------- All cataract surgeries were performed by a single surgeon (P.G.) in an ambulatory day surgery setting according to internal standard protocols. The surgeon alternated the use of eyeDRO gel and BSS on a daily basis. The procedures were performed under anesthesia with only 1 drop of 0.4% benoxinate hydrochloride (Alfa Intes Srl, Italy), 5 minutes before surgery, followed by intracameral injection of 200 μL of Tropicamide, phenylephrine, lidocaine solution (Mydrane Laboratoires Théa, France) through a small side port, before the first 2.2-mm corneal incision. Subsequently, a lid speculum was placed, and either 1 drop of eyeDRO corneal coating gel was applied before surgery (if necessary, reapplied during surgery) or 2 mL of the BSS irrigating solution (Alcon) was applied repeatedly during surgery by the nurse. For each patient, the number of applications of the appropriate hydrating substance was recorded. All patients underwent routine cataract extraction by phacoemulsification using Infinity equipment (Alcon), with a temporal incision and posterior chamber foldable intraocular lens implantation in the capsular bag. After surgery, eyeDRO residues were removed completely by rinsing with 20 mL of BSS. The postsurgery regimen included netilmicin and dexamethasone eye drops for 4 days, a transparent shield at night, and sunglasses during the day for 1 week. Intraoperative Assessment and Follow-up {#s1-3} --------------------------------------- Intraoperative and postoperative measurements were analyzed. Intraoperatively, the surgeon scored the clarity of the operative field and ease of manipulation using a 10-point arbitrary-unit grading scale at the end of surgery, adopted routinely with the aim of improving the procedure, with 10 being the highest and 1 being the lowest. Two hours after surgery, the cornea was stained using fluorescein ophthalmic strips (Optitech Eyecare, India) and assessed using a slit-lamp (Topcon Sl-7, Nikon, Japan); the extent of epithelial cell damage was scored by positive fluorescein staining (grades 1--3; 1 = no damage; 2 = punctate damage; and 3 = damaged area). Patients\' yes/no feedback to simple questions concerning postoperative pain, such as burning or itching, was used for the assessment of eye irritation or discomfort of foreign body sensation 2 and 24 hours after surgery. Postoperative observations were collected by medical staff other than the surgeon. Statistical Analysis {#s1-4} -------------------- The sample size estimate was obtained *a priori* based on a Mann--Whitney test using a Cohen d value of 0.9, an alpha error of 0.05, and a power of 80%, with an allocation ratio of 0.6 and increased by 10%. We performed the statistical analysis using Excel 2010 software. The box plots were generated by the BoxPlotR Web tool, <http://boxplot.tyerslab.com>. The Mann--Whitney *U* test was used to compare intraoperative parameters (clarity of the operative field, ease of manipulation, and the number of applications) between groups. Control of the family-wise error rate consequent to repeated statistical tests was performed using the Bonferroni correction. The χ^2^ test was used to analyze categorical variables (grade of fluorescent staining, eye irritation, and foreign body sensation) in 2-by-2 contingency tables. The results are presented as mean and SE. Differences yielding *P* \< 0.05 were considered statistically significant. RESULTS {#s2} ======= The analysis compared 20 patients who received BSS irrigation with 31 patients who received eyeDRO coating gel during cataract surgery. Mean age of the patients included in the analysis was 74.7 ± 1.3 years (range 45--89 yr). Seven patients were excluded from the analysis because of diabetes and 5 because of corneal surface irregularities. All surgeries were uneventful, with standard surgery times. All types of cataracts were included in both groups. The grade 4 cataracts corresponded to 4 (20%) and 6 (19%) patients in the BSS and eyeDRO groups, respectively. Application of eyeDRO significantly improved operative field clarity and ease of manipulation during ophthalmic surgery (Mann--Whitney *U* test, *P* \< 0.01 for both parameters) (Fig. [1](#F1){ref-type="fig"}). All eyeDRO-treated patients required only 1 application of the coating gel during the entire surgery. By contrast, 5.3 irrigations with BSS were needed on average (with the use of 13.7 ± 1.5 mL of BSS irrigating solution, on average) to complete surgery in BSS-treated patients (Fig. [1](#F1){ref-type="fig"}); the difference in the frequency of application of the 2 hydrating solutions was statistically significant (Mann--Whitney *U* test, *P* \< 0.01) (Fig. [1](#F1){ref-type="fig"}). {#F1} Two hours postoperatively, no epithelial damage was observed by fluorescein staining (grade 1) in 90.3% of eyeDRO-coated eyes, a significantly higher percentage (χ^2^ test, *P* \< 0.05) than in the BSS-treated patients, who showed 60.0% of eyes without epithelial damage (Fig. [2](#F2){ref-type="fig"}). Only 9.7% of eyeDRO-treated eyes showed punctate epithelial damage (grade 2) compared with 40.0% of BSS-irrigated eyes. The difference between both groups was statistically significant (χ^2^ test, *P* \< 0.05). No extended areas of damage (grade 3) were found in the eyes of any of the 51 patients (Fig. [2](#F2){ref-type="fig"}). {#F2} Two hours after surgery, eye irritation and foreign body sensation were experienced by 13.0% and 37.0% of eyeDRO-treated patients, respectively, and 65.0% and 100.0% of BSS-treated patients, respectively. The difference between both groups was statistically significant for both parameters (χ^2^ test, *P* \< 0.01) (Fig. [3](#F3){ref-type="fig"}A). Twenty-four hours postoperatively, eye irritation was still experienced by 9.0% and 25.0% of eyeDRO- and BSS-treated patients, respectively. This difference was not statistically significant. Foreign body sensation was experienced by 19.0% and 80.0% of patients treated with eyeDRO and BSS, respectively, which was a statistically significant difference (χ^2^ test, *P* \< 0.01) between both groups (Fig. [3](#F3){ref-type="fig"}B). {#F3} DISCUSSION {#s3} ========== Some of the most important functions of the precorneal tear film include lubricating the ocular surface and providing a smooth, regular optical surface for the eye.^[@R11],[@R12]^ Blinking of the eyelids prevents breaking up of the tear film; however, the eyelids must remain open during cataract surgery. As a consequence, several environmental factors (e.g., the use of anesthetics and topical mydriatic agents, long surgery, high temperature and/or low humidity in the operating room, and the intensity of the microscope illumination) may compromise lubrication of the ocular surface and leave the cornea hazy.^[@R6],[@R13]^ Therefore, agents that maintain optimal hydration and corneal transparency must be applied to the ocular surface to ensure a clear view for the surgeon and prevent damage and discomfort to the patient. BSS, which is the most commonly used corneal wetting agent, has the disadvantage of leaving the cornea hazy because it tends to evaporate quickly; therefore, frequent irrigations are required, which may harm the epithelium, stroma, and endothelium.^[@R7],[@R9]^ Similarly, cohesive and dispersive viscoelastics, which are intended for intraocular use and used off-label as superficial lubricants to provide epithelial protection, surface hydration, and optical clarity,^[@R5],[@R10]^ present some drawbacks.^[@R14]^ Based on our experience, cohesive agents do not provide an optimal thin coating layer because they form a sphere, which tends to slip on the wet corneal surface and is easily washed away. Dispersive viscoelastics do not cover the surface homogenously because they have a toothpaste-like consistency and must be spread on the corneal surface, thus resulting in very poor visibility of the operating field and difficult removal.^[@R14]^ Conversely, eyeDRO gel coated the entire corneal surface with a thin, uniform, and transparent layer that remained stable throughout surgery and also provided helpful magnification of the operating field. These effects could be due to the blend of 3 viscoelastic substances present in the eyeDRO formulation, in which each component contributes to the optimal rheological, thickening, stabilizing, bioadhesive, and magnifying properties of the gel.^[@R14]--[@R16]^ Our retrospective analysis showed that all the eyeDRO-treated patients required single application of gel to complete surgery, whereas 5 applications were required on average to maintain sufficient corneal hydration and optical clarity in patients treated with BSS. This corresponds to a slightly lower irrigation frequency than reported previously,^[@R5]^ in which 10 BSS irrigations were needed. Similarly, other studies indicated that second application of corneal wetting agents, such as the elastoviscous hylan surgical shield, 0.45% (HsS, Biomatrix, Canada) or 2.0% lignocaine gel, was required in many patients during surgery.^[@R7],[@R8]^ Even if based on subjective evaluation by the surgeon, our analysis clearly indicated that eyeDRO significantly improved corneal transparency and operative field clarity with respect to the use of BSS. Similarly, previous reports by Arshinoff and Khoury^[@R7]^ and Kalyanasundaram and Hasan^[@R8]^ found that HsS and 2.0% lignocaine gel, respectively, maintained corneal clarity longer than BSS. More recently, Chen et al^[@R5]^ compared a 2.0% HPMC corneal lubricant gel and BSS during cataract surgery under topical anesthesia and found that the use of a single dose (as a median, with a range of 1--8 applications) of HPMC gel showed significant advantages over BSS in both frequency of application and optical clarity for the surgeon. These findings are in agreement with our results, which were obtained with a formulation that, along with other polymers, is also HPMC based. The fact that some---although very few---patients in the study by Chen et al required second application of 2.0% HPMC might be related to the different composition of the 2.0% HPMC formulation with respect to that of the tripolymeric eyeDRO gel; nevertheless, it cannot be excluded that the larger population evaluated in the study by Chen et al might have accounted for this subtle difference in the frequency of application. Considering the previous studies on viscoelastic corneal lubricants, we obtained significantly better results in epithelial preservation and patient comfort, while using eyeDRO instead of BSS. Conversely, in the study by Arshinoff and Khoury,^[@R7]^ the use of HsS did not show significant advantages over BSS in postoperative corneal health. In the study by Chen et al,^[@R5]^ 1 hour postoperatively, the difference in fluorescein staining between the 2% HPMC-treated and control group patients was not significant, and the 1-hour postoperative subjective grading of patients\' discomfort and perception of dry eye was similar in both groups. Thus, it is tempting to speculate that the "enriched" formulation of the tripolymeric gel might have accounted for the more consistent advantages for both surgeon and patients compared with the 2.0% HPMC coating gel during cataract surgery because both xanthan gum and carrageenan-based formulations are known to have a protective effect on the corneal surface.^[@R17],[@R18]^ However, further studies are needed to test this hypothesis. Our study provided some important information on the postoperative safety and tolerability of eyeDRO corneal coating gel that, to our knowledge, has so far not been reported. In addition, we suggest that the combination of the intraoperative use of a single drop of eyeDRO with the preoperative use of an intraocular mydriatic agent, rather than mydriatic drops, might further simplify the whole surgical procedure. Obviously, this hypothesis needs further investigation. In conclusion, our retrospective analysis found that single application of 1 drop of eyeDRO at the beginning of surgery may be a preferable option in cataract surgery because it provides optimal corneal hydration, clear view of the surgical field during the entire surgery, less discomfort to the patients, and a simplified procedure. Cost--benefit analyses were outside the scope of this study. Nevertheless, the cost of eyeDRO gel is only slightly higher than that of a sterile BSS vial; however, considering the benefits to patients and surgeon, and the simplification of surgery, the use of the product has been implemented in our surgical practice. C. Gatto and J. D\'Amato Tóthová are employed by the company Alchilife S.r.l, which was involved in the development of the medical device discussed in this article. The remaining authors have no funding or conflicts of interest to disclose.

Citation {#SECID0EWHAC} ======== Laciny A, Zettel H, Kopchinskiy A, Pretzer C, Pal A, Salim KA, Rahimi MJ, Hoenigsberger M, Lim L, Jaitrong W, Druzhinina IS (2018) *Colobopsis explodens* sp. n., model species for studies on "exploding ants" (Hymenoptera, Formicidae), with biological notes and first illustrations of males of the *Colobopsis cylindrica* group. ZooKeys 751: 1--40. [https://doi.org/10.3897/zookeys.751.22661](10.3897/zookeys.751.22661) Introduction {#SECID0EKJAC} ============ The *Colobopsis cylindrica* (COCY) group likely represents a monophyletic clade containing Southeast Asian ant species with distinctive hypertrophied mandibular gland reservoirs. In territorial combat, minor workers use the sticky and irritant contents of their enlarged mandibular gland reservoirs to kill or repel rival arthropods. In species where this defensive behaviour is more advanced, this happens via the characteristic suicidal "exploding" by voluntary rupture of the gastral integument (autothysis) ([@B6]). This behaviour was first mentioned by Viehmeyer as early as 1916, and subsequently described in detail by [@B38], as well as [@B10], and [@B50]. The Bornean members of the COCY group have been the subject of various ecological (e.g., [@B6], [@B9], [@B7], [@B8]), morphological ([@B10], [@B34]) and chemical ([@B29], Hoenigsberger et al. in prep.) studies in the past. Based on the results of previous investigations, in 2014 an interdisciplinary research project started to explore the evolution and ecological significance of autothysis in the COCY group. From the surroundings of the Kuala Belalong Field Studies Centre (KBFSC) in Brunei, at least 15 species are known ([@B9]), most of which are probably new to science. One species, previously referred to as "yellow goo" ([@B9]) or "YG" ([@B8]) for the bright yellow colour of its mandibular gland secretion, was found to have a large colony just at the KBFSC. As this abundant species frequently exhibits characteristic autothysis behaviour and can be observed *in situ* and *in vitro*, it became the main object of behavioural and chemical experiments, and a model species for biological studies on "exploding ants". Preliminary taxonomic and molecular analyses revealed that this morphospecies is in fact an undescribed species. As the revision of the COCY group is still ongoing (I. Druzhinina et al. in prep.), the aim of this paper is to provide a valid name, *Colobopsis explodens* Laciny & Zettel, sp. n., for subsequent use in the various behavioural, chemical, microbiological, and evolutionary publications currently in preparation. Within this study, we employ the multidisciplinary concept of integrative taxonomy (*sensu* [@B49]) by combining morphometric, ecological, and molecular data. We provide a taxonomic description of all castes of *Colobopsis explodens* sp. n. including males. Illustrations and morphometric characterizations of males of the COCY group had not been previously published. We compare males of *Colobopsis explodens* sp. n. with the newly illustrated male of *C. badia* (Smith, 1857) to highlight species-specific characters in the complex. Morphological characters of the male, including genitalia, are also compared with other selected taxa of Camponotini. Based on field observations, the first records on the natural history and biology of *Colobopsis explodens* sp. n. are provided. Materials and methods {#SECID0E6OAC} ===================== Sampling-sites and imaging of living ants {#SECID0EDPAC} ----------------------------------------- The primary field research took place in the lowland dipterocarp rainforest at the Kuala Belalong Field Studies Centre (KBFSC), Temburong District, Brunei Darussalam (4°32\'48.2\"N, 115°09\'27.9\"E), where *Colobopsis explodens* sp. n. was sampled during five collecting trips (each of 30 days duration) encompassing different seasons from 2014 to 2016. The behaviour of *C. explodens* sp. n. was observed at multiple nesting sites on several height-levels, starting from the forest floor and understory up to the canopy and emergent layer. The activity of ants was recorded *in situ* and *in vitro* using a CANON 70D Digital SLR Camera with a CANON EF 100 mm macro lens and a Tamron AF 28--200 mm F/3.8--5.6 XR Di aspherical (IF) macro zoom lens (Suppl. material [2](#S2){ref-type="supplementary-material"}: S2a). For macro and close-up filming the Neewer adjustable LED light with LCD display was used. When necessary, the camera was mounted with the use of a Manfrotto Gorillapod 494RC2 tripod. The movie (Suppl. material [7](#S7){ref-type="supplementary-material"}) was annotated and cut using Corel VideoStudio X10 Software. Sampling of *Colobopsis badia* in southern Thailand was conducted by H. Zettel and W. Jaitrong in June 2016. The sampling site was located in the Khao Chong Botanical Garden, near the Ton Pliw Waterfall (07°32\'34\"N, 99°47\'33\"E); a single male specimen was caught at a light at the Botanical Garden headquarters. Host trees and activity assessment {#SECID0EIBAE} ---------------------------------- Nesting habits of *C. explodens* sp. n. were observed based on the model colony occupying several trees and an artificial nest (nest \#38, Fig. [9](#F9){ref-type="fig"}) in direct vicinity to the kitchen facility at KBFSC. The artificial nest consisted of a 100 cm tall and 6 cm wide square wooden stake, with a cavity of approximately 15 mm in diameter drilled into the centre and a 4 mm wide entrance hole in the top third of the stake. The nest was painted with green acrylic paint and fastened to a small tree with rope (for detailed method of construction, see [@B7] and [@B34]). The host trees were identified by comparison with type samples preserved in the herbarium of Universiti Brunei Darussalam, Brunei. The main host tree was DNA barcoded (see Suppl. material [6](#S6){ref-type="supplementary-material"} "accession numbers"). The activity of *C. explodens* sp. n. occupying artificial nest \#38 was observed from 14^th^ to 30^th^ November of 2015 at different times during the day, for 30 minutes each by counting the ants entering and leaving the nest. Temperature, barometric pressure, and weather conditions were recorded, as well as any observed noteworthy behaviour (see Fig. [9](#F9){ref-type="fig"}; Suppl. material [6](#S6){ref-type="supplementary-material"} "activity"). DNA Extraction, PCR amplification, and Sanger sequencing {#SECID0EJDAE} -------------------------------------------------------- DNA extraction, gene fragment amplification, and sequencing were performed for minor worker ants of five different taxa (*C. explodens* sp. n., *C. badia*, C. nr. saundersi, *C. aruensis* Karawajew, 1933, and *C. cylindrica* (Fabricius, 1798)), as well as for mandibular gland reservoir content of *C. explodens* sp. n., one symbiotic cricket (*Camponophilus* sp.) from artificial nest \#38, and the host plant of *C. explodens* sp. n. (*Shorea johorensis*). For DNA barcoding of ant specimens, DNA was extracted from ant legs using Qiagen's tissue QIAamp DNA Micro kit following the manufacturer's protocol (Qiagen, Venlo, Netherlands). To obtain sufficient DNA quantity for further processing, the amount of legs used per sample varied. For the DNA extraction of queens, a minimum of three legs of one individual were transferred into one 1.5 ml microcentrifuge tube and frozen with liquid nitrogen. Three legs of one individual was also the minimum amount for males and major workers. For minor workers, all legs from two to four individuals were pooled (see Suppl. material [6](#S6){ref-type="supplementary-material"} "accession numbers"). The frozen legs were ground into small pieces with disposable pestles attached to a pestle motor (Kimble, Vineland, NJ, USA). Subsequent steps were performed according to manufacturer\'s instructions with the following exceptions: sample lysis for 20 hours, final elution step with 25--50 µl elution buffer. To assess the purity of the extraction, DNA concentration and 260/280 nm ratio were measured with a NanoDrop ND-1000 Spectrophotometer (Software Version ND-1000 v.3.8.1, Thermo Fisher Scientific, MA, USA). For DNA barcoding of symbiotic crickets, DNA was extracted from one whole specimen applying the same procedure as used for ants legs, but with a pretreatment with an enzymatic lysis buffer (Tris·Cl 20 mM, pH 8.0, sodium EDTA 2 mM, Triton X-100 1.2%, add lysozyme to 20 mg/ml for 60 min). For DNA barcoding of the host plant, 100 mg of a leaf were ground with mortar and pestle under the use of liquid nitrogen and DNA was extracted using Qiagen's DNeasy Plant Mini Kit according to manufacturer's instructions. For the ants, the gene fragments cytochrome C oxidase subunit I and II (COI, COII), cytochrome B (cytB), and carbamoyl-phosphate synthase II (cad) were amplified, for the cricket only COI. Additionally, a fragment of 16S rRNA was amplified from the DNA extracted from the mandibular gland content of *C. explodens* sp. n. minor workers, to assess the presence of bacteria. For the plant, the gene fragment maturase K (matK) was amplified. Primer sequences and specific annealing temperatures are given in Tab. [1](#T1){ref-type="table"}. Final concentrations for PCR were 1× GoTaq Flexi Buffer (Promega, Madison, Wisconsin, USA), 0.16 mM dNTP's, 3 mM MgCl~2~ (Promega), 0.4 µM forward and reverse primer (Microsynth, Balgach, Switzerland), 0.8 Units GoTaqG2 Flexi polymerase (Promega) and 2--50 ng sample (diluted with HPLC water, ROTH), in a final volume of 50 µl. PCR was performed with a Biometra T3 Thermocycler (Biometra, Göttingen, Germany) with the following conditions: 2 min at 94 °C, 35 cycles of 1 min at 94 °C, 1 min at primer specific annealing temperature, 90 sec at 72 °C and finally 7 min at 72 °C. PCR products were separated by 1.5% agarose gel electrophoresis. PCR products were purified using mi-PCR Purification Kit (Metabion, Planegg, Germany) and one direction sequencing was performed at Microsynth (Austria). Sequences are deposited in NCBI GenBank. Accession numbers for ant specimens are given in Table [2](#T2){ref-type="table"}; see Suppl. material [6](#S6){ref-type="supplementary-material"} "accession numbers" for additional details. Sequences of non-ant material are deposited under [MG582639](MG582639) for COI of myrmecophilous crickets (*Camponophilus* sp.), [MF993320](MF993320) for matK of *Shorea johorensis* and [MF996752](MF996752) for 16S rRNA of cf. *Blochmannia* (Enterobacteriales). ###### Primers used in this study. ----------- ---------------------------------- ---------------------------- --------------- ------------------ ------------------------ ----------------------- ----------- Gene Name Sequence 5'--3' Length \[bp\] GC content \[%\] Fragment Length \[bp\] Annealing Temp \[°C\] Reference COI LCO1490-F GGTCAACAAATCATAAAGATATTGG 25 32 709 45 [@B5] HCO2198-R TAAACTTCAGGGTGACCAAAAAATCA 26 35 COII J2791-F ATACCHCGDCGATAYTCAGA 20 40--55 858 51 [@B5] H3665-R CCACARATTTCWGAACATTG 20 35--40 cytB CB11400-F TATGTACTACCHTGAGGDCAAATATC 26 35--42 485 45 [@B5] CB11884-R ATTACACCNCCTAATTTATTAGGRAT 26 27--35 cad CD1423EF AGGTRATACRATCGGARAGRCCDGA 25 40--60 800 55 [@B56] CD1910R CCGAGRGGRTCRACRTTYTCCATRTTRCAYAC 32 38--63 matK 472F CCCRTYCATCTGGAAATCTTGGTTC 25 44--52 750 47 [@B63] 1248R GCTRTRATAATGAGAAAGATTTCTGC 26 31--38 16S rRNA fD1 AGAGTTTGATCCTGGCTCAG 20 50 1500 56 [@B59] rP1 ACGGTTACCTTGTTACGACTT 21 43 ----------- ---------------------------------- ---------------------------- --------------- ------------------ ------------------------ ----------------------- ----------- ###### List of sequence accession numbers in NCBI GenBank. \* Nucleotide sequences from NCBI GenBank. ------- ----------- ------------------------- ------- -------------------------- -------------------------- ---------------------- ---------------------- TUCIM Other IDs Organism ng/µl COI COII cytB cad 5053 *C. explodens* sp. n. 14.7 [MF993252](MF993252) [MF993269](MF993269) [MF993286](MF993286) [MF993304](MF993304) 5056 *C. explodens* sp. n. 19.7 [MF993253](MF993253) [MF993270](MF993270) [MF993287](MF993287) [MF993305](MF993305) 5080 *C. explodens* sp. n. 10.4 [MF993254](MF993254) [MF993271](MF993271) [MF993288](MF993288) [MF993306](MF993306) 5098 *C. explodens* sp. n. 8.6 [MF993256](MF993256) [MF993273](MF993273) [MF993290](MF993290) [MF993308](MF993308) 5104 *C. explodens* sp. n. 16.6 [MF993257](MF993257) [MF993274](MF993274) [MF993291](MF993291) [MF993309](MF993309) 5148 *C. explodens* sp. n. 6 [MF993258](MF993258) [MF993275](MF993275) [MF993292](MF993292) [MF993310](MF993310) 5185 *C. explodens* sp. n. 7.3 [MF993259](MF993259) [MF993276](MF993276) [MF993293](MF993293) -- 5205 *C. explodens* sp. n. 29.8 [MF993260](MF993260) [MF993277](MF993277) [MF993294](MF993294) [MF993311](MF993311) 6600 *C. explodens* sp. n. 8 -- [MF993284](MF993284) -- -- 5855 *C. explodens* sp. n. 16.9 [MF993262](MF993262) [MF993278](MF993278) [MF993297](MF993297) [MF993314](MF993314) 5856 *C. explodens* sp. n. 28.2 [MF993263](MF993263) [MF993279](MF993279) [MF993298](MF993298) [MF993315](MF993315) 5942 *C. explodens* sp. n. 34.3 [MF993264](MF993264) [MF993280](MF993280) [MF993299](MF993299) [MF993316](MF993316) 5943 *C. explodens* sp. n. 142.1 [MF993265](MF993265) [MF993281](MF993281) [MF993300](MF993300) -- YG\* *C. explodens* sp. n. n.a. [EF634201](EF634201) -- -- 6461 *C. badia* 21.3 [MF993266](MF993266) [MF993282](MF993282) [MF993301](MF993301) [MF993317](MF993317) 6463 *C. badia* 5.4 [MF993267](MF993267) [MF993283](MF993283) [MF993302](MF993302) [MF993318](MF993318) 6601 *C. badia* 17.91 [MF993268](MF993268) [MF993285](MF993285) [MF993303](MF993303) [MF993319](MF993319) 5698 C. nr. saundersi 50.1 [KU975365.1](KU975365.1) [KU975366.1](KU975366.1) [MF993296](MF993296) [MF993313](MF993313) CH\* C. cf. cylindrica n.a. [EF634198](EF634198) -- -- 5086 *C. cylindrica* 26 [MF993255](MF993255) [MF993272](MF993272) [MF993289](MF993289) [MF993307](MF993307) 5300 CAMP004 *C. aruensis* 169.1 [MF993261](MF993261) -- [MF993295](MF993295) [MF993312](MF993312) Cflor36\* *Camponotus floridanus* n.a. [AY334397](AY334397) -- -- -- ------- ----------- ------------------------- ------- -------------------------- -------------------------- ---------------------- ---------------------- Phylogenetic analysis {#SECID0EBNAG} --------------------- GapStreeze v. 2.1.0 (<https://www.hiv.lanl.gov/content/sequence/GAPSTREEZE/gap.html>) was used for COI gene alignment with 95 % gap tolerance in order to retain only the conserved region. The individual gene alignments were subjected to best substitution model selection using the BIC criterion in SMS ([@B37]). Consecutively, HKY85, HKY85+I, HKY85+G, and GTR+G were chosen as best substitution models for genes cad, cytB, COI, and COII respectively. The concatenated alignment was partitioned for each locus using MrBayes v. 3.2.5 ([@B44]), and the respective substitution models were assigned to each partition. The substitution and branch length estimates were allowed to vary independently between each partition. Priors for an exponential distribution with mean 1 to all branch lengths and to all shape parameters were assigned for all four partitions. Metropolis-coupled Markov chain Monte Carlo (MCMCMC) sampling was performed using MrBayes v. 3.0B4 ([@B44]) with two simultaneous runs of four incrementally heated chains that performed 1 million generations. Bayesian posterior probabilities (PP) were obtained from the 50 % majority rule consensus of trees sampled every 100 generations after removing the first 25 % of trees using the "burnin" command. According to the protocol of Leache and Reeder (2002), PP values higher than 0.94 were considered significant. The phylogenetic trees were visualized in FigTree v. 1.4.3 ([@B42]) and then annotated using vector graphic software. Morphological methods {#SECID0E5NAG} --------------------- All specimens used for morphometry were card-mounted, individually numbered, and measured at magnifications from 25.6× up to 256× with a Nikon SMZ1500 binocular microscope. Genital structures of two male specimens were dissected and mounted separately. Results represent minimum and maximum values for each morph; in cases where a character could not be measured in all individuals, the number of measured specimens is given in parentheses. The complete dataset of measurements is provided in Suppl. material [6](#S6){ref-type="supplementary-material"} "measurements". **Measurements and indices (\* = only gynes and males)** **EL** Eye length. Maximum diameter of compound eye, measured in lateral view. **FeL** Femur length. Maximum length of metafemur, measured from base to apex. **FWL\*** Forewing length. Length of forewing, measured from tegula to distal tip. **HaL** Hair length. Length of the longest standing hair on first gastral tergite, measured from base to apex. **HL** Head length. Maximum length of head in full-face view, excluding mandibles, measured from anteriormost point of clypeus to posterior-most point of head vertex, parallel to midline. **HS** Head size. (HW + HL) / 2. **HW** Head width. Maximum width of head in full-face view (including eyes if protruding; only in gynes). **ML** Mesosoma length. Measured laterally from anterior surface of pronotum proper (excluding collar) to posterior extension of propodeal lobes. **MSW\*** Mesoscutum width. Maximum diameter of mesoscutum, measured dorsally. **NH** Node height. Height of petiolar node, measured laterally, from the intersection point of the axes of maximum height and length to dorsal apex **OcD\*** Ocellar distance. Minimum distance between lateral ocelli, measured between median borders. **OcW\*** Ocellus width. Maximum diameter of median ocellus. **OED\*** Ocellar eye distance. Minimum distance between lateral ocellus and outer border of compound eye. **PH** Petiole height. Maximum height of petiole in lateral view, measured from ventral-most point of petiolar sternum to dorsal apex. **PL** Petiole length. Maximum length of petiole in lateral view, measured from inflexion point of anterior constriction to posterior margin, perpendicular to axis of maximum height. **PS5** Length of maxillary palp segment V, measured from base to apex. **PS6** Length of maxillary palp segment VI, measured from base to apex. **SL** Scape length. Maximum length of antennal scape in dorsal view excluding basal neck and condyle. **SW** Scape width. Maximum width of antennal scape, measured dorsally. **TL** Total length. The added lengths of head (excluding mandibles), mesosoma, petiole, and gaster. **2r\*** Maximum length of 2^nd^ radial crossvein (see Figs [5e](#F5){ref-type="fig"}, [6b](#F6){ref-type="fig"}). **4Rs+M\*** Length of 4^th^ radial sector fused with median (see Figs [5e](#F5){ref-type="fig"}, [6b](#F6){ref-type="fig"}). **CI** Cephalic index. HW / HL × 100 **EI** Eye Index. EL / HW × 100 **FeI** Femur Index. FeL / HW × 100 **OI\*** Ocellar Index: OED / OcD × 100 **PI** Petiole Index. PH / PL × 100 **PSI** Palp Segment Index. (PS5 + PS6) / HS × 100 **SI** Scape index. SL / HW × 100 **SWI** Scape width index. SW / SL × 100 **WVI\*** Wing Vein Index. 4RsM / 2r × 100 Digital stacked images of most specimens (Figs [2](#F2){ref-type="fig"}--[6](#F6){ref-type="fig"}) were acquired with a Leica DFC camera attached to a Leica MZ16 binocular microscope with Leica Application Suite v3 and stacked with Zerene-Stacker 64-bit. Images of labels were taken with a Nikon D60 camera with an AF-S Micro Nikkor 105 mm objective and an EM-140 DG macro ring flash. Photographs of genital structures of males (Figs [7](#F7){ref-type="fig"}, [10c--f](#F10){ref-type="fig"}) as well as of the male *C. badia* specimen (Fig. [10 a, b](#F10){ref-type="fig"}) were created with the help of Leica Application Suite v3.8, using a Leica DFC450 camera attached to a Leica Z16APO optics carrier. All images were processed with Adobe Photoshop 7.0. Material examined {#SECID0EJ2AG} ----------------- Type material of *C. explodens*: **Holotype** (minor worker): Brunei, Temburong, Kuala Belalong Field Studies Centre, 04°33\'N, 115°09\'E, 60 m a.s.l., 10.XI.--5.XII.2015, leg. A. Laciny & A. Kopchinskiy ("YG Vienna Colony", specimen number COCY 01565). **Paratypes** (59 minor workers, 8 major workers, 16 gynes, and 6 males dry mounted; \> 500 imagines stored in 96 % ethanol): 19 minor workers, 2 major workers, 12 alate gynes, 4 dealate gynes, 6 males (including allotype) (all dry mounted), as well as 8 males, 2 alate gynes, ca. 500 minor workers (in alcohol) from the same nest sample as holotype; 1 major worker, same locality and collector as holotype, 17.IV.2015, "YG 373 main natural nest"; 1 major worker, same data as holotype ("YG doorkeeper \#19"); 2 major workers, same data as holotype except 20.IV.2015, leg. A. Kopchinskiy ("cf. YG 39 (351) artificial nest"); 8 minor workers, 2 major workers, same locality as holotype, 2002, leg. D.W. Davidson ("YG KB02-108"); 4 minor workers, same locality and collector as previous, no collection date, "YG 2025"; 5 minor workers, same locality and collector as previous, I.2012, "YG T-trail (202)"; 5 minor workers, same locality and collector as previous, 15.V.2014, "YG-2 (73)"; 7 minor workers, same locality and collector as previous, 15.V.2014, "YG-2 (49)"; 2 minor workers, same data as previous except Batu Apoi Forest Reserve, N04°32\', E115°10\', 200 m a.s.l., 25.XI.2004, ("CAYG A-370"); 15 minor workers (on 5 pins), same data as previous, except N04°55\', E115°19\', 60 m, 3.VII.2002, ("YG KB02-108 voucher"); 4 minor workers, Thailand, Chumphon Province, Krom Luang Chumphon W.S, 3.II.2002, leg. W. Jaitrong ("WJT02-TH-0116"); 5 minor workers, West Malaysia, Kelantan, 60 km NE Tanah Rata, Tanah Kerajaan, 1000 m a.s.l., 12.--30.IV2007, leg. P. Cechovský. Additional material: 3 pupae (Suppl. material [5](#S5){ref-type="supplementary-material"}) and 6 myrmecophilous crickets (*Camponophilus* sp., det. S. Ingrisch), from the same nest sample as the holotype. For unique identification numbers of all 90 dry mounted specimens (60 minor workers, 8 major workers, 16 gynes, and 6 males), as well as information on caste and colony affiliation, see Suppl. material [6](#S6){ref-type="supplementary-material"} "measurements". The holotype and a portion of the paratypes will be deposited at the Brunei Museum; additional paratypes will be housed in the Universiti Brunei Darussalam, the Natural History Museum Vienna, the University of California (Davis, USA), the Natural History Museum of Los Angeles County (Los Angeles, USA), the Thailand Natural History Museum (Technopolis, Thailand), and the collection of H. Zettel (Vienna, Austria). Molecular results {#SECID0E15AG} ================= The topology of the phylogram based on the concatenated alignment of 2757 bp was concordant with topologies of COI and COII and not contradicted by the topology of cytB. The phylogram based on cad was statistically unresolved (data not shown). The obtained Bayesian consensus tree (Fig. [1](#F1){ref-type="fig"}) shows conspecificity of the newly obtained *C. explodens* sp. n. specimens from Brunei and Thailand (TUCIM 6600) with a sequence previously deposited under "*Colobopsis cylindrica* s.l. YG". While there is some intraspecific variation within the analysed *C. explodens* sp. n. specimens, they form a clade distinctly separate from the closely related *C. badia*. The male of *C. badia* (TUCIM 6463) is clearly grouped with its conspecific workers from a nearby locality, thus confirming species identity. The herein examined representatives of the *C. saundersi* subclade, *C. explodens* sp. n., *C. badia* and the undescribed *C.* nrSA (see [@B34]), are clearly distinct from other members of the COCY group (e.g., *C. cylindrica*) and selected outgroup taxa of *Colobopsis* and *Camponotus*. {#F1} Taxonomic results {#SECID0EBEBG} ================= Colobopsis explodens -------------------- Animalia Hymenoptera Formicidae Laciny & Zettel sp. n. http://zoobank.org/DB4767B0-C745-4843-BE3F-B17DBCEB3A96 [Figs 2](#F2){ref-type="fig"} [, 3](#F3){ref-type="fig"} [, 4](#F4){ref-type="fig"} [, 5](#F5){ref-type="fig"} [, 6](#F6){ref-type="fig"} [, 7](#F7){ref-type="fig"} [, 8](#F8){ref-type="fig"} [, 9](#F9){ref-type="fig"} [; Suppl. materials 1](#S1){ref-type="supplementary-material"} [, 2](#S2){ref-type="supplementary-material"} [, 3](#S3){ref-type="supplementary-material"} [, 4](#S4){ref-type="supplementary-material"} [, 5](#S5){ref-type="supplementary-material"} [, 6](#S6){ref-type="supplementary-material"} [, 7](#S7){ref-type="supplementary-material"} 1. Camponotus (Colobopsis)sp. Yellow Goo: [@B9]: 470. 2. Camponotus (Colobopsis)sp. YG: [@B6]. [@B10]: 488. 3. Colobopsissp. YG: [@B8]: 518. [@B34]: 95. ### Etymology. Present participle of Latin *explodere*, referring to the "exploding"-like autothysis behaviour. ### Description of phenotypes. **Minor worker** (Figs [2](#F2){ref-type="fig"}, [4b--d](#F4){ref-type="fig"}; Suppl. material [1](#S1){ref-type="supplementary-material"}: S1a). Measurements of holotype minor worker: TL 6.78; HW 1.48; HL 1.67; HS 1.58; PS5 0.23; PS6 0.25; EL 0.42; SL 1.33; SW 0.14; ML 2.05; HaL 0.15; PH 0.55; PL 0.47; NH 0.33; FeL 2.05. Indices: CI 88; SI 90; SWI 11; EI 29; PI 116; FeI 139; PSI 30. Measurements of paratype minor workers: (n = 59): TL 4.74--7.21; HW 1.22--1.57; HL 1.30--1.78; HS 1.27--1.67; PS5 0.21--0.25 (20); PS6 0.20--0.26 (21); EL 0.33--0.43; SL 1.21--1.39; SW 0.11--0.16; ML 1.50--2.22; HaL 0.08--0.19; PH 0.41--0.56 (44); PL 0.33--0.49 (47); NH 0.24--0.38 (52); FeL 1.73--2.10. Indices: CI 85--94; SI 87--104; SWI 9--12; EI 27--29; PI 112--133 (41); FeI 123--151; PSI 28--35 (20). *Structures*: Head (Fig. [2a](#F2){ref-type="fig"}) subovate, longer than wide, narrower anteriorly; sides posteriorly convex, posterior cephalic margin roundly convex; microstructure consisting of very fine, isodiametric or transverse mesh-like reticules; intermixed punctures very fine and inconspicuous on face, larger but shallow laterally and ventrally. Eyes small compared to other castes (EI 27--29, vs. 28--31 in major workers and 35--37 in gynes), flat, positioned dorsolaterally. Ocelli lacking, in some larger specimens position of median ocellus indicated by shallow impression. Frons with very fine impressed midline; frontal carinae slightly converging anteriorly, not elevated. Median carina of clypeus not reaching anterior clypeal margin, especially in small specimens. Mandibles mostly smooth, with rather dense punctures; masticatory margin with five teeth. Maxillary palpi long (PSI 28--35). Antennal scape long, its length roughly equal to head width (SI 87--104), moderately flattened, slightly widened towards apex, integument punctate. Antenna 12-segmented; antennal segment III approx. 1/5 shorter than each IV and V, and approx. 2/5 shorter than II. Mesosoma slender, moderately low. Microreticulation isodiametric or slightly transverse, dorsally denser than laterally. Metanotal region delimited from mesonotum by a shallow groove; groove delimiting metanotum from propodeum indistinct or missing. Dorsal and posterior outline of propodeum rounded in lateral view, or meeting at an obtuse angle, dorsal face slightly convex, posterior face flat to shallowly concave. Legs slender. Petiole with isodiametric reticulation; petiolar node moderately high, its short, slightly convex anterior and its rather straight posterior face forming a triangular shape in lateral view, its apex not truncated, rather rounded; node narrow in dorsal view, a crest indistinct; a medial depression indicated in most specimens. Gaster: dorsum of tergites I--III with extremely fine, dense, transverse microreticulation, slightly shiny (Fig. [4b](#F4){ref-type="fig"}); mesh-like reticulation wider on lateral areas of tergites I--III , tergite IV, and sternites , therefore meshes appearing not so strongly transverse, and the integument shinier (Fig. [4c](#F4){ref-type="fig"}). Exposed parts of tergite V and sternite V with dense, almost isodiametric reticulation, dull; base of tergite V (usually covered by tergite IV) sculptured as tergite IV. {#F2} {#F3} {#F4} *Colour*: Body mainly reddish brown. Vertex of head, margins of clypeus, masticatory and lateral margins of mandibles, dorsum and ventral margins of mesosoma, mid portion of gastral tergites I--III, and legs slightly darker brown in most specimens; some specimens with darker area extending medially from head vertex to frons. Gastral tergites and sternites with very narrow hyaline margins. All gastral sternites, lateral fourths and posterior margins of tergites I--III, as well as entire tergites IV and V black. *Pilosity*: Dorsum of head with very short, inconspicuous, appressed and subdecumbent setae; a few very long, standing setae on frons near declivity to vertex, medial of frontal carinae, and on lateral portions of clypeus. Mesosoma and petiole with fine and short, whitish, velvety pilosity; long, standing, slightly undulated setae restricted to pronotum; declivity of propodeum and node of petiole with few very short standing setae. Gastral tergites with moderately dense, short whitish, decumbent setae and few slightly darker, longer standing setae, most of them in transverse rows near hind margins. Longest setae in transverse rows near hind margins of sternites and at base of gastral tergite V. *Notes*: Minor workers of *Colobopsis explodens* sp. n. show a continuous size variation across a remarkably wide range, similar to that found in the undescribed *Colobopsis* sp. nrSA (Fig. [8](#F8){ref-type="fig"}; compare with [@B34]). ### Phragmotic major worker (Figs [3](#F3){ref-type="fig"}, [4a](#F4){ref-type="fig"}; Suppl. material [1](#S1){ref-type="supplementary-material"}: S1b). Measurements of paratype major workers (n = 8): TL 7.30--8.71; HW 1.72--1.89; HL 2.25--2.58; HS 1.99--2.20; PS5 0.15--0.17 (6); PS6 0.15--0.17 (6); EL 0.50--0.56; SL 1.15--1.26; SW 0.17--0.20; ML 2.22--2.74; HaL 0.11--0.20; PH 0.59--0.69 (6); PL 0.45--0.51 (6); NH 0.40--0.45 (6); FeL 1.50--1.70. Indices: CI 71--77; SI 64--69; SWI 14--17; EI 28--31; PI 125--143 (6); FeI 87--95; PSI 14--17 (6). *Structures*: Integument mostly dull, only head and legs shiny. Head (Fig. [3a](#F3){ref-type="fig"}) large, subcylindrical, anteriorly truncated. On posterior areas of face punctation slightly stronger than in minor worker. Eyes somewhat larger and more distant from vertex compared to minor worker. Ocelli lacking, their positions often indicated by shallow grooves (Fig. [4a](#F4){ref-type="fig"}). Anterior part of head forming a large shield (Fig. [3a, b](#F3){ref-type="fig"}) formed by clypeal and genal components, limited by a sharp and elevated crest so that the shield surface appears concave in lateral view. Shield with fine isodiametric reticulation and rather variable, mostly longitudinal rugae; most prominent are a pair of rugae along sides of clypeus and a single median carina that does not reach the anterior margin, often reduced towards base. Genal part with curved rugae of variable number, length, and distinctiveness, but only exceptionally reaching onto the anteromedial triangle. Additional longitudinal rugae on clypeus often present, including a usually distinct pair of paramedian rugae running from base of clypeus over the crest anteriorly towards middle of shield; in specimens with short median carina, the area between these carinae more or less grooved. Longitudinal striation more regular and pronounced on frons and genae up to level of antennal insertions, laterally on genae similarly long and strong. Mandible with sharp and high ventrolateral ridge, coarsely punctate, its lateral face weakly rugose-striate; masticatory margin with acute apex and few (1--3) more or less distinct, very blunt teeth in distal half (Fig. [3b](#F3){ref-type="fig"}). Maxillary palpi very short (PSI 14--16). Antenna considerably shorter than head width (SI 64--69) and stouter than in all other morphs (Figs [3a](#F3){ref-type="fig"}, [8b](#F8){ref-type="fig"}); antennal scape distinctly widened towards apex. Mesosoma stouter and higher than in minors, especially promesonotum expanded; in lateral view dorsal and posterior face of propodeum forming an obtuse angle, somewhat less rounded than in minor workers, dorsally without concavity. Legs much shorter and stouter than in minors (Fig. [3c](#F3){ref-type="fig"}). Shape of petiole similar to minor workers, somewhat wider in dorsal view. Structures of gaster similar as in minor worker. *Colour*: Overall slightly darker than minor worker; head, legs and mesosoma reddish brown; gaster slightly darker chocolate-brown, becoming darker towards caudal apex, hyaline margins yellowish; elevated crest of frontal shield, anterior clypeal margin, frontal carinae, and masticatory and lateral margin of mandibles blackish brown. *Pilosity*: As in minor worker, except long setae on clypeus sides restricted to the area behind clypeal shield; mesonotum with standing setae which are approx. half the length of those on pronotum. *Notes*: The head shield with a sharp, elevated crest is typical for majors of the *Colobopsis saundersi* complex (Fig. [3b](#F3){ref-type="fig"}). ### Gyne (Fig. [5](#F5){ref-type="fig"}, Suppl. material [3](#S3){ref-type="supplementary-material"}: S3d). Measurements of paratype gynes (n = 16): TL 10.50--12.16; HW 1.74--1.83; HL 2.28--2.45; HS 2.02--2.14; PS5 0.19--0.21 (13); PS6 0.19--0.23 (13); EL 0.62--0.66; SL 1.33--1.45; SW 0.20--0.22; ML 4.11--4.63; HaL 0.14--0.29 (15); PH 0.77--0.92 (11); PL 0.54--0.67 (10); NH 0.40--0.54 (11); FeL 2.25--2.35; OcW 0.13--0.16; OED 0.34--0.38; OcD 0.54--0.64; FWL 9.72--10.50 (11); MSW 1.68--2.15; 2r 0.50--0.64 (12); 4Rs+M 0.14--0.32 (12). Indices: CI 73--77; SI 75--80; SWI 15--16; EI 35--37; PI 123--150 (8); FeI 125--132; PSI 18--20 (13); OI 54--69; WVI 26--58 (12). *Structures*: Head (Fig. [5a](#F5){ref-type="fig"}) large, subcylindrical, anteriorly truncated, similar to that in major worker with the following exceptions: eyes larger than in workers (EI 35--37) and breaking outline of head in full-face view. Ocelli fully developed, their colour ranging from almost clear to reddish amber. Head shield sharply delimited, but slightly smaller than in major worker, distinctly narrower than head width. Striation of clypeus, frons, and genae similar as in major, though somewhat more strongly developed on lateral parts of shield. Mandible with sharp ventrolateral ridge; its lateral face weakly rugose-striate, narrower than in major; dorsal-anterior face punctured; masticatory margin with acute apex and 3--4 blunt teeth in distal half, mandible basally with blunt ridges (Fig. [5b](#F5){ref-type="fig"}). Maxillary palpi moderately long (PSI 18--20). Antennal scape moderately long, slightly shorter than head width (SI 75--80), somewhat widened towards apex (Figs [5a](#F5){ref-type="fig"}, [8b](#F8){ref-type="fig"}). Mesosoma large, structures as typical for caste; propodeum large and evenly convex in lateral view. Cuticular microstructures dorsally consisting of very fine punctation, with intermixed larger punctures, laterally finely reticulated. Legs stout, but not as short as in major (Fig. [5c](#F5){ref-type="fig"}). Forewing venation strongly reduced, as in most Camponotini; M-Cu absent; Mf2+ interstitial (Fig. [5e](#F5){ref-type="fig"}). Petiole distinctly wider than in workers; node more rounded in lateral view, in some specimens its apex shallowly impressed medially, in others with two shallow lateral impressions forming a trilobed outline. Gastral tergites I--IV and sternites I--IV with extremely fine and dense microstructures consisting of strongly transverse meshes; only sides of tergites with wide mesh-like reticulation and shiny; tergite V with dense isodiametric reticulation. {#F5} *Colour*: Chiefly as in major worker. Head and pronotum reddish brown; ventral and posterior mesosoma, petiole, legs and gaster somewhat darker chocolate-brown; mandibles and ridges of clypeal shield blackish brown. Pronotum and mesonotum with very narrow yellow margins. Gastral tergites medially with very narrow hyaline margins; sternites with relatively broad posterior margins. Wings hyaline, but forewing cells along veins, as well as pterostigma darkened to brownish. On hind wing all veins pale yellow (Fig. [5c--e](#F5){ref-type="fig"}). *Pilosity*: Short pilosity and distribution of long setae on head, petiole, and gaster similar as in major worker, but that of mesosoma different; pronotum with few long, undulated setae. Medial part of mesonotum (between parapsidal furrows) with numerous long erect setae, scutellum with few long erect setae; lateral part of mesonotum in front of tegulae without setae. *Notes*: The head shield with a sharp, elevated crest is typical for gynes of the *Colobopsis saundersi* complex (Fig. [5b](#F5){ref-type="fig"}). ### Male (Figs [6](#F6){ref-type="fig"}, [7](#F7){ref-type="fig"}). This is the first detailed description and illustration of males from the *C. cylindrica* group. Measurements of allotype male: TL 7.11; HW 1.26; HL 1.20; HS 1.23; PS5 0.20; PS6 0.15; EL 0.44; SL 0.84; SW 0.11; ML 2.54; HaL n.a.; PH 0.46; PL.40; NH 0.29; FeL 1.83; OcW 0.18; OED 0.27; OcD 0.43; FWL 6.33; MSW 1.37; 2r 0.38; 4Rs+M 0.27. Indices: CI 105; SI 66; SWI 13; EI 35; PI 116; FeI 145; PSI 28; OI 62; WVI 70. Measurements of paratype males (n = 5): TL 6.46--6.85; HW 1.24--1.29 (4); HL 1.14--1.24; HS 1.20--1.27; PS5 0.17--0.21 (4); PS6 0.13--0.17 (4); EL 0.43--0.46; SL 0.80--0.85; SW 0.10--0.12; ML 2.38--2.87; HaL n.a.; PH 0.45--0.49 (4); PL 0.38--0.40 (4); NH 0.26--0.33 (4); FeL 1.71--1.86; OcW 0.18--0.19; OED 0.25--0.27; OcD 0.43--0.46; FWL 5.87--6.33; MSW 1.17--1.50; 2r 0.38--0.47; 4Rs+M 0.14--0.22. Indices: CI 104--110 (4); SI 64--67 (4); SWI 12--15; EI 35--36 (4); PI 113--123 (4); FeI 136--151 (4); PSI 27--30 (4); OI 53--62; WVI 31--53. *Structures*: Head (Fig. [6a](#F6){ref-type="fig"}) small, subtrapezoidal, eyes very large, round and protruding, EL more than one third of HL (EI 35--36). Ocelli very large, diameters larger than in gynes. Integument of head rather matt. Frons and genae finely reticulated, genae additionally finely punctured. Clypeus with some stronger punctures at margins (at base of setae), median carina weakly developed, present in proximal third of clypeus or entirely obsolete. Frons with impressed midline from median ocellus to level of antennal insertions. Frontal carinae weakly developed, converging more strongly than in minor worker. Mandible short with reduced dentition, masticatory margin with 2--3 blunt teeth; dorsal surface finely punctate. Maxillary palpi long (PSI 27--30). Antenna 13-segmented; scapes short (SI 64--67) and relatively slender (Fig. [8b](#F8){ref-type="fig"}). First funicular segment conspicuously enlarged distally, pear-shaped, 30--50% wider and ca. 20% longer than the following segment (Fig. [6a](#F6){ref-type="fig"}); all other funicular segments cylindrical, without modifications. Mesosoma large, structures as typical for alate ants. Mesoscutum anteriorly strongly convex with narrow impressed midline in posterior tenth. Scutellum moderately elevated; propodeum evenly convex. Cuticular microstructures of mesosoma consisting of a very fine reticulation with intermixed minute punctures at bases of short hairs, additionally with larger punctures dorsally at bases of erect setae. Legs very long and slender (FeI 136--151). Forewing venation strongly reduced, as in most Camponotini. M-Cu absent; 4Rs+M shortly developed or (more rarely) Mf2+ interstitial (Fig. [6b](#F6){ref-type="fig"}). Petiole small; in lateral view node more bluntly rounded than in female castes, anterior and posterior faces straight, not convex, apex not impressed medially in dorsal view. Gastral tergites I--IV and sternites I--IV with fine and dense microreticulation consisting of moderately transverse meshes; only sides of tergites with wide meshes and shiny; tergite V with almost isodiametric reticulation. Sternite VI posteriorly emarginated, sternite VII truncated. {#F6} *Genital structures* (Fig. [7](#F7){ref-type="fig"}): Genital capsule (Fig. [7a--c](#F7){ref-type="fig"}) approx. as long as wide in dorsal aspect (Fig. [7a](#F7){ref-type="fig"}), ventrally longer than dorsally, protruding from apex of gaster. Gonopod high, distally broadly rounded. Gonostylus (Fig. [7c](#F7){ref-type="fig"}) elongated and acuminated, with reticulated microstructure (only visible at very high magnification) and some long setae. Basivolsella (Fig. [7e](#F7){ref-type="fig"}) dorsally with roundish structure, ventrally with evenly distributed, comparatively short setae. Digitus (Fig. [7e](#F7){ref-type="fig"}) large, evenly widened towards apex; apex rounded but with rectangular corner ventrally. Penis valvae (Fig. [7d](#F7){ref-type="fig"}) in dorsal aspect broad at base, but very narrow distally. Valviceps leaf-shaped in lateral view, apically rounded; surface smooth; ventral margin with very fine serration. {#F7} {#F8} *Colour*: Mainly dark chocolate-brown. Head somewhat darker; eyes pale grey to blackish; ocelli translucent, ranging from almost clear to reddish amber. Antennae and legs lighter brown, fading into yellowish towards apices. Margins of mesoscutum, scutellum, and metanotum lighter yellowish brown. Gastral tergites medially with very narrow hyaline margins; sternites with relatively broad, indistinctly separated posterior margins. Wings almost hyaline, with a slight whitish tinge, but forewing cells along veins, as well as pterostigma darkened to brownish, all veins pale yellowish brown. On hind wing all veins pale yellow. *Pilosity*: On head setae sparsely distributed, short, inconspicuous, appressed, subdecumbent; a few very long standing setae on frons near vertex, and on anterior and posterior clypeal margins. Mandibles with dense short pilosity on lateral face, and few moderately long setae on anterolateral margin. Short pilosity and distribution of long setae on mesosoma, petiole, and gaster similar as in gyne, but pronotum lacking long, undulated setae. Medial part of mesonotum (between parapsidal furrows) with numerous long erect setae, scutellum with few long erect setae; lateral part of mesonotum in front of tegulae without setae. Tegulae with dense brush of setae. Petiole with a few stout setae anteroventrally (see insert Fig. [6c](#F6){ref-type="fig"}). Petiolar node lacking any standing setae; gastral tergite I without or with few subdecumbent, moderately long setae. Posterior gastral tergites and sternites (segments II and following) with sparse, relatively long, obliquely standing setae. #### Biological notes on *Colobopsis explodens* sp. n. {#SECID0EAUAI} Colonies of *C. explodens* sp. n. observed in the Ulu Temburong National Park are commonly polydomous and polygynous. This species was selected as a model for the study of the "exploding ants" because among the species with advanced autothysis behaviour it was the most abundant COCY taxon in the vicinity of KBFSC. *Colobopsis explodens* sp. n. frequently nests on dipterocarp trees and its colonies can contain thousands of individuals. The largest part of the studied colony lived on a 60 m tall *Shorea johorensis* Sym. (Dipterocarpaceae) tree identified morphologically and by DNA barcoding (matK, identical to GenBank accession number [KY973022](KY973022), E-value is zero; [@B27]). The colony's foraging ground included the canopy of the main tree, its direct vicinity, and also covered canopies of a 25 m tall *Horsfieldia wallichii* (Hook.f. & Th.) Warb. (Myristicaceae) tree and a smaller tree of *Shorea maxwelliana* King (9 m). Colony fragments on all trees were connected by ant trails either through the canopy or on the forest floor in the litter layer. The total area occupied by the colony was estimated to be at least 2500 m^2^. The colonies are distributed three-dimensionally, occupying any suitable nesting ground within the colony boundaries. On the main tree, we found four nesting sites of the examined colony in dead branches at heights ranging between 35 and 55 m above ground and two nesting sites in the living stem 50--60 m above ground. No nests in living branches were observed. At least five nest entrances were also seen in the stem of *S. maxwelliana*. No signs of necrosis of the plant tissue were observed around stem entrances. The translocation of a nest fragment in a fallen branch to the laboratory's terrace, 30 m away from its original location, resulted in the expansion of the colony's foraging ground to a neighbouring *Shorea* sp. tree where these ants were not previously present, while the connection to the colony on the original host tree was maintained. If provided with an appropriate artificial nest (Fig. [9c--d](#F9){ref-type="fig"}), *C. explodens* sp. n. ants will inhabit it within several weeks up to several months and even use it to rear brood. One artificial nest, mounted on the main host tree, was colonized one week after it was installed. For the activity assessment, the easily accessible artificial nest \#38 was observed. During behavioural monitoring, *C. explodens* sp. n. was observed to be mainly diurnal, foraging between 6:00 and 18:00 hrs, with peak activity around 9:00 and 16:00 hrs (Fig. [9a](#F9){ref-type="fig"}). The activity correlated positively with the temperature with lowest values at 24.2 °C and highest at 28.6 °C (Fig. [9b](#F9){ref-type="fig"}). The atmospheric pressure and clouds did not influence the activity of *C. explodens* sp. n. (Suppl. material [6](#S6){ref-type="supplementary-material"} "activity"); humidity was constant over the period of observations ranging from 86 to 88%. A slight rain on a warm day did not reduce the activity of ants near the nest but no activity was observed during heavy rains. However, if a shelter was provided, *C. explodens* sp. n. remained active also during the rain and even after sunset. A drastic reduction in the number of minor workers at the nest entrance was observed on the days of nuptial flight, when several alate gynes and males left the nest in the early evening (Suppl. material [6](#S6){ref-type="supplementary-material"} "activity"). Between one and six minor workers ("guards") were frequently positioned at the nest entrance, touching all incoming and outgoing workers with their antennae and seemingly monitoring the activity of foragers. In the early afternoon of the day with the highest activity, larvae were carried out of the nest. No carrying of larvae into the nest was ever observed. {#F9} Remarkably, during all observations the numbers of the minor workers leaving and entering the nest were almost equal. The fact that this proportion did not change over the day (Fig. [9a](#F9){ref-type="fig"}) suggests a tendency to maintain a constant number of individuals present inside the nest. After dusk, other species of Camponotini such as *Polyrhachis* spp., *Camponotus* spp., and *Dinomyrmex gigas* (Latreille, 1802) were observed on the trees in the vicinity of the artificial nest. Within the colonies, minor workers were by far the most abundant caste of *C. explodens* sp. n., whereas major workers (soldiers) were rare and almost never seen outside the nest. Alate gynes and males were observed leaving the nest during nuptial flight after dusk on two occasions during our field observations (Suppl. material [6](#S6){ref-type="supplementary-material"} "activity"). Several more alate sexuals were found inside a detached nest fragment (Suppl. material [3](#S3){ref-type="supplementary-material"}). The same nest fragment also contained symbiotic ant crickets of the genus *Camponophilus* Ingrisch, 1995. DNA barcoding of the cricket based on COI sequence resulted in the highest value of 80 % similarity to COI sequences of insects from several groups including Mann's ant cricket *Myrmecophilus manni* (Schimmer, 1911) ([EU938370](EU938370), [@B19]). Thus, the molecular identification of these crickets is currently not possible due to lack of reference sequences. Observations have shown that minor workers of *C. explodens* sp. n. display a characteristic, possibly defensive pose with raised gaster (Suppl. material [1](#S1){ref-type="supplementary-material"}: S1a) (compare with [@B9]) and are extremely prone to self-sacrifice when threatened. The mandibular gland content is released during autothysis by contracting the gaster until the integument ruptures, leading to the death of the ant (Suppl. material [4](#S4){ref-type="supplementary-material"}). The secretion is slightly viscous, sticky, and has a species-specific bright yellow colour and a distinctive spice-like odour (Hoenigsberger et al., in prep.). Minor workers of *C. explodens* sp. n. spend significant time on leaves, which has previously been hypothesized to contribute to their nutrition (compare with [@B9], [@B8]). However, the exact purpose of their activity on leaves is yet to be understood. Observations suggest a patrolling or monitoring behaviour aiming at the removal of debris from the phyllosphere (mainly adaxial leaf surface but also abaxial leaf and petiole surfaces) and possible deterrence of intruding arthropods. Similar "cleaning" behaviour was observed *in vitro*, as well. Another very specific behaviour was exhibited on the tree bark: Minor workers "graze" on the layers of epiphytes (mosses, lichens, algae, filamentous fungi, and yeasts) with their mandibles, often for up to 60 minutes. This behaviour differs from the cleaning behaviour on leaves and presumably contributes to the ants' nutrition. Preliminary feeding experiments using cultures of filamentous fungi isolated from the phyllosphere of the host trees remained unsuccessful, no fungal feeding was observed. Only a suspension of yeast in water was accepted *in vitro* (M. Rahimi, pers. obs.). However, minor workers of *C. explodens* sp. n. have been observed to feed on small dead insects, fruit, and fish when offered on the foraging grounds (A. Kopchinskiy, A. Laciny & M. Hoenigsberger pers. obs.). Commonly observed modes of behaviour of *C. explodens* sp. n. *in situ* and *in vitro* as well as a variety of nesting sites are documented in the Suppl. material [7](#S7){ref-type="supplementary-material"} (Video S7). The molecular analysis of the mandibular gland (MG) content of *C. explodens* sp. n. resulted in PCR amplification and sequencing of the 16S rRNA fragment of the bacteria *Blochmannia* sp. (Gammaproteobacteria), a genus of obligate symbiotic bacteria found in carpenter ants ([@B62]). We revealed four identical mOTUs originating from two different DNA extracts from samples composed of five pooled MG reservoirs of the minor workers each. The sequences of 728 nt were 99 % similar (11 SNP sites) to the "uncultured bacterium clone 193-11" [KC136854](KC136854) from *Camponotus* sp. voucher KC-A017-01 defined as *Blochmannia* sp. in [@B45]. More detailed data on autothysis, composition of mandibular gland secretion, biodiversity of the COCY-associated microorganisms, and experimental assessment of nutrition will be presented in future publications. Colobopsis badia ---------------- Animalia Hymenoptera Formicidae (Smith, 1857) [Figs 8](#F8){ref-type="fig"} [, 10](#F10){ref-type="fig"} 1. Formica badia: [@B51]: 54. 2. Camponotus badius: [@B43]: 3. 3. Colobopsis badia: [@B57]: 350. [@B4]. ### Type material examined. 1 lectotype minor worker (Oxford University Museum of Natural History, present designation), Singapore, "Formica badia", "Syntype", CASENT 0901897, "Lectotypus Formica badia Smith, 1857 des. Laciny & Zettel, 2017", 2 paralectotype minor workers (Oxford University Museum of Natural History) mounted on the same card, Sarawak ("Sar 32"), "Formica badia", "Syntype", "Paralectotypes Formica badia Smith, 1857". ### Additional material examined. 1 male (Natural History Museum Vienna), Thailand, Trang Province, Nayong District, Khao Chong Botanical Garden, at light of head quarter, 7°33\'N, 99°46\'E, 60 m a.s.l., 1--7.VI.2016, leg. H. Zettel (68); 10 minor workers (Natural History Museum Vienna), Thailand, Trang Province, Nayong District, Khao Chong Botanical Garden, trail to Ton Pliw Waterfall, N07°32\'34\", E99°47\'33\", 150 m a.s.l., 1--7.VI.2016, leg. H. Zettel (66-4). ### Description notes on the type specimens. *Lectotype*: Minor worker glued to a square cardboard, in relatively good condition; right hind leg missing; tarsi of middle legs and left hind leg broken; erect setae on dorsum probably lost. Structures agree well with other species of the *C. saundersi* complex, a few characteristic features are given: Setae on scape more decumbent than in *C. explodens* sp. n. Dorsal outline of mesosoma almost straight, only with weak indentation at meso-metanotal suture. Propodeum forming a distinct obtuse angle in lateral view. Petiolar node relatively short, apex acute in lateral view, its crest slightly indented in middle. Tergites I--III with very fine, strongly transverse microsculpture (lateral parts not visible). Colour relatively dark brown; appendages strongly infuscate; antennal segments III--XII, meso- and metafemora almost black. *Paralectotypes*: Two minor workers glued to the same square cardboard, in relatively poor condition. Left specimen with damaged head and gaster, lacking right middle leg; major parts of body covered by dirt or glue; most erect setae probably lost. Right specimen with slightly damaged head, lacking gaster and right hind leg; some parts of body covered by dirt or glue; most erect setae probably lost. The two specimens are probably conspecific, but conspecificity with the lectotype is uncertain. The combination of morphological features is intermediate between *C. badia* and *C. explodens* sp. n.: setae on scape similar to *C. explodens* sp. n., more erect than in the lectotype; dorsal outline of mesosoma intermediate, more structured than in the lectotype, but propodeum with angle; shape of node intermediate, apex more acute than in *C. explodens* sp. n. Colour almost as dark as in the lectotype. Measurements of lectotype minor worker: TL 6.13; HW 1.51; HL 1.63; HS 1.57; PS5 n.a.; PS6 n.a.; EL 0.40; SL 1.43; SW 0.15; ML 1.96; HaL 0.17; PH 0.54; PL 0.36; NH 0.32; FeL 2.22. Indices: CI 93; SI 95; SWI 10; EI 26; PI 150; FeI 147; PSI n.a. Measurements of paralectotype minor workers\* (n = 2): TL 6.13, n.a.; HW n.a., 1.36; HL n.a., 1.52; HS n.a., 1.44; PS5 n.a., 0.21; PS6 n.a., 0.25; EL 0.36, 0.37; SL 1.39, 1.40; SW 0.12, 0.13; ML 1.89; HaL 0.13, n.a.; PH n.a., 0.51; PL 0.37, 0.42; NH 0.27; FeL 2.15. Indices: CI n.a., 89; SI n.a., 103; SWI 8, 9; EI n.a., 27; PI n.a., 121; FeI n.a., 158; PSI n.a., 32. \*One specimen with strongly damaged head, one with missing gaster. Measurements of non-type minor workers (n = 10): TL 5.64--6.23; HW 1.46--1.59; HL 1.63--1.72; HS 1.54--1.65; PS5 0.24--0.25 (3); PS6 0.24 (3); EL 0.38--0.40; SL 1.37--1.43; SW 0.13--0.14; ML 1.96--2.22; HaL 0.13--0.19; PH 0.51--0.56 (7); PL 0.41--0.45; NH 0.31--0.37 (9); FeL 2.09--2.28. Indices: CI 88--92; SI 90--96; SWI 9--10; EI 25--27; PI 118--130 (7); FeI 141--148; PSI 30--31 (3). ### Male. *Notes on collecting and identification*: A single male collected at light was identified as a specimen of the *C. cylindrica* group. DNA barcoding revealed specific identity with a nest series of *C. badia* from the same botanical garden. The morphological identification of this nest series (Col. 66-4) was carried out by direct comparison to the lectotype of *C. badia*. *Description* (Fig. [10](#F10){ref-type="fig"}): Overall very similar to *C. explodens* sp. n. and differing by the following characters: Measurements of male (n = 1): TL 8.28; HW 1.35; HL 1.26; HS 1.30; PS5 0.20; PS6 0.15; EL 0.48; SL 0.83; SW 0.10; ML 3.07; HaL n.a.; PH 0.47; PL 0.41; NH 0.31; FeL 1.96; OcW 0.19; OED 0.29; OcD 0.46; FWL 7.43; MSW 1.37; 2r 0.41; 4Rs+M 0.37. Indices: CI 107; SI 61; SWI 13; EI 36; PI 116; FeI 145; PSI 26; OI 61; WVI 91. *Structures*: Size larger (TL ca. 8.3 mm). Integument rather shiny (Fig. [10a, b](#F10){ref-type="fig"}), especially on mesosoma. Clypeus with distinctly developed median carina, almost reaching anterior margin. Maxillary palpi (PSI 26) and antennal scapes (SI 61) relatively short. First funicular segment slightly more enlarged (30% wider than the following segment, Fig. [10a](#F10){ref-type="fig"}). Vein 4Rs+M of forewing long. Petiolar node slightly more widely rounded in lateral aspect. *Genital structures* (Fig. [10c--g](#F10){ref-type="fig"}) very similar to *C. explodens* sp. n., with the following exceptions: Gonostylus very narrow, with weaker reticulation of lateral surface (Fig. [10e](#F10){ref-type="fig"}). Basivolsella with extremely short ventral setae (Fig. [10f](#F10){ref-type="fig"}). Digitus with rounded apex, without ventroapical corner (Fig. [10f](#F10){ref-type="fig"}). Valviceps with slightly coarser ventral serration (Fig. [10g](#F10){ref-type="fig"}). {#F10} *Colour*: Head chiefly dark brown, with lighter area comprising frons between antennal insertions and clypeus. Eyes grey. Ocelli clear, almost colourless. Posterior and anterior clypeal margins, as well as proximal fourth of clypeal carina black. Gaster dark brown. Mesosoma, petiole, mandibles, antennae, and legs lighter brown, appendages becoming yellowish towards apices. Antennal insertions, mandibular bases, margins of thoracic sclerites (especially below tegulae) creamy yellow. Gastral tergites medially with very narrow hyaline margins; sternites with relatively broad posterior margins. Wings hyaline, forewing with a slight brownish tinge and cells along veins, as well as pterostigma darker brownish, all veins pale brown. On hind wing all veins pale yellow. *Pilosity*: Appressed and subdecumbent setae comparatively shorter and sparser, but difference less obvious on gaster. Standing setae on mesonotum and gaster shorter, on mesonotum less numerous. *Comparative notes*: The male of *C. badia* can be distinguished from males of *C. explodens* sp. n. by larger body size, differing colour pattern, more shiny integument, well-developed clypeal carina, differing proportions of wing venation, and relatively shorter scapes (Fig. [8b](#F8){ref-type="fig"}). In the genitalia, the most striking differences are in the narrower gonostylus and the more rounded digitus apex (compare Figs [7e](#F7){ref-type="fig"}, [10f](#F10){ref-type="fig"}). Discussion {#SECID0ED5BI} ========== Molecular results {#SECID0EH5BI} ----------------- In this study, three mitochondrial DNA loci and one nuclear DNA fragment were applied for the construction of a molecular phylogenetic tree (Fig. [1](#F1){ref-type="fig"}). The evolutionary analysis based on four loci showed that *C. explodens* sp. n. is clearly genetically distinct from morphologically similar species. A minor level of infraspecific polymorphism within the specimens from Brunei was observed for COI, COII, and in particular for cytB marker. While tree topologies based on single mitochondrial loci were concordant, the cad tree (nuclear locus) was not resolved for the entire COCY group. The comparison of COI sequences with several hundred COCY sequences available in our local database and 13 sequences deposited in NCBI (Nov. 2017) suggests that this marker can be used for the reliable molecular identification (DNA barcoding) of *C. explodens* sp. n., as COI sequences of the nearest COCY taxa share only 91% similarity (*C. badia*), and the similarity to the selected non-COCY species *C. aruensis* is 83%. The DNA extraction from the gastral parts of the mandibular gland reservoirs of *C. explodens* sp. n. minor workers resulted in drastically low yields indicating no abundant microbial symbionts present there. The successful 16S rRNA PCR amplification gave a sharp band that was sequenced with high reproducibility. The 16S rRNA fragment corresponding to the whole genome sequenced *Blochmannia* endosymbiont of North American *Colobopsis obliquus* strain 757 (NCBI GenBank accession number [CP010049](CP010049), [@B62]) was 92 % similar to mOTU revealed in this study resulting in 56 polymorphic sites. This confirms that *C. explodens* sp. n. also harbours these bacteria that usually colonize the midgut of Camponotini workers ([@B47]) and are considered to be beneficial for N-nutrition of these ants; they may also contribute to the general health of the workers and gynes. Thus, the detection of cf. *Blochmannia* bacteria rather indicated the contamination of the MG sample by fragments of the digestive system. In this respect, it is interesting to note that no *Wolbachia* (Alphaproteobacteria) mOTUs were recovered, but neither digestive tract nor ovaries were specifically investigated. Taxonomy {#SECID0EGDCI} -------- The treatment of *Colobopsis* as a genus separate from *Camponotus* is supported by molecular, morphological, and biological data ([@B3], [@B57]). Naked pupae ([@B60]; see Suppl. material [5](#S5){ref-type="supplementary-material"}) and presence of phragmotic soldiers and gynes are important features of *Colobopsis*, although unknown in many of the 94 valid species assigned to this genus by [@B57]. The morphological separation of minor workers of *Colobopsis* and *Camponotus* is chiefly based on head morphology, but complicated by extensive evolutionary changes within each group ([@B57]); however, the phylogeny of *Colobopsis* species has not been studied to date. The molecular data published by [@B3], obtained from only four species, do not allow an interpretation of the relationships of intrageneric clades. Attempts to classify the species by morphological characters ([@B16], [@B39]), although useful for a rough sorting of species, probably hardly reflect their evolutionary relationships. A first attempt of a comprehensive classification of the species of *Colobopsis* (as a subgenus of *Camponotus*) was done by Carlo Emery. In his outstanding treatment of Formicinae ([@B16]) he treated 58 species and established six groups to hold 49 of them (nine remained unclassified). He defined the \[Camponotus (Colobopsis)\] *cylindricus* group by a gradual variation between worker and soldier, interspecific variation of head in soldiers and females (from concave and marginate to oblique and obtuse), and generally large size. Emery included eight species presently classified as *Colobopsis* ([@B57]), of which *Colobopsis calva* Emery, 1920, *C. quadriceps* (Smith, 1859), and *C. smithiana* (Wheeler, 1919) are not presently assigned to this group (see below), whereas *C. badia* and *C. corallina* were not included (listed under incertae sedis). Although [@B16] correctly recognized the size variation of workers, he failed to recognize the unique characteristics of the soldier caste (see [@B34]). More species of the COCY group were subsequently described by [@B55], [@B41] and [@B30], [@B32]). A second attempt at classification was made by [@B39]: His Camponotus (Colobopsis) cylindricus group consists of species with "neck attached to head well below vertex" and is broader than [@B16] *cylindricus* group. It includes the following species (according to current classification) that do not fit the characteristics of the COCY group in the present sense: *Colobopsis anderseni* (McArthur & Shattuck, 2001), *C. brachycephala* Santschi, 1920, *C. cotesii* (Forel, 1893), *C. desecta* (Smith, 1860), *C. excavata* (Donisthorpe, 1948), *C. hosei* (Forel, 1911a), *C. mutilata* (Smith, 1859), and *C. quadriceps*, as well as *Camponotus dedalus* Forel, 1911b, and *Camponotus kutteri* Forel, 1915. According to our morphological studies the COCY group can be defined as such: polymorphic *Colobopsis* with at least three distinct female castes: (i) winged, phragmotic gynes, (ii) phragmotic soldiers (doorkeepers), and (iii) minor workers with a considerable size variation; intermorphic workers may occur in addition ([@B34]). Minor workers: Vertex highly raised above foramen. Eyes of minor worker small and flat, not breaking head sides in full-face view. Entire trunk with dense, reticulated microstructures; punctures of integument often reduced. Head with moderate, mesosoma with dense pubescence. Mesosoma (at least the pronotum, except in *C. clerodendri* Emery, 1887) dorsally with long undulated setae, never arranged in distinct rows. Gaster with appressed pubescence and two or three types of setae of different lengths (not arranged in rows, except at hind margin). Soldiers (not known of all taxa): differing from minor workers by enlarged heads and short appendages (antennae, palpi, legs); in most species with a clearly circumscribed head shield for phragmosis. Microsculpture and pilosity similar to minor worker. Following this definition, the COCY group presently comprises 17 names in the rank of species, subspecies or variations, which are partly in synonymy to each other. The strong intraspecific variation of minor workers, the frequently lacking knowledge on soldiers (or gynes), and the historical descriptions of taxa from different morphs (either workers or gynes) make the species taxonomy a true challenge. A preliminary analysis of morphological and molecular data (unpublished) supports the division of the group into four species complexes (molecular data of one complex not available). We restrict the following analysis to the species of the *C. saundersi* complex, which includes *C. explodens* sp. n. and can be defined by the following combination of characters observable in minor workers and soldiers: head always red or brown (not black); mesosoma moderately elongated and dorsally with some long undulated setae, at least on pronotum; node of petiole without long setae; gastral tergites with dense (in most species strongly transverse) micro-reticulation and with small hair pits (without large grooves). Soldiers and gynes (not known of all species) have a strongly truncated head, in most species with a well-defined (crested) head shield. The following names are available in this group (listed chronologically): *Colobopsis badia* (Smith, 1857), *C. corallina* Roger, 1863, *C. saundersi* Emery, 1889, C. badia var. krama Forel, 1912, *C. badia saginata* Stitz, 1925, *C. solenobia* (Menozzi, 1926), and *Colobopsis trieterica* (Menozzi, 1926), comb. n. *Colobopsis corallina* (=*C. solenobia* syn. n., =*C. trieterica* syn. n.) is an endemic species from the Philippines. Soldiers and gynes differ strongly from *C. explodens* sp. n. and other taxa of the complex (as far as such morphs are known) by a very obtusely margined, not crested head shield. Minors have a bright orange colour on head, mesosoma, and petiole, often extending to gastral tergite I. Morphometrically, the examined minor workers of *C. corallina* (n = 31) mainly differ from those of *C. explodens* sp. n. by a greater average length of appendages (SI 92--109 vs. 87--104; FeI 136--159 vs. 123--151; PSI 30--39 vs. 28--35). The greatest similarity is observed between *C. explodens* sp. n. and *C. saginata* (stat. n.), a taxon only known from a single alate gyne from Northern Borneo. The important structures of the head shield are almost identical. Although strongly different from *C. explodens* sp. n. by pale orange brown colour, this gyne differs only by subtle morphometric characters of which the long and distally wide scape seems to be the most reliable (SI 83 vs. 78--80). The length of vein 4Rs+M is considerably longer in *C. saginata* than in *C. explodens* sp. n. (WVI 65 vs. 26--58). Colobopsis badia var. krama, described from a soldier from Java ([@B24]) is a very poorly known taxon. We have not been able to study any further material from Java yet. The type (illustrated by [AntWeb.org](http://AntWeb.org) under CASENT0910610) differs from *C. explodens* sp. n. by a pale red head that strongly contrasts with the dark brown mesosoma, by a well-developed median carina of the head shield that reaches the foremargin of the clypeus, and by a stronger punctation of the preocellar area. *Colobopsis badia* was described by [@B51] from Singapore and Sarawak (Borneo). However, the original description is too brief to draw any meaningful taxonomic conclusions. [@B58] describes workers of this species in more detail, also noting the secretion of a sticky liquid upon capture. He mentions a strong variability in colouration (from red to almost black with reddish head) and propodeal shape. This raises the question whether all examined specimens were truly members of the same species or perhaps belonged to one of the other, similar species of the *C. saundersi* complex. We examined the three syntype minor workers of *C. badia* in the Oxford University Museum of Natural History. To fix the identity of this taxon, we have chosen the syntype from Singapore (imaged by [AntWeb.org](http://AntWeb.org) under CASENT0901897) as the lectotype of *Formica badia*. The two syntypes from Sarawak are in a relatively poor condition, which does not allow a complete morphometric analysis, and therefore the conspecificity with the *C. badia* type remains doubtful. We were not successful in obtaining fresh material of *C. badia* from Singapore, but a nest sample (minors only) from southern Thailand (Trang Province) which agrees well with the lectotype in morphology, especially morphometry, was available for a molecular analysis. It shows that *C. badia* and *C. explodens* sp. n. are closely related, but distinct (Fig. [1](#F1){ref-type="fig"}). Although very similar to *C. explodens* sp. n. in overall habitus and colouration, the examined *C. badia* minor workers are on average somewhat larger with less size variation (HW 1.22--1.57 vs. 1.46--1.59) and possess longer appendages (e.g., FeI 123--151 vs. 141--168; see Fig. [8a](#F8){ref-type="fig"}). We examined two syntype minor workers of *Colobopsis saundersi* from Myanmar ("Tenasserim, Thagata", one illustrated by [AntWeb.org](http://AntWeb.org) under CASENT0905463). Morphometric analysis revealed no differences between the types of *C. saundersi* and *C. badia*, suggesting that the two species should be synonymized. *Colobopsis saundersi* was considered a junior synonym of *C. badia* by Carlo Emery himself ([@B14]) but revived from synonymy by [@B2] without providing a reason. The large geographic distance of the type locality of *C. saundersi* and some minor differences in morphology led us to the decision to refrain from a formal synonymization at this time. A comparative molecular analysis of specimens from the type localities (Myanmar, Singapore) would most likely be necessary to corroborate this synonymy. Morphology of males {#SECID0EJADI} ------------------- The morphology of males of the tribe Camponotini is insufficiently studied, so that a complete comparison at generic level is not possible. The modified (enlarged) first funicular segment is presumably characteristic for males of *Colobopsis*. This characteristic has been described in the type species, *Colobopsis truncata* (Spinola, 1808), by [@B33] and has been equally observed in several species of the *C. cylindrica* group. Males of the COCY group have previously been described for three species (see below). However, these descriptions largely lack the necessary details to meaningfully compare taxa. No previous accounts of genital morphology or illustrations of male specimens have been found in the literature. *Colobopsis badia*: [@B58] gives a brief description of a male from Singapore. Colouration, size, proportions of head and ocelli, as well as the enlarged first funicular segment correspond well to the examined male from Thailand. *Colobopsis severini* (Forel, 1909): The extremely brief description of a male from the island Labuan (near Borneo) is not sufficient to draw any meaningful taxonomic conclusions. *Colobopsis leonardi* (Emery, 1889): [@B30] gives a rather detailed description of males collected within a nest-series on Sumatra. The correct species identification by Karawajew is uncertain; the series may belong to another species of the *C. leonardi* complex as well. The pattern of pilosity on the gaster, with standing setae only present on the posterior half, also corresponds to our observations in males of the *C. saundersi* complex. According to our knowledge, males of the *C. cylindrica* group can be distinguished from other Southeast Asian *Colobopsis* species by the relatively rich subdecumbent pilosity and the dense microreticulation of gastral tergites. Biology {#SECID0EUEDI} ------- The behavioural observations on *C. explodens* sp. n. conducted at KBFSC revealed multiple modes of behaviour that are either poorly studied or new to science. The diurnal activity pattern, as well as the positive influence of high temperatures correspond to the results of previous studies in related taxa (see [@B26]). Similarly large colonies containing several thousand individuals and extending to multiple trees and / or artificial nesting structures have also been described for other members of the genus ([@B18], [@B34]). However, it is still unclear whether individual workers are linked to certain parts of the colony or whether all foragers can move freely through the entire territory of the colony. An interesting and hitherto undescribed phenomenon in this regard is the presence of one or multiple "guards" at the artificial nest's entrance: These minor workers were frequently observed to touch any incoming or leaving workers with their antennae. In some instances, returning foragers were delayed or altogether denied entrance by the guarding ants. One reason for this may be that some workers are linked to different parts of the colony. Alternatively, the observed guarding behaviour may be related to the limited capacity of the artificial nest, which is also suggested by the conspicuously balanced numbers of workers entering and leaving the nest during times of foraging activity. These behavioural patterns are hitherto undescribed and must be investigated in future studies. A further noteworthy activity observed in foraging workers was so-called "grazing" behaviour, in which minor workers were frequently seen using their mandibles to pluck and chew at various mosses, lichens, and other epiphytes on the bark of trees or other surfaces. While this activity can last for up to one hour at a time, its exact purpose remains unclear. It is possible that minor workers cut and consume parts of the plants and microorganisms or merely ingest fluids. As previous analyses of nitrogen isotopes ([@B8]) suggest a largely plant-based diet for COCY ants, it seems likely that "grazing" contributes to their nutrition. However, other previously hypothesized modes of nutrition, such as tending of scale insects ([@B8]) were not observed, and recent investigations on *Colobopsis leonardi* (Emery, 1889) (Zettel et al., ms submitted to Asian Myrmecology) even suggest a higher prevalence of carnivory in COCY ants than previously suspected. Supplementary Material ====================== ###### XML Treatment for Colobopsis explodens ###### XML Treatment for Colobopsis badia We are thankful to Diane W. Davidson for introducing us to the diversity of *Colobopsis cylindrica* ants at KBFSC and for passing down her knowledge on the ecology of these organisms. We thank the administration of KBFSC and Universiti Brunei Darussalam (UBD) for project approval and Brunei's Forestry Department for permission to collect ants and use canopy walkways. The authors appreciate the help of UBD and KBFSC staff, especially Salleh Abdullah Bat, Teddy Chua, Masnah Mirasan, Rafhiah Kahar, Roshanizah Rosli, Rodzay Wahab, Chan Chin Mei, and Kushan Tennakoon for facilitating research and fieldwork in many ways. Sincere thanks are due to the authorities of the Khao Chong Botanical Garden, Trang, Thailand, for support and permission for fieldwork. We acknowledge the efforts of Komal Chenthamara (TU Wien, Austria) for her assistance in phylogenetic analysis and Marco Prusa (TU Wien, Austria) for the molecular DNA analysis of MG products. Thanks are owed to Heinz Wiesbauer (Vienna) for providing photographs of living specimens. We thank Sigfrid Ingrisch (Bad Karlshafen) for identifying the ant crickets. Finally, thanks are owed to James Hogan (Oxford University Museum of Natural History), Giar-Ann Kung (Natural History Museum of Los Angeles County), Phil Ward (University of California, Davis), and Roberto Poggi and Maria Tavano (Museo Civico di Storia Naturale "Giacomo Doria", Genova) for contributing important specimens. The research was supported by the WWTF LS13-048 grant to ISD. Supplementary materials ======================= ###### Figure S1. Living workers of *C. explodens* sp. n. on a detached branch containing a nest fragment Data type: multimedia Explanation note: **a** Minor worker with characteristically raised gaster **b** Major worker with phragmotic head to close nest entrances. Photo: H. Wiesbauer This is an open access article distributed under the terms of the Creative Commons Attribution License (CC BY 4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Laciny A, Zettel H, Kopchinskiy A, Pretzer C, Pal A, Salim KA, Rahimi MJ, Hoenigsberger M, Lim L, Jaitrong W, Druzhinina IS ###### Figure S2. Natural nest of *C. explodens* sp. n. in a dead branch in the high canopy of *S. johorensis* (the main host tree for the model colony) Data type: multimedia Explanation note: **a** Camera setup in the canopy, white arrow marks the nest entrance **b, c** Entrance and minor workers. This is an open access article distributed under the terms of the Creative Commons Attribution License (CC BY 4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Laciny A, Zettel H, Kopchinskiy A, Pretzer C, Pal A, Salim KA, Rahimi MJ, Hoenigsberger M, Lim L, Jaitrong W, Druzhinina IS ###### Figure S3. Interior of the natural nest of *C. explodens* sp. n. found in a dead tree branch of *S. johorensis* on the forest floor Data type: multimedia Explanation note: **a** Longitudinal section of the nest **b** Nest cavity. Sawdust is an artefact due to cutting. **c** Enlarged view of the nest cavity with a chamber made of dark carton **d** Dealate gyne and eggs found inside the chamber shown on **c**. The nest contained at least three such chambers. This is an open access article distributed under the terms of the Creative Commons Attribution License (CC BY 4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Laciny A, Zettel H, Kopchinskiy A, Pretzer C, Pal A, Salim KA, Rahimi MJ, Hoenigsberger M, Lim L, Jaitrong W, Druzhinina IS ###### Figure S4. Autothysis as defensive behaviour in an experimental setting Data type: multimedia Explanation note: A worker of the predatory species *Oecophylla smaragdina* is attacked by three minor workers of *C. explodens* sp. n.; all four animals died after the encounter. Black arrows indicate the yellow secretion expelled by rupture of the mandibular gland reservoirs. This is an open access article distributed under the terms of the Creative Commons Attribution License (CC BY 4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Laciny A, Zettel H, Kopchinskiy A, Pretzer C, Pal A, Salim KA, Rahimi MJ, Hoenigsberger M, Lim L, Jaitrong W, Druzhinina IS ###### Figure S5. Pupa of *C. explodens* sp. n. found inside the opened natural nest Data type: multimedia Explanation note: Note the absence of a pupal cocoon which is diagnostic for *Colobopsis*. This is an open access article distributed under the terms of the Creative Commons Attribution License (CC BY 4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Laciny A, Zettel H, Kopchinskiy A, Pretzer C, Pal A, Salim KA, Rahimi MJ, Hoenigsberger M, Lim L, Jaitrong W, Druzhinina IS ###### Table S6. Data type: specimens data Explanation note: "measurements": Complete dataset of measurements, indices, locality data and type status of all measured specimens of *C. explodens* sp. n. "activity": Activity of minor workers of *C. explodens* sp. n. at the entrance of the artificial nest \#38. "accession numbers": Complete list of accession numbers at NCBI GenBank for all organisms treated in this study. This is an open access article distributed under the terms of the Creative Commons Attribution License (CC BY 4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Laciny A, Zettel H, Kopchinskiy A, Pretzer C, Pal A, Salim KA, Rahimi MJ, Hoenigsberger M, Lim L, Jaitrong W, Druzhinina IS ###### Video S7. Video depicting habitat, nesting sites and behaviour of *C. explodens* sp. n. Data type: multimedia Explanation note: Video depicting habitat, nesting sites and behaviour of *C. explodens* sp. n. accessible under <http://explodingants.com/index.php/publications/colobopsis-explodens> This is an open access article distributed under the terms of the Creative Commons Attribution License (CC BY 4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Laciny A, Zettel H, Kopchinskiy A, Pretzer C, Pal A, Salim KA, Rahimi MJ, Hoenigsberger M, Lim L, Jaitrong W, Druzhinina IS [^1]: Academic editor: M. Borowiec

Introduction {#sec1-1} ============ Recently, the new trends of life style of the consumer showed an increasing demand of traditional dairy products manufactured at the small dairy level (Gaglio *et al.*, [@ref9]). Furthermore, the consumers require foods processed without chemical preservatives and characterized by low-fat and highnutritional values. Ricotta could respond to the demands of consumers because it is characterized by high quantity of proteins, low fat and salt content and other important components, which give this product a high nutritional value (Guatemim *et al.*, [@ref14]). Ricotta is a fresh dairy product that has soft, grainy, thick, lightly sour taste and presents a white colour. Usually, it can be eaten as a soft cheese but in Sicily is more frequently used as an ingredient in dishes and desserts. The name derived from the Latin word *re-cocta*, literally recooked or cooked twice. In Sicily it is generally manufactured from the ewes' whey remaining after the production of hard semi-cooked cheese (Pecorino Siciliano PDO) and fresh cheese (Vastedda della valle del Belìce PDO) but can be produced also with goat, cow and buffalo whey milk. Ricotta production technology uses the principle of coagulation and precipitation of the whey proteins such as globulin and albumin favoured by whey acidification at pH\<5.6 and by heating at 80-90°C. During the production of Sicilian ewes' ricotta cheese, the whey is enriched with whole raw ewes' milk (5-15%) and salt (0.5-1.5%) to increase the yield and improve the organoleptic characteristics. Fresh ricotta is characterized by a short *shelflife* because of its high moisture level, high concentration of residual sugars and an initial pH above 6.0 which make this product an excellent growth medium for a wide range of microorganisms (Mancuso *et al.*, [@ref18]). The attributed *shelflife* of ricotta cheese is generally comprised between 7-11 days under refrigerated storage (4°C). Some of the microorganisms that grow in dairy products are able to produce undesirable reactions that deteriorate the organoleptic characteristics of cheese, while other can potentially cause food-borne diseases (Lu *et al.*, [@ref16]). The presence of these microorganisms in dairy productions depends in some factors such as the quality of raw material, the respect of good practices of production and storage conditions (Ledenbach and Marshall, [@ref15]). In this context the present work was aimed to evaluate the microbiological quality of several samples of Sicilian ewes' ricotta cheese collected from dairy factories and small and large distribution. Materials and Methods {#sec1-2} ===================== Sample collections {#sec2-1} ------------------ A total of 1295 samples of ewes' ricotta cheeses were collected between those conferred to the Istituto Zooprofilattico Sperimentale della Sicilia (IZSSi) *Adelmo Mirri*, Palermo (Italy), by the competent authority during official control, by food business operator in HACCP systems and in research projects. The samples were collected from 36 bakeries and pastry shops; from 233 markets and large-scale retail trade and from 365 dairy factories. The samples were collected during fifteen years of investigation (2002-2016). All samples to be microbiologically investigated were transported under refrigeration at the laboratories of the above-mentioned institute. Microbiological analysis and pH Measurement {#sec2-2} ------------------------------------------- All samples were analyzed in accordance with international standard or validated method: Total mesophilic count (ISO 4833), sulphite reducing anaerobes (ISO 15213), *Enterobacteriaceace* (ISO 4832), *Escherichia coli* (ISO 16649-2); coagulasepositive staphylococci (CPS, ISO 6888-2); *Bacillus cereus* (ISO 7932); yeasts and moulds (ISO 21527-1), *Pseudomonas* (ISO 11059), enterococci on Kanamycin Aesculin Azide aerobically incubated at 37°C for 24-48 h, rod-shaped lactic acid bacteria (LAB) on de Man-Rogosa-Sharpe (MRS) agar, acidified to pH 5.4 with lactic acid (5 mol.L^−1^) and incubated anaerobically for 48 h at 37°C; and mesophilic and thermophilic coccus-shaped LAB on M17 agar and incubated aerobically for 48 h at 30 and 44°C, respectively. *Salmonella* spp. was detected using ISO 6579 or AFNOR BIO 12/23-05/07 and *Listeria monocytogenes* using ISO 11290-1 or AFNOR BIO 12/11-03/04. *Brucella* spp. using Manual of Diagnostic Tests and Vaccines for Terrestrial Animals, cap. 2.4.3.:2009. The pH value of ricotta samples was monitored electrometrically using a portable food and dairy pH meter HI3220-02 (Hanna Instruments, Woonsocket, Rhode Island, USA). Isolation, phenotypic characterization and identification of LAB {#sec2-3} ---------------------------------------------------------------- Some colonies of various shapes were picked up from count plates used for LAB (MRS and M17) enumeration. The isolates from MRS and M17 were transferred to the corresponding broth media. The isolates were purified by successive sub-culturing and the purity of the isolates, as well as the cell morphology were checked microscopically. Gram-positive and catalase-negative were stored in glycerol at --80°C until further experimentations. Lactic acid bacteria isolates were genetically identified by 16S rRNA gene sequencing. The DNA from ricotta LAB isolates was extracted using the InstaGene Matrix kit (Bio-Rad laboratories, Hercules, CA) according to the manufacturer's instructions. Crude cell extracts were used as templates for the polymerase chain reactions (PCRs). PCRs were performed as described by Weisburg *et al.* ([@ref26]) using the primers rD1 (5′-AAGGAGGTGATCCAGCC- 3′) and fD1 (5′- AGAGTTTGATCCTGGCTCAG-3′). DNA sequences were determined by using an ABI PRISM 3130 Genetic Analyzer (Applied Biosystems, Carlsbad, CA, USA) and compared using a BLAST search in the GenBank/ EMBL/DDBJ database (<http://www.ncbi.nlm.> nih.gov). Statistical analysis {#sec2-4} -------------------- Microbiological data of ricotta cheeses divided in three 5-year periods of investigation (2002-2006; 2007-2011; 2012-2016) were subjected to statistical analysis. The logarithm values of microbiological parameters were performed using 1-factor ANOVA model with a 5-year periods as fixed factor (SAS, 2010). Least square means were compared using the *t* Student test. Results and Discussion {#sec1-3} ====================== Microbiological counts and pH values of Ricotta samples {#sec2-5} ------------------------------------------------------- The microbial loads of the ricotta samples collected during fifteen years of investigation are reported in [Table 1](#table001){ref-type="table"}. The results reported in this table are referred only to the microbial group that showed a microbiological grow (positive samples). The total microbial count (TMC) and LAB concentrations detected in this study were, on average, 1 Log cycle higher than those reported in our previous work (Mancuso *et al.*, [@ref18]), who studied the sensory and microbiological evaluation of traditional ovine ricotta cheese in modified atmosphere packaging (MAP). Furthermore, Spanu and coworkers ([@ref23]) always in a study conducted in MAP packed ricotta cheese previously inoculated with commercial biopreservatives and analyzed at different time of refrigerated storage found at t~0~ (after 5 h from the production) a lower concentration of TMC and LAB. This may be due to different conditions of packaging and refrigerated storage used in the studies. *Enterobacteriaceae* and *E. coli* were found in c.a. 21% and 13% of samples, respectively as reported by Fadda *et al.* ([@ref6]) with an average values of the *Enterobacteriaceae* between 3 and 4 log CFU/g (Pala *et al.*, [@ref20]). Thirty four samples (43.6% on the total positive samples) showed a concentration of *E. coli* higher of 1000 UFC/g that represents the maximum enumeration limit for cheeses made from milk or whey that has undergone heat treatment according to Reg. (EC) 2073/2005. The highest level of *E. coli* detected in ricotta cheese samples was equal to 6.89 log CFU/g. *Enterobacteriaceae* and *E. coli* are not able to survive at the thermal treatment of the whey, to this purpose their presence in ricotta cheese is usually attributed to secondary contamination that depend exclusively to the low hygienic conditions during production. The level of enterococci ranged between 1.78 and 6.20 Log CFU/g with an overage value of 3.67 Log CFU/g as reported by Pala *et al.* ([@ref20]). The presence of these bacteria in dairy products is usually associated with inadequate hygiene practices as a consequence of fecal contamination. However, several authors (Foulquié Moreno *et al.*, [@ref7]; Gaglio *et al.*, [@ref11]) suggest that *Enterococcus* strains are involved in the development of the organoleptic characteristics especially in long ripened traditional cheeses and contribute to extend their *shelf life*. On the other hand, enterococci have assumed a major importance in clinical microbiology and their presence needs to be validated by the absence of risks for consumer in term of antibiotic resistance and virulence as well as cellular toxicity. Yeasts and moulds between 1 and 4 Log CFU/g as reported by Pintado and Malcata ([@ref21]). *Bacillus cereus* was found in ca. 16% of the samples analysed with a viable count of 3.79±0.95 Log CFU/g. This results was similar to previously reported by Cosentino *et al.* ([@ref2]) who studied the incidence and biochemical characteristics of *Bacillus* flora in Sardinian dairy products, but lower than those reported by Spanu *et al.* ([@ref22]) in ricotta salata cheese analyzed after 24 h from production (t~0~). However, the level of contaminations did not reach the threshold that leads to significant toxin production (EFSA [@ref4]). Sulphite reducing anaerobe was found just in a one sample of 194 analyzed. The presence of *B. cereus* and SRA in ricotta samples depend from contamination of milk during transformation and to the ability of the spores to survive at the thermal treatment of the whey. CPS were detected in a very low percentage of samples (2.19%) and their level ranged between 1.56 and 5.38 Log CFU/g. Only two samples showed a final concentration above 5.00 Log CFU/g that represents the threshold that leads to staphylococcal enterotoxins production (European Commission, [@ref5]). Furthermore, the production of enterotoxins were never detected in both samples. Spoilage psychotrophic bacteria such as *Pseudomonas* spp. were detected just in two samples with showed an mean values of 1.98±0.03 Log CFU/g lower than those reported by Spanu *et al.* ([@ref23]). This may be due to that the 91 samples tested for the presence of this microbial group were conferred to the IZSSi in a higher percentage from large-scale retail trade and in some cases from dairy factories characterized by high hygienic standards. Indeed, usually the presence of this microbial group is a result of a secondary contaminations and their level increase during refrigeration storage. Their presence in ricotta cheeses can overgrow the other microflora and cause an spoilage of the final quality and the sensory profile of the products (Carrascosa *et al.*, [@ref1]; Pala *et al.*, [@ref20]). *Listeria monocytogenes*, *Salmonella* spp. and *Brucella* spp. were never detected; this may be due to heat treatment of the whey used for the ricotta production that represent a killing step of these microorganisms and to the absence of secondary contamination from the processing environment. These results confirmed previous investigations performed in other Sicilian dairy environment and products where this pathogenic microorganisms were never detected (Cruciata *et al*., [@ref3]; Gaglio *et al.*, [@ref10]; Scatassa *et al.*, [@ref24]). Regarding the pH, the ricotta samples showed a mean values of 6.37±0.33 as reported by (Mucchetti and Neviani, [@ref19]) but during the study were registered values ranged between 6.04 and 6.74. Isolation, phenotypic characterization and identification of LAB {#sec2-6} ---------------------------------------------------------------- A total of 98 cultures were collected from 30 ricotta cheese representative of the three level of sampling. All of the cultures were inspected microscopically and classified as cocci (69) or rods (29). Gram determination and the catalase test indicated that 64 cocci and 27 rods could be considered putative LAB cultures. The 91 strains were identified by sequencing of the 16S rRNA gene. The strains were allotted into 10 species within the genera *Enterococcus*, *Lactobacillus*, *Lactococcus* and *Leuconostoc*. The species with the highest number of strains were *Lactococcus lactis* (n=23) e *Lactobacillus casei* (n=22). Other strains were identified as *Enterococcus faecium* (n=19), *Leuconostoc lactis* (n=7), *Lactococcus lactis* sub*. lactis* (n=6), *Leuconostoc mesenteroides* (n=6), *Lactobacillus sakei* (n=5)*. Enterococcus durans* (n=1), *Enterococcus faecalis* (n=1) and *Enterococcus gallinarum* (n=1). Except *L. lactis*, all species identified are commonly reported to be part of the non starter LAB (NSLAB) population in several cheeses (Gatti *et al.*, [@ref12])*.* Furthermore, all these species were found in other traditional ewes' Sicilian cheeses (Gaglio *et al.*, [@ref8]; Guarcello *et al.*, [@ref13]) and wooden equipment employed for cheese manufacture (Scatassa *et al.*, [@ref24]). Some works showed the ability of these LAB species to produce bacteriocin-like inhibitory substances (BLIS) active against *Listeria* and other pathogenic bacteria and this allows to improve the safety, control the fermentation microbiota, speed maturation and increase the *shelf life* of the final cheeses (Macaluso *et al.*, [@ref17]; Scatassa *et al.*, [@ref25]). Statistical analysis {#sec2-7} -------------------- The statistical analysis of the microbiological counts conducted on the ricotta samples during three 5-year periods of investigation (2002-2006; 2007-2011 and 2012-2016) are reported in [Table 2](#table002){ref-type="table"}. According to *t* Student test, significant statistical differences were found for the levels of Enterococci, *Enterobacteriaceae*, yeasts and moulds, *E. coli* and CPS between the ricotta samples collected during the first 5-years periods of investigation in comparison with the last two periods. A possible explanation to these observations is that in the last decade food business operators adopted the good hygiene and manufacturing practices, in compliance with the Commission Regulation (EC) No 2073/2005 on microbiological criteria for foodstuffs (European Commission, [@ref5]). Conclusions {#sec1-4} =========== The results reported in this study showed that from a hygienic point of view, the Sicilian ewes' ricotta represents a safe cheese production. In particular, during the entire periods of investigation the pathogenic bacteria (*L. monocytogenes*, *Salmonella* spp. and *Brucella* spp.) were never detected while the levels of undesired microorganisms such as *Enterobacteriaceae*, *E.coli* and CPS, during the last two 5-year periods of investigation were reduced in number in comparison with the first 5-year period of investigation. The last findings are not satisfactory since in that period the structural and hygienic conditions of dairies improved consistently. However, in order to improve the final quality of this product are needed further strategy such as the dairy makers training, with the aim to apply a good hygienic practices during the production. The Authors thanks V. Alio, A. Carrozzo, B. Ducato and G. Oliveri for their precious collaboration. ###### Microbial loads of ricotta samples. Microorganisms^[\*](#tfn1){ref-type="table-fn"}^ Number of samples Positive samples \% positive samples Mean Min. Max. SD -------------------------------------------------- ------------------- ------------------ --------------------- ------ ------ ------ ------ TMC 371 350 94.34 5.11 2 8.26 1.63 Rod LAB (37°C) 98 74 75.51 3.6 1.85 5.58 1.16 Coccus LAB (30°C) 98 82 83.67 3.73 2.00 6.83 1.38 Coccus LAB (44°C) 98 69 70.41 3.09 1.78 4.85 0.92 Enterococci 106 40 37.74 3.67 1.77 6.20 1.18 *Enterobacteriaceae* 371 78 21.02 3.68 1.43 7.00 1.26 *E. coli* 598 78 13.04 3.26 1.3 6.89 1.29 CPS 639 14 2.19 3.11 1.56 5.38 1.27 Yeasts and moulds 88 12 13.64 2.68 1.00 4.00 0.97 *B. cereus* 157 25 15.92 3.79 1.70 6.00 0.95 Pseudomonas 91 2 2.20 1.98 1.96 2 0.03 SRA 194 1 0.52 *L. monocytogenes* 1156 n.d. n.d. *Salmonella spp.* 998 n.d. n.d. *Brucella spp.* 721 n.d. n.d. ^\*^Units are log CFU/g. TMC, total mesophilic microorganisms; CPS, coagulase-positive staphylococci, SRA, Sulphite reducing anaerobe; n.d. not detected; SD, standard deviation. ###### Statistical analysis of microbial loads. Microorganisms[\*](#tfn2){ref-type="table-fn"} 2002-2006 2007-2011 2012-2016 ------------------------------------------------ -------------- -------------- -------------- TMC 5.33±1.13 6.45±1.12 4.78±0.75 Enterococci 2.60±0.28^a^ 0.01±0.70^b^ 0.81±0.22^b^ *Enterobacteriaceae* 1.79±0.24^a^ 0.50±0.17^b^ 0.68±0.10^b^ *E. coli* 1.33±0.09^a^ 0.24±0.08^b^ 0.19±0.07^b^ CPS 0.35±0.05^a^ 0.01±0.05^b^ 0.01±0.06^b^ Yeasts and moulds 1.94±0.40^a^ 0.15±0.12^b^ 0.49±0.17^b^ *B. cereus* \- 0.67±0.11 0.97±0.17 ^\*^Units are Log CFU/g. TMC, total mesophilic microorganisms; CPS, coagulase-positive staphylococci. On the row different letters are significant at P≤0.01 [^1]: Contributions: the authors contributed equally. [^2]: Conflict of interests: the authors declare no potential conflict of interests

I[NTRODUCTION]{.smallcaps} {#sec1-1} ========================== The incidence of primary hyperparathyroidism (PHPT) is increasing due to the widespread use of multichannel analyzers. The parathyroid lesions are mostly early adenomas, and fewer parathyroid hyperplasias. These two entities have overlapping histological features, \[[@ref1]\] which sometimes preclude a definitive frozen section diagnosis. The parathyroid hormone (PTH) is produced exclusively by the parathyroid glands. Hyperplasias, adenomas, and carcinomas, cause hypersecretion of this hormone and elevated serum PTH. The half-life of the PTH is short, three-to-five minutes, and excision of the parathyroid tissue is followed by a rapid fall in serum levels. PTH assays, with a turnaround time of 15-20 minutes, have made intraoperative PTH analysis a reality in various centers. \[[@ref2]\] Endocrine surgeon, George Irwin, is credited with first utilizing this phenomenon in parathyroid surgery. Successful surgery is determined by a drop of over 50% in the intraoperative plasma parathyroid hormone level when compared with the preoperative value. A short study of seven cases of parathyroidectomy with the use of intraoperative PTH is hereby presented. M[ATERIALS AND]{.smallcaps} M[ETHODS]{.smallcaps} {#sec1-2} ================================================= The intraoperative parathyroid hormone (IOPTH) assay was performed in seven patients undergoing parathyroidectomies for primary hyperparathyroidism. The blood sample was drawn from a peripheral vein 10 minutes after surgical excision of the abnormal gland and sent to the laboratory. Care was taken to prevent dilution with intravenous fluids. The PTH levels were estimated by the use of a rapid Electrochemiluminescence immunoassay (ECLIA) on the Cobas 6000 combi analyzer. The excised gland was also sent for frozen section examination. The Elecsys assay for determining intact PTH employs a Sandwich Test Principle, in which, the biotinylated monoclonal antibody reacts with the N-terminal fragment (1-37) of the parathyroid hormone and a monoclonal antibody labeled with a ruthenium complex reacts with the C-terminal fragment (38-84). The antibodies used in this assay are reactive with epitopes in the amino acid regions 26-32 and 37-42. \[[@ref3]\] R[ESULTS]{.smallcaps} {#sec1-3} ===================== The results are tabulated in [Table 1](#T1){ref-type="table"}. In case No. 1, multiple frozen section examinations were done for 'suspected' nodules excised from different areas, including the mediastinum. The nodules excised were lymph nodes. A hemithyroidectomy for goiter, done in the same sitting, revealed that the parathyroid adenoma was ectopically located within the left lobe of the thyroid. The intraoperative PTH showed a fall from 128.8 to 5.8 mgm%, indicating removal of all hyperfunctioning parathyroid tissue. The postoperative period was uneventful and both the serum PTH and calcium levels normalized. ###### Comparison of IOPTH assays, frozen and paraffin section results  In the next two and the last three patients, the frozen section revealed an enlarged hypercellular parathyroid gland consistent with either adenoma or hyperplasia. The IOPTH values showed a fall of more than 50%, confirming removal of all hyperfunctioning tissue, and therefore, confirming an adenoma. No further exploration was done, and no second gland was excised. In the fourth patient, only IOPTH was done and frozen section examination was not performed. The IO-PTH dropped to \>50% in all the seven cases, with an average intraoperative fall of 73%. The operative time was increased by 15 minutes on an average. In all seven patients, the final histopathology on the paraffinized tissue was consistent with a parathyroid adenoma and correlated with the postoperative normalization of both PTH and calcium values. D[ISCUSSION]{.smallcaps} {#sec1-4} ======================== Primary hyperparathyroidism is most often caused by a single adenoma, often small in size and difficult to localize with imaging techniques. Diffuse hyperplasias account for only 9% of all patients with PHPT. In secondary and tertiary hyperparathyroidism, the diagnosis is self evident, \[[@ref4]\] due to the underlying disease. Hypercalcemias seen in malignancy can be due to the PTH-related plasma protein (PTHrP). This protein is not detected by the technique used for the PTH assay. Serum PTH levels are, conversely, low in paraneoplastic hypercalcemia related to PTHrP. \[[@ref5]\] The standard surgical technique for hyperparathyroidism is a bilateral neck exploration to identify the parathyroid adenoma and the other three normal glands. The suspected adenoma and one more 'normal' gland are excised and sent for frozen section examination. Cellularity and size are compared. If one is hypercellular and the other normocellular, the hypercellular gland is an adenoma and surgical treatment is complete. If both are hypercellular the diagnosis is hyperplasia, and subtotal parathyroidectomy is performed, that is, three-and-a-half glands are removed. Histologically, hypercellularity is defined as replacement of the normal glandular fat by parathyroid cells followed by gradual gland enlargement. When it is understood that normal aglandular fat varies from 17 to 50% and normal glands vary in size from 3-6 mm, the problems encountered in frozen section diagnosis become evident. The additional pitfalls are: \[[@ref6]\] (1)Although parathyroid glands usually number from four to five, they can range from two to twelve. Thus, removal of two glands could mean loss of all parathyroid tissue and excision of three-and-a-half glands in hyperplasia could still leave several hyperfunctioning glands(2)Fifteen to twenty percent of the parathyroid glands are often ectopically located, for example, in the mediastinum and intrathyroid, and this causes difficulty in intraoperative identification(3)Thyroid follicular adenomas, lymph nodes, and the like, can show freezing artifacts leading to misidentification as parathyroid adenomas and vice versa. On the other hand, IOPTH is a specific and sensitive guide, as a 50% drop in serum hormone levels signals removal of all hyperfunctioning parathyroid tissue. The chances of operative failure are reduced. \[[@ref7]\] Procedures like minimally invasive parathyroidectomy, with excision of only a single adenoma are slowly replacing the traditional four gland bilateral exploration, leading to a decrease of the surgical field, reduced hospital stay, and improved cosmesis. \[[@ref8]\] Surgical failures \[[@ref9]\] in IOPTH can be due to excessive pre-excision manipulation, leading to a falsely elevated initial reading, false post-excision drop in levels, or rarely, due to a dominant tumor suppressing other tumors in the multiple endocrine neoplasia (MEN) syndrome. \[[@ref10]\] C[ONCLUSION]{.smallcaps} {#sec1-5} ======================== The IOPTH assay is a useful test in parathyroid surgery because of its unique ability to, (1)Identify the complete removal of hyperfunctioning tissue. The intraoperative persistence of elevated PTH would indicate residual hyperfunctioning tissue and signal the need for further exploration and thus operative failure is prevented(2)Facilitate minimally invasive parathyroidectomy for single parathyroid adenomas, which in turn improves the cost-effectiveness, duration of hospital stay, and cosmetic outcome. Following implementation of this test, parathyroid excision in this institute has had a zero surgical failure rate, reduced the number of frozen section requests, and decreased the field of operation, with improved cosmesis to the patient. It is agreed that the successful use of IOPTH in the seven cases presently reported is a small number, but as the validity of this procedure is corroborated in several institutes worldwide, it was thought important to convey to pathologists and surgeons alike, the feasibility of conducting IOPTH in their respective institutes in this country and avail of its multiple advantages. **Source of Support:** Nil **Conflict of Interest:** None declared.

Introduction {#s1} ============ With age, older adults may experience declining memory, reasoning, and speed of processing as well as peripheral vision and dynamic visual acuity (Park et al., [@B49]; Salthouse, [@B53]; Muiños and Ballesteros, [@B41], [@B42]). In addition, normal aging is associated with reduced age-related white matter integrity and gray matter volume in the caudate, cerebellum, hippocampus and the prefrontal cortex (O'Sullivan et al., [@B46]; Hedden and Gabrieli, [@B26]; Raz et al., [@B51]). Increasing age is also associated with a higher incidence of neurodegenerative disorders such as mild cognitive impairment (MCI), Alzheimer's disease (AD), and Parkinson's disease (PD; Blennow et al., [@B7]; Levy, [@B32]; Herrup, [@B27]; Luck et al., [@B35]). These changes affect the quality of daily life of older adults and aggravate the socioeconomic burden (Williams, [@B62]; Mitchell et al., [@B40]). Several beneficial interventions have been considered to delay cognitive decline in older adults, including memory strategy training (Li et al., [@B33]), dancing (Coubard et al., [@B16]), aerobic exercise (Erickson and Kramer, [@B18]), music training (White-Schwoch et al., [@B61]) and video game playing (Toril et al., [@B56]; Wang et al., [@B57]). Of these approaches, video games have received growing attention from researchers because of their popularity and encouraging transfer effects (Zelinski and Reyes, [@B66]; Oei and Patterson, [@B44]). Increasing evidence shows that video games are beneficial to the cognitive-perceptual domain, including reaction time (Qiu et al., [@B50]), visual short-term memory, processing speed and attention tasks (Ballesteros et al., [@B2]; McDermott et al., [@B38]) as well as higher-level cognitive functions such as executive function, reasoning and planning (Basak et al., [@B3]; Oei and Patterson, [@B45]; Buitenweg et al., [@B10]; Wang et al., [@B58]). Recent studies have provided some evidence of the neural basis for superior behavioral performance in video game players (VGPs). Kühn and Gallinat ([@B30]) found that the amount of lifetime video gaming was positively associated with cortical thickness in the left dorsolateral prefrontal cortex and left frontal eye fields. Other studies have shown significantly larger gray matter volume in the right posterior parietal cortex in VGPs compared with non-video game players (NVGPs; Tanaka et al., [@B54]). Another study reported that video game experience altered the visual cortical network (Granek et al., [@B23]). Moreover, Bavelier et al. ([@B4]) found that VGPs showed less recruitment of the frontoparietal attentional network during a visual search task compared to NVGPs, while VGPs showed increased activation in frontoparietal regions in a flanker task (Wang et al., [@B59]). In recent resting-state functional magnetic resonance imaging (fMRI) studies, Gong et al. ([@B21]) found that VGPs showed increased functional connectivity between the attentional and sensorimotor networks, and they further found that VGPs had enhanced functional integration both within and between the salience network and the central executive network (Gong et al., [@B22]). By studying spontaneous fluctuations in blood oxygen level-dependent contrasts, resting-state fMRI studies can provide valuable insights into investigating the differences between VGPs and NVGPs (Biswal et al., [@B6]; Fox and Raichle, [@B20]). To assess spontaneous brain activity in attention deficit hyperactivity disorder, Zang et al. ([@B65]) proposed studying the amplitude low-frequency fluctuations (ALFF), which is defined as the square root calculated at each frequency of the power spectrum within a specific low-frequency band. Mennes et al. ([@B39]) demonstrated that resting-state ALFF had robust predictive value for behavior. Wei et al. ([@B60]) demonstrated that individual differences in ALFF could predict the conceptual processing capacity of subjects. These findings suggested that the assessment of ALFF could be used to investigate the neural basis of video game experience in older adults. To the best of our knowledge, no resting-state fMRI study has examined the ALFF difference between VGPs and NVGPs in older adults. In the current study, we mainly aimed to investigate the ALFF differences between older VGPs and older NVGPs in combination with behavioral performance and sought to determine whether the differences in ALFF were associated with video game experience. We hypothesized that older VGPs would present better behavioral performance and increased ALFF activity in comparison with older NVGPs. Materials and Methods {#s2} ===================== Participants {#s2-1} ------------ Older adult volunteers were recruited by means of posters and online advertisements. All participants were healthy, had normal or corrected-to-normal vision and had more than 6 years of education. In addition, the inclusion criteria for all participants included having no medical disorders, no neurological disease, and no metallic implants in the body that could interfere with or cause injury due to the magnetic field. Participants were divided into two groups: those with video game playing experience, classified as VGPs, had played video games at least 2 h per week (more than 0.5 h per day and at least 4 days) over the previous 6 months; the time that VGPs dedicated to game-playing was approximately 395 h per year on average. Participants with no video game experience in their lifetimes were classified as NVGPs. Thirty-seven participants participated in a resting-state fMRI scan, and four participants (two VGPs and two NVGPs) were excluded due to poor registration between functional and structural images or excessive head motion. Ultimately, 15 VGPs (nine females, median = 65.20 years, range from 55 to 74 years old) and 18 NVGPs (six females, median = 65.06 years, range from 55 to 78 years old) were included for further analysis, and no significant gender difference was found between the two groups (*χ*^2^ = 2.347, *p* = 0.126). The Mini-Mental State Examination (MMSE; Folstein et al., [@B19]) and the activities of daily living scale (ADL; Katz et al., [@B29]) were used to assess all participants. Moreover, we also investigated the hobby hours of every participant, which included calligraphy, chess, playing cards and other hobbies other than video games. The study was approved by the local ethics committee of the Institute of Psychology, Chinese Academy of Sciences (IPCAS). We obtained written informed consent from all participants prior to the study. Data Acquisition and Analysis {#s2-2} ----------------------------- ### Scanning Procedure {#s2-2-1} Magnetic resonance imaging scans were conducted on a General Electric 3T scanner (GE Discovery MR750). Plastic earplugs were used to reduce scanner noise. The participants were instructed to lie quietly with their eyes closed and to keep their head as still as possible. Resting-state functional images were acquired using gradient-echo sequences with the following parameters: repetition time (TR) = 2,000 ms, echo time (TE) = 30 ms, flip angle = 90°, the number of slices = 43, slice thickness = 3.5 mm, slice gap = 0.5 mm, voxel size = 3.75 × 3.75 × 3.5 mm^3^, field of view (FOV) = 240 × 240 mm^2^, matrix size = 64 × 64, axial orientation. A T1 structural image was recorded using spoiled gradient echo sequences with the following settings: TR = 39.9 ms, TE = 24.76 ms, flip angle = 10°, the number of slices = 256, slice thickness = 1 mm, voxel size = 1 × 1 × 1 mm^3^, FOV = 256 × 256 mm^2^, sagittal orientation. Task-based fMRI images were acquired after the resting-state functional images and T1 structural images. All participants performed the flanker task during the scanning, and the task-based fMRI results were reported elsewhere (Wang et al., [@B59]). The flanker paradigm consisted of strings of arrows, and participants were required to focus on the central arrow and to ignore surrounding arrows that could be congruent (e.g., \>\>\>\>\>) or incongruent (e.g., \<\<\>\<\<) to the central arrow. The participants used the number button "1" or "4" on the response pad to indicate the left or right direction, respectively, of a central arrow after a fixation cross was presented, and their reaction time and accuracy were recorded. More details can be found in our previous study (Wang et al., [@B59]). The flanker task performances were used as behavioral performance in the current study. ### Data Preprocessing {#s2-2-2} Resting-state fMRI data were carried out using RESTplus V1.2[^1^](#fn0001){ref-type="fn"} based on MATLAB2014a. Preprocessing included the following steps: (1) Discarding the first 10 time points to allow the tissue to reach steady state magnetization and to allow participants to adapt to the scanning noise; (2) Slice timing to temporally correct the interleaved slice acquisition; (3) Aligning each volume to the mean EPI image to correction between head movements; (4) Normalizing using T1 image unified segmentation; (5) Spatial smoothing with a 6 mm full width at half maximum Gaussian kernel; (6) Removing the linear trend within the time series; (7) Regressing out the nuisance covariates, which included 6 head motion parameters, a global mean signal, a white matter signal and a cerebrospinal fluid signal; and (8) Temporal filtering (0.01--0.1 Hz) was performed on the time series of each voxel to reduce the effect of low-frequency drifts and high-frequency noise. ### ALFF Calculation {#s2-2-3} The filtered time series was converted to the frequency domain using the Fast Fourier Transform. Then, the square root of the power spectrum at each frequency was averaged across a frequency band of 0.01--0.1 Hz. This averaged square root was taken as the ALFF, and the result of normalized ALFF values of each participant was adopted for statistical analysis. The procedure for calculating ALFF has been described in previous studies (Yang et al., [@B64]; Zang et al., [@B65]). ### Statistical Analysis {#s2-2-4} A two-sample *t*-test was conducted on the ALFF between VGPs and NVGPs. Multiple comparisons were corrected using familywise error correction (FWE) based on Gaussian random-field theory, which yielded a voxel level of *p* \< 0.01 and a cluster level of *p* \< 0.05. For the flanker task, reaction time from incorrect trials was excluded, and the difference in the average reaction time in the congruent condition subtracted from that in the incongruent condition was used as behavioral performance. Considering the relatively small sample size in each group, we used the Mann-Whitney *U* test to compare the demographic variables and flanker task performances between older VGPs and NVGPs with SPSS 19.0. ### Correlation Analyses {#s2-2-5} To investigate the relationship between the ALFF values and the behavioral measures, including hobby hours, MMSE scores, and flanker task performance, we used the brain regions with significant differences in the two-sample*t*-test as regions of interest (ROI), then extracted the signals from the ROI of each participant and computed the Pearson's correlation coefficients between ROI signals and behavioral performance for all participants. Results {#s3} ======= The results showed that the two groups did not differ in age, education years and ADL. However, Older VGPs showed significant higher score in MMSE and more hobby hours than older NVGPs ([Table 1](#T1){ref-type="table"}). For flanker task performance, the mean reaction time of the VGPs was significantly shorter than that of the NVGPs (*U* = 192, *Z* = 2.06, *p* = 0.04). In addition, the effect size reached 0.816, which was considered large (Cohen, [@B15]). ###### Demographic information on video game player (VGP) and non-video game player (NVGP) participants. VGP (*N* = 15) NVGP (*N* = 18) *Z* *p* ------------------- ----------------- ----------------- ----------------- -------- ------- ------ Age (years) 65.20 (5.92) 55--74 65.06 (7.17) 55--78 −0.16 0.87 Education (years) 13.13 (2.92) 9--18 11.78 (2.82) 9--17 −1.38 0.19 MMSE 28.87 (1.64) 25--30 27.72 (1.60) 24--30 −2.28 0.02 ADL 8.13 (0.35) 8--9 8.00 (0.00) 8--8 −1.57 0.53 Hobby hours 225.80 (163.99) 7--560 125.11 (211.56) 0--843 −2.54 0.01 *Note: N, sample number; MMSE, Mini-Mental State Examination; ADL, Activities of Daily Living Scale; Hobby hours, total hours spent on hobbies other than video games*. Group Differences in ALFF {#s3-1} ------------------------- The results showed that VGPs showed increased ALFF in the left inferior occipital gyrus, left cerebellum, and left lingual gyrus compared with NVGPs ([Figure 1](#F1){ref-type="fig"}, [Table 2](#T2){ref-type="table"}). No decreased ALFF regions were found in VGPs compared with NVGPs. However, when MMSE, hobby hours, gender, and years of education were controlled as covariates, the significant results disappeared. {#F1} ###### Regions showing ALFF differences between VGPs and NVGPs. Brain regions Peak MNI coordinates Peak *T*-value ----------------- ---------------------- ---------------- ----- ----- ------ Left IOG 38 −27 −87 −9 3.79 Left cerebellum 30 −36 −78 −21 4.06 Left LG 29 −27 −90 −12 3.48 *Note: x, y, and z are coordinates of primary peak locations in the MNI space. T is the statistical peak value of the voxel showing ALFF differences between VGPs and NVGPs (positive values: VGPs \> NVGPs), IOG, inferior occipital gyrus, LG, lingual gyrus. Voxel level: *p* \< 0.01, cluster level: *p* \< 0.05, corrected for FWE*. Correlation Analysis {#s3-2} -------------------- We found positive correlations between MMSE scores and the ALFF in the left inferior occipital gyrus (*r*~(33)~ = 0.356, *p* \< 0.05) and left lingual gyrus (*r*~(33)~ = 0.36, *p* \< 0.05; [Figure 2](#F2){ref-type="fig"}) across all participants. No significant correlations were found between ALFF and hobby hours or flanker task performance. We also did not find significant correlations between ALFF and the behavioral measures in the VGP and NVGP groups separately. {#F2} Discussion {#s4} ========== In this study, we investigated behavioral performance and differences in spontaneous brain activity by measuring the ALFF of resting-state fMRI signals between older VGPs and older NVGPs. We found that older VGPs outperformed older NVGPs in the behavioral task and presented increased ALFF in the left inferior occipital gyrus, left cerebellum and left lingual gyrus compared with NVGPs. Older VGPs showed significantly better performance in the flanker task than older NVGPs. During the flanker task, participants needed to ignore the peripheral arrows while concentrating on the center arrows, and executive control was required to identify whether the targets and distractors were in conflict. This result was consistent with previous findings that VGPs have better performance than NVGPs in the flanker task (Green and Bavelier, [@B24]). VGPs have also responded more quickly than NVGPs in an attention network test (Dye et al., [@B17]). Other studies have also shown that video game training enhances cognitive control and selective visual attention in older adults (Anguera et al., [@B1]; Belchior et al., [@B37]). In addition, a recent study found that higher-order executive function skills were improved after training on a physics-based puzzle game (Oei and Patterson, [@B45]). VGPs have an advantage in top-down distractor suppression (Chisholm and Kingstone, [@B14]), which may be the underlying mechanism of superior performance in executive control. VGPs showed increased ALFF in the left inferior occipital gyrus and left lingual gyrus. In our recent study, we found that VGPs compared with NVGPs showed increased activation in the left lingual gyrus in the flanker task (Wang et al., [@B59]). Functional imaging studies have indicated that the amount of lifetime video game playing is positively associated with occipital gray matter volume (Kühn and Gallinat, [@B30]). The occipital lobe and lingual gyrus have been reported to be involved in visual processing (Cai et al., [@B12]). Several studies have found that the enhancement of visual function, including visuospatial attention (Green and Bavelier, [@B25]), contrast sensitivity (Li et al., [@B34]), and visual search (Castel et al., [@B13]), is associated with video game experience. Jung et al. ([@B28]) revealed that a larger volume of lingual gyrus predicted better performance in neuropsychological tests. An ERP study showed increased amplitudes in the occipital cortex after playing video games for 10 h (Wu et al., [@B63]). We also found that VGPs showed increased ALFF in the left cerebellum in comparison with NVGPs. Recent studies have substantiated that gray matter volume in the cerebellum increases significantly in the experimental group after playing a video game for 2 months (Kühn et al., [@B31]). The cerebellum has been considered to contribute to motor control (Manto et al., [@B36]). In a clinical study, it has been reported that video game exercises are beneficial for subjects' balance control (Betker et al., [@B5]). In the current study, we found positive correlations between ALFF values in the left inferior occipital gyrus and left lingual gyrus with MMSE scores. The results suggest that increased ALFF activity is associated with better cognitive function. Previous studies have shown that older adults have cognitive and brain plasticity (Park and Bischof, [@B47]; Brehmer et al., [@B9]). The scaffolding theory of aging and cognition suggests that the adaptive aging brain engages in compensatory scaffolding in response to the challenges posed by declining neuronal structure and function (Park and Reuter-Lorenz, [@B48]; Reuter-Lorenz and Park, [@B52]). Video game playing is considered an effective way to bring about broad cognitive transfer (Zelinski and Reyes, [@B66]; Boot et al., [@B8]; Oei and Patterson, [@B44]; Toril et al., [@B56]). When immersed in video games, older VGPs need to engage more cognitive resources, and this may help them construct compensatory scaffolding due to changing visual stimulation and repetition experiences, enhancing their brain plasticity and protecting them against the cognitive decline caused by aging. Although we speculate that older VGPs construct compensatory scaffolding to underlie cognitive and neural plasticity, the underlying biological mechanisms of ALFF activity are not clear. A recent study revealed significant positive correlations between ALFF and gamma-aminobutyric acid (GABA; Nugent et al., [@B43]). GABA is a main inhibitory neurotransmitter in the central nervous system (Paulsen and Moser, [@B100]). GABA modification plays an important role in cortical plasticity (Bütefisch et al., [@B11]; Ziemann et al., [@B67]). In the current study, we observed increased ALFF activity in older VGPs, which might be the consequence of GABA modification and increased GABA in the brain. The increased GABA in older VGPs may lead to inhibitory signaling and therefore improve inhibition ability; consequently, it may help older VGPs achieve better performance in the flanker task. There are some limitations in the current study. The genre of the video games may be a factor that impacts diversified cognitive abilities. We could not analyze this factor due to the relatively small sample size. The current study was a cross-sectional study, and we could not exclude the contribution of other variables to the increased behavioral performance and brain activity. Future video game training studies, especially studies focusing on certain game genres, are warranted to further reveal the video game training-related effects on behavioral performance and brain activity. Moreover, demographic variables might influence the ALFF results. When we controlled for the MMSE, hobby hours, gender, and years of education as covariates, we could not observe significant ALFF results between VGPs and NVGPs; therefore, this finding has to be treated cautiously. In the future, when increasingly older adults fulfill the inclusion criteria of VGPs and meet the magnetic resonance imaging scanning criteria, researchers could recruit large numbers of participants and further reveal the video game experience-related effects on behavioral and neural changes. Conclusion {#s5} ========== The present study revealed that VGPs outperformed NVGPs in the flanker task and that they present increased ALFF in the left inferior occipital gyrus, left cerebellum and left lingual gyrus compared with NVGPs. These results suggest the positive effects of video game playing on brain plasticity in older adults, suggesting that video games may be a promising tool to delay cognitive decline in older adults. Future studies are recommended to train video game naive participants and investigate ALFF effects over time. Author Contributions {#s6} ==================== H-JL and H-YH conceived the idea and wrote the manuscript. X-ZJ guided the data analysis. PW, SH and J-XZ collected the data and contributed towards writing the article. Conflict of Interest Statement {#s7} ============================== The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. **Funding.** This work was supported by the National Basic Research Program (973 Program; 2015CB351702), the Youth Innovation Promotion Association CAS (2016084) and National Natural Science Foundation of China (31871143). ^1^<http://www.restfmri.net> [^1]: Edited by: Panteleimon Giannakopoulos, Université de Genève, Switzerland [^2]: Reviewed by: Tomoki Arichi, King's College London, United Kingdom; Xintao Hu, Northwestern Polytechnical University, China

Introduction {#sec1} ============ Biomaterials for application in medical implants and devices often undergo treatments requiring organic solvents, a high temperature, pressure, or voltage (e.g., solution or melt electrospinning, 3D printing, injection molding). The decoration of biomaterials with functional proteins such as growth factors, cytokines, and signaling molecules can be poorly compatible with such processing techniques. Therefore, bioorthogonal conjugation chemistries have been implemented for the postprocessing modification of biomaterial surfaces. A family of reactions referred to as "click-chemistry" has gained popularity among material scientists because of its hallmark features of fast kinetics, high yields, easily removable harmless byproducts, high selectivity without a catalyst, and insensitivity to water or oxygen.^[@ref1],[@ref2]^ Examples of applications of click-chemistry in biomedical sciences are reported using photoactivated thiol--ene/yne chemistry,^[@ref3]−[@ref5]^ Michael additions,^[@ref6],[@ref7]^ oxime and hydrazones formation via reaction of aldehydes/ketones with respectively alkoxyamines or hydrazides,^[@ref8],[@ref9]^ strain promoted azide--alkyne cycloadditions (SPAAC),^[@ref10]−[@ref12]^ and inverse electron demand Diels--Alder cycloadditions (iEDDAC).^[@ref13]−[@ref17]^ The latter class comprises the reaction occurring between a tetrazine (Tz) and a strained *trans*-cyclooctene (TCO). Reported rate constants for this reaction are as high as 3.3 × 10^6^ M^--1^ s^--1^ (in H~2~O at 25 °C), thereby making the Tz/TCO conjugation the fastest bioorthogonal reaction known so far.^[@ref18],[@ref19]^ This class of reactions has been found to be very suitable for targeting biomolecules such as proteins and antibodies in complex environments. iEDDAC chemistry has been used for the labeling of antibodies with nanoparticles and radioactive probes for cell targeting,^[@ref20],[@ref21]^ rapid *in vivo* protein conjugation,^[@ref22]^ or *in vivo* fluorescent labeling of antibodies.^[@ref23]^ Recombinant protein G presents two immunoglobulin (IgG)-binding domains. The major binding site in IgG is located in the Fc domain of the antibody.^[@ref24]^ Protein G binds strongly all human IgG subclasses with a *K*~d~ ∼ 2 × 10^--10^ M,^[@ref25]^ thereby finding several applications as an affinity matrix for purification and detection of antibodies. The Fc-binding properties of protein G have been exploited for the immobilization of Fc-fusion proteins,^[@ref26],[@ref27]^ including Notch ligands Dll1, Dll4, and Jagged1.^[@ref28]−[@ref30]^ As a model Fc-fusion bioactive protein, we chose the Notch ligand Jagged1. Jagged1-mediated Notch signaling has been recently proposed as a target for cardiovascular regenerative medicine as the Notch signaling pathway is a powerful orchestrator of development, homeostasis, and wound healing of such tissues.^[@ref31],[@ref32]^ Jagged1-functionalized materials based on poly(ethylene glycol) (PEG) have been proposed as artificial niches for stem cells, which find applications in myocardial regeneration^[@ref33],[@ref34]^ and pancreatic islet cell immunoprotection^[@ref35]^ and for elucidating cell-matrix interactions.^[@ref36],[@ref37]^ Furthermore, Notch signaling has been found to have varied roles in cancer,^[@ref38],[@ref39]^ with the Notch-Jagged signaling representing a target for studying the origin of metastasis.^[@ref40]^ Recently, we combined the iEDDAC click reaction with the dynamic traits of a supramolecular material system based on ureido-pyrimidinone (UPy).^[@ref41]^ Materials based on the strong, directional interactions between hydrogen-bonding UPy groups represent an excellent framework for ECM-mimicking biomaterials.^[@ref42]−[@ref44]^ These supramolecular systems exhibit both macromolecular properties of thermoplastic elastomers and the dynamic character that can be found in highly complex biological systems. In solid UPy-polymers, lateral stacking of UPy-dimers and flanking urea units leads to the formation of a nanoscale fibrous assembly.^[@ref45]^ Cell-instructive additives (e.g., peptides or antifouling molecules) can be robustly incorporated into the pristine material by means of matching supramolecular moieties to generate modular and easily processable materials for regenerative medicine applications.^[@ref46]−[@ref49]^ In a recent study, supramolecular UPy-poly(ε-caprolactone) (UPy-PCL) was modified with a UPy-conjugated Tz (UPy-Tz) additive.^[@ref41]^ We were able to detect fluorescent TCO-modified enhanced yellow fluorescent protein (TCO-EYFP) immobilized on a three-dimensional fiber mesh fabricated with electrospinning, reporting the first evidence of successful decoupling of material processing and functionalization on a UPy-based modular system. The functionalization of UPy-PCL with reactive UPy-Tz additives allowed the introduction of antifouling coatings on the material surface to reduce protein and cell adhesion.^[@ref50]^ Nevertheless, a biological read-out that proves the efficacy of the postmodification approach in immobilizing proteins has not been provided yet. Here, for the first time, we report the activity in a biological context of a supramolecular UPy-based material loaded with a TCO-modified protein, namely, protein G ([Scheme [1](#sch1){ref-type="scheme"}](#sch1){ref-type="scheme"}). Recombinant protein G is modified with TCO-moieties, and the resulting TCO-protein G variant (TCO-pG) is characterized to verify its structural integrity and cytotoxicity. UPy-PCL containing UPy-Tz is processed into cast films and electrospun fibers, and the biocompatibility, functionality, and bioactivity of the materials are investigated in comparison with materials created by random adsorption of unmodified recombinant protein G (pG). {#sch1} Materials and Methods {#sec2} ===================== Materials {#sec2.1} --------- UPy-PCL (*M*~n~ ≈ 2700 g/mol) and UPy-Tz (MW = 1190 g/mol) were synthesized by SyMO-Chem BV (Eindhoven, NL) as previously described.^[@ref50]^ TCO-OEG~4~-NHS ester (equatorial isomer) was obtained from Jena Bioscience (Jena, DE). 1,1,1,3,3,3,-Hexafluoroisopropanol (HFIP) was obtained by FluoroChem (Hadfield, UK). Other chemicals were obtained from Sigma-Aldrich and used without further purification, unless otherwise specified. Protein G Conjugation and Analysis by LC-QTOF-MS {#sec2.2} ------------------------------------------------ Recombinant protein G (ThermoFisher, 21 444 g/mol; pG) in a 1:1 mixture of water/glycerol v/v was reacted with 10 equiv of TCO-OEG~4~-NHS ester at 20 °C and 550 rpm for 1 h. The reaction was repeated three times, for a total of 30 equiv of TCO-OEG~4~-NHS ester. Afterward, the mixture was purified using 10 000 MWCO Amicon filters by diluting the reaction mixture with phosphate buffered saline (PBS) solution at pH 7.4 and centrifuging at 13 400 rpm for 5 min. The conjugated protein remained in the filter and was obtained by inverted spinning and analyzed using a Waters Xevo G2 Quadrupole Time of Flight (QToF) Liquid Chromatograph--Mass Spectrometer equipped with an Agilent Polaris C18A reverse phase column (ID 2.0 mm, length 100 mm). Proteins were flowed (0.3 mL/min) over the column using a 15 vol % to 75 vol % water/acetonitrile gradient with 0.1 vol % formic acid prior to analysis in positive mode in the mass spectrometer. Deconvolution of the *m*/*z* spectra was done using the MaxENT1 algorithm in the MassLynx software. Preparation of Drop Cast Films {#sec2.3} ------------------------------ For the preparation of drop cast films, 5 mol vol % of UPy-Tz was with the dry UPy-PCL from stock solutions in HFIP, and the amount of solvent was adjusted to reach a final UPy-PCL concentration of 50 mg/mL. Drop casting was performed by distributing 50 μL of solution on 14 mm Ø glass coverslip. Afterward, the films were dried overnight in a vacuum at 40 °C. For cell culture experiments, the films were sterilized with UV light for 10 min. Electrospinning {#sec2.4} --------------- Electrospinning was carried out on climate EC-CLI electrospinning equipment from IME Technologies (Waalre, NL). A 300 mg/mL solution of UPy-PCL with the addition of 5 mol vol % UPy-Tz was dissolved in an 8:2 mixture of CHCl~3~/HFIP v/v overnight before loading in a syringe equipped with a 19G flat-tipped nozzle. The solution was fed at a constant flow rate of 55 μL/min with a voltage of 13 kV and a tip-collector distance of 15 cm. Electrospun fibers were collected on a rotating (100 rpm) cylindrical collector covered in aluminum foil. After electrospinning, the meshes were dried overnight in vacuo. For cell culture experiments, samples were sterilized with UV light for 10 min. Scanning Electron Microscopy (SEM) {#sec2.5} ---------------------------------- Scanning electron microscopy (SEM) was performed using an FEI Quanta 600 and Xt Microscope Control software. Mesh samples were mounted on a metal stub by using double sided carbon tape. The samples were visualized under a low vacuum with an accelerating voltage of 10 kV and a working distance of 10 mm. The fiber diameters were determined from multiple high magnification images using ImageJ software. Circular Dichroism (CD) {#sec2.6} ----------------------- The secondary structures of recombinant pG and TCO-pG were evaluated with CD spectroscopy using a JASCO J-815 spectrometer and a quartz cuvette with 1 cm path length. Spectra were collected between 200 and 300 nm at room temperature. Data are expressed as molar residual ellipticity (MRE), which is calculated from the measured ellipticity using the equation:^[@ref51]^where θ is the ellipticity in millidegrees, *m* is the molecular weight in grams per mole, *c* is the protein concentration in miiligrams per milliliter, *l* is the path length of the cuvette in centimeters, and *n*~r~ is the number of amino acids in the molecule. Water Contact Angle Measurements {#sec2.7} -------------------------------- Water contact angle (WCA) measurements on drop cast films (*n* = 5) were performed on an OCA 30 system from Dataphysics using SCA20 software. A 5 μL drop of deionized water was placed on the films, and images were captured 10 s after placement of the water drop. Water contact angles were determined from the recorded images. Activation of Notch Signaling by Immobilized Ligands {#sec2.8} ---------------------------------------------------- For induction of Notch signaling with immobilized extracellular domain of the Jagged1 ligand, UPy-PCL drop cast films or electrospun meshes containing 5 mol vol % UPy-Tz were incubated with a 50 μg/mL solution of TCO-pG in PBS at room T for 1 h. As a control for aspecifically adsorbed protein, substrates were incubated with a 50 μg/mL solution of unconjugated recombinant pG in PBS at room temperature for 1 h. After coating, substrates were washed three times with PBS and further blocked with 10 mg/mL of Bovine Serum Albumin (BSA) in PBS for 2 h. As recombinant protein G lacks albumin-binding domains, this blocking step prevents aspecific adsorption of Fc-ligands at a later stage to surfaces that are not covered with protein G (conjugates). The blocked surfaces were washed with PBS and incubated with recombinant Fc-Jagged1 chimera (R&D systems) or only Immunoglobulin G (IgG) and Fc fragments (Jackson ImmunoResearch) at concentrations of 1 μg/mL in BSA 1 mg/mL in PBS for 3 h. After washing, cells were immediately seeded on the coated surfaces.^[@ref52]^ Inhibition was performed by the addition of 10 μM N-\[N-(3,5-difluorophenacetyl)-[l]{.smallcaps}-alanyl\]-S-phenylglycine t-butyl ester (DAPT) to the culture medium from stock solutions in dimethyl sulfoxide (DMSO). For comparison, all other groups were treated with the same amount of DMSO. Protein Quantification {#sec2.9} ---------------------- The amount of pG or TCO-pG deposited on either cast films or electrospun meshes was quantified by QuantiPro micro-BCA assay (Sigma-Aldrich) by directly incubating the protein-functionalized materials with the kit's working solution at 37 °C for 2 h, followed by absorbance measurements at 562 nm. The quantification of immobilized FJagged1 on electrospun meshes (*n* = 3) was carried out using a similar procedure as described for protein G. The absolute absorbance of meshes coated with pG/BSA or TCO-pG/BSA was subtracted from the absorbance of samples bioactivated by incubation with an FcJagged1 solution at 1 μg/mL. Cell Culture {#sec2.10} ------------ Human embryonic kidney (HEK) 293T cells stably expressing full-length Notch1 (HEK293 FLN1) were cultured in complete medium consisting of Dulbecco's Modified Eagle Medium (DMEM, Gibco) supplemented with 5 vol % heat inactivated fetal bovine serum (HI-FBS) (Invitrogen) and 1 vol % penicillin--streptomycin solution (Invitrogen), at 37 °C and 5 vol % CO~2~ in a humidified atmosphere. HEK293 FLN1 were cultured on tissue culture treated polystyrene under puromycin selection (1 μg/mL), passaged at 80 vol % confluency, and seeded at a concentration of 1.0 × 10^5^ cells/cm^2^ on drop cast films or electrospun meshes. Primary human coronary artery smooth muscle cells (HCASMC, Lonza) were purchased and placed at passage 3, and experiments were carried out with cells at passage 5 or 6. Cells were cultured in 231 basal medium (Gibco) supplemented with 5 vol % smooth muscle growth supplement (SMGS, Gibco) and 1 vol % penicillin--streptomycin solution (Invitrogen), at 37 °C and 5 vol % CO~2~ in a humidified atmosphere on cell culture treated polystyrene. As a control for expression of α-smooth muscle actin in the differentiated state, the basal medium was supplemented with 1 vol % smooth muscle differentiation supplement (SMDS, Gibco) instead of SMGS. HCASMC was passaged at 80 vol % confluency and seeded at a concentration of 5.0 × 10^3^ cells/cm^2^. Cytotoxicity, Cell Viability Assay {#sec2.11} ---------------------------------- Cytotoxicity of different concentrations of TCO-pG in solution for the HEK293 FLN1 cell line was assessed using an LDH Cytotoxicity Assay kit (Pierce) following the manufacturer's instructions. Cell viability was assessed as metabolic activity on different materials using an XTT in vitro Toxicology Assay kit (Sigma-Aldrich) following the manufacturer's instructions. Luciferase Reporter Assay {#sec2.12} ------------------------- Notch signaling activity in HEK293 FLN1 cells was measured by transfecting cells with a 12xCSL-Luciferase reporter construct^[@ref53]^ (250 ng per 10^5^ cells) 1 day before seeding. Poly(ethylene imine) (PEI) was used as a transfection vector in a 2:1 ratio with the DNA construct. Luminescence intensity was detected after 48 h from seeding with a Luciferase Assay Kit (Promega) following the manufacturer's instruction (*n* ≥ 6). Luminescence intensity was normalized for cell amount, which was measured with the CyQuant Assay kit (ThermoFisher) following the manufacturer's instruction. Gene Expression Analyses {#sec2.13} ------------------------ For gene expression analyses, HCASMC was cultured on the different substrates (*n* = 6) for 48 h. Total RNA was isolated using the RNAeasy isolation kit (Qiagen). cDNA was synthesized with 150 ng of RNA per sample using the M-MLV Reverse Transcriptase enzyme/kit (Bio-Rad). cDNA samples were subjected to quantitative polymerase chain reaction (qPCR) using iQ SYBR Green Supermix (Bio-Rad) and the Bio-Rad IQ5 detection system (Version 1.6). mRNA expression was normalized to glyceraldehyde 3-phosphate dehydrogenase (GAPDH) as a reference gene and expressed as a fold increase with respect to the control group. Primer sequences are available in the [Supporting Information](http://pubs.acs.org/doi/suppl/10.1021/acsapm.9b00334/suppl_file/ap9b00334_si_001.pdf). Fluorescence Microscopy {#sec2.14} ----------------------- Cells were first washed with PBS, fixated in 3.7 vol % formaldehyde (Merck) for 10 min, washed with PBS, and permeabilized with 0.5 vol % Triton X-100 (Merck) for 15 min. Cells were then incubated with phalloidin-Atto488 (Sigma-Aldrich) in the dark for 45 min, followed by incubation with DAPI (0.1 μg/mL) in PBS for 5 min. Finally, samples were washed and mounted on cover glasses with Mowiol (Sigma). Samples were imaged with a Zeiss Axiovert 200 M epifluorescence microscope. Statistical Analyses {#sec2.15} -------------------- Data are expressed as average ± standard deviation. Statistically significant differences were determined using a nonparametric Kruskall--Wallis test followed by Dunn's posthoc test. Asterisks in the figures indicate significant differences as follows: \**p* \< 0.05, \*\**p* \< 0.01, and \*\*\**p* \< 0.001. All statistical analyses were performed using GraphPad Prism 5 Software (GraphPad Software, Inc.). Results and Discussion {#sec3} ====================== Synthesis and Characterization of Reactive TCO-OEG~4~-Protein G Conjugate {#sec3.1} ------------------------------------------------------------------------- The protein G-mediated immobilization of Fc-tagged Jagged1 has been demonstrated to be beneficial compared to the random adsorption of the ligand on cell culture substrates.^[@ref54]^ The difference is attributed to a better orientation of the Fc-ligand on protein G with respect to uncontrolled adsorption. Therefore, a Tz-reactive TCO-functionalized variant of protein G was obtained by reacting the recombinant protein with a TCO-OEG~4~-NHS ester. The reaction conditions for TCO conjugation of recombinant protein G ([Figure [1](#fig1){ref-type="fig"}](#fig1){ref-type="fig"}A) were optimized with the aim of minimizing unconjugated protein, in order to exclude the influence of randomly adsorbed protein on the following experiments. In a precedent study, EYFP was reacted with 3 equiv of TCO-OEG~4~-NHS ester for 1 h, eventually yielding one to five TCO moieties per protein molecule.^[@ref41]^ In the case of recombinant protein G, the same reaction conditions led to mostly unconjugated protein, with a minor fraction of one-time conjugated protein (data not shown). Eventually, by feeding 10 equiv of TCO-OEG~4~-NHS ester three times to the reaction mixture, protein G functionalized with one to five TCO moieties was obtained, and no unconjugated protein could be detected anymore ([Figure [1](#fig1){ref-type="fig"}](#fig1){ref-type="fig"}B). {#fig1} The modified protein G was characterized by means of CD spectroscopy in order to compare signals coming from unmodified pG and TCO-pG. The shape of the two traces is found to be comparable, and it is consistent with the spectra reported in the literature ([Figure [1](#fig1){ref-type="fig"}](#fig1){ref-type="fig"}C), indicating that the modification did not affect the folding state of the protein.^[@ref55]^ In order to study the cytotoxicity of TCO-pG, a HEK293 FLN1 cell line expressing Notch1 receptors was chosen in view of the upcoming Notch signaling activity studies. TCO-pG was found to have no or low cytotoxicity for HEK293 FLN1 cells at a concentration up to 10 μM, with cell viability higher than 80 vol % ([Figure [1](#fig1){ref-type="fig"}](#fig1){ref-type="fig"}D). Cell viability decreased to about 60 vol % when the TCO-pG reached concentrations of 100 μM, although it is unlikely that such an elevated concentration is found in the applied context. Cell Response to pG or TCO-pG Functionalized Films {#sec3.2} -------------------------------------------------- Drop cast films of UPy-PCL with 5 mol vol % UPy-Tz were employed as a simple model to test cell response to materials in which protein G was incorporated either via adsorption (pG) or specific iEDDAc reaction (TCO-pG). First, the WCA of functionalized materials was measured to ascertain whether the short oligo(ethylene glycol) spacer present in TCO-pG might affect the material hydrophilicity and therefore cell behavior. The WCA decreased upon protein incorporation from 70° ± 1° for UPy-PCL/UPy-Tz to 67° ± 3° and 63° ± 4° for UPy-PCL/UPy-Tz with pG and TCO-pG, respectively ([Figure [2](#fig2){ref-type="fig"}](#fig2){ref-type="fig"}A). The measured WCA values are consistent with previously reported values for comparable systems,^[@ref41]^ and they are not likely to affect the cell response on bioactive films. {#fig2} HEK293 FLN1 cells adhered on films with aspecifically adsorbed pG, while they were repelled by surfaces on which TCO-pG was reacted with the Tz moieties present at the surface ([Figure [2](#fig2){ref-type="fig"}](#fig2){ref-type="fig"}B). As the viability of HEK293 FLN1 cells that did not adhere is not different from the viability of cells adhered on control UPy-PCL/UPy-Tz film or on randomly adsorbed protein G ([Figure [2](#fig2){ref-type="fig"}](#fig2){ref-type="fig"}C), toxicity effects of surfaces modified with iEDDAC possibly being the cause of the unfavorable cell response toward materials functionalized with reactive TCO-pG can be excluded. After protein immobilization via either adsorption or "click" reaction, the films were functionalized with the Fc-Jagged1 ligand (or Fc fragment only as a negative control) to measure Notch signaling activity. Consistently with the observed cell response, cells stimulated on materials with adsorbed protein G/Fc-Jagged1 exhibited a significant 2.8-fold increase in Notch activity, while no significant effect was detected on materials with TCO-pG/Fc-Jagged1 ([Figure [2](#fig2){ref-type="fig"}](#fig2){ref-type="fig"}D). To discriminate between the effects of the iEDDAC reaction and the TCO-modified protein G on cell adhesion, we observed the behavior of HEK293 FLN1 cells and CASMC, a good model for cardiovascular applications, on a film of UPy-PCL without any UPy-Tz that have been incubated for 1 h with either pG or TCO-pG. For both cell types, the response in terms of adhesion and morphology is consistent with the one observed in the presence of UPy-Tz. On UPy-PCL and UPy-PCL/pG films, the cells formed healthy confluent monolayers, while they formed few clustered aggregates on UPy-PCL with adsorbed TCO-pG ([Figure S1](http://pubs.acs.org/doi/suppl/10.1021/acsapm.9b00334/suppl_file/ap9b00334_si_001.pdf)). In light of this, it is concluded that the negative cell response to TCO-pG, either reacted or adsorbed, might be due to conformational changes of the protein in contact with the relatively hydrophobic material, which might lead to the exposure of cell-repellent portions of pG. Although the CD spectroscopy characterization did not reveal structural changes of TCO-pG in solution with respect to pG ([Figure [1](#fig1){ref-type="fig"}](#fig1){ref-type="fig"}C), we cannot exclude that the TCO-modification could have affected the stability of the protein's folding state when in the proximity of the material. Although the CD spectroscopy characterization in solution did not reveal structural changes of TCO-pG with respect to pG ([Figure [1](#fig1){ref-type="fig"}](#fig1){ref-type="fig"}C), this measurement is performed in solution only, and it is not indicative of the interaction of the protein with the material surface during adsorption. We cannot exclude that the stability of the protein's folding state in the proximity of the solid material surface is compromised because of the presence of TCO moieties. WCA measurements were performed to get insights on the material surface characteristics after functionalization. The introduction of TCO-pG on the material's surface decreases the average WCA of 7° ± 4° only ([Figure [2](#fig2){ref-type="fig"}](#fig2){ref-type="fig"}A), and this value is almost identical to the WCA decrease caused by pG adsorption, which instead supports cell-adhesion. Therefore, we are unable to make a connection between surface hydrophilicity and cell-repulsion based on the available data. Cell Response on Functionalized Electrospun Meshes {#sec3.3} -------------------------------------------------- The proposed material modification was translated from a two-dimensional film to a porous fibrous mesh prepared with electrospinning, as this is a more relevant structure for applications such as scaffolds for soft tissue engineering. A solution of UPy-PCL with 5 mol vol % UPy-Tz was electrospun into meshes with an average fiber diameter of 2.3 ± 0.7 μm ([Figure [3](#fig3){ref-type="fig"}](#fig3){ref-type="fig"}A). The reaction of TCO-pG and coating of the unmodified pG were carried out as described for the cast film. Quantification of the amount of protein G deposited on the electrospun meshes revealed that the iEDDAC reaction of TCO-pG on the material surface leads to a 7.2-fold increase in the amount of protein presented on the scaffold with respect to the amount of pG present after a specific adsorption ([Figure [3](#fig3){ref-type="fig"}](#fig3){ref-type="fig"}B). Therefore, it can be concluded that the immobilization of pG via TCO-Tz chemistry allows loading of the pristine material with a wider range of bioactive protein concentrations than mere adsorption. Protein quantification was attempted for cast films as well, but the amount of protein G turned out to be below the detection limit of the assay (data not shown). The difference can likely be attributed to the large surface area that characterizes porous electrospun meshes compared to two-dimensional films. A higher amount of immobilized protein G is hypothesized to cause a higher amount of immobilized Fc-fusion protein. To test this hypothesis, the amount of FcJagged1 bound to electrospun scaffolds functionalized with either pG or TCO-pG was quantified as well. The results show an average 9.6-fold increase in FcJagged1 loading ([Figure S2](http://pubs.acs.org/doi/suppl/10.1021/acsapm.9b00334/suppl_file/ap9b00334_si_001.pdf)). Although the increase is not statistically significant due to the large spread in the values, the data suggest that the iEDDAC-mediate functionalization might yield a higher Fc-fusion protein loading on the scaffold. {#fig3} The HEK293 FLN1 cells' response notably improved on the electrospun substrates with respect to cast films. Cells were able to adhere on UPy-PCL/UPy-Tz meshes with protein G either adsorbed or immobilized via specific iEDDAC reaction, although they showed slight clustering behavior on the latter ([Figure [4](#fig4){ref-type="fig"}](#fig4){ref-type="fig"}A), indicating that the cells still prefer cell--cell contact instead of cell-material contact. Nevertheless, cell metabolic activity on the protein-loaded material was found to be similar to that on the pristine UPy-PCL/UPy-Tz ([Figure [4](#fig4){ref-type="fig"}](#fig4){ref-type="fig"}B), indicating that HEK293 FLN1 cells do not suffer cytotoxic effects of the protein-modified material. After introduction of the bioactive Fc-Jagged1 ligand on the material, activation of Notch signaling in HEK293 FLN1 cells was detected on both substrates. The immobilized Fc-Jagged1 ligand elicited a 2.9-fold and 5.7-fold increase in luminescence intensity on materials with pG/Fc-Jagged1 and TCO-pG/Fc-Jagged1, respectively ([Figure [4](#fig4){ref-type="fig"}](#fig4){ref-type="fig"}C). {#fig4} The mechanism of action of the Notch signaling pathway involves a number of inter- and intracellular processes that occur after ligand--receptor binding, such as endocytosis, glycosylation, and feedback regulation of ligand and receptor levels.^[@ref56]−[@ref59]^ These processes are essential for the translation of the signal into a functional response of the cell; therefore we found it relevant to investigate whether the complexity of the biological process could be recapitulated on the bioactive material, besides the ligand--receptor binding phenomena demonstrated in [Figure [4](#fig4){ref-type="fig"}](#fig4){ref-type="fig"}C. Primary human vascular smooth muscle cells endogenously expressing the Notch3 receptor were cultured on UPy-PCL/UPy-Tz electrospun meshes modified with TCO-pG/Fc-Jagged1 to track the expression of genes regulating transcription downstream of NICD translocation to the nucleus, namely, HEY1 and HES1, and genes related to the regulation of ligand and receptor levels, namely, JAG1 and NOTCH3. Gene expression analyses reveal upregulation of expression levels of all the mentioned genes with respect to control samples presenting the TCO-pG/Fc fragment only. Consistently, the addition of DAPT, a γ-secretase inhibitor, decreases expression levels for all the investigated genes ([Figure [5](#fig5){ref-type="fig"}](#fig5){ref-type="fig"}). Interestingly, the expression intensity of all the investigated genes on TCO-pG/Fc-Jagged1 functionalized substrates is higher than on meshes coated aspecifically with pG/Fc-Jagged1. This might be due to a higher amount of Fc-Jagged1 immobilized on the material surface, as it is shown in [Figure [3](#fig3){ref-type="fig"}](#fig3){ref-type="fig"}B that performing the iEDDAC reaction allowed the immobilization of more protein G with respect to random absorption. {#fig5} On the basis of the findings illustrated in [Figures [4](#fig4){ref-type="fig"}](#fig4){ref-type="fig"}C and [5](#fig5){ref-type="fig"}, we concluded that a functional cell response can be induced upon introduction of reactive protein G variant that binds to an Fc-fusion protein on supramolecular UPy-PCL. Notwithstanding the successful activation of the Notch signaling pathway and its downstream effects, the issue of compromised cell adhesion on substrates modified with TCO-pG is still present on electrospun meshes, although significantly less severe than on drop cast films. We speculate that the TCO-functionalization of protein G might have affected the stability of the protein, which in contact with the material surface can undergo structural changes that make it cell-repellent. However, it is still unclear how materials that have been functionalized in exactly equal ways, but have been processed by either solvent casting or electrospinning, lead to drastically different cellular responses ([Figures [2](#fig2){ref-type="fig"}](#fig2){ref-type="fig"}B and [4](#fig4){ref-type="fig"}). One explanation might be that during polymer processing many parameters related to the technique can influence the dynamics of the polymer matrix and additive, thus leading to differences in the amount and modes of presentation of additives on the surface. Hence, it is hypothesized that a modified UPy-Tz exposure might be at the origin of the differential biocompatibility of drop cast films and electrospun meshes, which counteracts the cell-repellent effect of TCO-pG in the latter case. Another factor that might contribute to an improved cell adhesion on the electrospun material is the availability of a microporous environment which closely represents the natural cellular environment.^[@ref60],[@ref61]^ In this case, microstructural parameters might be dominant over surface chemistry in dictating cell adhesion. In this study, the ability of TCO-pG to bind and orient Fc-fusion ligands was tested by performing bioactivation of the surface with FcJagged1, a Notch ligand. The results in [Figure [3](#fig3){ref-type="fig"}](#fig3){ref-type="fig"}B and [Figure S2](http://pubs.acs.org/doi/suppl/10.1021/acsapm.9b00334/suppl_file/ap9b00334_si_001.pdf) show that the reactive immobilization results in a 7.2-fold higher loading of TCO-pG, and a (not significant) 9.6-fold increase in FcJagged1 binding. Nonetheless, no statistical difference in Notch signaling intensity was detected between chemical immobilization and physical adsorption ([Figures [4](#fig4){ref-type="fig"}](#fig4){ref-type="fig"}C and [5](#fig5){ref-type="fig"}), even though it is known that Notch signaling activity increases linearly with the concentration of surface-bound FcJagged1 in this cell model.^[@ref62]^ One explanation for this might be that the covalently bound TCO-pG does not provide the correct orientation of the Fc-fusion ligand. Improvements on this aspect could be done by developing site-directed protein modification chemistry, thereby providing control on the targeted residues. Regardless, the iEDDAC-mediated protein immobilization offers the advantage of a covalent anchoring of the protein to the material, thus providing stable functionalization in complex environments such as body fluids. Here, a physically adsorbed protein is susceptible to displacement from the surface by other proteins with larger affinity for the material surface. Taken together, our results represent the first evidence of successful postprocessing bioactivation of a supramolecular UPy-based solid material by means of click chemistry for the immobilization of Fc-fusion ligands. A large number of cytokines, enzymes, and receptor extra-cellular domains are commercially available as Fc-fusion proteins due to advances in the techniques for their large scale production.^[@ref63]^ Furthermore, up to date, nine Fc-fusion proteins have been approved as drugs by the FDA, and a large number are currently in clinical trials^[@ref64]^ since the Fc domain is often used to prolong the half-life of their conjugate counterpart *in vivo* or enhance their therapeutic efficacy.^[@ref65]^ Besides Fc-fusion proteins, the system described in this work can be potentially used for binding antibodies, which are frequently used in pharmacology to target receptors and cellular processes with high specificity.^[@ref66]^ In light of this, it is expected that the development of substrates postmodified to incorporate Fc-tagged protein and antibodies will find broad applicability in therapeutic strategies that require processed biomaterials. Conclusions {#sec4} =========== A versatile and modular platform for introduction of bioactive Fc-fusion proteins on supramolecular solid materials based on UPy was developed via immobilization on Fc-binding protein G. TCO modification of recombinant protein G allowed functionalization of UPy-PCL/UPy-Tz surfaces via a fast iEDDAC reaction. The introduction of TCO-pG on cast or electrospun surfaces turned out to affect the biocompatibility of the material in a processing-dependent manner. Films of UPy-PCL/UPy-Tz functionalized with TCO-pG turned out to be cell-repellent, while mere adsorption of protein G did not compromise cell adhesion. Instead, UPy-PCL/UPy-Tz fibers created by electrospinning supported cell adhesion, viability, and functionality. Bioactivation of supramolecular substrates was carried out by incorporation of Fc-Jagged1, a Notch ligand of interest in tissue regeneration strategies, on the Fc-binding material surface. The Fc-fusion ligand retained its functionality on TCO-pG modified electrospun materials, indicating that complex cellular processes such as Notch signaling can be recapitulated on the modified material in cell models expressing Notch1 and Notch3 receptors. In conclusion, a convenient strategy based on biorthogonal click-chemistry is found to be suitable to modify supramolecular material surfaces. We envision that the system described in this work potentially has broad applications in the display of bioactive proteins, such as Fc-fusion proteins and therapeutic antibodies, onto processed solid materials. The Supporting Information is available free of charge on the [ACS Publications website](http://pubs.acs.org) at DOI: [10.1021/acsapm.9b00334](http://pubs.acs.org/doi/abs/10.1021/acsapm.9b00334).List of primers used for gene expression; fluorescence micrographs of HEK293 FLN1 and CASMC on UPy-PCL cast films with adsorbed pG or TCO-pG; quantification of FcJagged1 immobilization on pG or TCO-pG functionalized electrospun meshes ([PDF](http://pubs.acs.org/doi/suppl/10.1021/acsapm.9b00334/suppl_file/ap9b00334_si_001.pdf)) Supplementary Material ====================== ###### ap9b00334_si_001.pdf The authors declare no competing financial interest. Carlijn Bouten is acknowledged for valuable discussions and Maaike Schotman for assistance with CD measurements. The ICMS Animation Studio is acknowledged for the design of some of the cartoons. This work was funded by the European Research Council (FP7/2007-2013) Grant Agreement 604514 and ERC Grant Agreement 308045 and the Ministry of Education, Culture, and Science (Gravity Program 024.001.03).

1. Introduction {#sec1} =============== Cytopenia is a frequent hematological disorder in patients with human immunodeficiency virus (HIV) infection. The most common manifestation is reduction of any of the blood cell lines, leading to neutropenia, anemia, and thrombocytopenia. Previous studies have reported prevalence rates of anemia, neutropenia, and thrombocytopenia of 1.3%--95% \[[@B1]\], 10%--85% \[[@B2], [@B3]\], and 7%--21% \[[@B4], [@B5]\], respectively; the wide span reflects the different definitions, geographical locations, race/ethnicity of patients, and stages of disease. Lower CD4^+^ T-cell counts, higher viral loads, advanced disease stages, and side effects of medicines used for HIV were risk factors for cytopenias in HIV-infected patients \[[@B5]\]. In HIV-infected patients, cytopenia has been associated with the progression to AIDS (anemia), high mortality (anemia), hospitalization (all), and secondary infections caused by bacterial and fungal pathogens (neutropenia) or drugs used for HIV \[[@B6]--[@B11]\]. Highly active antiretroviral therapy (HAART) has been proven to reduce AIDS-related mortality significantly and leads to improvements in cytopenia \[[@B9], [@B12]--[@B16]\]. However, some antiretroviral compounds such as zidovudine (AZT) also exert bone marrow cytotoxicity and contribute to cytopenia in HIV-infected patients \[[@B9], [@B13], [@B17]\]. Monitoring changes in hematological parameters in HIV-infected patients while they are on HAART could be a useful indicator of the patient response to HAART. In China, Dai et al. found that the prevalence of anemia was 9.76% \[[@B6]\], while Fan et al. reported a thrombocytopenia rate of 4.5% among 1730 patients \[[@B18]\]. Shen reported that the prevalence rates of leukopenia and thrombocytopenia were 33.2% and 15.6% \[[@B19]\]. However, the prevalence rates and risk factors of cytopenia and the effect of HAART on cytopenia have not been well documented. This study was designed to investigate the prevalence and risk factors of cytopenia among HAART-naïve patients. Furthermore, we aimed to evaluate the effect of HAART on cytopenia. 2. Materials and Methods {#sec2} ======================== 2.1. Study Subjects {#sec2.1} ------------------- In total, 5047 HIV-infected patients who were on follow-up after HAART at the Department of Infectious Diseases, Beijing Ditan Hospital, Capital Medical University, from November, 2004 to August, 2016, were included in this study. This study was approved by the ethics committees of Beijing Ditan Hospital, Capital Medical University, and the study conformed to the Declaration of Helsinki. The profiles of 722 participants were excluded because of neutrophilia (*n* = 108) or incomplete information (*n* = 613). Finally, a total of 4325 patients were included. The demographic and clinical data for the HAART-naïve patients including sex, age, height, weight, route of transmission, the initial time of HAART treatment, and WHO stage were collected. Laboratory data such as absolute neutrophil numbers, hemoglobin concentration, absolute platelet count, and absolute CD4^+^ T-cell count were collected at baseline. To investigate the effect of HAART on cytopenia, 824 HIV-infected patients with cytopenia were followed up to 24 months after the initiation of HAART. The exclusion criteria were all patients without cytopenia *n* = 3501, patients who were lost at follow-up *n* = 48, fatalities *n* = 23, irregular follow-up times *n* = 73, and lack of follow-up time points (\>2 follow-up time points) *n* = 69. In total, 213 patients were excluded and 611 HIV-infected patients with cytopenia were included. The number of neutrophils, the platelet count, the CD4^+^ T-cell count, and the hemoglobin concentration at 6, 12 and 24 months after the initiation of HAART were recorded. The flow chart is shown in [Figure 1](#fig1){ref-type="fig"}. 2.2. Definition {#sec2.2} --------------- Conditions were defined based on the following parameters. Anemia: hemoglobin \<11 g/dL in women or \<12 g/dL in men \[[@B20]\]; neutropenia: neutrophil count \<2000 cells/*μ*L; thrombocytopenia: platelet count \<100,000 cells/*μ*L. Patients with isolated anemia, thrombocytopenia or neutropenia were defined as patients with unicytopenia. Bicytopenia was defined as a patient in whom any two of the three lineage cell counts (neutrophils, hemoglobin, or platelets) were below the levels designated above. Pancytopenia was defined as having three lineage cell counts (neutrophils, hemoglobin, or platelets) below the levels designated above. 2.3. Data Collection {#sec2.3} -------------------- Retrospective data were collected from medical records, including demographic, clinical, and laboratory characteristics. Variables include age, sex, HIV transmission route, coinfection with HBV, coinfection with HCV, WHO staging, HIV viral load, the use of AZT and stavudine (D4T), and body mass index (BMI). Age was categorized as \<40 and over 40 years. Based on Chinese guidelines for diagnosis and treatment of HIV/AIDS, CD4 T-cell count was categorized as \<200 cells/*μ*l and over 200 cells/*μ*l \[[@B21]\]. Based on the WHO clinical staging of HIV disease in adults and adolescents, subjects were classified in to stage I, stage II, stage III, and stage IV \[[@B22]\]. The cutoff for BMI for defining underweight was \<18.5 kg/m^2^ and overweight and obesity was \>24.0 kg/m^2^ \[[@B23], [@B24]\]. HIV viral load was categorized as \<100000 copies/ml and over 100000 copies/ml \[[@B25]\]. 2.4. Statistical Analysis {#sec2.4} ------------------------- All statistical analyses were performed using SPSS 22.0 software (SPSS, Chicago, IL, USA) and GraphPad 7 (GraphPad Software, La Jolla, CA, USA). Data were presented as the medians with interquartile ranges (IQR) due to skewed statistical distributions, while categorical variables were presented by percentages. The differences between two groups were analyzed using the Mann--Whitney *U* test (nonparametric). A Pearson\'s test was used to evaluate differences in categorical variables. Binary logistic regression analysis or ordinal logistic regression analysis was used to evaluate risk factors associated with cytopenias and persistent cytopenia after 24 months of HAART. Demographic variables (sex and age) and clinical/laboratory characteristics (CD4^+^ T-cell count, BMI, WHO staging, viral load, transmission routes, coinfection with HBV, and coinfection with HCV) were included to investigate for the risk factor of cytopenia. The risk factors of persistent cytopenia investigated included demographic variables (sex and age) and clinical/laboratory variables (CD4^+^ T-cell count at baseline and after 24 months HAART, BMI, WHO staging at baseline, coinfection with HBV, and coinfection with HCV, viral load, and the use of AZT and D4T). Statistical significance was set at a level of *p* \< 0.05. 3. Results {#sec3} ========== 3.1. The Demographic and Clinical Characteristics of HAART-Naïve Patients {#sec3.1} ------------------------------------------------------------------------- A total of 4325 individuals were included in this study, of which 26.1% had a CD4 cell count of \<200 cells/*μ*L. The median age was 36 years, the ratio of males to females was 15 : 1, the median BMI was 21.5 (19.7, 23.7) kg/m^2^, and the median CD4^+^ T-cell count was 249 (151, 356) cells/*μ*L. Sexual transmission was mainly via the transmission route, which accounted for 91% of cases. According to WHO staging, stages II and III were present in 2541 (81.9%) individuals ([Table 1](#tab1){ref-type="table"}). 3.2. The Prevalence and Distribution of Cytopenia in HAART-Naïve Patients {#sec3.2} ------------------------------------------------------------------------- Cytopenia was detected in 824 (19.1%) patients in our cohort, of which 662 (15.3%) patients had unicytopenia, 144 (3.3%) had bicytopenia, and 18 (0.4%) had pancytopenia. Among the unicytopenia cases, 287 (6.63%) patients had isolated anemia, 306 (7.7%) had isolated neutropenia, and 69 (1.6%) had isolated thrombocytopenia. Among the bicytopenia cases, 96 (2.2%) had anemia and neutropenia, 26 (0.6%) had anemia and thrombocytopenia, and 22 (0.5%) had thrombocytopenia and neutropenia. The prevalence and distribution of cytopenia are shown in [Figure 2](#fig2){ref-type="fig"}. 3.3. The Risk Factors for Cytopenia in HAART-Naïve Patients {#sec3.3} ----------------------------------------------------------- As seen in [Figure 3](#fig3){ref-type="fig"}, multivariate logistic regression analyses were performed to investigate the risk factors of cytopenia in HAART-naïve patients. We found that a CD4 cell count \<200 cells/*μ*L (OR: 3.57, 95% CI: 2.83--4.5, *p* \< 0.001), femaleness (OR: 2.92, 95% CI: 2--4.28, *p* \< 0.001), WHO stage IV (OR: 1.94, 95% CI: 1.45--2.57, *p* \< 0.001), coinfection with hepatitis B virus (HBV) (OR: 1.8, 95% CI: 1.12--2.9, *p*=0.007), BMI \<18.5 kg/m^2^ (OR 1.61, 95% CI: 1.1--2.32, *p*=0.01), a viral load \>100,000 (OR: 1.55, 95% CI: 1.24--1.92, *p* \< 0.001), and age ≥40 years (OR: 1.51, 95% CI: 1.21--1.9, *p* \< 0.01) were all risk factors of cytopenia ([Figure 3(a)](#fig3){ref-type="fig"}; [Table 2](#tab2){ref-type="table"}). Cytopenia was not associated with HCV coinfection or transmission route. Considering the distribution of cytopenia, we chose the ordinal logistic regression model to test the risk factors of blood cell line reduction. Our results showed that lower CD4 cell count \<200 cells/*μ*L (OR: 3.64, 95% CI: 2.9--4.58, *p* \< 0.001), femaleness (OR: 2.9, 95% CI: 2--4.3, *p* \< 0.001), WHO stage IV (OR: 2.11, 95% CI: 1.6--2.79, *p* \< 0.001), coinfection with hepatitis B virus (HBV) (OR: 1.8, 95% CI: 1.18--2.71, *p*=0.01), BMI \<18.5 kg/m^2^ (OR 1.4, 95% CI: 1.1--1.82, *p*=0.048), a viral load \>100,000 (OR: 1.6, 95% CI: 1.28--2, *p* \< 0.001), and age ≥40 years (OR: 1.55, 95% CI: 1.25--1.93, *p* \< 0.001) were all risk factors of cytopenias ([Figure 3(b)](#fig3){ref-type="fig"}; [Table 3](#tab3){ref-type="table"}). 3.4. Effect of HAART on Cytopenia {#sec3.4} --------------------------------- Obvious resolution of cytopenia was observed after 24 months of HAART. The percentage of cytopenia recovery is shown in [Table 4](#tab4){ref-type="table"}. We found that the proportions of neutropenia, anemia, thrombocytopenia, bicytopenia, and pancytopenia were all decreased after 6, 12, and 24 months of HAART, as shown in [Figure 4(a)](#fig4){ref-type="fig"}. After antiviral treatment, with the increase in the CD4^+^ T-cell count, the absolute count of neutrophils, the absolute platelet counts, and the hemoglobin concentration in cytopenia patients increased sharply up to 6 months of HAART and then slowly after 6 months of therapy (Figures [4(b)](#fig4){ref-type="fig"}--[4(d)](#fig4){ref-type="fig"}). 3.5. The Risk Factors for Persistent Cytopenia after 24 Months HAART {#sec3.5} -------------------------------------------------------------------- To investigate the risk factors for these patients without persistent cytopenia, a logistic regression analysis was performed. We found that a CD4 cell count of \<200 cells/*μ*L at baseline (OR: 2.42, 95% CI: 1.2--4.86, *p*=0.013), femaleness (OR: 5.54, 95% CI: 2.47--12.42, *p* \< 0.001), coinfection with HBV (OR: 4.84, 95% CI: 1.47--15.91, *p*=0.01), and treatment with AZT (OR: 3.71, 95% CI: 1.16--11.9, *p*=0.009) were all risk factors for unrecoverable hematological parameters after 24 months of antiviral therapy ([Table 5](#tab5){ref-type="table"}). 4. Discussion {#sec4} ============= In the present study, we found that HAART was an effective treatment for cytopenia patients before the initiation of HAART. Along with the recovery of the CD4^+^ T-cell count, blood cells (neutrophils, platelets, and hemoglobin) were also increased. Femaleness, the lower CD4^+^ T-cell count at baseline, an advance stage at baseline, and coinfection with HBV were risk factors for persistent cytopenia after HAART, which suggested the importance of starting HAART early in cytopenia patients with HIV infection. There have been a large number of studies about the prevalence of anemia, thrombocytopenia, leucopenia, and neutropenia \[[@B26]--[@B30]\], and HIV-1-infected patients often show a reduction in two or three blood cell lines. Furthermore, pancytopenia was linked with higher mortality than nonpancytopenia \[[@B31]\]; however, there have been few studies to investigate this fully. In this study, we found that the most common type of cytopenia was unicytopenia, followed by bicytopenia and pancytopenia. Among unicytopenia, the neutropenia was the most common cytopenia and neutropenia and anemia was the most common bicytopenia. In this study, CD4 \<200 cells/*μ*l, advanced WHO clinical stage, higher HIV viral load, age ≥40 years, and coinfection with HBV at baseline was identified as the risk factors for cytopenia in the multivariate analysis, a finding consistent with that of the previous studies \[[@B32]\]. The association of low CD4 cell count with cytopenia may be due to the effect of HIV on the function of early hematopoietic progenitor cells \[[@B33]\]. Previous studies have reported that HAART is an effective treatment for anemia, thrombocytopenia, and neutropenia \[[@B16], [@B28], [@B34]\] in HIV-1-infected patients, and the results of our study were consistent with this. We found that the neutrophil count, the platelet count, and the hemoglobin concentration were increased and the proportions of neutropenia, anemia, and thrombocytopenia were decreased after 6 months of HAART. The resolution of cytopenia was associated with the improvement in the CD4^+^ T-cell count, which indicated that HIV-related cytopenia was caused by HIV-1 and immunosuppression and that virus suppression and immunological reconstitution could prompt the normalization of blood cells. In accordance with a previous study, we also found that a low CD4^+^ T-cell count and advanced staging before the initiation of HAART were predictors of patients with persistent cytopenia after 24 months of HAART. Levine et al. found that HAART use and higher CD4^+^ T-cell counts were associated with the resolution of neutropenia \[[@B34]\]. Berhane et al. reported that a CD4 cell count \<200 cells/*μ*L, HIV RNA \>5000 copies/mL, and erythrocyte mean corpuscular volume \<80 fl were independent predictors of anemia after 12 months of HAART \[[@B14]\]. O\'Bryan et al. demonstrated that patients with a higher HIV load were at risk of persistent thrombocytopenia after HAART \[[@B15]\]. In the present study, we also found that a low CD4^+^ T-cell count was associated with HAART-naïve patients with cytopenias, which suggested the advanced stage before initiation of HAART was associated with the risk of cytopenia and the recovery ratio of cytopenia after treatment. More importantly, early detection of HIV could decrease the prevalence of cytopenias, while early initiation of HAART was very important for HIV-1 infected patients with cytopenia. A previous study found that the incidence of cytopenia was 63.2% in HIV-infected patients in South Africa \[[@B35]\]. Kyeyune et al. reported that the prevalence of cytopenia was 65%, of which 21.9% of patients had bicytopenia and only 2/400 patients had pancytopenia \[[@B5]\]. In the present study, we found that the prevalence of cytopenia was 19.1%, and the difference between this result and that of the former study might be due to differences in the population and the geographical location. The present study found that the ratio of recovery was 68.2% in patients with cytopenia after 6 months HAART, and 76.4% of patients with cytopenia returned to normal blood cell levels after 24 months HAART. After 6 months of antiviral therapy, the increase in blood cells obviously slowed down. The main reasons for this were that viral replication was suppressed and immunity was improved by 6 months of HAART, and HIV-1 related cytopenia caused by myelosuppression was relieved. Other factors such as malnutrition and a patient\'s economic situation may play a role in persistent cytopenia after 6 months of HAART. It is not only HIV-1, but also HAART such as AZT, that can lead to persistent hematopoietic suppression and resulting cytopenia \[[@B3], [@B17]\]. In our study, among cytopenia patients with persistent cytopenia after 24 months of HAART, AZT was used in 28 patients. We also found that persistent cytopenia was associated with the effects of AZT. There are some limitations to our study. First, there was potential inherent bias in a retrospective observational study. Second, most of the participants in the present study were urban residents of Beijing, who are not necessarily representative of all HIV-infected patients in China. 5. Conclusion {#sec5} ============= The prevalence of cytopenia in patients before the initiation of HAART was 19.1%, of which the prevalence of unicytopenia, bicytopenia, and pancytopenia was 15.3%, 3.3%, and 0.4%, respectively. Cytopenia was associated with femaleness, a lower BMI, a lower CD4^+^ T-cell count, a higher viral load, WHO stage IV, and coinfection with HBV. HAART was an effective treatment for cytopenia, and along with the CD4^+^ T-cell count, the blood cell count was also increased. Persistent cytopenia after 24 months of HAART was associated with femaleness, coinfection with HBV, a CD4^+^ T-cell count at baseline, and WHO stage IV at baseline. Early initiation of HAART and combination antiretroviral therapy without AZT seems to promote the recovery of HIV-infected patients with cytopenia. We acknowledge the work of HIV health care providers for their diagnosis, nursing, and treatment of HIV/AIDS patients in Ditan Hospital. We also thank International Science Editing (<http://www.internationalscienceediting.com>) for editing this manuscript. This work was supported by the grants from the National Natural Science Foundation of China (No. 81672000) and the Thirteen-fifth Key Project (No. 2018ZX10715-005). Data Availability ================= The clinical data used to support the findings of this study have not been made available because of patient privacy. Disclosure ========== The funders played no role in study design, data collection, analysis, the decision to publish, or preparation of the manuscript. Conflicts of Interest ===================== The authors declare no conflicts of interest regarding the publication of this article. Authors\' Contributions ======================= Hongxin Zhao conceived and designed the experiments; Lina Fan and Cuilin Li were involved in the acquisition of data and analyzed the data. Lina Fan wrote the manuscript. {#fig1} {#fig2} {#fig3} {#fig4} ###### The demographic and clinical characteristics of participants. Characteristics Total (*n* = 4325) Noncytopenia (*n* = 3501) Cytopenia (*n* = 824) ------------------------------ -------------------- --------------------------- ----------------------- Age (years)  \<40 2796 (64.6%) 2442 (67.6%) 354 (49.8%)  ≥40 1530 (35.4%) 1173 (32.4%) 357 (50.2%) Sex  Male 4048 (93.6%) 3340 (95.4%) 708 (85.9%)  Female 277 (6.4%) 161 (4.6%) 116 (14.1%) WHO stage  I 3307 (76.5%) 2795 (79.8%) 512 (62.1%)  II 234 (5.4%) 188 (5.4%) 46 (5.6%)  III 281 (6.5%) 217 (6.2%) 64 (7.8%)  IV 503 (11.6%) 301 (8.6%) 202 (24.5%) BMI (kg/m^2^)  \<18.5 337 (9.8%) 240 (8.1%) 97 (19.7%)  18.5--24 2248 (65.3%) 1938 (65.7%) 310 (62.9%)  \>24 858 (24.9%) 772 (26.2%) 86 (17.4%) CD4^+^ T-cell counts (/*μ*l)  \<200 1475 (35.6%) 949 (22.9%) 526 (66.2%)  ≥200 2667 (64.4%) 2398 (71.6%) 269 (33.8%) Viral load (copies/ml)  ≥100000 979 (30%) 699 (26.2%) 280 (48%)  \<100000 2280 (70%) 1977 (73.9%) 303 (52%) Infection routes  Sexual 3933 (91%) 3258 (93.1%) 675 (81.9%)  Others 392 (9.9%) 243 (6.9%) 149 (18.1%) HCV-coinfection  No 3481 (80.5%) 2934 (81.2%) 547 (76.9%)  Yes 77 (19.5%) 681 (15.7%) 164 (23.1%) HBV-coinfection  No 4111 (95%) 3474 (96.1%) 637 (89.6%)  Yes 215 (5%) 141 (3.9%) 74 (10.4%) Note: data are presented as *n* (%); WHO, World Health Organization; BMI, body mass index; variable had missing values: BMI = 883; HIV RNA = 1066; CD4^+^ T-cell counts = 183. ###### The risk factors of cytopenia in HAART-naïve patients.   df Univariate analysis OR (95% CI) *p* value Multivariate analysis AOR (95% CI) *p* value ------------------ ---- --------------------------------- ----------- ------------------------------------ ----------- Age (years) 1          \<40   Ref --- Ref ---  ≥40   1.5 (1.21, 1.9) \<0.01 1.51 (1.21, 1.9) \<0.01 Sex 1          Male   Ref --- Ref ---  Female   2.9 (2, 4.3) \<0.001 2.92 (2, 4.28) \<0.001 WHO stage 3   \<0.001   \<0.001  I   Ref --- Ref ---  II   0.73 (0.44, 1.22) 0.064 0.73 (0.44, 1.22) 0.23  III   1.03 (0.67, 1.59) 0.65 1.03 (0.67, 1.59) 0.9  IV   1.94 (1.45, 2.58) \<0.001 1.94 (1.45, 2.57) \<0.001 HCV-coinfection 1          No   Ref --- Ref ---  Yes   1.8 (1.18, 2.76) 0.015 1.8 (1.12, 2.87) 0.007 CD4 (cells/*μ*l) 1          ≥200   Ref --- Ref ---  \<200   3.57 (2.8, 4.5) \<0.001 3.57 (2.83, 4.5) \<0.001 HIV RNA 1          \<100000   Ref --- Ref ---  ≥100000   1.55 (1.24, 2.58) \<0.001 1.55 (1.24, 1.92) \<0.001 BMI 2   0.034   0.034  18.5--24   Ref --- Ref ---  \<18.5   1.6 (1.12, 2.31) 0.01 1.61 (1.1, 2.32) 0.01  \>24   1.2 (0.9, 1.54) 0.23 1.18 (0.9, 1.54) 0.23 Infection routes 1          Others   Ref ---      Sexual   1.27 (0.6, 2.65) 0.53     HCV-coinfection 1          No   Ref ---      Yes   1.94 (0.98, 3.82) 0.056     Note: df, degree of freedom; WHO, World Health Organization; BMI, body mass index; Ref, reference. ###### The ordinal logistic regression model of cytopenia in HAART-naïve patients.   B AOR (95% CI) *p* value ------------------ -------- ------------------- ----------- Age (years)  \<40 Ref --- ---  ≥40 0.44 1.55 (1.25, 1.93) \<0.001 Sex  Male Ref --- ---  Female 1.064 2.9 (2, 4.3) \<0.001 WHO stage  I Ref --- ---  II −0.3 0.74 (0.45, 1.23) 0.249  III 0.071 1.07 (0.7, 1.65) 0.74  IV 0.748 2.11 (1.6, 2.79) \<0.001 HBV-coinfection  No Ref --- ---  Yes 0.58 1.8 (1.18, 2.71) 0.01 CD4 (cells/*μ*l)  ≥200 Ref --- ---  \<200 1.29 3.64 (2.9, 4.58) \<0.001 HIV RNA  \<100000 Ref --- ---  ≥100000 0.47 1.6 (1.28, 2) \<0.001 BMI  18.5--24 Ref --- ---  \<18.5 0.3 1.4 (1, 1.82) 0.048  \>24 −0.145 0.9 (0.66, 1.13) 0.28 Infection routes  Others Ref --- ---  Sexual −0.246 0.78 (0.6, 1.11) 0.17 HCV-coinfection  No Ref --- ---  Yes 0.51 1.66 (0.85, 3.22) 0.14 Note: WHO, World Health Organization; BMI, body mass index; Ref, reference. ###### The proportion of cytopenias at baseline and 6, 12, and 24 months after HAART. Cytopenia Baseline At 6 months of HAART At 12 months of HAART At 24 months of HAART ------------------ ------------- ---------------------- ----------------------- ----------------------- Neutropenia 285 (46.6%) 97 (18%) 92 (15.1%) 69 (11.3%) Anemia 349 (57.1%) 91 (14.9%) 56 (9.2%) 48 (7.9.%) Thrombocytopenia 118 (19.3%) 27 (4.4%) 26 (4.3%) 15 (2.5%) Unicytopenia 483 (79.1%) 155 (25.4%) 133 (21.8%) 108 (17.7%) Bicytopenia 111 (18.2%) 21 (3.4%) 16 (2.6%) 12 (2%) Pancytopenia 16 (2.6%) 6 (1%) 3 (0.5%) 1 (0.2%) ###### The risk factors of cytopenia with persistent cytopenia after 24 months HAART.   df Univariate analysis, OR (95% CI) *p* value Multivariate analysis, OR (95% CI) *p* value -------------------------------------------- ---- ---------------------------------- ----------- ------------------------------------ ----------- AZT 1          No   Ref --- Ref ---  Yes   4.89 (1.49, 16.07) 0.009 3.71 (1.16, 11.9) 0.009 Sex 1          Male   Ref --- Ref ---  Female   5.42 (2.35, 12.5) \<0.001 5.54 (2.47, 12.42) \<0.001 HBV-coinfection 1          No   Ref --- Ref ---  Yes   3.8 (1.05, 13.95) 0.042 4.84 (1.47, 15.91) 0.01 CD4^+^T-cell counts at baseline 1          ≥200   Ref --- Ref ---  \<200   2.47 (1.04, 5.85) 0.04 2.42 (1.2, 4.86) 0.013 CD4^+^ T-cell counts at 24 months of HAART 1          ≥200   Ref ---      \<200   0.58 (0.23, 1.48) 0.257     WHO stage at baseline 3   0.4      I   Ref ---      II   1.13 (0.51, 2.5) 0.77      III   4.37 (0.93, 20.54) 0.124      IV   0.72 (0.12, 4.21) 0.72     Age 1          \<40   Ref ---      ≥40   0.97 (0.44, 2.15) 0.93     D4T 1          No   Ref ---      Yes   1.4 (0.72, 2.73) 0.32     HCV-coinfection 1          No   Ref ---      Yes   1.01 (0.1, 9.96) 0.993     Note: df, degree of freedom; WHO, World Health Organization; BMI, body mass index; Ref, reference; D4T, stavudine; AZT, zidovudine. [^1]: Academic Editor: Surender Khurana

Introduction {#sec1-1} ============ Intraocular medulloepithelioma arises from the primitive medullary epithelium, a derivative of the inner layer of the optic cup.\[[@ref1][@ref2][@ref3][@ref4][@ref5]\] This tumor most commonly appears as a white, gray, or yellow-colored ciliary body tumor.\[[@ref2]\] In rare instances, it can also originate from the retina or optic nerve.\[[@ref3]\] This childhood tumor is diagnosed at a median age of five years.\[[@ref2]\] Histologically, it appears as a network of sheets and cords of neuroepithelial cells.\[[@ref2][@ref4]\] The growth of medulloepithelioma is slow and it is locally invasive.\[[@ref1][@ref2]\] Herein, we report a case of malignant teratoid medulloepithelioma presenting with a triad of leukocoria, lens changes, and a white cystic ciliary body mass in a young child. Case Report {#sec1-2} =========== A 22-month-old Caucasian female twin presented with leukocoria and poor vision in the left eye (OS). Birth and family history were unremarkable. Visual acuity was fix and follow in the right eye (OD) and no fix or follow OS. Examination under anesthesia revealed normal findings OD and a mass in OS. Superotemporally OS, there was posterior synechiae of the iris overlying a gray-white ciliary body mass with slight sub-luxation of the cataractous lens and subtle lens notch \[[Figure 1a](#F1){ref-type="fig"}\]. There was no evidence of iris neovascularization or glaucoma. Intraocular pressures were 22 mm Hg OD and 18 mm Hg OS. There was no view of the fundus OS secondary to dense cataract. Transillumination disclosed a focal 12 mm round ciliary body shadow superotemporally abutting the limbus \[[Figure 1b](#F1){ref-type="fig"}\]. Ocular ultrasonography and ultrasound biomicroscopy displayed a solid, dome-shaped, 10 mm thick ciliary body mass with numerous intratumoral cysts \[[Figure 1c](#F1){ref-type="fig"}\] and areas of calcification. There was no evidence of extrascleral extension. Based on the clinical triad of leukocoria, lens changes, and a white cystic ciliary body mass in a young child, ultrasonographic, and transillumination features, the lesion was diagnosed as a non-pigmented ciliary epithelial medulloepithelioma. {#F1} After informed consent, enucleation was performed. Gross examination of the sectioned globe revealed a white mass arising from the ciliary body with multiple cystic areas. Patent hyaloid artery inserting into Bergmeister papillae over the optic disc was also noted \[[Figure 1d](#F1){ref-type="fig"}\]. Histopathological examination revealed interlinking and interweaving bands and cords of neuroepithelial cells with scant cytoplasm in the bulk of the tumor \[[Figure 1e](#F1){ref-type="fig"}\]. There were prominent foci of hyaline cartilage, an area of calcification, scattered rosettes, a few foci of sheet-like proliferation of neuroepithelial cells resembling retinoblastoma, and no extrascleral component \[[Figure 1f](#F1){ref-type="fig"}\]. The diagnosis of malignant teratoid medulloepithelioma of the non-pigmented ciliary epithelium was confirmed. There was no evidence of tumor recurrence or systemic metastasis at three years follow-up. Discussion {#sec1-3} ========== Intraocular medulloepithelioma arises from the primitive medullary epithelium, a derivative of the inner layer of the optic cup.\[[@ref1][@ref2][@ref3][@ref4][@ref5]\] This tumor most commonly appears as a white, gray, or yellow-colored ciliary body tumor.\[[@ref2]\] In rare instances, it can also originate from the retina or optic nerve.\[[@ref3]\] This childhood tumor is diagnosed at a median age of five years.\[[@ref2]\] In a clinicopathologic study of 56 cases of intraocular medulloepithelioma by Broughton and Zimmerman, poor vision (41%) and pain (30%) were the predominant symptoms.\[[@ref2]\] Leukocoria was the presenting sign in 18% of the cases.\[[@ref2]\] The most common clinical signs included cyst or mass in iris, anterior chamber or ciliary body (56%), glaucoma (48%), and cataract (26%).\[[@ref2]\] Many of these salient features were observed in our patient. Glaucoma with medulloepithelioma is classically due to iris neovascularization. Broughton and Zimmerman found that 61% (11/18) of eyes with secondary glaucoma displayed histopathologic evidence of iris neovascularization.\[[@ref2]\] Iris neovascularization and/or neovascular glaucoma in a child with a normal fundus examination is highly suggestive of an occult ciliary body medulloepithelioma.\[[@ref1][@ref5]\] Shields and co-workers studied 10 patients with ciliary body medulloepithelioma and noted the following lenticular changes: lens notch or coloboma (90%), lens subluxation (70%), cataract (60%), and retrolental neoplastic cyclitic membrane (60%).\[[@ref5]\] They emphasized that lens coloboma could be the earliest clinical sign of medulloepithelioma, and, furthermore, any child with an unexplained retrolenticular cyclitic membrane could be harboring an occult ciliary body medulloepithelioma.\[[@ref5]\] Intratumor cysts, another important feature suggestive of medulloepithelioma, is found in 60% of cases and are best distinguished on ultrasonography.\[[@ref5]\] An interesting finding with medulloepithelioma is the presence of persistence hyperplasia of the primary vitreous (PHPV) in nearly 20% of eyes. In our case, the persistent hyaloid artery was evidence of PHPV. Based on histopathological examination, medulloepithelioma can be classified as non-teratoid or teratoid, and benign or malignant.\[[@ref2]\] The non-teratoid medulloepithelioma results from the proliferation of primitive medullary epithelium.\[[@ref2]\] The teratoid medulloepithelioma shows heteroplastic elements of hyaline cartilage, rhabdomyoblasts, undifferentiated mesenchymal cells, or neuroglial cells in addition to the proliferation of the medullary epithelium.\[[@ref1][@ref2]\] Histopathologic criteria for malignancy include one or more of the following: Presence of areas composed of poorly differentiated neuroblastic cells resembling those of retinoblastoma with or without rosettes, sarcomatous areas resembling chondrosarcoma, rhabdomyosarcoma, or embryonal sarcoma, greater pleomorphism or mitotic activity, and invasion of uvea, cornea, and sclera with or without extrascleral extension.\[[@ref2]\] Based on these features, approximately 66% of medulloepitheliomas show malignant changes, but rarely do they metastasize.\[[@ref2]\] The treatment of choice for medulloepithelioma is enucleation.\[[@ref1][@ref2][@ref3][@ref5]\] Medulloepithelioma shows mild response to standard dose radiotherapy, but plaque radiotherapy with a dose of 4000 cGy or above remains an alternative for smaller tumors.\[[@ref2][@ref5]\] Local resection by partial lamellar sclerouvectomy can be considered for small circumscribed tumors (\<3 clock hours) but is associated with possible tumor recurrence.\[[@ref1][@ref2][@ref3][@ref5]\] In a study by Broughton and Zimmerman, 80% (8/10) of the patients displayed tumor recurrence after iridocyclectomy, ultimately requiring enucleation. In a study of ten patients by Shields and coworkers, four patients had enucleation as primary treatment and six patients were managed by local resection. All six patients that were managed by local resection (iridocyclectomy) as primary treatment developed tumor recurrence, requiring additional treatment of enucleation in four, exenteration in one, and cryotherapy in one patient.\[[@ref5]\] Intraocular medulloepithelioma with extraocular extension can metastasize to regional lymph nodes and distant sites and is associated with low mortality.\[[@ref1][@ref2]\] Extraocular extension is a key prognostic factor for tumor-related death.\[[@ref2][@ref3][@ref5]\] Death occurs mostly in cases with orbit involvement or intracranial extension.\[[@ref1][@ref2][@ref5]\] **Source of Support:** Eye Tumor Research Foundation, Philadelphia, PA (CLS). The funders had no role in the design and conduct of the study, in the collection, analysis and interpretation of the data, and in the preparation, review or approval of the manuscript. Carol L. Shields, M.D. has had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis **Conflict of Interest:** None declared.

Introduction ============ A novel coronavirus (SARS-CoV-2) that can infect humans and cause severe respiratory tract illness was first identified in Wuhan, China in December of 2019 ([@deaa137-B19]). Since this virus emerged, it has spread throughout the world, with the first case in the USA reported on 19 January 2020 ([@deaa137-B9]). The virus has since spread to all 50 states, resulting in shortages of life-saving medical supplies and equipment. This has led to unprecedented changes to the healthcare system, including limitations on non-urgent patient visits and procedures in order to preserve resources needed to combat the virus and slow its spread ([@deaa137-B15]; [@deaa137-B1]; [@deaa137-B8]; [@deaa137-B12]). In response to the SARS-CoV-2 pandemic, the American Society for Reproductive Medicine (ASRM) convened a task force to provide interim guidance for patient management (Coronavirus/COVID-19 TaskForce of the American Society for Reproductive Medicine). One of the key recommendations was to 'suspend initiation of new treatment cycles, including ovulation induction, intrauterine inseminations, in vitro fertilization including retrievals and frozen embryo transfers, as well as non-urgent gamete cryopreservation'. In addition, a joint statement from the ASRM, the American College of Obstetricians and Gynecologists, the American Association of Gynecologic Laparoscopists, the American Urogynecologic Society, the Society of Family Planning, the Society of Gynecologic Oncology, the Society of Gynecologic Surgeons and the Society for Maternal-Fetal Medicine was released that endorses the suspension of medically indicated procedures except in cases where a delay in treatment would negatively affect the health and safety of the patient ([@deaa137-B5]). There is disagreement among reproductive endocrinologists about whether delaying IVF treatment for a few months can negatively affect patient outcomes. In particular, some clinicians are concerned that women with diminished ovarian reserve may experience worse pregnancy outcomes if their treatment is delayed during this pandemic. Since estimates of peak infections vary depending on the adoption of and compliance with measures used to combat spread, such as stay-at-home orders and social distancing, estimates of infertility treatment delays range from weeks to months. Currently, there are no data to indicate whether patients are negatively affected by delays in infertility treatment for up to 6 months. This study was performed to determine whether women with diminished ovarian reserve have worse pregnancy outcomes when their treatment is delayed up to 180 days compared to women who undergo immediate infertility treatment. Materials and methods ===================== Study population and design --------------------------- This was a retrospective cohort study of women at our institution who underwent their first IVF cycle resulting in an oocyte retrieval with a planned fresh embryo transfer from 1 January 2012 to 31 December 2018. For each patient, all subsequent cycles that occurred in this time range were excluded. Since all cycle outcomes evaluated were from fresh embryo transfers, no genetically tested embryos were included in this study. Data were collected from our hospital electronic medical record system. IVF cycle and embryology data are prospectively entered into the electronic medical chart. Key data points were verified in the electronic medical records. This study was approved by the institutional review board at Weill Cornell Medical College. Definition of study groups -------------------------- All patients with diminished ovarian reserve, defined as an anti-Müllerian hormone (AMH) level of \<1.1 ng/ml, at the time of initial presentation to our office were included in this study. At our center, the AMH level was determined using the AMH Gen II ELISA (Beckman Coulter Inc., Brea, CA, USA) up until 10 May 2016, at which time we transitioned to using the Access 2 AMH assay (Beckman Coulter Inc., Brea, CA, USA). A minority of patients had their AMH level processed at an outside laboratory (14.6%) for which information about assay type was not available. Patients were stratified by whether they initiated an IVF cycle immediately after their first visit or whether their treatment was delayed. At our center, once patients are given a full evaluation and a treatment recommendation is made, they can start their IVF cycle on a timeline of their choice. Some patients begin their IVF cycle as soon as possible after their initial visit, while others may take longer to complete their evaluation or schedule their treatment cycle. In our cohort, immediate treatment was defined as initiating their first IVF cycle within 90 days of their first patient visit, and delayed treatment was defined as initiating their first IVF cycle between 91 and 180 days after their first patient visit. Clinical protocols ------------------ Controlled ovarian stimulation, trigger timing and dose, oocyte retrieval, embryo culture, embryo transfer and cryopreservation of supernumerary oocytes and/or embryos were performed per the standard protocols at our institution. Stimulation protocols utilized GnRH antagonists or GnRH agonist flare protocols ([@deaa137-B14]; [@deaa137-B2]). Letrozole or clomiphene citrate was added to some GnRH antagonist protocols at the treating physician's discretion ([@deaa137-B16]; [@deaa137-B18]). Patients who received priming in the mid-luteal phase of the cycle preceding the ovarian stimulation cycle were treated with either 0.1 mg estradiol patches or oral contraceptive pills ([@deaa137-B6]). Once two leading follicles had reached 17 mm in size (20 mm for letrozole or clomiphene-based protocols), oocyte maturation was triggered with either an hCG (Novarel \[Ferring Pharmaceuticals Inc., Parsippany, NJ, USA\] or Pregnyl \[Merck, Whitehouse Station, NJ, USA\]) trigger or a dual trigger with hCG and GnRH agonist (leuprolide acetate) according to a sliding-scale regimen for hCG dose ([@deaa137-B11], [@deaa137-B10]). Oocyte retrieval was performed transvaginally under ultrasound guidance ∼35--37 h after the trigger was administered. The method of insemination, embryo evaluation and day of embryo transfer was determined based on internal protocols, patient history, embryo numbers and embryo development assessed using morphokinetics. Luteal support was provided with 50 mg daily i.m. progesterone in oil injection beginning the day after oocyte retrieval. Patients who received a dual trigger with \<3300 units of hCG also received two 0.1 mg estradiol patches every other day. Luteal support was continued until documentation of a fetal heartbeat identified around 7 weeks of gestation. Demographics and outcomes ------------------------- Key demographic variables were collected ([Table I](#deaa137-T1){ref-type="table"}). Clinical pregnancy was defined as the presence of at least one intrauterine gestational sac observed on ultrasound. Spontaneous abortion was defined as a failed pregnancy after the observation of at least a gestational sac on ultrasound. Live birth was defined as delivery of a live-born infant at ≥24 weeks of gestation. ###### Demographic characteristics for immediate and delayed treatment groups. Characteristics Immediate treatment (1--90 days), *n* = 1115 Delayed treatment (91--180 days), *n* = 675 --------------------------------------------- ---------------------------------------------- --------------------------------------------- Time from initial visit to IVF start (days) 50.5 ± 21.9 128.8 ± 25.9 Age at IVF start (years) 39.1 ± 4.4 38.9 ± 4.3 BMI (kg/m^2^) 24.8 ±5.3 24.3 ± 4.7 Race  Caucasian 538 (48.3%) 342 (50.7%)  Asian 137 (12.3%) 100 (14.8%)  Black 43 (3.9%) 21 (3.1%)  Other/declined 397 (35.6%) 212 (31.4%) AMH (ng/ml) 0.56 ± 0.29 0.57 ± 0.29 Antral follicle count  0--5 520 (46.6%) 317 (47.0%)  6--10 494 (44.3%) 301 (44.6%)  11--15 79 (7.1%) 45 (6.7%)  ≥16 22 (2.0%) 12 (1.8%) Prior IVF cycles at outside institutions 1.3 ± 2.1 0.8 ± 1.6 Stimulation protocol  Gonadotrophin/antagonist 847 (76.0%) 504 (74.7%)  Gonadotrophin/agonist flare 145 (13.0%) 105 (15.6%)  Gonadotrophin+CC or letrozole/antagonist 115 (10.3%) 56 (8.3%)  Other 8 (0.7%) 10 (1.5%) Day of embryo transfer  None 133 (11.9%) 69 (10.2%)  Day 3 913 (81.9%) 555 (82.2%)  Day 5 69 (6.2%) 51 (7.6%) Number of embryos transferred 2.0 ± 1.3 2.1 ± 1.2 Oocytes retrieved 6.3 ± 3.9 6.6 ± 4.4 Oocytes mature 4.9 ± 3.4 5.3 ± 3.7 Oocytes fertilized 3.5 ± 2.8 3.7 ± 3.1 Data are mean ± SD or *n* (%). AMH, anti-Müllerian hormone; CC, clomiphene citrate. Statistical analysis -------------------- The primary outcome of this study was live birth. The secondary outcomes included implantation, biochemical and miscarriage rates. The size of our cohort is powered to detect a 5.7% difference in live birth per cycle with a 5% level of significance and 80% power. Logistic regression analysis was used to estimate the odds ratio (OR) with a 95% CI among the study groups for pregnancy outcomes listed in [Tables II](#deaa137-T2){ref-type="table"}, [III](#deaa137-T3){ref-type="table"} and [IV](#deaa137-T4){ref-type="table"}. This analysis was adjusted *a priori* for patient age, and all pregnancy outcomes were additionally adjusted for the number of embryos transferred. A subgroup analysis of more severe forms of diminished ovarian reserve was performed to evaluate outcomes for patients with an AMH \<0.5 ng/ml ([Table III](#deaa137-T3){ref-type="table"}) and for patients \>40 years old with an AMH \<1.1 ng/ml (Bologna criteria for diminished ovarian reserve; [Table IV](#deaa137-T4){ref-type="table"}) ([@deaa137-B7]). Statistical analyses were performed using StataSE 16 (StataCorp LLC, College Station, TX, USA). ###### The association between time to treatment and IVF treatment outcomes. Outcome Immediate treatment (1--90 days), *n* = 1115 Delayed treatment (91--180 days), *n* = 675 --------------------------------------------------------------------------- ---------------------------------------------- --------------------------------------------- No transfer\* 133 (11.9%) 69 (10.2%) 1.00 0.84 (Ref) (0.62, 1.15) Pregnancy rate among all IVF cycles 385 (34.5%) 264 (39.1%) 1.00 1.23 (Ref) (0.99, 1.51) Live birth rate among all IVF cycles[^a^](#tblfn5){ref-type="table-fn"} 235 (21.1%) 155 (23.0%) 1.00 1.11 (Ref) (0.88, 1.42) If embryo transfer (*n* = 982) (*n* = 606)  Pregnancy rate after embryo transfer 385 (39.2%) 264 (43.6%) 1.00 1.20 (Ref) (0.97, 1.48)  Live birth rate after embryo transfer[^a^](#tblfn5){ref-type="table-fn"} 235 (23.9%) 155 (25.6%) 1.00 1.08 (Ref) (0.85, 1.38) If clinically pregnant[^b^](#tblfn6){ref-type="table-fn"} (*n* = 385) (*n* = 264)  SAB[^c^](#tblfn7){ref-type="table-fn"} 66 (17.1%) 43 (16.3%) 1.00 0.96 (Ref) (0.62, 1.48)  Live birth[^a^](#tblfn5){ref-type="table-fn"} 235 (61.0%) 155 (58.7%) 1.00 0.91 (Ref) (0.65, 1.26) Data are *n* (%) with OR (95% CI). Logistic regression models adjusted *a priori* for age and number of embryos transferred to estimate the OR of pregnancy outcomes. Adjusted for age only. The reason for no transfer was due to unplanned upfront cryopreservation in six patients in the immediate treatment group and in six patients in the delayed treatment group. The reason for no transfer in all other patients was due to a lack of oocytes, sperm or embryo development. Live birth was defined as delivery at ≥24 weeks of gestational age. Clinical pregnancy was defined as the visualization of at least one gestational sac on ultrasound. Spontaneous abortion (SAB) was defined as a failed pregnancy after the observation of at least one gestational sac on ultrasound. ###### The association between time to treatment and IVF treatment outcomes in patients with AMH \<0.5 ng/ml. Outcome Immediate treatment (1--90 days), *n* = 506 Delayed treatment (91--180 days), *n* = 279 ---------------------------------------------------------------------------- --------------------------------------------- --------------------------------------------- No transfer[\*](#tblfn9){ref-type="table-fn"} 76 (15.0%) 38 (13.6%) 1.00 0.90 (Ref) (0.59, 1.37)  Pregnancy rate among all IVF cycles 154 (30.4%) 86 (30.8%) 1.00 1.01 (Ref) (0.72, 1.41)  Live birth rate among all IVF cycles[^a^](#tblfn10){ref-type="table-fn"} 81 (16.0%) 46 (16.5%) 1.00 1.02 (Ref) (0.67, 1.54) If embryo transfer (*n* = 430) (*n* = 241)  Pregnancy rate after embryo transfer 154 (35.8%) 86 (35.7%) 1.00 0.99 (Ref) (0.70, 1.39)  Live birth rate after embryo transfer[^a^](#tblfn10){ref-type="table-fn"} 81 (18.8%) 46 (19.1%) 1.00 0.99 (Ref) (0.65, 1.51) If clinically pregnant[^b^](#tblfn11){ref-type="table-fn"} (*n* = 154) (*n* = 86)  SAB[^c^](#tblfn12){ref-type="table-fn"} 35 (22.7%) 18 (20.9%) 1.00 0.97 (Ref) (0.50, 1.89)  Live birth[^a^](#tblfn10){ref-type="table-fn"} 81 (52.6%) 46 (53.5%) 1.00 0.99 (Ref) (0.57, 1.72) Data are *n* (%) with OR (95% CI). Logistic regression models adjusted *a priori* for age and number of embryos transferred to estimate the OR of pregnancy outcomes. Adjusted for age only. Live birth was defined as delivery at ≥24 weeks of gestational age. Clinical pregnancy was defined as the visualization of at least one gestational sac on ultrasound. SAB was defined as a failed pregnancy after the observation of at least one gestational sac on ultrasound. ###### The association between time to treatment and IVF treatment outcomes in patients \>40 years old. Outcome Immediate treatment (1--90 days), *n* = 524 Delayed treatment (91--180 days), *n* = 305 ---------------------------------------------------------------------------- --------------------------------------------- --------------------------------------------- No transfer[\*](#tblfn14){ref-type="table-fn"} 60 (11.5%) 39 (12.8%) 1.00 1.18 (Ref) (0.76, 1.83) Pregnancy rate among all IVF cycles 135 (25.8%) 85 (27.9%) 1.00 1.11 (Ref) (0.79, 1.55) Live birth rate among all IVF cycles[^a^](#tblfn15){ref-type="table-fn"} 57 (10.9%) 39 (12.8%) 1.00 1.19 (Ref) (0.76, 1.87) If embryo transfer (*n* = 464) (*n* = 266)  Pregnancy rate after embryo transfer 135 (29.1%) 85 (32.0%) 1.00 1.13 (Ref) (0.81, 1.59)  Live birth rate after embryo transfer[^a^](#tblfn15){ref-type="table-fn"} 57 (12.3%) 39 (14.7%) 1.00 1.21 (Ref) (0.77, 1.91) If clinically pregnant[^b^](#tblfn16){ref-type="table-fn"} (*n* = 135) (*n* = 85)  SAB[^c^](#tblfn17){ref-type="table-fn"} 43 (31.9%) 16 (18.8%) 1.00 0.51 (Ref) (0.26, 0.98)  Live birth[^a^](#tblfn15){ref-type="table-fn"} 57 (42.2%) 39 (45.9%) 1.00 1.10 (Ref) (0.63, 1.93) Data are *n* (%) with OR (95% CI). Logistic regression models adjusted *a priori* for age and number of embryos transferred to estimate the OR of pregnancy outcomes. Adjusted for age only. Live birth was defined as delivery at ≥24 weeks of gestational age. Clinical pregnancy was defined as the visualization of at least one gestational sac on ultrasound. SAB was defined as a failed pregnancy after the observation of at least one gestational sac on ultrasound. Results ======= This study consisted of 1790 patients with diminished ovarian reserve. Overall, 1115 patients initiated an IVF cycle within 90 days of initial evaluation (immediate), and 675 patients initiated an IVF cycle between 91 and 180 days of initial evaluation (delayed). Among the study cohort, 785 patients (43.9%) had an AMH \<0.5 ng/ml, and 829 patients (46.3%) were \>40 years old and had an AMH \<1.1 ng/ml. Demographic characteristics are shown in [Table I](#deaa137-T1){ref-type="table"}. The mean ± SD and median (interquartile range) number of days from initial presentation to the start of an IVF cycle was 50.5 ± 21.9 and 48 (33--69) days in the immediate treatment group and 128.8 ± 25.9 and 125 (107--149) in the delayed treatment group, respectively. [Figure 1](#deaa137-F1){ref-type="fig"} displays the length of duration, in 30-day increments, from the initial visit to IVF cycle start for the study cohort. A plot of each study year that displays the length of duration from the initial visit to IVF cycle start for both groups is shown in [Fig. 2](#deaa137-F2){ref-type="fig"}. The mean AMH was 0.56 ± 0.29 ng/ml in the immediate treatment group and 0.57 ± 0.29 in the delayed treatment group. The number of oocytes retrieved was 6.3 ± 3.9 (immediate) and 6.6 ± 4.4 (delayed), and the number of mature oocytes was 4.9 ± 3.4 (immediate) and 5.3 ± 3.7 (delayed). The number of fertilized oocytes was 3.5 ± 2.8 (immediate) and 3.7 ± 3.1 (delayed). {#deaa137-F1} {#deaa137-F2} Among the 1790 patients, an embryo transfer was not performed in 133 cycles (11.9%) in the immediate treatment group and in 69 cycles (10.2%) in the delayed treatment group. There were six cycles per group that resulted in unplanned cryopreservation of embryos. Otherwise, the reason for no embryo transfer was an unexpected lack of oocytes, sperm or embryo development. Among all patients with an AMH \<1.1 ng/ml, the pregnancy rate was comparable between the immediate and delayed treatment groups when evaluated among all patients in the cohort (34.5% versus 39.1%; OR 1.23, 95% CI 0.99--1.51) and when evaluated among only patients who had an embryo transfer (39.2% versus 43.6%; OR 1.20, 95% CI 0.97--1.48). The live birth rate was also comparable between the immediate and delayed treatment groups when evaluated among all patients in the cohort (21.1% versus 23.0%; OR 1.11, 95% CI 0.88--1.42) and when evaluated among only patients who had an embryo transfer (23.9% versus 25.6%; OR 1.08, 95% CI 0.85--1.38). For women who achieved a pregnancy, the live birth rate was also similar between the immediate (61.0%) and delayed treatment groups (58.7%) (OR 0.91, 95% CI 0.65--1.26). Additionally, there were no differences between the proportion of biochemical pregnancies or miscarriages that were observed in pregnant patients ([Table II](#deaa137-T2){ref-type="table"}). Subgroup analysis for women with an AMH \<0.5 ng/ml is displayed in [Table III](#deaa137-T3){ref-type="table"}. The pregnancy rate was comparable between the immediate and delayed treatment groups when evaluated among all patients in the cohort (30.4% versus 30.8%; OR 1.01, 95% CI 0.72--1.41) and when evaluated among only patients who had an embryo transfer (35.8% versus 35.7%; OR 0.99, 95% CI 0.70--1.39). The live birth rate was also comparable between the immediate and delayed treatment groups when evaluated among all patients in the cohort (16.0% versus 16.5%; OR 1.02, 95% CI 0.67--1.54) and when evaluated among only patients who had an embryo transfer (18.8% versus 19.1%; OR 0.99, 95% CI 0.65--1.51). For women who achieved a pregnancy, the live birth rate was also similar between the immediate (52.6%) and delayed groups (53.5%) (OR 0.99, 95% CI 0.57--1.72). Additionally, there were no differences between the proportion of biochemical pregnancies or miscarriages that were observed in pregnant patients. Subgroup analysis for women who met the Bologna criteria for diminished ovarian reserve by age \>40 years old and an AMH \<1.1 ng/ml is displayed in [Table IV](#deaa137-T4){ref-type="table"}. The pregnancy rate was comparable between the immediate and delayed treatment groups when evaluated among all patients in the cohort (25.8% versus 27.9%; OR 1.11, 95% CI 0.79--1.55) and when evaluated among only patients who had an embryo transfer (29.1% versus 32.0%; OR 1.13, 95% CI 0.81--1.59). The live birth rate was also comparable between the immediate and delayed treatment groups when evaluated among all patients in the cohort (10.9% versus 12.8%; OR 1.19, 95% CI 0.76--1.87) and when evaluated among only patients who had an embryo transfer (12.3% versus 14.7%; OR 1.21, 95% CI 0.77--1.91). For women who achieved a pregnancy, the live birth rate was also similar between the immediate (42.2%) and delayed treatment groups (45.9%) (OR 1.10, 95% CI 0.63--1.93). Additionally, the proportion of biochemical pregnancies was similar between the immediate (24.4%) and delayed treatment groups (34.1%) (OR 1.75, 95% CI 0.95--3.24). However, the proportion of pregnancies that resulted in a miscarriage was significantly higher for women in the immediate (31.9%) compared to the delayed treatment groups (18.8%) (OR 0.51, 95% CI 0.26--0.98). Discussion ========== The aim of this study was to identify whether a delay in initiating IVF treatment has a negative effect on outcomes for women with diminished ovarian reserve (AMH \<1.1 ng/ml). In our cohort of patients who initiated an IVF cycle within 6 months of their first consultation, there was no difference in the live birth rate among women who initiated their IVF cycle within 90 days of their first visit compared to those who did so 91--180 days after initial consultation. There was also no difference in the live birth rate for the immediate or delayed treatment groups for women with an AMH \<0.5 ng/ml or for women who met Bologna criteria for diminished ovarian reserve who were \>40 years old with an AMH \<1.1 ng/ml. In clinical practice, treatment delays can occur due to medical, logistical or financial reasons. In a more extreme scenario, treatment delays also can occur as the result of natural disasters or pandemics, at which time resources are often temporarily reallocated to manage the disaster at hand. The SARS-CoV-2 pandemic is one such example that prompted both the ASRM and ESHRE to recommend the suspension of new infertility treatment cycles, and that fertility patients should avoid becoming pregnant until more information about the virus is known ([@deaa137-B3]; [@deaa137-B4]). Regardless of the reason for treatment delay, both patients and providers express concern when infertility treatment cycles are unable to start in a timely manner. This is particularly true for patients with diminished ovarian reserve due to the shorter timeframe these patients have to achieve a pregnancy compared to age-matched women with normal ovarian reserve. It is well understood that over an interval of several years, women experience a gradual decline in ovarian reserve and fecundability ([@deaa137-B13]; [@deaa137-B17]). However, in infertile patients, the length of time that it takes for a clinically meaningful decline in ovarian reserve or the likelihood of a successful treatment outcome is not known. The results of our study are reassuring in that women with an AMH \<1.1 ng/ml were not observed to have a decline in pregnancy rate or live birth rate after a delay of up to 180 days in initiating IVF treatment. This observation remained true when pregnancy and live birth rates were assessed among all patients in the cohort, among only patients who had an embryo transfer and among only patients who achieved a pregnancy. These findings suggest that a short delay in treatment does not have a clinical effect on implantation or fetal and pregnancy development. A similar proportion in both groups did not have an embryo transfer due to no oocytes retrieved, no sperm available or poor embryo development. This finding suggests that embryo development is also not affected by a short delay in IVF treatment. Furthermore, for patients with an AMH \<0.5 ng/ml and for patients who were \>40 years old with an AMH \<1.1 ng/ml, there were no differences observed for pregnancy or live birth rate analyses. Since both of these subgroups may be considered particularly poor-prognosis patients, providers may feel there is an urgent need to start their treatment as soon as possible. While prioritizing treatment for all patients, including those with diminished ovarian reserve, is important in the normal clinical setting, it is reassuring that a delay in treatment of up to 180 days did not lead to a decline in pregnancy outcomes, even in patients who are among the highest risk for poor response to ovarian stimulation. The increased rate of miscarriages observed in the immediate treatment group in patients \>40 years old was an unexpected finding and warrants a more detailed evaluation in a larger patient cohort. In terms of the study objective, it is reassuring that this outcome was not observed in women in the delayed treatment group. Furthermore, the overall similar pregnancy and live birth rates between the two groups in these women support the overall findings of this study that a short delay in infertility treatment does not affect the goal outcomes of infertility treatment. We acknowledge the limitations of this study. There is the potential for selection bias with regard to the patients who started their IVF cycle within 90 days compared to 91--180 days after initial consultation. It is possible that for patients with severely diminished ovarian reserve, there may have been an urgency in starting their treatment. However, the mean AMH levels and the proportion in each antral follicle count category were comparable between the two groups, which partially mitigates this concern. In addition, we did not include patients who were seen for initial evaluation but did not progress to IVF treatment with oocyte retrieval; therefore, our results should only be applied to patients with diminished ovarian reserve who complete an IVF cycle. We were unable to determine the assay used in cases where the AMH level was processed at an outside laboratory, which may lead to some inter-assay variation between the reported AMH levels. Finally, since we excluded patients who started their IVF cycle greater than 180 days from their first visit, it is not known how such a delay in treatment affects pregnancy outcomes in IVF cycles. Conclusion ========== In summary, a delay in initiating IVF treatment in patients with diminished ovarian reserve (AMH \<1.1 ng/ml) up to 180 days from the initial visit does not affect pregnancy outcomes. This observation remains true for women who are considered the highest risk for poor response to ovarian stimulation (either having an AMH \<0.5 ng/ml or being \>40 years old with an AMH \<1.1 ng/ml). Overall, these results are reassuring to providers that when short-term treatment delays are deemed necessary for medical, logistic or financial reasons, treatment outcomes will not be affected. Additionally, these results are reassuring to patients who may feel anxious to begin their treatment and become frustrated when unexpected delays occur. Future studies should seek to identify the length of time over which clinical outcomes are affected by treatment delays, particularly to help counsel patients in whom a treatment delay of greater than 180 days is expected. We would like to thank Alexandra MacWade for her help in proofreading the manuscript. Authors' roles ============== P.A.R.: Participated in study design, execution, analysis planning, manuscript drafting and editing. P.B.: Participated in study design, analysis planning, manuscript drafting and editing. Z.R.: Participated in study design, analysis planning and manuscript editing. G.L.S.: Participated in study design, analysis planning and manuscript editing. Funding ======= No financial support, funding or services were obtained for this study. Conflict of interest ==================== The authors do not have any conflict of interest disclosures.

Generative models of neural circuits may help to create a link between neural mechanisms and observable data. We propose a model of rat\'s cortex using a neural field model containing biologically plausible anatomical connections from tractography based on dwMRI data and from the neural morphological database NeuroMorpho \[[@B1]\]. There are three principal types of anatomical connections in the cortex: Local, long-range and distal connections \[[@B2]\]. For specifying local connections we use neural morphologies from \[[@B1]\]. We consider each voxel in the model as a neural mass and distribute randomly drawn neurons from the database therein. After that we use bootstrap methods to determine the total number and variability of synaptic contacts. For the distal connectivity we estimated the degree of anatomical connectedness using white matter tractography on the basis of diffusion weighted MRI \[[@B3]\]. Our neural field consists of 5 layers. For each layer we assume three different neural masses: pyramidal cells, excitatory and inhibitory interneurons \[[@B4]\]. The mutual interactions between neural masses will be described by a system of integral differential equations: where **V** is the state vector (mean membrane potentials), **T** the time delays in the dendritic arbors, **S** the sigmoidal output function, **W** the connection coefficients between the neural elements, **I** the input, and t^(d)^ is the distance dependent time delays. Figure [1A](#F1){ref-type="fig"} shows the estimated spatial dependency of local connectivity, which is in accordance with anatomical observations \[[@B2]\]. The distal connectivity map of 28 regions is displayed in Fig. [1B](#F1){ref-type="fig"}. An example map of simulated activity on the cortex in response to a stimulus to somatosensory cortex is illustrated in Fig. [1C](#F1){ref-type="fig"}. {#F1} To summarize, we developed a method for estimating the local connectivity and constructed a neural field model of the entire cortex enriched by estimated local and distal connectivities. With this model we are able to simulate spatio-temporal activity patterns. This is a first step for a comprehensive dynamic brain model and thereby for understanding complex brain processes.

1. Introduction {#sec1-ijerph-17-00018} =============== Air pollution is the largest single environmental risk to health, responsible for an estimated 7 million premature deaths every year globally and about 556,000 in the European Region. As people spend a considerable amount of time indoor, either at work or at home, indoor air quality (IAQ) plays a significant part in their general state of health, wellbeing, and human performances \[[@B1-ijerph-17-00018]\]. This is particularly true for children, elderly people, and other vulnerable groups. Household indoor air pollution is one of the goals of the 2030 Agenda for Sustainable Development (<https://ec.europa.eu/environment/sustainable-development/SDGs/index_en.htm>), and, in particular, the Sustainable Development Goals (SDGs) that are related to the topics of health, sustainable cities, industrialization, and mitigating the effects of climate change, are numerous. IAQ in household environments is therefore mandatory in private homes, schools, workplaces, etc. Inside houses, offices, etc., there are many indoor sources ranging from cooking to domestic works \[[@B2-ijerph-17-00018],[@B3-ijerph-17-00018]\]. In particular, house cleaning operations are one of the main sources since generate indoor pollutants in both gaseous and particle phases. Among the house cleaning operations, the use of vacuum cleaners could represent a potential source of health risk due to the enormous quantity of particles that are dispersed during its use. Due to the increase of allergic rhinitis, the utility of indoor environmental management deserves increasing attention \[[@B4-ijerph-17-00018]\] and the use of air filtration systems is spreading \[[@B5-ijerph-17-00018]\]. Aerosol particulate matter (PM) is among the main pollutants both outdoor and indoor. PM~10~ (particles with aerodynamic diameter less than 10 μm), PM~2.5~ (particles with aerodynamic diameter less than 2.5 μm), and PM1 (particles with aerodynamic diameter less than 1 μm) are the most harmful air pollutants together with ground-level ozone, nitrogen oxides, and sulfur oxides that are also of concern being the main precursors of secondary PM (particulate matter) in the atmosphere. Despite reductions in emissions of PM~10~, the majority (50--92%) of the urban population in monitored European countries was exposed to concentrations above the air quality goal. For example, levels of PM~2.5~ exceeded the goal established by WHO at 75% of stations in European Region in 2015, and exposure to PM reduces the life expectancy of every person by an average of almost 1 year, mostly because of the increased risk of cardiovascular and respiratory diseases, and lung cancer, with an enormous cost of the health care with a prevalence of diseases among children and adolescents. Exposure to air pollution has been associated with numerous health effects that range from minor upper respiratory irritation to chronic respiratory and heart disease, lung cancer, and acute respiratory infections \[[@B6-ijerph-17-00018]\]; in particular, exposure has been associated with respiratory infections, asthma and allergic symptoms, ear inflammation, and deficits in lung function, as well as cognitive impairment \[[@B6-ijerph-17-00018]\]. Exposure to PM during pregnancy has also been associated with adverse to birth outcomes, such as preterm births and low birthweights \[[@B7-ijerph-17-00018]\]. The negative effects of PM on human health are linked, as it is well known from the literature, not only to the particles' concentration but also to their chemical composition, which should be considered \[[@B6-ijerph-17-00018],[@B8-ijerph-17-00018]\] in order to highlight health relevant components, including the measurements of organic and elemental carbon \[[@B9-ijerph-17-00018]\]. A large numbers of studies have been carried out on this topic also focusing on sources identification \[[@B10-ijerph-17-00018],[@B11-ijerph-17-00018],[@B12-ijerph-17-00018]\], size-resolved measurements \[[@B13-ijerph-17-00018]\], and on specific markers investigation \[[@B8-ijerph-17-00018],[@B14-ijerph-17-00018],[@B15-ijerph-17-00018]\]. As far as indoor environments are concerned, there are no limits for air quality, even though we spend about 80% of our time indoors \[[@B1-ijerph-17-00018]\]. The U.S. Environmental Protection Agency (EPA) has shown that the quality of indoor air is 5--10 times worse than outdoor because of both penetration of pollutants from outdoor and, as stated before, because of the presence of specific indoor sources. In China, heavy pollution episodes are frequent and strongly influence indoor air quality \[[@B16-ijerph-17-00018]\]. Our homes are full of particles including also dust mites, which are one of the most common type of indoor allergens and the most important factor in causing asthma in children \[[@B17-ijerph-17-00018]\]. Due to poor IAQ, the use of air purification devices, like that tested in the present paper, is increasing worldwide \[[@B5-ijerph-17-00018]\], even though, ventilating and air conditioning systems are often employed in many countries to achieve an improvement of indoor air quality. In this study, a commercial household vacuum cleaner has been tested in order to verify its particle abatement capacity. The evaluation has been carried out using a portable optical particle counter instrument and simulating the working conditions typical of a household environment. Particle mass concentrations of PM~10~, PM~2.5~, PM~1~, and particle number concentration in 7 size-fractions between 0.3 and \>10 μm have been measured. 2. Materials and Methods {#sec2-ijerph-17-00018} ======================== In the present paper, a commercial vacuum cleaner that can also be used as air purification system has been tested. Traditional household vacuum cleaners for air purification generally have a filter or a dust bag as a barrier for dust collection. However, after use, collecting all the materials can stick to the surface of the filter or accumulate in the dust bag. The main drawback of this system is that dust can block or even damage the filter affecting the suction and the air flow rate. The device tested in the present study is a water-based cleaning system and it can be used both as vacuum cleaner and air purification system.It can effectively filter dust particles and micro-organisms. Since the separator produces a strong centrifugal force, dust, dirt, and thin allergens can be separated in water and no filters or dust bags are required. The maximum suction flow rate is 156 m^3^/h. The measurements were carried out by placing the device in parallel with a P-DustMonit (Contec, Milan, Italy) instrument inside a test chamber measuring 335 × 470 × 290 (H) cm (about 45 m^3^). The test chamber was ventilated with outdoor air before starting of the test, and, before proceeding with the start of the air purification device, it was expected that conditions were stabilized, i.e., that the trend of the curves relative to the three fractions of interest (PM~10~, PM~2.5~, and PM~1~) reached a plateau. The P-DustMonit unit is a portable complete system for continuous monitoring of particulate concentration in the air. It is based on the light scattering principle: a constant flow pump draws air in through a radial symmetric probe where each particle is hit with a laser. The energy reflected by each particle, proportional to its dimension, is measured by a high-velocity photodiode, which generates counting signals corresponding to the particle's dimensions. Instrument sample flow is 1 L/min, and the device is able to work in the temperature range −10--40 °C. The instrument allows: (1) measurement of the concentrations of the fine particulate expressed as PM~10~--PM~2.5~--PM~1~ in μg/m^3^ (in real time and at the same time); (2) measurement in μg/m^3^ (in real time and simultaneously) of the concentrations of the inhalable, thoracic, and breathable dust, as defined by current regulations; (3) measurement of the particle number in real time classifying them simultaneously up to 15 different size classes. During the experiment, the relative humidity (RH) and temperature were also registered by P-DustMonit unit---the temperature increased from about 22.5--24.1 °C while relative humidity (RH) decreased when air ventilation was performed. P-DustMonit unit can be assumed as a valid system to track air quality. Optical particle count systems, like that one employed in this work, are often used to quantify fine and ultra-fine particles in both indoor and outdoor environments \[[@B3-ijerph-17-00018],[@B18-ijerph-17-00018]\]. With regards to the intercomparison with other instruments, it is reported in the literature \[[@B18-ijerph-17-00018]\] that optical particle counters can be successfully used to analyze diurnal average trends in particle number and mass concentrations; furthermore, the high temporal resolution showed that optical detectors could be very useful for air quality applications and to investigate specific pollution events. However, especially, if they are used for evaluating mass concentrations (PM~1~, PM~2.5~, or PM~10~), it is worth to notice that it is necessary to take into account RH effects, even if the experimental conditions registered during our experiments were such that they were not a problem. 3. Results and Discussion {#sec3-ijerph-17-00018} ========================= In order to test the vacuum cleaner's efficiency as an air purification system, measurements have been carried out putting in parallel the device and the P-DustMonit unit inside a chamber without air exchange but not perfectly tight in order to simulate what can happen in a room of a house with the windows closed. The aim was to quantify both PM and particle sizes in different size-fractions. PM~10~ concentrations inside the measurement chamber, after stabilization of the environmental conditions, was approximately 20 μg/m^3^. The device was started at the maximum speed corresponding to a suction flow rate of 156 m^3^/h. In [Figure 1](#ijerph-17-00018-f001){ref-type="fig"}, it is clearly observable how PM concentrations (PM~10~, PM ~2.5~, and PM~1~) rapidly decrease when the instrument is switched on at 10:20 a.m. The concentration of the three size-fractions after about 40 min has been reduced considerably, and, in particular, PM~10~ was reduced from 14 to about 7 μg/m^3^, and PM~2.5~ and PM~1~ from 13 to about 6 μg/m^3^. At 10:40 a.m., a reduction of 50% of PM concentration has been achieved. It is also possible to hypothesize that by operating at the minimum speed (52 m^3^/h) in 2 h, it would be possible to achieve the same result that was achieved in 20 min. At 11:25 a.m., the instrument was turned off and the room was ventilated by opening the windows. From the graph, it can be seen how, in particular, PM10 concentration rises very quickly. Subsequently, at 11:38 a.m., after obtaining a stabilization of the concentrations, the system was again switched at the minimum speed (52 m^3^/h), and, after 20 min, it was observed that the concentration decreased from about 16 μg/m^3^ to about 12 μg/m^3^ with a reduction of 25%. Therefore, even when operating with the minimum speed, a significant decrease in PM is appreciated in a short time. Furthermore, it is worth noting that a further advantage of the air purification system tested here is its long-term performance, thanks to particle abatement in water that allows to avoid worsening of performances due to dust loading, which normally happens for air cleaners based on filtration \[[@B19-ijerph-17-00018],[@B20-ijerph-17-00018]\]. The measurement performed by P-DustMonit unit was carried out outdoor in the same time interval as indoor (i.e. between 10:20 and 11:40) in order to compare with the corresponding indoor PM values, and the outdoor recorded values showed some variations in PM concentrations mainly due to the influence of wind ([Figure 2](#ijerph-17-00018-f002){ref-type="fig"}). As normally expected, indoor values are lower if the typical indoor sources are not present (cleaning operation, cooking, etc.). As a consequence, before vacuum cleaner was working, initial PM concentrations were about 10 μg/m^3^ lower than the outdoor values. Nevertheless, it is worth to notice that in our indoor experiment, the finer fractions (PM ~2.5~ and PM~1~) are quite similar in concentration to PM10, while, in outdoor, the prevalent fraction is represented by PM~10~. It is worth noting that a lower PM~10~/PM~2.5~ (or PM~10~/PM~1~) ratio represents a greater health risk, with the finer fractions being more penetrating in the pulmonary alveoli. [Figure 3](#ijerph-17-00018-f003){ref-type="fig"}, [Figure 4](#ijerph-17-00018-f004){ref-type="fig"} and [Figure 5](#ijerph-17-00018-f005){ref-type="fig"} report the trends of the numerical concentrations of the particles within 7 size classes. The higher particles' indoor concentration (more than 60,000 particles/L, i.e., 60 particles/cm^3^) have been registered for the fraction formed by particles with diameter \>0.3 μm ([Figure 5](#ijerph-17-00018-f005){ref-type="fig"}). In outdoor, the concentration level for the same fraction is double (with a maximum of 17,000 particles/L). The values registered in outdoor ambient air are in good accordance with data from the environmental protection Agency (ARPA) of Lombardy Region (Northern Italy) for Milan area \[[@B21-ijerph-17-00018]\], where our measurements have been carried out. Our values are also in fair accordance with the results obtained in other studies for the same particles' dimensional range \[[@B3-ijerph-17-00018],[@B22-ijerph-17-00018]\]. It is also interesting to notice that if ultra-fine particles (with diameter \<0.1 μm) would be considered (this dimensional range cannot be measured with the instrument used in the present study, since P-DustMonit's lower particle limit is 0.3 μm, an enormous increase in the number would have been observed \[[@B23-ijerph-17-00018]\] reaching values of the order of particle thousands per cm^3^. Furthermore, it has been observed that UFP indoor concentrations are generally higher than outdoor ones because of the presence of numerous and intermittent indoor sources (cooking, cleaning operations, etc.) \[[@B3-ijerph-17-00018]\] with the consequence that people who spend most of the time at home have a high risk to be exposed to indoor pollutants in the finest fraction, i.e., the most dangerous ones. From the comparison among the particle number trends reported in [Figure 3](#ijerph-17-00018-f003){ref-type="fig"}, [Figure 4](#ijerph-17-00018-f004){ref-type="fig"} and [Figure 5](#ijerph-17-00018-f005){ref-type="fig"}, it can be observed how the vacuum cleaner seem to be more efficient in reducing fraction \> 0.3 µm. In order to make a more realistic comparison among the different size-fractions, within each size-fraction particle number was divided for the corresponding maximum value in that specific range. The results obtained are shown in [Figure 6](#ijerph-17-00018-f006){ref-type="fig"}a,b depending on the dimensions. In addition, in this case, it is evident how the vacuum cleaner is more effective in the reduction of particles with diameter \> 0.3 µm while when ventilation with outdoor air is performed, particles with diameter \> 3 µm show the strongest change while a less significant decrease is observable for the other fractions. Therefore, the vacuum cleaner tested in this study has proved to be particularly efficient in reducing the number of finest particles, which is the greatest concern from the health point of view, being the one able to deeply penetrate the respiratory system until reaching the pulmonary alveoli. So from a health point of view, since the finest particles are those that cause the most concern, their reduction in number is a very significant point to be achieved. 4. Conclusions {#sec4-ijerph-17-00018} ============== It has been demonstrated that the tested air cleaner can significantly reduce the indoor PM concentrations (PM~10~, PM~2.5~, and PM~1~) to values below WHO guideline level that correspond to 10 μg/m^3^. It is also worth noting that the particles that are more efficiently removed are those having the finest size, i.e., particles in the range 0.3--0.5 μm. PM being one of the priority causes that lead to the onset of diseases affecting the respiratory system, the use of such devices could contribute to significantly improving the household air quality. In fact, people that spend most of their time at homes or, in general, in indoor environments, may have a higher risk of acute or chronic exposure to typical indoor pollutants, and the possibility to use devices that allows a significant improvement of IAQ certainly contributes to improving health and wellbeing. Conceptualization, P.F. and A.M.; methodology, P.F., V.C. and L.F.; validation, P.F. and L.F.; formal analysis, P.F., V.C. and V.G.; resources, P.F. and A.M.; data curation, V.C.; writing---original draft preparation, P.F.; writing---review and editing, P.F., V.C.; supervision, P.F. and A.M.; project administration, P.F. and A.M.; all authors have read and agree to the published version of the manuscript. This research received no external funding. The authors declare no conflict of interest. {#ijerph-17-00018-f001} {#ijerph-17-00018-f002} {#ijerph-17-00018-f003} {#ijerph-17-00018-f004} {#ijerph-17-00018-f005} {#ijerph-17-00018-f006}

1. Introduction {#sec1} =============== Alzheimer\'s disease (AD) is one of the utmost prevalent age-related neurodegenerative disorders affecting older people \[[@B1], [@B2]\]. The main risk for the development and progression of AD is age, and in people over the age of 65, the risk becomes double nearly every 5 years \[[@B3]--[@B5]\]. In 2015, it was estimated that around 44 million people were affected by this neurodegenerative disorder. However, this number is projected to become double by the year 2050 \[[@B6], [@B7]\]. It has been found that most of the AD patients (i.e., over 95%) have sporadic AD (SAD) or late-onset AD (LOAD), a multifactorial disease in which genetic predisposition and environmental factors have significant contribution in the pathology \[[@B8], [@B9]\]. On the other hand, less than 5% of the AD people is found to have either early-onset AD (EOAD) or familial AD (FAD). It has been found that the aforesaid forms of AD (i.e., EOAD and FAD) might take place because of the mutations in any of the presenilin-1 (*PSEN1*), presenilin-2 (*PSEN2*), and amyloid precursor protein (*APP*) genes \[[@B10]--[@B12]\]. The neuropathological characteristics of AD include an aberrant buildup of the amyloid beta (A*β*) peptide in amyloid plaques and aggregation of hyperphosphorylated tau in intracellular neurofibrillary tangles (NFTs) \[[@B13]--[@B16]\]. Furthermore, dystrophic neurites (DNs), neuropil threads, related astrogliosis, and microglial activation are found to often coexist \[[@B17], [@B18]\]. The downstream effects of these pathological mechanisms involve neurodegeneration with neuronal and synaptic loss, which can lead to macroscopic atrophy \[[@B19]\]. It has been suggested in various studies that AD-related alterations in the brain might start even 20 or more years before the symptoms even appear \[[@B20]--[@B23]\]. It has been found that when the initial alterations take place, the brain compensates for them and permits people to continue to function normally. However, when the neuronal damage continues to rise, the brain can no longer compensate for the alterations, and patients exhibit subtle cognitive impairment \[[@B24]\]. At a later stage, neuronal damage becomes so substantial that people exhibit clear cognitive impairment, including symptoms for instance loss of memory or confusion regarding the place or time \[[@B25], [@B26]\]. Currently, the available drugs only provide symptomatic treatment and do not have the proper ability to affect the advancement of the AD \[[@B27], [@B28]\]. Henceforth, it is essential to develop treatments that can slow, delay, or prevent the advancement of the disease and also target the AD-related pathological processes. Interestingly, it has been observed that if the onset of AD could be delayed by five years, the overall frequency of AD would be reduced by 40% \[[@B29]\]. Over the course of time, A*β* has been regarded as the key therapeutic target for AD \[[@B30]--[@B32]\]. Furthermore, a number of pharmaceutical/biopharmaceutical companies are trying to develop therapeutic compounds (i.e., as small molecules or immunotherapies) to reduce the accumulation of A*β* with possible positive action on AD pathology. In this review, we have focused on the role of APP-cleaving secretase-based novel molecules as therapeutic targets for AD. Furthermore, auspicious molecules targeting A*β* accumulation and phosphorylation signaling in APP processing are also presented. 2. Approved Anti-Alzheimer\'s Drugs {#sec2} =================================== Out of the five Food and Drug Administration (FDA) -approved drugs ([Figure 1](#fig1){ref-type="fig"}) for AD therapy, four of them are acetylcholinesterase inhibitors (AChEIs) and one is N-methyl-D-aspartic acid (NMDA) receptor antagonist \[[@B33], [@B34]\]. The AChEIs (i.e., galantamine, rivastigmine, donepezil, and tacrine) have been found to play a role in improving the cholinergic neurotransmission \[[@B34], [@B35]\]. Memantine is the only noncholinergic and NMDA receptor antagonist drug. Memantine plays roles in the restoration of the A*β*-stimulated calcium imbalance (i.e., intracellular buildup of Ca^2+^) and the associated reduction of neuronal death \[[@B36]\]. Unfortunately, memantine and currently approved cholinesterase inhibitors (ChEIs) only provide relief of the dementia symptoms, and these agents possess limited clinical effects \[[@B37], [@B38]\]. Over the last 20 years, a number of drug candidates have been developed with the aim of hitting different and novel targets of this disease. However, effective disease-modifying therapeutic agents are yet to be found \[[@B39]--[@B41]\]. Other than the cholinergic recovery, researchers are also searching for other AD targets, for instance, pathogenic A*β* and tau aggregates \[[@B42], [@B43]\] as well as dysregulated metal ions \[[@B44]\]. 3. Amyloid Precursor Protein Processing {#sec3} ======================================= APP has been found to be generated in large quantities in neurons and is metabolized very rapidly \[[@B45], [@B46]\]. In fact, APP processing is reliant on secretase enzymes (i.e. *α*-, *β*-, and *γ*-secretases), which yield products that are secreted into the extracellular space or which stay within or related to the cell. Amyloidogenic and nonamyloidogenic are the 2 pathways of APP processing \[[@B23]\] as shown in [Figure 2](#fig2){ref-type="fig"}. Three enzymes, which are members of the ADAM family (i.e., a disintegrin and metalloprotease), play a role in harboring the activity of *α*-secretase; these include ADAM17, ADAM10, and ADAM9 \[[@B47]\]. These enzymes are all primarily membrane-bound and cell-surface glycoproteins. Furthermore, they have a contribution in cell fusion, degradation of matrix protein, ectodomain shedding, and cell adhesion \[[@B48]--[@B52]\]. *β*-Secretase (BACE-1) is regarded as the rate-limiting enzyme in the proteolytic processing of APP. *β*-Secretase is a type I transmembrane protein and belongs to the pepsin family of aspartyl proteases \[[@B53]--[@B58]\]. *β*-Secretase possesses an *N*-terminal catalytic domain, harboring a transmembrane domain, two catalytic aspartic residues, and a short C-terminal cytoplasmic tail. *γ*-Secretase (GS) is a heterogeneous protein complex that possesses four transmembrane proteins including anterior pharynx-defective 1 (APH1), presenilin enhancer 2 (PEN2), nicastrin (NCT), and presenilins (PS1 and PS2) \[[@B59]--[@B61]\]. It has been observed that the nonamyloidogenic APP-processing pathway initiates with *α*-secretase-arbitrated APP cleavage at amino acid 687 (i.e., in the APP 770 isoform) which has been found to release the soluble APP alpha (sAPP*α*) into the extracellular space. Therefore, in the plasma membrane, a C-terminal fragment (CTF) of APP that is 83 amino acids in length (CTF 83) stays embedded. It has been found that the GS-mediated cleavage of CTF 83 then results in the APP intracellular domain (AICD) release into the cytoplasm and a small p3 fragment into the extracellular space. On the other hand, in the amyloidogenic APP processing, *β*-secretase cleaves APP which leads to the generation of the CTF that is 99 amino acids in length (CTF 99). It has also been observed that GS-mediated CTF 99 cleavage causes AICD release into the cytoplasm and A*β* into the extracellular space \[[@B62]\]. The activity of *α*-secretase is arbitrated via one or more enzymes from a disintegrin and metalloprotease (ADAM) family, with ADAM19, 17, 10, and 9 being the most possible candidates \[[@B48], [@B63], [@B64]\]. In the brain, BACE-1 is the main *β*-secretase \[[@B65]\], whereas GS is a multiprotein complex \[[@B66], [@B67]\]. 4. Anti-Alzheimer\'s Molecules Targeting *α*-Secretase {#sec4} ====================================================== It has been suggested by recent studies that the detected *α*-secretases exhibit an increased level of redundancy. In addition, it is still unclear exactly which *α*-secretases are accountable for cleavage of APP in neurons and other brain cells \[[@B51], [@B68]\]. Henceforth, it is essential to overcome the aforesaid ambiguity in order to develop effective compounds which will directly activate the *α*-secretases. An indirect and alternative approach to promote the *α*-secretase-arbitrated APP cleavage might be the induction of one or more of the signal transduction pathways associated with the regulation of the *α*-secretase activity. A number of pathways including mitogen-activated protein kinases, tyrosine kinases, protein kinase C, and calcium-mediated pathways have been found to play roles in terms of regulation of *α*-secretase activity. Indeed, it is possible to develop compounds that will have the capacity to induce the activity of *α*-secretase via these pathways \[[@B69]\]. It has been proposed that derivatives of retinoic acid have the ability to elevate the level of ADAM10 transcription. In addition to this, these derivatives might also have the potential to indirectly induce the APP cleavage mediated by *α*-secretase \[[@B70]\]. However, as an AD drug treatment, the development of a direct *α*-secretase activator appears unlikely, at least in the short-term. In clinical trials, the drug treatments that indirectly stimulate the activity of *α*-secretase are already being studied. Evidence reporting that the indirect activation of *α*-secretase is effectively attained through the agonists of *α*-7-nicotinic acetylcholine receptor (*α*7-nAChR), an agonist of 5-hydroxytryptamine receptor 4 (5-HT~4~) and a modulator of the gamma-aminobutyric acid (GABA) receptor, has been utilized as support to justify the clinical development of these agents \[[@B71]\]. However, there is a scarcity of data to support the fact that these compounds can elevate the APP cleavage mediated via *α*-secretase and decrease the levels of A*β*. But, in spite of such concerns, the approach is appealing from a theoretical perspective, according to the fact that these drugs might be approved based on their symptomatic benefits. More extensive studies can be further carried out regarding their potential to modify the disease \[[@B72]--[@B74]\]. 4.1. GABA~A~ Receptor Modulators and PDE4 Inhibitors {#sec4.1} ---------------------------------------------------- EHT-0202 ([Figure 3](#fig3){ref-type="fig"}) is a modulator of the GABA type A (GABA~A~) receptor and an inhibitor of phosphodiesterase 4 (PDE4) that has been developed as an AD treatment. EHT-0202 has the unique property of stimulating the activity of *α*-secretase via elevating the generation of procognitive and neurotrophic sAPP*α* fragments of APP. Preclinical studies revealed that EHT-0202 has the ability to protect the cortical neurons against A*β*42 and related stresses and that the provided neuroprotection is linked with the induction of sAPP*α*. In addition to this, it has also been revealed that EHT-0202 exhibits precognitive activities in several preclinical models such as memory impairment related to age and amnesia induced by scopolamine. Preclinical studies in guinea pigs and rats have shown that chronic oral administrations of EHT-0202 can lead to decreased levels of A*β*42 in cerebrospinal fluid (CSF) and brain \[[@B75]\]. It has been reported by Désiré et al. \[[@B76]\] on the identification in AD patients of blood-based transcriptomic signatures related to treatment response of EHT-0202 in a placebo-controlled, three-month, phase IIA study for estimating the exploratory efficacy, tolerability, and clinical safety of EHT-0202 (i.e., 80 and 40 mg twice a day) as adjunctive therapy to one cholinesterase inhibitor in mild-to-moderate AD individuals. Additionally, this pilot study also confirmed the value of using blood transcriptomic signatures as biomarkers for monitoring efficacy or predicting the response of patients, for a given therapeutic drug in the case of AD. For EHT-0202 or other AD drugs, such biomarkers might provide help to enhance the approaches in order to comprehend the drug mechanism efficacy, detect suitable patient populations for treatment, and/or elucidate the inherent subjectivity in most clinical endpoints utilized in Alzheimer\'s treatment \[[@B76]\]. 4.2. Partial *α*7-nAChR Agonists {#sec4.2} -------------------------------- EVP-6124 (encenicline, [Figure 3](#fig3){ref-type="fig"}) is a partial agonist of *α*7-nAChR which is under study to treat AD. EVP-6124 can cause activation of *α*7-nAChR at low nanomolar brain concentrations and can also improve memory performance in rats \[[@B77]\]. In the case of phase I and II clinical studies performed in individuals with mild-to-moderate AD, it has been observed that the EVP-6124 treatment was tolerated and exhibited momentous enhancements as compared to placebo on functional and cognitive measures \[[@B77]\]. However, a clinical hold was imposed on EVP-6124 by FDA in 2015 due to the reported gastrointestinal side effects observed in two phase III Alzheimer\'s studies \[[@B78]\]. 4.3. Partial 5-HT~4~ Receptor Agonists {#sec4.3} -------------------------------------- Activation of 5-HT~4~ receptors can lead to possible memory improvement and also changes in brain 5-HT~4~ receptors in AD \[[@B79]\]. On the other hand, 5-HT~4~ receptor stimulation is found to be linked with both effects on acetylcholine release and APP processing \[[@B79]\]. For AD, PRX-03140 ([Figure 3](#fig3){ref-type="fig"}) is a partial agonist (i.e. 18% compared with 5-HT) of the 5-HT~4~ receptor which is being developed by EPIX Pharmaceuticals \[[@B80]\]. It has been observed in a phase 2a clinical trial of PRX-03140 (as a single agent and in combination with donepezil) in mild AD individuals administered daily oral 150 mg doses of PRX-03140 as monotherapy achieved a mean 3.6-point enhancement on the Alzheimer\'s Disease Assessment Scale-Cognitive Subscale (ADAS-Cog) versus a 0.9-point worsening in individuals receiving placebo (*P* = 0.021). When PRX-03140 was used as a monotherapy, it was found to be well tolerated. In addition to this, no significant drug-related adverse effects were observed, when PRX-03140 was used in combination with donepezil \[[@B81]\]. Unfortunately, PRX-00023 development was discontinued by EPIX Pharmaceuticals in March 2008 because of the lack of efficacy exhibited in a recently completed phase 2b trial in individuals with major depressive disorder \[[@B82]\]. 4.4. Natural *α*-Secretase Modulators {#sec4.4} ------------------------------------- ### 4.4.1. Retinoids {#sec4.4.1} The retinoid family involves retinol (i.e., vitamin A; [Figure 4](#fig4){ref-type="fig"}) and its natural derivatives. Retinoids have a significant contribution in regulating various activities of the adult brain including long-term potentiation, release of neurotransmitters, neurite growth, and neuronal differentiation \[[@B83]\]. The study of Corcoran et al. \[[@B84]\] first confirmed that decrease in the brain retinoid acid signaling pathway can lead to deposition of A*β* in the adult rat brain. In a different study, it was observed that tamibarotene (i.e. an agonist of the retinoid receptor) reduced the levels of A*β*~42~ and A*β*~40~ by upregulating the expression of *α*-secretase in A*β*PP23 transgenic mice \[[@B85]\]. Recently, Cummings et al. \[[@B86]\] carried out a placebo-controlled, randomized, double-blind, parallel-group study of a single dose (300 mg per day) of bexarotene (a highly specific agonist of retinoid X receptor) in individuals with AD and both noncarriers and carriers of the apolipoprotein E (*APOE*) allele. Individuals who received treatment for four weeks and were of the noncarrier group experienced decreased A*β* levels in the brain as compared to those of the *APOE4* carrier group \[[@B86]\]. Ghosal et al. \[[@B87]\], in another clinical study, assessed the activity of bexarotene in the alteration of A*β* metabolism in cognitively healthy participants. This study reported a significant problem with bexarotene treatment because of its poor CNS-penetrating capacity in cognitively healthy individuals \[[@B87]\]. ### 4.4.2. Epigallocatechin-3-Gallate {#sec4.4.2} Epigallocatechin-3-gallate (EGCG, [Figure 4](#fig4){ref-type="fig"}) is a polyphenol found in the leaves of tea plants (i.e. *Camellia sinensis*). This polyphenol possesses antioxidant activities that may be beneficial in the case of AD as oxidative stress plays a vital role in the development of this devastating disorder \[[@B88]\]. A study by de la Torre et al. \[[@B89]\] described that when EGCG is administered in parallel with cognitive training to patients with Down\'s syndrome at a dose of 9 mg/kg/day for 12 months, beneficial actions were observed on memory as compared to placebo-receiving participants or the placebo plus cognitive training group in cognitive tests. Although this clinical study has been carried out in individuals displaying Down\'s syndrome only, in order to assess the ability of EGCG in improving cognitive functions and to observe its activities on APP and DYRK1A, the activity of this polyphenol needs also to be evaluated in the case of AD \[[@B89]\]. The activity of a formulation of EGCG named Sunphenon EGCg was studied in individuals with early-stage of AD in a clinical study, but till now, the findings regarding the potential disease-modifying activities have not been posted \[[@B90]\]. ### 4.4.3. Curcumin {#sec4.4.3} Curcumin ([Figure 4](#fig4){ref-type="fig"}), the major curcuminoid of turmeric, comes from the turmeric rhizome *Curcuma longa*. Because of its small size, curcumin can easily cross the blood-brain barrier (BBB), and this natural product has been recommended as a promising therapy for AD \[[@B91]\]. Amino acid conjugates of curcumin were developed by Narasingapa et al. \[[@B92]\]. In their study, they observed that these conjugates had a potent *α*-secretase stimulatory effect \[[@B92]\]. The efficacy of this compound in the case of AD was also evaluated through multiple clinical studies. The potential effects of curcumin and *Ginkgo biloba* leaf extracts on AD progression have also been evaluated in a clinical study \[[@B93]\]. However, the findings revealed that the combination of curcumin and extracts was unsuccessful in reducing the levels of A*β* in the blood of AD-affected individuals or in improving their cognitive functions \[[@B93]\]. In a different clinical study, curcumin\'s tolerability, safety, and activity were evaluated on individuals with mild-to-moderate AD by Ringman et al. \[[@B94]\]. This study did not report any beneficial action of curcumin in reducing the levels of A*β* in CSF. Moreover, it was suggested that the inefficacy of curcumin may be linked to its poor bioavailability \[[@B94]\]. ### 4.4.4. Bryostatin-1 {#sec4.4.4} Bryostatin-1 ([Figure 4](#fig4){ref-type="fig"}) is a naturally occurring macrocyclic lactone derived from the marine bryozoan (i.e. *Bugula neritina*). Bryostatin-1 has the ability to induce the activity of *α*-secretase by protein kinase C (PKC) activation and subsequent elevation of sAPP*α* release \[[@B95]\]. Multiple clinical studies have assessed the efficacy of this compound in individuals with AD. Participants were administered a single intravenous placebo dose or 25 *μ*g/m^2^ bryostatin-1 in a single-dose randomized double-blind phase IIA clinical study \[[@B95]\]. In that study, pharmacodynamics, pharmacokinetics, and efficacy of bryostatin-1 were also studied. It was observed that bryostatin-1 was well tolerated in individuals with AD. This compound also increased the Mini-Mental State Examination (MMSE) score, and no adverse events were observed with the use of this compound \[[@B95]\]. In a different phase II clinical study, bryostatin-1 was administered intravenously at the doses of 20 and 40 *μ*g in order to treat patients with moderately severe to severe AD \[[@B96]\]. The primary endpoint of this study was not significant in the full data set \[[@B96]\]. 5. Anti-Alzheimer\'s Molecules Targeting *β*-Secretase {#sec5} ====================================================== Following the discovery of BACE-1, many attempts have been made in order to develop small-molecule BACE-1 inhibitors. Interestingly, peptide-based mimetics of the APP *β*-cleavage site were the first generation of BACE-1 inhibitors, which contained the APP *β*-site scissile amide bond replaced with a nonhydrolyzable transition state analog, for example, statin \[[@B55]\]. In recent times, nonpeptidergic compounds containing high affinities for BACE-1 have been produced \[[@B97], [@B98]\]. BI-1181181 is an orally active and first-generation BACE-1 inhibitor. Unfortunately, because of its low BBB penetration and low oral bioavailability, BI-1181181 has been found to fail in phase I clinical trials. Subsequently, in clinical trials, second-generation BACE-1 inhibitors including LY-2886721, LY-2811376, and RG-7129 have also failed due to liver toxicity \[[@B99], [@B100]\]. Fortunately, the third-generation BACE-1 inhibitors including JNJ-54861911, CNP-520, AZD-3293, and E-2609 have exhibited encouraging clinical data and satisfactory pharmacokinetic (PK) parameters in ongoing studies. Even though verubecestat (MK-8931) has the ability to decrease the A*β* level of CNS in AD individuals and animal models \[[@B101]\], nevertheless, in February 2017, Merck (an American multinational pharmaceutical company) declared that they will stop the clinical trial in mild-to-moderate AD individuals due to its lack of efficacy \[[@B102]\]. However, APECS trials in individuals with prodromal AD are still ongoing. As a disease-modifying AD treatment, CTS-21166 is a small-molecule BACE-1 inhibitor that is being developed. It has been reported that CoMentis (an American biotech company) has started a phase I clinical trial of its orally bioavailable small-molecule CTS-21166. As mentioned by CoMentis, CTS-21166 is a highly selective, efficacious, and potent brain-penetrating BACE-1 inhibitor \[[@B103]\]. Nonetheless, initial clinical outcomes of CTS-21166 were found to be unsatisfactory \[[@B104], [@B105]\]. 5.1. Acyl Guanidine-Based Inhibitors {#sec5.1} ------------------------------------ Interestingly, via using high-throughput screening (HTS), Wyeth (an American pharmaceutical company) has discovered a number of acyl guanidine-based BACE-1 inhibitors \[[@B106]\]. In addition, an X-ray crystal structure of compound 1 ([Figure 5](#fig5){ref-type="fig"}) complexed with the catalytic domain of BACE-1 showed that the acyl guanidine moiety forms 4 key hydrogen interactions with the catalytic aspartic acids including Asp228 and Asp32 \[[@B106]\]. Moreover, a structural alteration in BACE-1 upon binding with this compound was detected \[[@B106]\]. In order to improve potency, later substitutions were made to compound 1. On the other hand, compound 2 ([Figure 5](#fig5){ref-type="fig"}), comprising a *para*-propyloxyphenyl moiety in the unsubstituted aryl ring and propyl alcohol in the third guanidine nitrogen, permitted an enhancement of about 30-fold in potency as compared to compound 1. However, it has been observed that the 2-chloro group of compound 1 does not play a significant role in the case of potency \[[@B106]\]. Bristol-Myers Squibb (an American pharmaceutical company) has also worked in an acyl guanidine series. They started with hit compound 3 ([Figure 5](#fig5){ref-type="fig"}), and developed compound 4 with a good potency against BACE-1 (IC~50~ = 5.0 nM) while remaining inactive against the other aspartyl proteases studied, i.e., pepsin (IC~50~ \> 100.000 nM), cathepsin E (CatE), and cathepsin D (CatD) \[[@B107]\]. In rats, the optimized compound 4 was assessed to evaluate its action on the levels of A*β*40 in CSF, the brain, and plasma. A dose-dependent and significant decrease in the level of A*β*40 (about 80%) in plasma was observed, but no substantial decrease in CSF and the brain was attained (\<20%). This deficiency of efficacy in CSF and the brain was attributed to the efflux of P-glycoprotein (P-gp). Therefore, further enhancement was essential in order to optimize its PK properties \[[@B107]\]. Interestingly, Array BioPharma (an American biopharmaceutical company) in cooperation with Genentech (a biotechnology corporation-subsidiary of Roche) developed a number of chromane-based spirocyclic acyl guanidine-derived BACE-1 inhibitors leading to compound 5 ([Figure 5](#fig5){ref-type="fig"}). Indeed, this compound exhibited a good selectivity towards BACE-1. Additionally, this compound also had the ability to decrease the levels of A*β*40 in CSF from 53% to 63% in rats and cynomolgus monkeys, respectively. Furthermore, this compound exhibited a high efflux ratio of P-gp \[[@B108]\]. 5.2. 2-Aminopyridine-Based Inhibitors {#sec5.2} ------------------------------------- The discovery of 2-amino heterocycles is regarded as the major progress in the development of small BACE-1 inhibitor molecules. It has been found that these inhibitors commonly possess a better physicochemical profile and can improve *in vivo* efficacy and brain penetration \[[@B109]\]. On the other hand, as an extension of their study on acyl guanidine inhibitors, Wyeth also developed a number of pyrrolyl 2-aminopyridines. It has been suggested that the inhibitors of acyl guanidine are polar due to the presence of the acyl guanidine moiety, which leads to poor permeability (\<5%) of BBB. Interestingly, it has been observed that the bioisosteric replacement of the acyl guanidine moiety by an aminopyridine (compound 6) in compound 1 leads to the enhancement of its permeability while preserving the hydrogen-bonding interactions with the two aspartic acids in the catalytic site of BACE-1 \[[@B110]\]. Through X-ray studies, the similarity of the hydrogen interaction in between these 2 compounds was verified, which further established the interaction of the 2-aminopyridine moiety with the aspartic acids Asp32 and Asp228 of the catalytic site of BACE-1 \[[@B110]\]. Interestingly, permeability was shown to be enhanced through the modulation of the total polar surface area. Compound 6 ([Figure 5](#fig5){ref-type="fig"}) indeed exhibited a good central drug exposure with a brain-to-plasma ratio of 1.7 in comparison with the 0.04 ratio attained with the acyl guanidine-based compound 1 \[[@B110]\]. 5.3. Aminothiazine- and Aminooxazoline-Based Inhibitors {#sec5.3} ------------------------------------------------------- A number of aminooxazoline/aminothiazine-based inhibitors have been developed by researchers of different pharmaceutical companies. Unfortunately, out of these compounds, only few of the compounds were able to reach the clinical evaluation stage. Indeed, the first clinical BACE-1 inhibitor which was developed by Eli Lilly (an American pharmaceutical company) is an aminothiazine-based compound, LY-2811376 (compound 7, [Figure 5](#fig5){ref-type="fig"}). This pharmaceutical company initiated phase I clinical trials of this compound in 2009. At single doses from 5 to 500 mg as oral capsules, this BACE-1 inhibitor was administered to 61 healthy participants in order to assess its tolerability and safety \[[@B111]\]. It was reported that LY-2811376 exhibited well tolerability with no serious adverse effects \[[@B112]\]. In addition to this, peak concentration was achieved within two hours postdose in the plasma and five hours postdose in CSF. Interestingly, a dose-dependent decrease in the level of A*β*42 and A*β*40 was noticed. Following a single dose of 90 mg, an A*β*40 decrease of 80% was reported within 7 hours and 54% within 12--14 hours, in the plasma and CSF, respectively. On the other hand, in parallel with the phase I clinical trials, a toxicological study was performed in rats. A retinal pathology was observed in these animals at doses ≥ 30 mg/kg, and the pathology was characterized by cytoplasmic buildups of finely granular autofluorescent material dispersed within the retinal epithelium. Therefore, the ongoing clinical studies of LY-2811376 (compound 7) were terminated. Thus, LY-2811376 did not enter into the phase II clinical trial. The aforesaid toxic effect might take place due to the off-target actions of LY-2811376 against other aspartic acid proteases, viz., CatD, as confirmed through a subsequent study by means of LY-2811376 in *BACE-1* knockout mice \[[@B112]\]. After the end of the trial, all the participants were subjected to follow-up study as a safety measure. Luckily, no participants exhibited any clinically significant observations \[[@B112]\]. The second clinically studied compound was an aminothiazine-based inhibitor called LY-2886721 (compound 8, [Figure 5](#fig5){ref-type="fig"}). This compound was the first BACE-1 inhibitor that reached phase II trials. In fact, LY-2886721 was studied in six phase I trials to evaluate its pharmacology, safety, and tolerability effects on 150 healthy participants. Interestingly, the administration of this compound for 14 days at a dose of 70 mg/day caused a reduction of CSF A*β*42 by 71% and CSF A*β*40 by 74%. Up to six weeks, no apparent safety concerns were observed with the used dosage \[[@B113]\]. Due to the satisfactory findings obtained in phase I clinical studies, Eli Lilly started a phase II clinical trial in March 2012 to evaluate the tolerability, pharmacodynamic (PD) properties, and safety of LY-2886721 in mild AD patients \[[@B114]\]. In the course of the study, routine safety monitoring identified aberrant elevations in the liver enzyme level in 4 out of 70 patients. Therefore, Eli Lilly terminated the phase II trial and clinical development of LY-2886721. It has been observed that the toxicity exerted by this BACE-1 inhibitor was found to be an off-target action of this compound which was unrelated to the inhibition of BACE-1 \[[@B113], [@B115]\]. The aminooxazoline-based compound, RG-7129 (RO-5598887, compound 9, [Figure 5](#fig5){ref-type="fig"}) was developed by Roche (a Swiss multinational healthcare company). This compound entered into a phase I trial in September 2011. Preclinical studies revealed that it presents high potency against BACE-1 (IC~50~ = 30 nM). Except against BACE-2 (IC~50~ = 40 nM), this compound exhibited selectivity against pepsin, CatE, CatD, and renin (\>1000-fold). Between 2012 and 2013, three phase I trials have been completed. Unfortunately, no results were reported in this period. In addition, Roche terminated the development of RG-7129 without providing any clarification \[[@B116]\]. However, in a study published in 2014, Roche started the use of a combination treatment of the anti-A*β* antibody gantenerumab and the BACE-1 inhibitor RG-7129 (compound 9) in transgenic mice, recommending a future clinical assessment of the combination of these two compounds \[[@B117]\]. 5.4. Aminoimidazole-Based Inhibitors {#sec5.4} ------------------------------------ It has been revealed through molecular modeling studies that the amino group of the imidazole heterocycle was found to be accountable for the binding with the catalytic aspartic acid residues of BACE-1. Potent inhibitors including compound 10 ([Figure 6](#fig6){ref-type="fig"}) were obtained as a result of 2-methoxy-5-nitro substituents on the benzyl subunit. Subsequently, compound 11 ([Figure 6](#fig6){ref-type="fig"}) was synthesized, and this latter was found to cause a significant decrease in the P-gp efflux ratio (ER). Therefore, Merck (an American multinational pharmaceutical company) carried out further studies in order to improve this inhibitor\'s potency. It has been observed that the addition of a conformational constraint in compound 12 ([Figure 6](#fig6){ref-type="fig"}) allowed a five-time potency increment because of additional hydrophobic interactions with the flap area of BACE-1 \[[@B118]\]. Furthermore, the addition of a fluorine group in compound 13 ([Figure 6](#fig6){ref-type="fig"}) permitted additional hydrophobic interactions as well as a six-times potency increase to an IC~50~ (BACE − 1) value of 63 nM. Moreover, this compound was suggested to exhibit brain penetration capacity due to its decreased P-gp ER of 3.6 \[[@B118]\]. AstraZeneca developed compound AZD-3293 (LY-3314814, Lanabecestat, compound 14 ([Figure 6](#fig6){ref-type="fig"})), which is an aminoimidazole-based compound, and this compound reached phase I trials in 2012 \[[@B119]\]. Currently, the findings obtained from the first two phase I clinical trials are available \[[@B120]\]. These clinical trials involved single ascending dose studies in order to evaluate the doses of 1--750 mg with a food-effect component (*n* = 72). The trials also involved a two-week several ascending dose studies in order to evaluate the doses of 50 or 15 mg once per day or 70 mg once per week in elderly participants (i.e. part 1; *n* = 31) and doses of 150, 50, or 15 mg once per day in individuals with mild-to-moderate AD (i.e. part 2; *n =* 16). Findings obtained from these studies revealed that AZD-3293 was well tolerated up to the highest doses provided. In addition, ≥70% reduction in mean plasma concentrations of A*β*42 and A*β*40 was observed with single doses of ≥5 mg. On the other hand, at the highest dose level studied, prolonged inhibition for up to three weeks has been observed. With the use of multiple doses, a vigorous reduction was observed in A*β* levels of CSF (i.e. ≥76% at ≥50 mg and ≥51% at 15 mg) and plasma (i.e. ≥78% at ≥50 mg and ≥64% at 15 mg). In addition to this, prolonged inhibition was observed even with a once-a-week dosing regimen \[[@B120]\]. Four additional phase I clinical studies of compound 14 were conducted with a total of 175 healthy participants in 2015 and 2016. In these studies, a new tablet formulation was used evaluating possible interaction of this compound with a number of commonly prescribed drugs (including donepezil, simvastatin, midazolam, dabigatran, and warfarin) in the elderly \[[@B121]\]. Presently, Eli Lilly and AstraZeneca are sponsoring the two phase III clinical trials of AZD-3293 (compound 14) \[[@B119]\]. At daily doses of 20 or 50 mg for 18 to 24 months, these placebo-controlled, double-blind, multicenter, randomized studies are evaluating the disease-modifying potential of AZD-3293 in over 4,000 individuals with early and mild AD \[[@B119], [@B120]\]. 5.5. Iminothiadiazinane Dioxide-Based Inhibitors {#sec5.5} ------------------------------------------------ A series of iminothiadiazinane dioxide-based inhibitors represented by verubecestat (compound 15, [Figure 6](#fig6){ref-type="fig"}) has been developed by Merck. Interestingly, to explore a new intellectual property space and improve binding affinity, the starting point was an iminopyrimidinone scaffold (compound 16, [Figure 6](#fig6){ref-type="fig"}) which was modified \[[@B105]\]. The compound 15 (verubecestat) exhibited good PD/PK properties in animal models. Furthermore, this compound caused sustained and marked decrease in the levels of CSF A*β*40 in cynomolgus monkeys (i.e., 81% and 72% decrease at 10 and 3 mg/kg, respectively). In telemetered monkeys, this compound did not exhibit any action on the QT interval in a single-dose cardiovascular study. On the other hand, it caused no stimulation of expression of CYP1A2 or CYP3A4 in human hepatocytes \[[@B122]\]. In 2011, compound 15 entered into phase I trials, where its PD, PK, and safety were assessed in healthy participants as well as in individuals with mild-to-moderate AD. In the case of AD and healthy individuals, both the multiple and single doses were well tolerated and decreased A*β* levels in the CSF. Furthermore, in any of the phase I trials, no alterations were observed in hair or skin pigmentation due to the inhibition of BACE-2. Pigmentation alterations became apparent for prolonged treatment times \[[@B122]\]. The compound 15 advanced to phase II/III trials in 2012, and the trials involved individuals with mild-to-moderate AD (i.e. the EPOCH trial). Additionally, a phase III trial of this compound started in 2013 for prodromal AD (i.e. the APECS trial) \[[@B109]\]. Because of the lack of efficacy as well as the presence of marked neuronal and synaptic losses, the EPOCH trial was terminated in February 2017 \[[@B123]\]. On the other hand, the APECS trial was also terminated in February 2018. Furthermore, Merck excluded verubecestat from its research pipeline. As compared to the placebo group, volunteers who were administered verubecestat scored worse on the cognitive test Alzheimer\'s Disease Cooperative Study-Activities of Daily Living (ADCS-ADL). More studies are essential to understand the reason of this negative effect \[[@B124], [@B125]\]. 6. Anti-Alzheimer\'s Molecules Targeting *γ*-Secretase {#sec6} ====================================================== It has been found that unlike BACE1, GS has demonstrated to be a highly manageable target for AD, at least in terms of the development of orally bioavailable GS inhibitors (GSIs) with brain-penetrating capacity \[[@B66]\]. For mouse and human brains, GSIs have been found to reduce the production of A*β*. In the case of APP mouse models, the deposition of A*β* was found to be decreased due to the chronic administration of GSIs \[[@B126]--[@B129]\]. So far, several orally bioavailable GSIs presenting brain-penetrant properties have been found. Nevertheless, a vital problem in the development of GSIs is target-based toxicity. This target-based toxicity was observed on the basis of preclinical studies \[[@B71]\]. Along with APP, a number of other type 1 transmembrane proteins are also processed by the GS enzyme. In this regard, for instance, GS processing of Notch1 is essential for Notch signaling \[[@B130]\]. In mice, *PESN1* deletion is embryonically lethal. Furthermore, these mice exhibit a phenotype---protuberant brain and skeletal abnormalities---that is also obvious in mice lacking the notch \[[@B131], [@B132]\]. In addition to this, treatment with nonselective GSIs can lead to various Notch inhibition-related toxicities, including (but not limited to) abnormalities of the integumentary, immune, and gastrointestinal systems \[[@B133]--[@B135]\]. In phase III trials, semagacestat (a GS inhibitor) failed to attain the primary goals due to observed worsening symptoms in some patients \[[@B85]\]. The study of avagacestat was terminated due to the serious adverse events observed in a phase II trial including nonmelanoma skin cancer, dose-dependent glycosuria, and cerebral microbleeds. Interestingly, EVP-0962, the selective, small molecule, orally available GS modulator (GSM), decreases the production of A*β*~1-42~ via shifting the cleavage of APP toward the generation of less toxic and shorter A*β*, without affecting the cleavage of Notch \[[@B136]\]. On the other hand, the naturally occurring cyclic sugar alcohol NIC5-15 has the ability to modulate the activity of GS to lower the production of A*β*. NIC5-15 is currently in phase II clinical trials of Alzheimer\'s treatments \[[@B137]\]. 6.1. NSAID-Derived *γ*-Secretase Modulators {#sec6.1} ------------------------------------------- The first generation of GSMs was discovered from an epidemiological study. This study reported a decreased prevalence of AD among the individuals who were administered nonsteroidal anti-inflammatory drugs (NSAIDs) \[[@B138]\]. Subsequently, Weggen et al. \[[@B138]\] stated that NSAIDs reduced the level of A*β*42 along with the increment of an A*β*38 isoform, which further indicated that the activity of GS can be modulated by NSAIDs without causing a marked disturbance of Notch cleavage or other APP-processing mechanisms \[[@B138]\]. The aforesaid alteration in the pattern of cleavage could be explained by several processes including an elevation in the GS\'s cleavage activity (defined by the catalytic constant, *κ*~cat~) or a reduction in the possibility of releasing longer A*β* from the enzyme-substrate complex (defined as a dissociation constant, *κ*~d~). Chávez-Gutiérrez et al. \[[@B139]\] through the use of CTF 99 as a substrate showed that the ratios of A*β*38/A*β*42 and A*β*40/A*β*43 were reduced via all the *PESN* mutations studied, further recommending a weakening of the 4 cleavage cycles of GS (GS epsilon-cleavage). In addition to this, Okochi et al. \[[@B140]\], using A*β*42 as a substrate, estimated the kinetic constants for the *γ* cleavage. These researchers also stated that GSMs elevated *κ*~cat~ and reduced *κ*~d~, whereas mutations of PS lead to the opposite effect. Collectively, these findings indicated that the GS modulation effect might be a useful technique to reverse the action of PS mutations through the restoration of the normal ratios of the formed A*β*, in this manner modifying the progression and pathology of the disease \[[@B141]\]. However, the use of NSAID as GSMs can induce some risks of renal and gastrointestinal toxicity because of the effects of NSAIDs against the cyclooxygenase 1 (COX-1) enzyme. This aforesaid problem of NSAID is compromising its usage for long-term therapeutic solutions \[[@B142]\]. Luckily, the activity of COX-1 inhibition was found to be not dependent on the activity of GS modulation. For example, flurbiprofen is a COX-1 inhibitor which was administered as a racemate for clinical trials. Nevertheless, tarenflurbil (compound 17, [Figure 7](#fig7){ref-type="fig"}) exhibited a decreased activity of COX-1 inhibition while preserving its action as a GSM \[[@B142]\]. This latter compound has entered into phase III trials. However, data exhibited no differences between the placebo and tarenflurbil. The cause of this failure was attributed to its inadequate PD properties. In fact, weak CNS penetration of tarenflurbil was formerly stated in preclinical studies in rodents, with a CSF/plasma ratio of 1.3%. The clinical development of tarenflurbil was terminated in view of these unsatisfactory findings \[[@B143]\]. Subsequently, 2 NSAID carboxylic acid derivatives were developed by Chiesi Farmaceutici (an Italian family-controlled global pharmaceutical company) (CHF-5074, compound 18, [Figure 6](#fig6){ref-type="fig"}) and FORUM Pharmaceuticals (EVP-0015962, compound 19, [Figure 7](#fig7){ref-type="fig"}) and also studied in clinical studies. Based on the scaffold of tarenflurbil (compound 17), Chiesi Farmaceutici developed CHF-5074 (compound 18). Interestingly, there was complete cessation of COX inhibition (at 100 *μ*M and 300 *μ*M) by a cyclopropyl group. It has been observed that the inhibition of A*β*42 generation displayed a 7-fold improvement due to the introduction of chlorine substituents on the terminal phenyl ring as compared to tarenflurbil (compound 17) \[[@B144]\]. It was proposed that the carboxylic acid group interacts with a lysine residue of APP situated close to the membrane interface, with the lipophilic substituents playing a role as membrane anchors \[[@B145]\]. In 2012, EVP-0015962 (compound 19) entered into phase II trials, but findings were not posted publicly on ClinicalTrials.gov. 6.2. Non-NSAID-Derived *γ*-Secretase Modulators {#sec6.2} ----------------------------------------------- In 2004, NeuroGenetic Pharmaceuticals (an American pharmaceutical company) developed one of the first GSM classes not presenting a carboxylic acid moiety (non-NSAID), which further lead to the discovery of NGP-555 \[[@B146]\]. NGP-555 (compound 20, [Figure 7](#fig7){ref-type="fig"}) exhibited a reduction in CSF A*β*~42~ levels between 20 to 40% and an elevation of the shorter forms in rodent studies. In addition to that, this compound confirmed neuroprotection from cognitive impairment in two independent mice experiments by employing various learning and memory tasks \[[@B147]\]. In 2015, NGP-555 was taken into phase I trials. In January 2017, NeuroGenetic Pharmaceuticals revealed that this compound exhibited well-tolerability and safety in healthy participants \[[@B148]\]. However, detailed findings and future clinical trials with this compound have not been revealed yet. Interestingly, the work on the cinnamide series from Eisai Pharmaceuticals leads to the discovery of the clinical compounds E-2012 (compound 21, [Figure 7](#fig7){ref-type="fig"}) and E-2212 (compound 22, [Figure 7](#fig7){ref-type="fig"}) \[[@B146]\]. In a dose-dependent manner, E-2012 reduced the levels of A*β*42 and A*β*40 in rat plasma, brain, and CSF *in vivo*. It has been reported that the IC~50~ values of E-2012 for A*β*42 and A*β*40 were 92 and 330 nM, respectively \[[@B149]\]. On the other hand, E-2012 markedly reduced the levels of A*β*42 in rat CSF by 47.2% and 16.6% at doses of 30 and 10 mg/kg, respectively. Furthermore, E-2012 caused a reduction of the levels of A*β*42 and A*β*40 and elevated the levels of shorter A*β*s (i.e. A*β*37 and A*β*38), without altering the total amount of A*β* \[[@B150]\]. E-2012 (compound 21) was moved into the clinical trials in 2006 as the first noncarboxylic acid compound. Furthermore, during a phase I clinical trial, this compound exhibited efficacy in terms of decreasing the A*β*42 levels in plasma (\~50%) \[[@B151]\]. Nevertheless, in parallel to the phase I clinical trial, lenticular opacity was seen in a high-dose group of a 13-week preclinical safety study in rats leading to the termination of the clinical trial. However, ocular toxicity was not observed in monkeys, when follow-up studies were done with up to the highest tolerated dose for E-2012 \[[@B152]\]. Nonetheless, Eisai Pharmaceuticals decided to further develop the improved compound E-2212 (compound 22) \[[@B141]\]. In 2010, E-2212 (compound 22) was taken into a phase I clinical trial. This compound exhibited a similar pharmacological profile as E-2012 (compound 21) as well as improved safety profile, along with no clinically important ophthalmologic findings \[[@B153]\]. In addition to this, the PD response measured in the plasma elevated with the dose and was revealed to perform a 54% decrease in the level of A*β*42 at the dose of 250 mg. To date, Eisai Pharmaceuticals has not revealed any reports regarding the further development of E-2212 (compound 22). Furthermore, there is no update regarding possible new studies on ClinicalTrials.gov. The anticipated structure of E-2212 (compound 22) containing a high cLogP (5.5), high molecular weight (480 g/mol), and four aromatic rings \[[@B141]\] might have hardened its further development because of its poor drug-like properties. 7. Anti-Alzheimer\'s Molecules Targeting A*β* Accumulation {#sec7} ========================================================== Multiple enzymes can cause degradation of A*β* and have been considered for new drug development \[[@B154], [@B155]\]. The newly developed anti-AD drug sodium oligomannurarate (GV-971, [Figure 8](#fig8){ref-type="fig"}) by Shanghai Green Valley Pharmaceuticals can bind with several sites of amyloid and can cause destabilization and inhibition of A*β* aggregation, which can eventually elevate the clearance of A*β* \[[@B156]\]. The activity of GV-971 was also evaluated in individuals with mild-to-moderate AD in phase III clinical trials \[[@B157]\]. An alteration in the Alzheimer\'s Disease Assessment Scale-Cognitive Subscale (ADAS-Cog) score was observed as the main endpoint of that clinical study. Additionally, other findings of this study also indicate the significant beneficial effects of GV-971 on cognitive functions. The China National Medical Product Administration (NMPA) provisionally permitted (on November 2, 2019) the use of GV-971 to treat individuals with mild-to-moderate AD \[[@B157]\]. The use of immunotherapy to clear A*β* also seems a reasonable approach \[[@B158]\]. Multiple clinical studies are carried out with monoclonal antibodies that target A*β*. Aducanumab is a monoclonal antibody that targets A*β* aggregations \[[@B159]\]. Nonetheless, Biogen and Eisai declared on March 21, 2019, that they would stop the 221AD301 phase III study of aducanumab (BIIB037) in individuals with an early stage of AD (ENGAGE) and the 221AD302 phase III study of aducanumab (BIIB037) in individuals with an early stage of AD (EMERGE). This decision was taken according to the results of an interim analysis anticipating that ENGAGE and EMERGE were likely to miss their main endpoints \[[@B160]\]. Furthermore, Biogen reported on April 24, 2019, that it would not start the projected phase III secondary prevention program with aducanumab and terminated further studies. However, Biogen declared on October 22, 2019, that the interim analysis was incorrect. Moreover, following analysis of a larger data set, Biogen rather revealed that EMERGE had met its main endpoint. Later, Biogen declared the plan to apply for regulatory approval in early 2020 for aducanumab in the US \[[@B160]\]. Crenezumab is an anti-A*β* monoclonal antibody with a specific affinity towards all fibrillary, oligomeric, and pentameric amyloids \[[@B161]\]. Crenezumab is being assessed in a clinical trial of crenezumab versus placebo to assess the safety and efficacy in individuals with prodromal-to-mild AD (CREAD). Nevertheless, Roche terminated both the CREAD 1 and CREAD 2 studies. Following an interim analysis, it was indeed observed that the trial was less likely to achieve its primary endpoint of reducing the decrease on the Clinical Dementia Rating Sum of Boxes (CDR-SB) data \[[@B162]\]. 8. Anti-Alzheimer\'s Molecules Targeting the Phosphorylation Signaling in APP Processing {#sec8} ======================================================================================== Like A*β*, hyperphosphorylated tau has been detected in the brains of individuals with AD; therefore, protein kinase inhibitors can play an important role to reduce AD by primarily targeting pathogenic tau \[[@B163], [@B164]\]. Hyperphosphorylation can cause loss of tau solubility and can lead to paired helical filament (PHF) formation, which can further cause NFT formation \[[@B165]--[@B168]\]. The glycogen synthase kinase (GSK)3*β* has a significant contribution to the phosphorylation of tau. Furthermore, GSK3*β* causes phosphorylation of around 31% of the pathological phosphorylation sites of tau \[[@B166]\]. GSK3*β* activity can be increased by toxic A*β*, which can eventually increase the production of A*β* through the phosphorylation of tau \[[@B169], [@B170]\]. Thus, GSK3*β* forms a relationship in between tau pathology and A*β* toxicity \[[@B166]\]. The cyclin-dependent kinase 5 (CDK5) is another vital tau protein kinase associated with pathophysiological tau phosphorylation. Typically, p35 (i.e., the CDK5 regulating protein) is found truncated to 25 amino acids in the brain of individuals with AD. Interestingly, hyperphosphorylated tau dissociates from the microtubules and produces NFTs on CDK5 stimulation by p25 \[[@B171], [@B172]\]. The activity of CDK5 can also cause NFT phosphorylation \[[@B173], [@B174]\]. It has been observed that lithium can inhibit GSK3*β* and lead to a decreased rate of tau phosphorylation, which ultimately averted tau pathology in animal models \[[@B175]\]. However, no improvement in cognitive functions was observed, when used in individuals with early-stage of AD \[[@B176]\]. These findings are associated with unaltered biomarkers of AD in CSF of the individuals including toxic A*β*, phosphorylated tau, and total tau \[[@B176]\]. Multiple GSK3*β* inhibitors are also being developed that belong to the indirubin, paullone, and maleimide ([Figure 9](#fig9){ref-type="fig"}) families. Nevertheless, none of these inhibitors have entered into clinical studies until now. The cytotoxic effects of these inhibitors are regarded as the main reasons for their failure \[[@B177]\]. In another study, administration of the CDK5 inhibitor was reported to decrease the levels of A*β* and increase the levels of phospho-tau in very old and nontransgenic mice \[[@B178]\]. However, in the elderly mice, CDK5 inhibitors might not be useful in targeting tau phosphorylation \[[@B178]\]. Fyn physically associates with tau and can cause phosphorylation of tyrosine residues, such as Tyr18, close to the amino terminus \[[@B179]--[@B182]\]. Tyr18 is also phosphorylated in NFTs in the brains of AD individuals, which denotes a probable clinical relevance \[[@B179]\]. Masitinib ([Figure 9](#fig9){ref-type="fig"}) is an inhibitor of tyrosine kinase. Masitinib shows selectivity towards c-Kit, platelet-derived growth factor receptor, and, to a lesser extent, Lyn and Fyn. Masitinib was studied in a 24-week, placebo-controlled, randomized, phase II study involving 34 mild-to-moderate AD patients. Masitinib was found to be reasonably well tolerated and was linked with improved cognitive functions at 12 and 24 weeks, therefore suggesting inhibition of tyrosine kinase as a therapeutic approach for AD. There is also an ongoing phase III clinical study to assess the safety and efficacy of various doses of masitinib as compared to placebo \[[@B183]\]. 9. Conclusions {#sec9} ============== For rational anti-AD drug design, the characterization and identification of the secretase enzymes that cleave APP provides a molecular framework. In terms of physiological importance, the three secretases have extensively been studied. Thus, we now have the appropriate understanding and knowledge regarding the potential adverse events related to the use of drugs that modulate the secretases effects. Currently, multiple compounds targeting each of the 3 secretases are under clinical studies. The use of high throughput and virtual screening, followed by chemical optimization is trying to accommodate some physicochemical nuances and discover new compounds that will have the required safety and efficacy to modify the progression of AD. Henceforth, based on the currently available information, *α*-, *β*-, and *γ*-secretases can be considered as safe and efficacious targets for the reduction of AD pathology. Collectively, the future of AD will depend on the development of safe, potent, and selective compounds that will have the ability to delay the development and progression of AD. Furthermore, it is essential to identify the specific biomarkers which will allow an effective and early pharmacological intervention for AD. This work was funded by the Deanship of Scientific Research at Princess Nourah bint Abdulrahman University through the Fast-Track Research Funding Program. Conflicts of Interest ===================== The authors proclaim no competing interests. Authors\' Contributions ======================= MSU conceived of the original idea and designed the outlines of the study. MSU and MTK wrote the draft of the manuscript. MSU and BM prepared the figures for the manuscript. PJ edited the whole manuscript and improved the draft. GMA, AP, MNB-J, SAM, and MMA-D performed the literature review and aided in revising the manuscript. All authors have read and approved the final manuscript. {#fig1} {#fig2} {#fig3} {#fig4} {#fig5} {#fig6} {#fig7} {#fig8} {#fig9} [^1]: Academic Editor: Fabiana Morroni

Introduction {#Sec1} ============ One of the criteria used for evaluation of digital cameras is the image sensor used in the device. In general, a larger image sensor size allows a wider field of view to be acquired, while a smaller pixel size enables acquisition of finer images. The wide variety of image sensors that are available also leads to a wide variety of cameras, from the cameras installed in handy smartphones to high-end single-lens reflex (SLR) cameras. The situation is the same for X-ray image acquisition. In the soft X-ray (SX) region with its energy range from 30 eV to 3 keV, back-illuminated charge-coupled device (CCD) sensors are used as the image sensors. At present, commercially available CCD cameras for the SX region have pixel dimensions of 13.5 × 13.5 μm^2^ and contain 1024 × 1024 pixels. These values have remained constant for the past 10 years and we have no alternative options for image sensor configuration^[@CR1]^. Because of this lack of options, the freedom of X-ray optical systems equivalent to that of camera lenses has narrowed; therefore, the methods available for X-ray measurement have been limited to date. As the types of available X-ray image sensors increase, the latitude of X-ray optical systems will increase accordingly and a variety of X-ray measurement methods will then be obtained. Scintillators show luminescence in the visible wavelength region when exposed to high-energy light, e.g., X-rays or γ-rays, and these devices are widely used as two-dimensional (2D) detectors in applications such as computed tomography equipment^[@CR2]^. In a 2D detector, the scintillator plate converts an X-ray image into a visible (VI) image, which can then be magnified using a VI microscope. In recent years, some scintillators have demonstrated the stimulated emission depletion (STED) phenomenon under UV light excitation, such as Tb:LSO scintillators, and the possibility of realising a super-resolving microscope using a scintillator has been proposed^[@CR3]^. If this STED phenomenon can be applied to detection of the luminescence image of the scintillator plate, the spatial resolution of the luminescence image will increase and exceed the Abbe diffraction limit^[@CR4]^. One scintillator, which is composed of a Lu~2~SiO~5~ crystal doped with cerium atoms (Ce:LSO), is a phosphor that exhibits luminescence over a wide energy range from the soft X-ray (SX) to hard X-ray (HX) regions^[@CR5]^. The Ce:LSO luminescence spectrum shows its maximum luminescence intensity at a photon energy of approximately 3.10 eV (*λ* = 400 nm) and shows a tail structure on the long-wavelength side from the peak^[@CR2]^. Because the Lu atoms in the LSO crystal are located at two different sites and the Ce atoms that are introduced randomly displace Lu atoms, the Ce atoms are thus also coordinated to the two different sites^[@CR1]^. The Ce^3+^ ions have two occupied Ce 4 *f* states: the ^2^*F*~5/2~ and ^2^*F*~7/2~ orbitals, as shown in Fig. [1](#Fig1){ref-type="fig"} ^[@CR6]^. The dominant transitions of the luminescence are from the lowest unoccupied Ce 5*d* state to the occupied Ce 4 *f* states. Therefore, the Ce:LSO luminescence spectrum shows four different luminescence peaks: the two peaks that originate from the site difference and the other two peaks that originate from the ionization levels of the Ce ions. In this paper, we demonstrate the STED phenomenon in the luminescence of Ce:LSO that has been excited by SX light and the area limits of the luminescence region using vector light at the weak luminescent wavelength of Ce:LSO.Figure 1Possibility of STED phenomenon in Ce:LSO when excited by SX light. In the LSO crystal, two Lu atom sites exist and are denoted by Lu(1) and Lu(2)^[@CR7]^. The Ce atoms that replace the Lu atoms at the Lu(1) and Lu(2) sites are also denoted by Ce(1) and Ce(2), respectively, in the figure. The spatial resolution of a STED microscope is represented by the modified Abbe diffraction limit^[@CR4]^. When the spatial resolution *δd*~*N*~ of an ordinary objective lens is given by the Abbe diffraction limit, *δd*~*N*~ = *λ*/2NA, where NA is the numerical aperture of the objective lens and the observation wavelength is *λ*, the spatial resolution *δd*~STED~ given by the modified Abbe diffraction limit is$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{\delta {d}_{{\rm{STED}}}}{\delta {d}_{N}}=\frac{1}{\sqrt{1+{I}_{{\rm{STED}}}/{I}_{{\rm{sat}}}}},$$\end{document}$$where *I*~sat~ is the fluorescent saturation intensity of the phosphor and *I*~STED~ is the light intensity of the stimulated emission light at the focal plane. Hereinafter, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${I}_{{\rm{STED}}}/{I}_{{\rm{sat}}}$$\end{document}$ will be referred to as the normalized laser intensity. The best reported spatial resolution for the STED microscope to date is approximately 6 nm, recorded when observing nitrogen impurities in diamond^[@CR4]^. Results {#Sec2} ======= Ce:LSO luminescence spectrum {#Sec3} ---------------------------- The luminescence spectra of Ce:LSO, when excited using 800 eV SX light, were measured at room temperature (RT) and at the liquid nitrogen temperature (LNT). The spectral shape obtained was similar to the previously reported spectral shapes^[@CR5]^. When the luminescence spectrum shape obtained at RT was compared with that of the spectrum obtained at the LNT, the two spectra showed almost the same profile, apart from the shoulder structure at the LNT in the figure, which was designated peak B. A difference between the sample temperatures caused the shoulder structure to disappear. Next, the spectra obtained were deconvoluted using Gaussian peaks. As described in the Introduction, the peak structure was expected to comprise four luminescence levels; therefore, the luminescence spectra were deconvoluted and separated according to expectations. The deconvoluted curves are distinguished by broken curves in Fig. [2](#Fig2){ref-type="fig"} and Table [1](#Tab1){ref-type="table"} summarises the parameters of the peaks.Figure 2Luminescence spectrum of Ce:LSO obtained by SX excitation at (**a**) LNT and (**b**) RT. Solid curves show the measurement results and broken curves represent the fitting results.Table 1Peak profiles of the deconvolution results in the luminescence spectra of Ce:LSO shown in Fig. [2](#Fig2){ref-type="fig"}.QuantityABCDEnergy \@RT (eV)3.13.02.92.7FWHM \@RT (eV)0.14 ± 0.010.20 ± 0.030.31 ± 0.030.55 ± 0.02Integrated Int. \@RT16.324.536.622.6Peak Int. \@RT0.50 ± 0.020.51 ± 0.070.50 ± 0.060.17 ± 0.02Energy \@LNT (eV)3.12.92.82.5FWHM \@LNT (eV)0.16 ± 0.010.25 ± 0.10.3 ± 0.30.50 ± 0.06Integrated Int. \@LNT36.131.620.611.8Peak Int. \@LNT0.84 ± 0.020.47 ± 0.50.27 ± 0.60.09 ± 0.06 Stimulated emission depletion by UV excitation {#Sec4} ---------------------------------------------- Over the range from 2.58 eV (480 nm) to 1.97 eV (630 nm), measurements were taken to verify whether or not the luminescence of the Ce:LSO was depleted and to determine the excitation photon energy of the STED phenomenon under SX excitation conditions. The occurrence of the depletion phenomenon was confirmed over the photon energy region from 2.58 eV to 2.25 eV. Because the STED light drawn as the green area in the figure was azimuthally-polarized vector light, there is no central light intensity in this spot. The blue area represents the luminescence from the Ce:LSO that was excited by the UV light. The excitation light illuminated the entire scintillator plane and thus the entire observed area emitted blue light, apart from the area illuminated using the STED laser light. Figure [3(b)](#Fig3){ref-type="fig"} shows the light intensity profiles of the luminescence and the STED light beams indicated by the line A-B in Fig. [3(a)](#Fig3){ref-type="fig"}. The maximum and minimum values are normalized with respect to the profile curves. The luminescence intensity varies in accordance with the increase and reduction of the STED light intensity and the excitation light and luminescence light intensity profiles have shapes that are the inverse of each other.Figure 3Luminescence image of UV excitation with the spot of the azimuthally polarized laser light. (**a**) Beam profiles at the broken line A--B. (**b**) Luminescence image of SX excitation (**c**), and the beam profiles at the broken line C--D. (**d**) In all figures, green colour represents the laser spot, and blue colour is the luminescence of the scintillator. In (**a**,**c**), "Reflection" means the reflection beam spot of the STED laser from the scintillator surface. Figure [4](#Fig4){ref-type="fig"} shows the changes in the spot diameter ratio, which is the spot size of the luminescence at the dark centre of the STED light spot when normalized with respect to the STED laser spot diameter. In the figure, points with bars represent the measurement results and the curves represent the fitting results from Eq. ([1](#Equ1){ref-type=""}). The spot diameter ratios decrease as the normalized laser intensity increases in all figures.Figure 4Normalized laser intensity dependences of the spot diameter ratio. The photon energies of the STED laser were (**a**) 2.58 eV (480 nm), (**b**) 2.43 eV (510 nm), and (**c**) 2.33 eV (532 nm), and the UV excitation light energy was 3.40 eV (365 nm). The photon energy of the STED laser was (**d**) 2.33 eV (532 nm), while that of the SX excitation light was 800 eV (1.55 nm). Circles with bars represent the measurement results with errors, and solid curves represent fitting results obtained using Eq. ([1](#Equ1){ref-type=""}) as a guide for the eye. The spot diameter of the STED light in the beam profile shown in Fig. [3(b)](#Fig3){ref-type="fig"} was approximately 9.2 μm, while that of the luminescence light from the scintillator observed through the dark spot was 2.8 μm. Although the STED light diameter remained constant, even if the STED light intensity was changed, the diameter of the small bright spot increased as the luminescence intensity increased. The luminescence spot diameter also varied with the STED wavelength. The luminescence spot size changed from 1 to 4 μm when the STED photon energy was 2.58 eV, but changed from 2 to 8 μm at a photon energy of 2.25 eV. We observed a tendency where the normalized laser intensity decreased in tandem with reduction of the STED photon energy. Stimulated emission depletion by SX light excitation {#Sec5} ---------------------------------------------------- The results in the previous section showed that the STED phenomenon occurred in the photon energy range from 2.58 eV (480 nm) to 2.25 eV (550 nm). We investigated the STED phenomenon in the SX-excited luminescence using the photon energy of 2.33 eV (532 nm) because of the ease with which suitable laser diodes (LDs) could be obtained. As a result, the same STED phenomenon and the same light intensity profile were obtained as in the UV excitation case, as shown in Fig. [3(c,d)](#Fig3){ref-type="fig"}. Because the spot size of the incident SX light was smaller than that of the UV light, the black area at the bottom of Fig. [3(c)](#Fig3){ref-type="fig"} represents the weakly irradiated area. In the figure, light reflection from the scintillator surface (opposite the side that was under X-ray irradiation) is observed and the spots are clearly separated. Figure [3(d)](#Fig3){ref-type="fig"} shows the results for the intensity profiles when evaluated in the same manner as Fig. [3(b)](#Fig3){ref-type="fig"}. The normalized laser intensity dependence of the spot diameter ratio was also evaluated in Fig. [4(d)](#Fig4){ref-type="fig"}. As in the UV excitation case, the spot size ratio for SX excitation decreased as the normalized laser intensity increased. Beam diameter and luminescence spot diameter {#Sec6} -------------------------------------------- The STED wavelength at which the smallest luminescence spot beam diameter was obtained in the experiment was 480 nm. The effective NA was 0.067 when taking the incident light beam diameter into consideration and the spot diameter was 4.4 μm at the Abbe diffraction limit predicted using this NA value. The actual STED light beam diameter was approximately equal to this expected value. The luminescence spot diameter was 1 μm, which was approximately 1/4 of the value determined from the Abbe diffraction conditions. By performing the same evaluation for the SX-excited STED, the luminescence spot diameter was shown to be approximately 1/10 of the spot diameter of the STED light. Discussion {#Sec7} ========== The diameter of the luminescence spot within which the luminescence area is limited by the STED phenomenon is affected by two parameters: the size of the luminescence region along the depth direction from the surface upon which the X-rays are incident and the depth of focus (DOF) of the objective lens. The luminescence region along the depth direction is considered to be almost equal to the X-ray penetration depth, and the X-ray penetration depth is dependent on the photon energy of that X-ray. These penetration depths are less than 1 μm in most materials when the photon energy of the incident X-ray is less than 1 keV. In the case of Ce:LSO, however, the penetration depth remains less than 1 μm when the photon energy is less than 3 keV^[@CR7]^. The DOF of the objective lens is expressed using Berek's formula^[@CR8]^. The objective lens used in this study has an NA of 0.2 and the observation wavelength of approximately 400 nm (Fig. [2](#Fig2){ref-type="fig"}); therefore, the value of the DOF is calculated to be 5.9 μm. This DOF is sufficiently longer than the attenuation length of Ce:LSO at the photon energy of 800 eV. In addition, the DOF of the objective lens used for observation of the Ce:LSO luminescence is 1 μm or less when the NA value is greater than 0.95, but most commercially available, non-immersion objective lenses have NA values of less than 0.95. In summary, as long as the STED luminescence of Ce:LSO is observed using a non-immersion lens in the photon energy region below 3 keV, the size of the STED luminescence point remains unaffected by the X-ray penetration depth. In the photon energy region above 3 keV, the thickness of the STED scintillator should be reduced to be below the penetration depth of the incident X-rays. The simplest microscopes that can be realized by combining scintillator luminescence with use of the STED phenomenon are used in radiography. The spatial resolution of a two-dimensional detector with the same configuration used in this research was reported to be 200 nm^[@CR9]^. By replacing this detector with a detector that uses a STED scintillator, the spatial resolution value can then be increased significantly. However, the light intensity detected by a pixel will decrease in inverse proportion to the square of the pixel size. Therefore, when it is assumed that the pixel size of the detector decreases to 1/10 of its previous size, approximately 100 times the previous measurement time or use of a strong X-ray light source such as an undulator will be required to acquire the same number of photons. An SX image in which the spatial resolution was evaluated as 3 μm at a photon energy of 91.8 eV would require a photon number of 4.8 × 10^10^ photons/mm^2^/s^7^. Therefore, an SX image with spatial resolution of 20 nm will require a photon number of 1.1 × 10^15^ photons/mm^2^/s or more, based on our previous results. In addition, the STED point must be scanned to obtain an X-ray image when using the STED scintillator. Because the scanning method takes the same measurement time for each of the considerable number of pixels, a multiple point scanning procedure will be required to reduce the measurement time. Use of a commercially available Nipkow disk with 1000 pinholes will allow the measurement time to be reduced to 1/1000 of the initial value^[@CR10]^. Because the Ce(1) and Ce(2) atoms at the Lu(1) and Lu(2) sites where the Lu atoms have been replaced by Ce atoms, respectively, are considered to have Oh symmetry with the crystal structure^[@CR11]^, the electronic structure of the Ce^3+^ ions splits into ^2^*E*~2g~ and ^2^*T*~2g~ in accordance with the crystal field splitting mechanism^[@CR12]^. The Ce 4 *f* orbitals also split into the ^2^*F*~7/2~ and ^2^*F*~5/2~ states in accordance with the spin-orbit splitting mechanism^[@CR13]^. The luminescence of the Ce atoms is caused by a dipolar transition from the lowest Ce 5*d* orbital to the Ce 4 *f* orbitals^[@CR14]^ and thus peaks A and B of the luminescence spectrum shown in Fig. [2(a,b)](#Fig2){ref-type="fig"} can be assigned to ^2^*F*~7/2~ and ^2^*F*~5/2~ of the Ce(1) site, respectively. Similarly, peaks C and D are assigned to ^2^*F*~7/2~ and ^2^*F*~5/2~ of the Ce(2) site, respectively. These deconvolved peaks represented by A -- D have different widths among their peaks and different split widths between the Ce(1) and Ce(2) sites (Table [1](#Tab1){ref-type="table"}). These differences may be caused by differences between the electronic structures of the Ce sites. The photon energy required to cause stimulated emission has the same value as the luminescence peak; therefore, the photon energy that is equal to the value at each peak position in Fig. [2](#Fig2){ref-type="fig"} is required to cause stimulated emission in Ce:LSO. Because these peaks overlap each other, as indicated in Fig. [2](#Fig2){ref-type="fig"}, the contributions of these transitions to the stimulated emission will depend on the STED laser's photon energy. The stimulated emission of Ce:LSO can thus be classified using the peak luminescence positions while considering their respective peak widths. When Ce:LSO is irradiated by light with photon energy of 2.58 eV, the transitions contribute to the stimulated emission at peaks C and D, i.e., to an overlap of the transitions Ce 5*d* → ^2^*F*~7/2~ and Ce 5*d* → ^2^*F*~5/2~ at the Ce(2) site. Similarly, in the case where stimulated emission is produced by light with photon energy of less than 2.50 eV, the transition is that of peak D, i.e., Ce 5*d* → Ce 4*f   *^2^*F*~5/2~, at the Ce(2) site. As the photon energy decreases, the stimulated emission intensity also decreases according to the peak luminescence intensity. The reason why the STED phenomenon was hardly observed under 1.97 eV light irradiation is believed to be that no *4 f* level was contributing to the luminescence. In summary, the number of transition levels that contribute to the stimulated emission decreases as the photon energy of the irradiation decreases. This decreasing number of transition levels may thus be the origin of the decreasing intensity of the normalized laser beam. Summary {#Sec8} ------- An investigation of whether the luminescence of a Ce:LSO crystal excited by UV light would be depleted or not when using photons from the energy region from 2.58 eV (480 nm) to 1.97 eV (630 nm) was performed. The normalized laser intensity at which the saturated luminescence intensity normalized the laser light intensity was shown to decrease as the photon energy decreased. In addition, the normalized spot diameter ratio that represents the spatial resolution of the STED microscope increased as the normalized laser light intensity decreased. The spot diameter ratio was obtained by normalizing the luminescence spot diameter of Ce:LSO using the spot diameter of the illumination light. This spot diameter ratio increases as the normalized laser light intensity decreases and this behaviour is explained well by the modified Abbe diffraction limit. The reduction in the normalized laser light intensity is considered to be dependent on the reduction in the density of states of the Ce*4f* electrons that contribute to the luminescence of Ce:LSO. The STED phenomenon was also observed in the luminescence of Ce:LSO when excited by soft X-rays. In this case, the normalized laser light intensity at SX excitation was smaller than the UV excitation value, which caused the spot diameter ratio of the luminescence to increase. These results show that a scintillator using the STED phenomenon can be applied as a high-resolution two-dimensional detector in the X-ray region. Methods {#Sec9} ======= To identify the Ce:LSO luminescence levels, the SX-excited luminescence spectrum was measured. It was then determined whether laser-induced emission depleted the Ce:LSO luminescence when the luminescence was excited using UV light. The STED phenomenon's wavelength dependence was also investigated by varying the laser wavelength. Finally, the Ce:LSO luminescence excited by soft X-rays was confirmed to be suppressed, as in the UV excitation case. Luminescence spectra of Ce:LSO {#Sec10} ------------------------------ The SX-excited luminescence spectrum was measured using optics evaluation beamline BL11D, Photon Factory, KEK^[@CR15]^. The experimental setup and method were the same as in our previous paper^[@CR7]^. Because the Photon Factory was operating in "Storage Mode," where the ring current decreased with time, the incident SX intensity decreased with time because it was proportional to the ring current. Because the luminescence spectra were normalized using the incident SX light intensity for comparison, incident SX intensities were measured before and after energy distribution curve (EDC) measurement of the luminescence intensity; these intensities were averaged as the SX intensity of *I*~0~(*E*). Finally, the luminescence intensity EDC, *I*~*m*~(*E*), was normalized using exposure time *t*, *I*~0~(*E*), the reflectance and transmittance of the optical elements, and the spectrometer quantum efficiency. The luminescence spectrum *I*~*L*~(*E*) was obtained^[@CR7]^. STED light source {#Sec11} ----------------- A diode-pumped solid-state (DPSS) laser diode (532 nm, 40 mW; DJ 532--40, Thorlabs) was used as the STED light source for SX excitation measurement. A high-brightness supercontinuum (SC) generator was used as the STED light source for 3.40 eV excitation measurement^[@CR16]^. The initial pulse was produced using a typical all-normal-dispersion (ANDi) mode-locked fibre oscillator^[@CR16],[@CR17]^. Output pulses were detected using a Pockels cell in a regenerative amplifier (RGA). The Yb:YAG thin-disk RGA output power was 0.3 W at a 1 kHz repetition rate. The 1 kHz pulse was compressed using a transmission grating compressor and chirped mirrors to 300 fs at a 1030 nm carrier wavelength. To produce the supercontinuum beam, a 500 fs pulse was propagated through a 2-cm-long SiO~2~ crystal column. Typical supercontinuum beam spectra ranged from 400 to 1000 nm, and an interferometric filter was used at 532 nm. The laser wavelength was confirmed using a spectrometer (Bluewave, Stellar Net) after passing through the vector optical system. The laser light intensity was adjusted using a power meter (PDA 200 C, Thorlabs) and measured using a microscope camera (DS-Fi1, Nikon) after passing through an optical filter. Optical system and vector beam converter {#Sec12} ---------------------------------------- Laser light was launched into the vector light system after being monochromatized using a wavelength filter. Figure [5](#Fig5){ref-type="fig"} shows the STED beam concept, with azimuth polarization reducing the geometric phase. The optical configuration for the STED beam with azimuth polarization and uniform phase, consisting of a vortex beam generator and a vectorial vortex beam generator, is shown in Fig. [5(a)](#Fig5){ref-type="fig"}. The vortex beam generator comprises two quarter-wave plates and a space-variant wave plate, which were fabricated as sub-wavelength structures on SiO~2~ plates. The vectorial vortex beam generator incorporates a quarter-wave plate, a radial polarizer and two half-wave plates as polarization rotators. Fig. [5(b)](#Fig5){ref-type="fig"} shows the electric fields of the generated beams, which were determined via Jones calculus at points "A" to "G" in Fig. [5(a)](#Fig5){ref-type="fig"}. The beam at points "A" and "B" has a Gaussian profile. After passing through the space-variant wave plate, a doughnut-shaped beam is generated using the left-hand vortex phase of −*iθ* at point "C." A left-hand circularly polarized beam is generated using the quarter-wave plate at point "E." The vortex phase is cancelled by the inverse vortex phase generated by the radial polarizer. The vector beam is radially polarized. The polarization rotator, which consists of two half-wave plates, converts it into an azimuthally polarized beam with uniform π/4 phase. The Jones vectors of the generated beams are shown in Fig. [5(c)](#Fig5){ref-type="fig"}. The phase factor of exp \[−*i*(*θ* − π/4)\] indicates another geometric phase caused by cyclical polarization element changes. After passing through the radial polarizer, the vector beam becomes radially polarized with uniform phase of exp (−*i*π/4). The phase factor of exp (−*iθ*) is cancelled by the inverse optical vortex. Finally, the polarization rotator is adjusted to be azimuthally polarized with uniform phase. Cyclic polarization changes caused by polarization elements shown in Fig. [5(a)](#Fig5){ref-type="fig"} generate a geometric phase, as shown in Fig. [5(d)](#Fig5){ref-type="fig"}. A Poincare sphere is used for geometric phase calculations. Polarization changes from "A" to "G" are described on the Poincare sphere. After passing through "C," the generated geometric phase is illustrated by the green area on the sphere. The geometric phase caused by the right-hand circularly polarized beam is 2π rad, because it is half of the sphere's surface area. However, the geometric phase caused by the left-hand circularly polarized beam is also 2π rad. The geometric phase illustrated as the blue area is generated via two rounds on yellow circuits. The two vortex phases cancel each other.Figure 5Concept of STED beam with azimuth polarization reducing the geometric phase. (**a**) Optical configuration of STED beam with azimuth polarization and uniform phase. (**b**) Electric field of the generated beams as determined using Jones calculus at points "A" to "G" in Fig. 5(a). (**c**) Jones vectors of generated beams. (**d**) Cyclic polarization changes caused by the polarization elements in Fig. 5(a) generate the geometric phase. A Poincare sphere is used for the geometric phase calculation. Sample preparation {#Sec13} ------------------ The commercial Ce:LSO sample was optically polished to 0.2 mm thickness and had dimensions of 5 × 5 mm^2^. SX beamline {#Sec14} ----------- The SX light source for SX-excited STED experiments was beamline BL11D, as used in the luminescence spectral measurements^[@CR7]^. Because the Photon Factory operated in "Top-Up Mode," the ring current was constant and the incident SX intensity remained constant. Therefore, no normalization was performed using the incident SX intensity in the STED experiment. The incident SX light wavelength resolution was *λ*/*δ*λ = 1000. A four-dimensional slit was installed in the beamline and adjusted to produce a sufficiently small luminescence point. Microscope system {#Sec15} ----------------- For STED light spot reduction and scintillator image enlargement, a self-made microscope system was used, as shown in Fig. [6(a,b)](#Fig6){ref-type="fig"}. The reflected STED light and the scintillator luminescence were separated from the sampled light using a dichroic mirror (DMSP 425, Thorlabs) and an optical filter (FGL 400, Thorlabs) inserted between an infinity-corrected objective (NA = 0.2; UV10X, Union Optical) and an imaging lens (UV Tube lens, Union Optical). The camera head (DS-Fi1, Nikon) for image detection has 2560 × 1920 pixels and deposited colour filters. The camera head colour filter operates in the blue (390--510 nm), green (480--600 nm), and red (580--680 nm) wavelength ranges when transmittance of 0.3 is the boundary condition. The camera head can separate the image colours and thus can separate the excitation light and the luminescence light if their colours are different. The intensities of the STED light and the scintillator luminescence were normalized using the measurement time and the number of pixels. The STED light beam diameter on entering the objective was 6 mm for 3.40 eV excitation and 4 mm for SX excitation. The effective NA of the objective was 0.067 under UV excitation and 0.044 under SX excitation.Figure 6Layout of optics used in (**a**) the microscope unit for UV excitation and (**b**) the microscope unit for SX excitation. Measurements of stimulated emission depletion in Ce:LSO {#Sec16} ------------------------------------------------------- This section describes the STED experiments on the scintillator using a supercontinuum laser and vector optics. The lasers and optics used are described elsewhere^[@CR18],[@CR19]^. An interference filter monochromatized the supercontinuum laser light. When the laser photon energy was adjusted to 2.43 eV (*λ* = 510 nm), Fig. [7(a)](#Fig7){ref-type="fig"} shows the light intensity spectrum measured via a spectrometer (Bluewave, Stellar Net). The peak photon energy observed was 2.43 eV (*λ* = 511 nm) and the width was 0.07 eV (FWHM), as determined by Gaussian peak fitting. The estimated photon energy error was ±0.1 eV (±2 nm).Figure 7(**a**) Supercontinuum laser light after monochromatization. (**b**) Emission spectrum of Ce:LSO after passing through the DM filter. Supercontinuum laser light with a 400--1000 nm wavelength width was introduced into the vector-light generation optics and evaluation showed an elliptical distribution and the major-axis direction distribution of the exit light. Figure [8](#Fig8){ref-type="fig"} shows evaluation of the azimuthally polarized beam performed via polarization analysis. The elliptical distribution of the measured light was ≤0.3 and the major-axis direction distribution showed that the light was azimuthally-polarized. The correlation percentage between measured and ideal values was ≥90%. Stokes parameters were measured by controlling the geometrical phase based on the presence or absence of light-vortex generation optics^[@CR18],[@CR19]^. Evaluation of the geometrical phase from the Stokes parameters showed a uniform phase and a clockwise eddy phase.Figure 8Evaluation of the generated azimuthally polarized beam via polarization analysis. (**a**,**c**) show the azimuth and ellipticity of the generated vector beam, respectively. The cross-sections at the broken lines in (**a**,**c**) are described in (**b**,**d**), respectively. For the azimuth and ellipticity, the geometric phase is determined as shown in (**e**). (**f**) Shows the cross-sectional profile of the obtained geometric phase. As a result, the generated vector beam is shown to have azimuthal polarization and uniform phase. The dichroic mirror described in section 2.2 can reflect light at wavelengths of 425 nm or more and guide it to the objective. The dichroic mirror and optical filter combination can transmit light with a 400--425 nm wavelength width and transmittance of ≥95% and transmits from the objective to the imaging lens. The light intensity of the 445--800 nm wavelength width is reduced to 1/50 or less by this combination. The scintillator light has two possible origins: the scintillator luminescence and reflection of the illuminating STED light from the scintillator. These two light beams have almost identical intensities after transmission using the dichroic mirror and optical filter combination and are separated into the STED light image and the luminescence image using the light filter on the camera head. The luminescence was measured via a spectrometer (Bluewave, Stellar Net) after the transmission, and Fig. [7(b)](#Fig7){ref-type="fig"} shows the spectrum. The luminescence light photon energy was 3.08 eV (*λ* = 402.8 nm) with a FWHM width of 0.17 eV (wavelength conversion value: 22.1 nm). **Publisher's note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. The measurements were performed with the approval of the Photon Factory Program Advisory Committee (Proposal No. 2018G072). The authors are deeply indebted to Mr. K. Anraku, Mr. R. Kaneda, Mr. S. Roy, and Mr. J. Tsuda of Utsunomiya University for their helpful technical support. The authors are particularly grateful to Prof. K. Mase and the staff at the Photon Factory, KEK for their assistance in performing the measurements, and also wish to express our gratitude to Mr. Y. Nakamura of the technical staff of IMRAM, Tohoku University for his assistance in sample preparation. This work was supported by JSPS KAKENHI (grant numbers 16H03902 and 17H190210). T.E. and T.W. conceived the concept for reduction of the vortex phase in the vector beams. T.E., T.W. and T.H. designed the experiments. N.S. and M.S. produced the laser system, S.C. generation system, and polarized beam conversion system. T.E., T.W. and T.H. performed the experiments and analysed the data. T.E., T.W., N.S., M.S., G.H. and T.H. discussed the results. T.E. supervised the project. T.E., T.W., N.S., M.S., G.H. and T.H. contributed to the writing of the paper. The data generated and analysed during the current study will be made available from the corresponding author on reasonable request. The authors declare no competing interests.

**Erratum to: Cancer Metastasis Rev** **DOI 10.1007/s10555-014-9531-3** The original version of this article unfortunately contained a mistake in the author group. Instead of "Robert A. Gatenby", it was captured as "Bob Gatenby". The correct author's name now appears above. The online version of the original article can be found at 10.1007/s10555-014-9531-3.

EBOV is a negative-stranded RNA virus, which belongs to the *Filoviridae* family and causes severe systemic disease in humans and non-human primates (NHP) with high case-fatality ratios. Epidemiology studies indicate that direct contact with infected body fluids is the main mode of transmission between humans[@b1][@b2], which points out at the skin and the mucosal surfaces as main portals of EBOV entry[@b2]. Previous studies have demonstrated replication of EBOV in macrophages and dendritic-like cells in NHP tissue sections early after infection[@b3][@b4], and suggested that macrophages and DCs were early virus targets. The notion that DC infection is an important event for EBOV pathogenesis has been further substantiated by the finding that EBOV infection impairs DC function *in vitro*[@b5][@b6]. Based on these previous findings and due to the migratory properties of DCs, the current hypothesis is that EBOV infection of DCs at the portals of virus entry leads to virus dissemination through the lymphatic system and blood[@b7][@b8]. However, this hypothesis has been difficult to test *in vivo* due to the lack of small animal models of EBOV infection. Inbred laboratory mice are completely resistant to infection with filoviruses and only mouse models with different degrees of immunosuppression are susceptible to infection with non-adapted EBOV[@b9]. This lack of immunocompetent animal models has precluded endpoint studies to elucidate the kinetics of EBOV infection *in vivo*. The last decade of research has generated important advances in our knowledge of DC biology. It is now clear that DCs comprise many subsets with overlapping and non-overlapping functions and with different ontogeny[@b10]. Classic DCs are characterized by their dependence on FMS-like tyrosine kinase 3 (Flt3) and its ligand and are formed by several subsets in peripheral as well as lymphoid tissues. Plasmacytoid DCs are also dependent on Flt3 but are morphologically and functionally distinct. Finally, inflammatory DCs are derived from activated monocytes that infiltrate tissues during inflammation or infection[@b11]. Thus, previous *in vitro* experiments of EBOV infection of monocyte-derived DCs do not reflect the variety of DC subsets in living organisms. Currently, it is not known whether EBOV is equally capable of infecting all DC subsets *in vivo*. Here we have investigated the kinetics of EBOV infection using chimeric mice that retained fully competent hematopoiesis. Upon mucosal exposure, EBOV productively infected resident macrophages and CD11b^+^ DCs in peripheral tissues and draining lymph nodes. Conversely, monocytes, neutrophils and cross-presenting tissue-resident CD103^+^ DCs were spared from infection. Inflammatory (monocyte-derived) CD11b^+^ DCs were preferred virus target cells at the onset of inflammation, after which both moDCs and conventional CD11b^+^ DCs supported EBOV infection. CD11b^+^ cells showed evidence of infection at both the mucosa and the draining lymph nodes. Depletion of T cells resulted in loss of control of local viral replication, virus dissemination and fatal Ebola virus disease (EVD). Our results identified infected CD11b^+^ DCs as putative viral vessels and underscored a key role of T cells as controllers of EBOV infection at the initial points of virus entry. Results ======= Hematopoietic immune competence is required to control EBOV dissemination ------------------------------------------------------------------------- Previous studies have highlighted the important role of type I interferon (IFN-I) on host resistance to EBOV infection, particularly in the mouse model[@b9][@b12]. Consistently, IFN-I receptor knockout mice (IFNAR^−/−^) are susceptible to non-adapted EBOV via several routes. To study the kinetics of EBOV infection in an immune competent environment we generated bone marrow chimeras via whole body irradiation of IFNAR^−/−^ mice and transplantation of WT bone marrow progenitor cells from C57BL/6 donor mice. In these mice, hereafter referred to as WT → IFNAR^−/−^, IFN-I deficiency is restricted to the radio-resistant compartment ([Fig. 1a](#f1){ref-type="fig"}). We hypothesized that partial IFN-I deficiency would allow EBOV replication to proceed and cause disease after which the hematopoietic-driven adaptive immune response would be essentially WT as we previously described for Lassa virus infection[@b13]. In fact, WT → IFNAR^−/−^ chimeras were susceptible to mucosal (intranasal) EBOV infection, displayed significant weight loss and survival rates of around 50% ([Fig. 1b](#f1){ref-type="fig"}). WT → WT control chimeras were entirely resistant to infection while IFNAR^−/−^ → IFNAR^−/−^ succumbed to EBOV within the first 10 days after infection. These controls reflected the features of WT and IFNAR^−/−^ mice respectively and indicated that the results observed were not due to transplantation ([Fig. 1b](#f1){ref-type="fig"}). Interestingly, WT → IFNAR^−/−^ chimeras did not show viremia ([Fig. 1c](#f1){ref-type="fig"}) and had significantly lower virus titers in most peripheral organs compared to IFNAR^−/−^ → IFNAR^−/−^ control chimeras ([Fig. 1d](#f1){ref-type="fig"}). These results strongly suggested that immune competence in the hematopoietic compartment was required to prevent virus dissemination from the initial point of virus entry. Kinetics of EBOV infection *in vivo* ------------------------------------ Our results indicated that loss of IFN-I responsiveness in the hematopoietic compartment resulted in virus dissemination suggesting a putative correlation between infection of IFN-deficient cells and virus spread in our model. To test this hypothesis we sought to determine the kinetics of EBOV infection after intranasal virus inoculation of WT → IFNAR^−/−^ chimeras as well as IFNAR^−/−^ → IFNAR^−/−^ mice. To track infected cells *in vivo* we utilized Alexa Fluor 488-conjugated anti-EBOV glycoprotein (GP) antibodies (clones 5D2 and 5E6)[@b14] in combination with multiparametric flow cytometry ([Supplementary Fig. S1](#S1){ref-type="supplementary-material"}). This strategy allowed identification of immune cell subsets productively infected with EBOV via detection of EBOV GP in the cell surface. Serial flow cytometric analysis of lung tissue from infected mice revealed that infection of alveolar resident macrophages and DCs were detectable via anti-GP staining at day 4 post-infection and was observable in both chimeras until the humane endpoint for IFNAR^−/−^ → IFNAR^−/−^ (day 9). Strikingly, the pattern of infection was not dependent on IFN-I competence but was restricted to DCs and macrophages ([Fig. 2a](#f2){ref-type="fig"}). We did not detect expression of surface GP in other leukocyte populations such as neutrophils, monocytes, T cells and B cells, as well as in CD45^−^ stromal cells. These findings suggested that these cell subsets were not infected productively with EBOV, even though it is possible that they support levels of viral replication below the detection limit of our technique ([Fig. 2a](#f2){ref-type="fig"} and [Supplementary Fig. S2](#S1){ref-type="supplementary-material"}). Tissue-resident DCs are composed at least by two subsets. CD103^+^ DCs express the C-type lectin langerin and play a major role cross-presenting viral antigens to CD8 T cells[@b15]. Conversely, CD11b^+^ DCs express DC-specific ICAM-3-grabbing non-integrin (DC-SIGN) and are heterogeneous, in particular during inflammation where this subset is formed by resident DCs as well as CD11b^+^ DCs derived from inflammatory monocytes[@b16]. In order to further determine the nature of EBOV-infected cells *in vivo*, we next sought to characterize the phenotype of virus-infected DCs. Flow cytometry analysis of infected DC populations allowed detection of surface EBOV GP in CD11b^+^ DCs but not in CD103^+^ DCs strongly suggesting that these cells were spared from infection, even though other possibilities such as non-productive infection cannot be ruled out ([Fig. 2b](#f2){ref-type="fig"}). In any case, productive EBOV infection was restricted to the CD11b^+^ compartment perhaps reflecting the important role of DC-SIGN as cell attachment factor for EBOV[@b17]. Although the kinetics of infection was similar in WT → IFNAR^−/−^ and IFNAR^−/−^ → IFNAR^−/−^chimeras, the frequency of infected CD11b^+^ DCs was higher in the total absence of IFN-I competence at early time points after infection (day 4) ([Fig. 2c and d](#f2){ref-type="fig"}). These results indicated that, while IFN-I did not influence the ability of EBOV to infect target cells, it restricted early viral replication. Due to the fact that the CD11b^+^ compartment is heterogeneous, we next sought to identify the nature of EBOV-infected cells within this population. Staining of cells with CD64 and MAR-1 allowed specific discrimination between inflammatory monocyte-derived DCs (MAR-1^+^ CD64^+^) and conventional resident CD11b^+^ DCs (MAR-1^low^ CD64^−^) as previously described[@b18] ([Supplementary Fig. S1](#S1){ref-type="supplementary-material"}). As shown for other models of acute inflammation, the peak of EBOV replication coincided with the detection of high numbers of moDCs in the lung, which outnumbered resident CD11b^+^ DCs within the CD11b^+^ cell compartment at days 7--9 post-infection ([Fig. 3a](#f3){ref-type="fig"}). At earlier time points (day 5) where the frequency of conventional CD11b^+^ DCs was significantly higher than that of moDCs (50% and 20% respectively) EBOV infection was detected to a higher extent in moDCs, suggesting that these cells were preferred early EBOV targets ([Fig. 3a--c](#f3){ref-type="fig"} and [Supplementary Fig. S3](#S1){ref-type="supplementary-material"}). Staining of tissue sections of mediastinal lymph nodes (mLNs) at the experimental endpoint (day 11) also revealed infection of CD11b^+^ cells ([Fig. 3c](#f3){ref-type="fig"}). These results suggest that infected cells in the lungs may carry infectious virus to the mLNs or that infection of lymphoid CD11b^+^ cells, for example due to passive drainage of infectious EBOV from the lung, also serves to amplify EBOV infection outside the initial replication site. Taken together, our data suggested that productive EBOV infection was restricted to the CD11b^+^ compartment and that infiltration of inflammatory monocytes and their differentiation into moDCs provided early viral targets that resulted in virus amplification. These findings are consistent with monocytes being refractory to EBOV infection and permissive only after their differentiation into moDCs[@b19]. T cells are required to prevent EBOV dissemination -------------------------------------------------- EVD is a systemic disease and the levels of viremia are strongly correlated with fatal outcome. Thus, we sought to elucidate the mechanisms by which EBOV disseminates from the initial infection site to the body. Our results pointed out at inflammatory moDCs as important EBOV targets at the mucosa, and suggested that cross-presenting CD103^+^ DCs could retain T cell priming capacity during infection. Therefore, we specifically addressed the role of moDCs and T cells on EBOV dissemination via specific depletion of these cell subsets *in vivo*. To block monocyte infiltration into the respiratory tract upon infection we engineered chimeric mice in which hematopoietic cells harbored mutations in the C-C chemokine receptor type 2 (CCR2). CCR2 is required for efficient egress of monocytes from the bone marrow[@b20] and for infiltration of monocytes into peripheral tissues during inflammation[@b21]. Perhaps surprisingly, blockade of monocyte infiltration did not result in reduction of EBOV replication in the lung ([Supplementary Fig. S4](#S1){ref-type="supplementary-material"}). These findings suggested that while moDCs are EBOV targets, other cells in the respiratory tract such as alveolar macrophages and resident CD11b^+^ DCs were sufficient to support EBOV replication. To deplete T cells *in vivo* we utilized intraperitoneal administration of monoclonal antibodies against CD8 and/or CD4 in WT → IFNAR^−/−^ chimeras and compared the effect of specific T cell depletion in these mice with those treated with isotype control antibody. Individual depletion of CD4 and CD8 T cells resulted in moderate increase of viremia, but did not significantly influence survival and morbidity. However, depletion of both CD4 and CD8 T cells completely abolished protection and resulted in uniformly lethal EVD ([Fig. 4a](#f4){ref-type="fig"}). In addition, T cell depletion resulted in viremia and virus replication in peripheral organs ([Fig. 4b and c](#f4){ref-type="fig"}). Furthermore, depletion of CD8 T cells alone or in combination with CD4 T cell depletion resulted in significant increase of EBOV replication in the lungs ([Fig. 4d](#f4){ref-type="fig"}). These results indicated that T cell immunity was essentially required to control local EBOV replication and to prevent systemic virus dissemination. Discussion ========== Previous research has shown that EBOV infects DCs and macrophages derived *in vitro* from hematopoietic progenitors which was consistent with early studies showing the presence of EBOV virions in dendritic-like cells in tissue sections[@b3][@b4][@b5][@b6]. However, DCs derived *in vitro* with GM-CSF only mimic features of inflammatory monocyte-derived DCs but not of conventional Ftl-3-dependent DCs[@b22][@b23]. Thus, it is not possible to infer from these previous studies whether EBOV infects all DCs or only discrete subsets. Due to the fact that DC subsets have non-overlapping functions, this question is of key importance to understand the pathophysiology of EVD. Here we describe the kinetics of EBOV infection *in vivo* after mucosal exposure, a presumed natural route of infection. We show that the only populations infected in the respiratory mucosa were alveolar macrophages, conventional CD11b^+^ DCs and inflammatory moDCs. Surface EBOV GP staining was not detected in monocytes and neutrophils *in vivo*, in agreement with these cells being refractory and non-productively infected respectively[@b19][@b24]. We also failed to detect EBOV infection in any other cell subset, including non-hematopoietic CD45^−^ cells as well as in T and B lymphocytes. While it is known that T cells are spared from infection, other cell types including fibroblasts, endothelial and epithelial cells have been described as EBOV targets[@b2]. Indeed, since virus titers grow steadily in the lung for several days before viremia, there must be one or several cell types that support initial EBOV replication. These could be resident antigen-presenting cells such as CD11b^+^ DCs and macrophages, or other cells where the infection levels are below our detection limit. Additional research technology, which may be difficult to implement in BSL4 containment (i.e. individual cell sorting), may be necessary to address the early kinetics of EBOV infection *in vivo*. The fact that inflammatory moDCs but not monocytes supported productive viral replication is noteworthy and points out to an important role of inflammation for EBOV infection and pathogenesis. Our results support the finding that while EBOV attaches to the surface of monocytes it can only enter these cells during their differentiation to moDCs which depends on the inflammatory environment[@b19]. However, abrogation of monocyte entry into the infected tissue in the CCR2^−/−^ model did not prevent virus dissemination and disease, which indicated that despite moDCs being preferred early EBOV targets, ferrying of infectious viruses by other migratory populations such as conventional CD11b^+^ DCs, or draining of free virus through the lymphatic system, may be sufficient for viral spread from the infection site to the body. Interestingly, one of the main migratory DC subsets, the CD103^+^ DCs does not seem to play a role on EBOV dissemination since we did not detect expression of viral GP in the surface of CD103^+^ DCs. Interestingly, there are substantial parallelisms between our study and a recent HIV report which indicated that productive infection of DC-SIGN^+^ but not langerin^+^ DCs by HIV depended on the C-type lectin receptor usage[@b25]. Since CD103^+^ DCs express langerin[@b26] but not DC-SIGN, it seems worthwhile to determine whether EBOV can actually infect CD103^+^ DCs or rather whether the virus cannot replicate in these cells. CD103^+^ DCs are essential to cross-prime CD8 T cells, and thus, it would be important to assess the functionality of these cells *in vivo* both in animal models and the CD141^+^ human equivalent[@b27]. Our findings point out at CD11b^+^ DCs (bona-fide and monocyte-derived) as putative vessels for EBOV dissemination from the initial point of entry and replication to the draining lymph nodes. This hypothesis is substantiated by the mobility of these cells and previous studies showing their capacity to transport pathogens to the lymphoid tissues[@b11][@b16][@b19]. We cannot rule out however, that other mechanisms such as drainage of free virus through the lymphatic system may be responsible, at least to some extent, for EBOV dissemination. Many studies have shown that, in particular in the mouse model, susceptibility is associated to a great extent to the capacity of the mice to mount IFN-I dependent antiviral responses[@b9]. Consistently, IFNAR^−/−^ → IFNAR^−/−^ control chimeras reproduced the 100% lethality observed in IFNAR^−/−^ infected with non-adapted EBOV[@b9][@b12]. IFNAR^−/−^ → IFNAR^−/−^ control chimeras also displayed significantly higher levels of early virus replication in the lung than their WT → IFNAR^−/−^ counterparts, suggesting that IFN-I-dependent control of early EBOV replication is a key determinant of disease outcome. However, strikingly the pattern of EBOV infection did not depend on IFN-I competence. This was different from previous studies with, for example, influenza virus where the capacity of DCs to respond to IFN-I was determinant for infection of these cells *in vivo*[@b28]. These findings suggest that the susceptibility of IFNAR^−/−^ and STAT1^−/−^ mice to EBOV depends not only on the effects of IFN-I the coordination of in innate immune responses[@b12][@b29], but also on the effect of this cytokine in adaptive immunity. This is also consistent with our observation that non-adapted EBOV replicated in the lungs of WT mice. It is conceivable that attachment factors such as C-type lectin receptors may determine EBOV tropism after which the action of virus-encoded IFN-I antagonists may be sufficient to ensure virus replication in target cells[@b29][@b30][@b31]. In line with the importance of adaptive immunity for EBOV pathogenesis, we demonstrated that T cells are chief controllers of EBOV dissemination. As demonstrated before in other model systems including NHPs, both CD4 and CD8 T cells were required for protection in agreement with the idea that EBOV survival requires both arms of the adaptive immune response, namely, cytotoxic T cells and antiviral antibodies[@b2][@b3][@b4]. The importance of T cell function for protection is also consistent with recent findings from our group which indicated that fatal EVD was related with dysregulation of T cell homeostasis in humans[@b32]. Our results point out at T cell immunity as an important correlate of protection against EVD, and suggest that harnessing of T cell function via immunotherapy merits further investigation. Methods ======= Mice and bone marrow chimeras ----------------------------- IFNAR^−/−^ mice (C57Bl/6 background) were obtained from the Friedrich Loeffler Institute, Isle of Riems, Germany and bred in the animal facility of the Bernhard Nocht Institute for Tropical Medicine. CD45.1^+^ congenic C57Bl/6-Ly5.1 and CCR2^−/−^ mice (C57Bl/6 background) were purchased from Jackson Laboratories and bred in the animal facility of the Heinrich Pette Institute. Bone marrow chimeras were generated at the Heinrich Pette Institute. Four to eight weeks old female recipient mice were irradiated with a lethal dose (2 × 7 Gray, 2 hr apart) in a Cs^137^ irradiator and then transplanted with 3 × 10^6^ bone marrow cells from donor mice. Engraftment of donor cells was analyzed in peripheral blood four weeks post-transplantation. Bone marrow chimeric mice with an engraftment of 85% and higher were utilized for the experiments. Experimental EBOV infection --------------------------- This study was carried out in strict accordance with the recommendations of the German Society for Laboratory Animal Science and under the supervision of a veterinarian. The protocol was approved by the Committee on the Ethics of Animal Experiments of the City of Hamburg (permit no. 125/12). All efforts were made to minimize the number of animals used for experiments and to alleviate suffering of animals during experimental procedures. All staff carrying out animal experiments passed an education and training program according to category B or C of the Federation of European Laboratory Animal Science Associations. All experimental infections described in this study were performed within the biosafety level 4 (BSL4) facility at the Bernhard Nocht Institute for Tropical Medicine in Hamburg in accordance with institutional safety guidelines. Personnel wore appropriate protective equipment (biosafety suits). Ebola virus H.sapiens-tc/COD/1976/Yambuku-Mayinga was used for all infection experiments. Mice were anesthetized with isoflurane and infected intranasally (i.n.) with 1000 focus forming units (FFU) of EBOV in 50 μl PBS and monitored daily for signs of disease. Body weight and rectal body temperature were measured daily and blood (30--50 μl) was drawn via the tail vein every 2--7 days over a period of three weeks for clinical chemistry and viremia. Animals with severe symptoms such as temperature \<28 °C or weight loss \>20% were euthanized with isoflurane anesthesia followed by cervical dislocation. T cell depletion ---------------- CD4 and CD8 T cells were depleted via intraperitoneal administration of anti-CD4 (YTS191, BioXcell) and anti-CD8 (YTS169.4, BioXcell) depleting antibodies three and one day prior to infection (total of 300 μg per application). A control group received an isotype control antibody (LFT-2, BioXcell) at the same time and dose. Flow cytometry -------------- For flow cytometry analysis lungs of infected animals were harvested and digested with Collagenase D (2 mg/ml, Roche) and DNAseI (50 μg/ml, Sigma). Single cell suspensions were treated with BD Pharm Lysing Buffer (BD Bioscience); cells were blocked with CD16/CD32 Fc-Block antibody followed by staining with an antibody cocktail. All fluorochrome-conjugated antibodies were purchased from BD Biosciences, eBiosciences or BioLegend. Monoclonal antibodies against EBOV GP (5D2 and 5E6) have been previously described[@b14]. A mix of equal parts 5D2 and 5E6 was conjugated with Alexa Fluor 488 using an antibody labeling kit (life technologies). Formaldehyde inactivated samples were aquired using a LSRFortessa (BD Bioscience) and analysed with FlowJo Software (FlowJo, LLC). Virus Titrations ---------------- Organs were homogenized in 1 ml of DMEM containing 2% FCS using Lysing Matrix D tubes (MP Biomedical) and a FastPrep homogenizer. To quantify infectious particles in supernatants of homogenized organs and blood samples a focus formation assay was utilized as described elsewhere. Clinical chemistry ------------------ To determine the levels of aspartate aminotransferases (AST) serum was diluted 1:10 or higher with a 0.9% saline solution and measured with a commercial kit from Roche and a Reflotron. The normal range for mice was determined in 20 unifected mice and was 40--60 U/l. Statistical analysis -------------------- Statistical analyses were done using Graphpad Prism 5 software. Differences in organ titers and infection rates of immune cells were analyzed using non-parametric statistics (Kruskal-Wallis test followed by Dunn's post-test). The levels of significance were represented as follows: ns (not significant) when *p* \> 0.05, \*(*p* ≤ 0.05), \*\*(*p* ≤ 0.01) and \*\*\*(*p* ≤ 0.001). Immunofluorescence analysis --------------------------- Tissues were fixed in 4% buffered formalin and processed for paraffin embedding. For the detection of viral protein, mounted sections (2 μm) were dewaxed and antigen retrieval was performed for 30 min at 96 °C in 10 mM citrate buffer pH 6.0. Sections were washed once and blocked in blocking buffer (Protein-Free T20 (TBS) Blocking buffer \#37071 Thermo Fischer) for 1 h. Anti-Ebola antibodies (5E6 and 5D2) and respective isotype control antibody (mouse IgG2) were both conjugated to Alexa488 according to the manufacturer's protocol (Molecular Probes). Directly conjugated EBOV-Alexa488 or isotype-control-Alexa488 antibody (1:40 in blocking buffer each), respectively, was applied on the sections over 3 nights at 4 °C with gentle agitation. Afterward, sections were intensively washed and anti-CD11b antibody (1:2,000 in blocking buffer; \#ab133357 Abcam) was applied overnight at 4 °C. Sections were washed again, followed by incubation with Alexa555-conjugated secondary anti-rabbit antibody for 1.5 h at room temperature. After repeated washing, sections were mounted with DAPI-Fluoromount-G (SouthernBiotech, Birmingham, USA). Data acquisition was performed using a Leica Sp5 confocal microscope and Leica application suite software (LAS-AF-lite). Additional Information ====================== **How to cite this article:** Lüdtke, A. *et al*. Ebola virus infection kinetics in chimeric mice reveal a key role of T cells as barriers for virus dissemination. *Sci. Rep.* **7**, 43776; doi: 10.1038/srep43776 (2017). **Publisher\'s note:** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Supplementary Material {#S1} ====================== ###### Supplementary Information We thank the animal caretakers at the Heinrich Pette Institute and the Bernhard Nocht Institute in Hamburg for excellent support. We also thank Kristin Hartmann from the University Medical Center Hamburg-Eppendorf for technical support. This work was supported with funds from the EBOKON program of the German Center for Infection Research (DZIF) and partially supported by DZIF project TTU 01.702_00 (to CMF and SG). This work was also partially supported by Public Health Agency of Canada and Canadian Safety and Security Program (CSSP CP1017) to GPK and XQ. AL is a recipient of a predoctoral fellowship from the Leibniz Graduate School. The Heinrich-Pette-Institute is financed by the German Federal Ministry of Health and the state of Hamburg. The authors declare no competing financial interests. **Author Contributions** A.L. performed all the experiments in this study, prepared the figures and contributed to study design with C.M.F., P.R., D.M.W., E.P., S.W., S.B. and S.G.M. contributed to mouse monitoring and data gathering in the BSL4 laboratory. X.Q. and G.P.K. generated monoclonal antibodies, helped with experimental design and edited the manuscript. E.R. generated bone marrow chimeras. S.K. performed histology analyses. S.G., J.I. and L.O. designed experiments and edited the manuscript. C.M.F. designed the study and wrote the manuscript. {#f1} {#f2} {ref-type="supplementary-material"}. Discrimination of moCD11b^+^ DCs and coCD11b^+^ DCs was achieved using the markers MAR-1 and CD64. EBOV infection of moCD11b^+^ DCs and coCD11b^+^ DCs was determined using monoclonal anti-GP antibodies. Representative plots of infected CD11b^+^ DC subsets at day 5 are depicted in (**b**). Kinetics of infection of both DC subsets over time. Numbers represent the frequencies of infected cells within either co or moCD11b^+^ DCs. Statistics was assessed via Kruskal-Wallis test followed by Dunn's post-test. ns (not significant) when *p* \> 0.05, \*(*p* ≤ 0.05), \*\*(*p* ≤ 0.01) and \*\*\*(*p* ≤ 0.001). (**c**) Mediastinal lymph node section of an EBOV infected IFNAR−/− mouse was stained with AF488-conjugated mononclonal anti-GP antibodies and anti-CD11b antibody followed by a AF555-conjugated secondary anti-rabbit antibody. Two individual stainings of the same lymph node harvested 11 days post-infection are shown as well as staining of EBOV infected lymph nodes with AF488-conjugated isotype control antibody and staining of a mock mouse with AF488-conjugated mononclonal anti-GP antibodies (**d**).](srep43776-f3){#f3} {#f4}

Introduction ============ There is intense interest in the use of photoelectrochemical (PEC) systems for the production of solar fuels. Amongst the more promising photoanodes for PEC water oxidation is hematite (α-Fe~2~O~3~), a non-toxic, abundant, low-cost, and relatively inert material. The bandgap and band energies of α-Fe~2~O~3~ (∼2.1 eV) lead to a maximum theoretical solar-to-hydrogen (STH) efficiency of ∼15%,[@cit1] however actual achieved STH efficiencies of α-Fe~2~O~3~ photoelectrodes are substantially below this value and are typically ∼1--2%.[@cit2] This has been proposed to be due to multiple limiting factors, including; poor conductivity,[@cit3] short electron--hole lifetimes,[@cit4] slow oxygen evolution reaction kinetics[@cit5] and a low visible light absorption coefficient coupled to a short hole diffusion length (2--4 nm).[@cit6] Numerous approaches to improving the activity of α-Fe~2~O~3~ have been explored. Higher photocurrent densities have been achieved through nanostructuring which has the aim of increasing the concentration of charges generated close to the semiconductor liquid junction (SCLJ) to overcome the short hole diffusion length.[@cit7] In such nanostructured electrodes, dopants are known to be critical for photoelectrochemical activity with un-doped α-Fe~2~O~3~ electrodes being electrically insulating[@cit7] and often photoelectrochemically inactive.[@cit8] A wide variety of extrinsic dopants have now been explored including Si,[@cit8]--[@cit10] Sn,[@cit11]--[@cit13] Ti,[@cit13]--[@cit15] Pt [@cit16],[@cit17] and these lead to both enhanced long range charge transport and the formation of a sufficient electric field for charge separation at the SCLJ within the nanostructured domains.[@cit8] The introduction of oxygen vacancies (V~O~) has also recently been shown to be an effective approach to improve the activity of α-Fe~2~O~3~ photoelectrodes.[@cit18]--[@cit20] A report by Li *et al.* on the decomposition of β-FeOOH nanowires in an oxygen deficient atmosphere demonstrated the highly active nature of oxygen deficient hematite (α-Fe~2~O~3--*x*~) photoelectrodes for water oxidation, with a photocurrent of 3.4 mA cm^--2^ being achieved under 100 mW cm^--2^, and in this paper we examine the factors behind the high activity of these same electrodes.[@cit19] The enhanced activity, ease of inclusion of V~O~, ability to process at temperatures as low as 350 °C [@cit21] and reports of a synergistic effect of extrinsic dopants with intrinsic V~O~ sites[@cit20],[@cit22] has led to a surge of interest in the controlled inclusion of V~O~ in hematite over the last 2 years. Given the potential utility of this approach it is important that the fundamental processes associated with V~O~ inclusion are elucidated.[@cit23]--[@cit26] The higher incident photon to current efficiency (IPCE) of α-Fe~2~O~3--*x*~ (64% at 1.50 V~RHE~) compared to the air annealed sample (α-Fe~2~O~3~, 0.57% at 1.50 V~RHE~) in the report of Li *et al.* correlated with the presence of Fe^2+^ sites as measured by XPS and a large increase in the measured donor density (*N*~d~).[@cit19] It has been known for over 25 years that the inclusions of V~O~/Fe^2+^ sites within α-Fe~2~O~3~ leads to formation of a donor band ∼80 meV below the conduction band.[@cit27] The inclusion of V~O~\'s is suggested to improve activity through a number of mechanisms including; improved charge transport, higher charge separation yields and decreased contact resistances at the semiconductor/transparent conducting oxide interface, with various studies citing differing contributions from each effect.[@cit13],[@cit21],[@cit23]--[@cit26] However, to date research has concentrated on material development and few direct measurements of the actual mechanism of enhanced activity for α-Fe~2~O~3--*x*~ are reported. A further complication is that although large increases in photocurrent have been achieved through the inclusion of V~O~, the onset potentials remain relatively high for α-Fe~2~O~3--*x*~, typically 1.0 V~RHE~ or greater[@cit19],[@cit21] compared to as low as ∼0.6 V~RHE~ for an ALD α-Fe~2~O~3~ electrode following a high temperature (800 °C) treatment, leading to typical solar to fuels efficiencies for oxygen deficient hematite that are significantly lower than other state of the art hematites.[@cit28],[@cit29] It has been proposed by several authors that a possible cause of the high onset potentials in α-Fe~2~O~3--*x*~ is the increased concentration of surface defect states.[@cit30],[@cit31] Surface defects in hematite are known to lead to Fermi level pinning and increased levels of trap-mediated recombination, and the selective passivation of such states in stoichiometric α-Fe~2~O~3~ has proven to be a highly effective approach to improving photoelectrode activity.[@cit30] Surface passivation treatments on extrinsically doped photoelectrodes have included high temperature annealing steps and overlayer depositions.[@cit29],[@cit32],[@cit33] The hypothesised deleterious role of surface states in oxygen deficient hematite is apparently contradicted by a recent study on hydrogen treated α-Fe~2~O~3~ which reported improved linear sweep photocurrents (recorded at 50 mV s^--1^) for samples only containing Fe^2+^ defect states at the surface, compared to photoelectrodes containing both bulk and surface V~O~.[@cit24] It is therefore apparent that currently no consensus exists regarding the mechanism of operation of oxygen deficient hematite. In order to systematically address the high onset potential of α-Fe~2~O~3--*x*~ samples an improved understanding of the role of both bulk and surface V~O~ sites and their effects on the charge carrier kinetics is required. Transient absorption (TA) spectroscopy and transient photocurrent (TPC) measurements offer a route to directly measure the effect of material modifications and treatments on the yield and dynamics[@cit34],[@cit35] of photogenerated charges within a PEC cell.[@cit36] Previous TA spectroscopic studies of α-Fe~2~O~3~ photoelectrodes have examined numerous aspects of the photophysics and chemistry of extrinsically doped α-Fe~2~O~3~, including the role of co-catalysts on hole kinetics,[@cit31],[@cit32],[@cit37],[@cit38] and the effect of bias on charge trapping and recombination, from the fs--ms timescale.[@cit39],[@cit40] Of particular significance has been the realization that a key requirement for water splitting is the ability to accumulate very long-lived holes at the SCLJ, with apparent rate constants for water splitting ranging from 0.1--6 s^--1^,[@cit41],[@cit42] correlating with a measured significant thermal barrier to hole transfer.[@cit43] The slow hole transfer enables recombination between bulk electrons and surface accumulated holes on the millisecond timescale in Si--Fe~2~O~3~ at \<1.2 V~RHE~ which significantly lowers photoelectrochemical activity. Here we report the first TA study of a α-Fe~2~O~3--*x*~ photoelectrode in a PEC cell, with the aim of identifying the key design rules required to develop more efficient defect rich photoanodes. We have chosen to study samples prepared as originally reported by some of us[@cit19] as they remain amongst the most active oxygen deficient electrodes reported, with current densities reaching *ca.* 4 mA cm^--2^ under 1 sun illumination. Our detailed kinetic analysis allows to (i) elucidate the observed bias dependent activity of α-Fe~2~O~3--*x*~, yielding a complete mechanistic model of the operation of oxygen deficient electrodes and (ii) to provide critical mechanistic insights into the role of ALD Al~2~O~3~ overlayers. ALD Al~2~O~3~ is found to be particularly effective in shifting the photocurrent onset potential of α-Fe~2~O~3--*x*~ (∼0.2 V), representing a significant enhancement in photoelectrochemical water splitting efficiency. Results and discussion ====================== For this study fresh α-Fe~2~O~3~ (air annealed) and α-Fe~2~O~3--*x*~ (oxygen deficient) films were prepared as previously described, details can be found in the ESI.[†](#fn1){ref-type="fn"} [@cit19] The photoelectrochemical activities of both α-Fe~2~O~3~ and α-Fe~2~O~3--*x*~ were assessed through linear sweep voltammograms measured in 1 M NaOH using a portion of the output of a 75 W Xe lamp (*ca.* 0.1 sun) scanning at 5 mV s^--1^. We employ this low power light source in our mechanistic study as it allows us to measure the photoelectrochemical response of the sample inside the transient spectrometer using the same cell as in the TA measurements. In line with previous reports we observe a strong photocurrent from α-Fe~2~O~3--*x*~ assigned to photoelectrochemical water oxidation, [Fig. 1](#fig1){ref-type="fig"}.[@cit41],[@cit44] It is known that IPCE values for these α-Fe~2~O~3--*x*~ films exceed 60% at 1.5 V~RHE~, leading to photocurrents of *ca.* 4 mA cm^--2^ under 100 mW cm^--2^, however a key concern remains the high onset potential,[@cit19] observed here at *ca.* 1.15 V~RHE~, [Fig. 1](#fig1){ref-type="fig"}. In contrast to α-Fe~2~O~3--*x*~, the air annealed (550 °C) α-Fe~2~O~3~ sample shows no significant photocurrent (0.6--1.5 V~RHE~). The lack of activity of this control α-Fe~2~O~3~ sample, chosen due to its similar morphology, preparation route and light harvesting properties, indicates that the presence of V~O~ is critical for enabling photoelectrochemical water splitting in these otherwise un-doped films. {#fig1} In order to rationalize the observed photoelectrochemical response of the hematite samples we have measured TA spectra following UV excitation (355 nm, 6 ns, 100 μJ cm^--2^) at a range of applied biases, [Fig. 2](#fig2){ref-type="fig"}. The excitation energy employed leads to photo-generated carrier densities several orders of magnitude lower than the calculated oxygen vacancy density (ESI, S7[†](#fn1){ref-type="fn"}). A broad positive transient absorption is measured at wavelengths greater than 600 nm in all of the spectra. In line with numerous past TA studies on Si-doped and Nb-doped α-Fe~2~O~3~ photoelectrodes,[@cit40],[@cit43],[@cit45] the TA feature at *λ* \> 600 nm is assigned to photoholes in α-Fe~2~O~3~ (see Fig. S4[†](#fn1){ref-type="fn"} for a schematic explanation) and this assignment is further affirmed in TA experiments employing a hole scavenger (H~2~O~2~, *vide infra*) which show a decreased signal at *λ* \> 600 nm. The bias dependent TA spectra also contain a bleach (decrease in optical density) centred at *ca.* 580 nm which is seen to increase in magnitude with applied bias. It has been proposed that the feature at this wavelength is due to localized states close to the band edge, possibly related to the presence of oxygen vacancies.[@cit27],[@cit40],[@cit46] Under positive bias the trap state occupancy is lowered, enabling the promotion of a valence band electron to the vacant trap state upon absorption of a visible photon (580 nm). Using both fs and μs TA measurements, Pendlebury *et al.* have shown that following UV excitation of α-Fe~2~O~3~ held at potentials significantly positive of the flat band potential, rapid photoelectron trapping can occur leading to a bleaching of this 580 nm feature, (see Fig. S4[†](#fn1){ref-type="fn"}).[@cit39],[@cit40] We also assign the bleach at ∼580 nm to photoelectron trapping at localized states which is primarily occurring on the sub-microsecond timescale. {#fig2} Initially we concentrate our study on the yield and kinetics of photoholes in α-Fe~2~O~3~ and α-Fe~2~O~3--*x*~. At all potentials investigated we note an increased yield of photoholes in Fe~2~O~3--*x*~ at the earliest timescales studied (2 μs), [Fig. 2](#fig2){ref-type="fig"}. The increased photohole yield at 2 μs may indicate more efficient initial charge separation in the oxygen deficient hematite, however a detailed study of the photohole kinetics at a single potential (1.4 V~RHE~) shows only a ∼35% difference in photohole yield at 2 μs (assuming a similar extinction coefficient for both materials), [Fig. 3](#fig3){ref-type="fig"}. This suggests that a previously proposed enhancement[@cit19],[@cit38] in initial charge separation yield in V~O~ rich materials with higher electron densities is not likely to be a significant factor in rationalizing the differences in activity of α-Fe~2~O~3~ and α-Fe~2~O~3--*x*~. Instead, of greater significance is the rate of photohole decay in α-Fe~2~O~3~ (*t*~50%~ = 0.27 ms) and α-Fe~2~O~3--*x*~ (*t*~50%~ = 1.20 ms) at 1.4 V~RHE~.[@cit47] Previous studies have indicated that the improved conductivity of α-Fe~2~O~3--*x*~ would be expected to aid both initial charge separation and electron transport/extraction which reduce bulk electron--hole recombination and thereby increase the photohole lifetime.[@cit21],[@cit48] {#fig3} In order to asses if such a change in photoelectron dynamics is the cause of the improved activity of α-Fe~2~O~3--*x*~ we have also measured the transient photocurrent (TPC) between the hematite working electrodes and the counter electrode following UV excitation. At 1.4 V~RHE~ the TPC traces of both α-Fe~2~O~3~ and α-Fe~2~O~3--*x*~ are relatively similar at early times (\<0.5 ms) indicating that, contrary to expectations, initial photoelectron extraction to the external circuit can occur effectively even in the un-doped α-Fe~2~O~3~ sample.[@cit19] Instead we find that the difference in photohole kinetics and photoelectrochemical activity is due to a slow back electron transfer from the external circuit into the air annealed α-Fe~2~O~3~, which we see as a negative current on the timescale of 1--10 ms after photon absorption, [Fig. 4](#fig4){ref-type="fig"}. For α-Fe~2~O~3~ samples the total charge re-injected from the external circuit approximately matches that initially extracted from the α-Fe~2~O~3~ photoelectrode at all potentials studied (0.8--1.4 V), leading to minimal net photocurrent in our transient study, in line with the steady state photocurrent measurements, [Fig. 4](#fig4){ref-type="fig"} and S5.[†](#fn1){ref-type="fn"} {ref-type="fn"}) and (b) the converted total charge passed following UV excitation (355 nm, 6 ns).](c5sc00423c-f4){#fig4} An overlay of the kinetics of charge reinjection (TPC) and hole decay (TA) on α-Fe~2~O~3~ (*τ* = 10 ms at 1.4 V~RHE~, [Fig. 4](#fig4){ref-type="fig"} inset and Fig. S6[†](#fn1){ref-type="fn"}) shows an excellent agreement indicating that bulk electrons are recombining with the accumulated photoholes measured in the TA experiment, leading to reinjection of electrons from the external circuit. Experiments in the presence of H~2~O~2~ in the following section allow us to assign this recombination to be occurring with surface trapped holes. This is in agreement with a very recent study on Si--Fe~2~O~3~ where TPC measurements showed the presence of a recombination process between surface trapped holes and bulk α-Fe~2~O~3~ electrons on the milliseconds timescale, that led to a flow of electrons back into the photoelectrode from the external circuit.[@cit41] In contrast, with α-Fe~2~O~3--*x*~ electrodes we do not observe any slow re-injection of electrons into the film, Fig. S5.[†](#fn1){ref-type="fn"} The lack of back electron transfer into α-Fe~2~O~3--*x*~ electrodes can be rationalized by Mott--Schottky measurements of the two films (Fig. S7[†](#fn1){ref-type="fn"}) that show a higher donor density (*N*~d~) in the V~O~ rich photoelectrode (α-Fe~2~O~3--*x*~*N*~d~ = 1.2 × 10^20^ cm^--3^, α-Fe~2~O~3~*N*~d~ = 6.7 × 10^19^ cm^--3^) with flat-band potentials of 0.4 V~RHE~ (α-Fe~2~O~3--*x*~) and 0.34 V~RHE~ (α-Fe~2~O~3~).[@cit19] Whilst absolute values of donor densities for a nanostructured film obtained through a Mott--Schottky analysis should be treated with caution, the relative change between these two samples with similar morphologies (Fig. S3[†](#fn1){ref-type="fn"}) is significant. This two-fold increase in *N*~d~ in the α-Fe~2~O~3--*x*~ electrode significantly decreases the width of the space charge layer (*W*~sc~), Fig. S7[†](#fn1){ref-type="fn"} for illustrative calculations. Our TAS studies confirm that greater localised band bending in the surface region of α-Fe~2~O~3--*x*~ is a significant factor in rationalising the enhanced activity of α-Fe~2~O~3--*x*~, as hypothesised in several previous studies.[@cit19],[@cit21],[@cit30],[@cit31],[@cit38] However it is not found to be due to more efficient initial charge separation or injection kinetics but instead the blocking of the back flow of electrons from the bulk towards the SCLJ preventing recombination with surface accumulated photoholes. A simple model of the kinetic competition between the bias dependent back electron--hole recombination pathway and the rate of water oxidation has been shown elsewhere to account for observed photoelectrochemical activity of Si--Fe~2~O~3~ electrodes without the need for the inclusion of inter-band trap states.[@cit41] Specifically the photocurrent onset potential of a Si-doped α-Fe~2~O~3~ photoelectrode, with a similar *N*~d~ (∼10^20^ cm^3^) to the α-Fe~2~O~3--*x*~ samples examined here (1.2 × 10^20^ cm^--3^), correlated with the potential at which back electron--hole recombination became slow enough to enable surface hole accumulation and water oxidation to occur (*ca.* 1.0 V~RHE~). Here we find that back electron--hole recombination in α-Fe~2~O~3--*x*~ is blocked at potentials as low as 0.8 V~RHE~ (Fig. S5[†](#fn1){ref-type="fn"}) and no correlation is noted with the photocurrent onset potential, *ca.* 1.15 V~RHE~ ([Fig. 1](#fig1){ref-type="fig"}). This gives rise to an intriguing question, given that our TPC and TA measurements show that holes can be accumulated at potentials as low as 0.8 V~RHE~ and that the slow back electron--hole recombination pathway has been prevented, why is the photocurrent onset potential for water oxidation so positive (1.15 V~RHE~) with α-Fe~2~O~3--*x*~ photoelectrodes? Site of photohole accumulation in α-Fe~2~O~3--*x*~ -------------------------------------------------- It is therefore important to identify if the photoholes measured in α-Fe~2~O~3--*x*~ using TA spectroscopy on the μs--ms timescale are accumulating at or close to the SCLJ, as hole trapping elsewhere in this defect rich material may account for the high photocurrent onset potential. In order to distinguish these two cases we explore the response of the hole dynamics to the presence of a hole scavenger. Hydrogen peroxide is a commonly used hole scavenger due to its near unity efficiency for the removal of holes in α-Fe~2~O~3~ that are at or close to the SCLJ.[@cit49] In the presence of 0.5 M H~2~O~2~ we observe a photocurrent at potentials as low as 0.7 V~RHE~ on both α-Fe~2~O~3~ and α-Fe~2~O~3--*x*~ confirming that (i) initial charge separation is effective in both materials at potentials well below the water oxidation photocurrent onset potential and (ii) that in both samples the photoholes are able to reach the surface to participate in oxidation reactions, Fig. S8.[†](#fn1){ref-type="fn"} TA experiments in the presence of H~2~O~2~ show a decrease in the hole yield at the earliest time-scales studied (2 μs) in Fe~2~O~3--*x*~ when compared to experiments in the absence of a hole scavenger. This indicates that holes are present at the SCLJ and are transferring into solution at time scales earlier than studied here, providing a lower limit for the lifetime of transport and accumulation of holes at the SCLJ (∼2 μs), [Fig. 5](#fig5){ref-type="fig"}. {#fig5} Role of trap state mediated recombination ----------------------------------------- We now turn to the potential role of electron trap states in rationalizing the behaviour of α-Fe~2~O~3--*x*~ as it has been proposed that oxygen vacancies may act as a "mixed blessing"[@cit38] with the enhanced electrical properties being balanced with a potential increase in trap-mediated electron--hole recombination at the defect sites introduced.[@cit30] Following absorption of a UV photon we initially observe a bleaching at 580 nm for α-Fe~2~O~3--*x*~ which is assigned to photoelectron trapping at this inter-band state, [Fig. 6(b)](#fig6){ref-type="fig"}.[@cit39] The slow recovery on the millisecond timescale of the trap state feature is due to the subsequent de-trapping of the photoelectrons, as shown by the dynamics in [Fig. 6(b)](#fig6){ref-type="fig"}. In line with this assignment, the change in occupancy of the 580 nm trap state at 1.4 V~RHE~ is well fitted to the integrated rate equation for an intermediate species in a consecutive reaction scheme (A → B → C) ([Fig. 6(b)](#fig6){ref-type="fig"}), eqn (1).where the rate constants *k*~trap~ and *k*~detrap~ correspond to the rate of photoelectron trapping and de-trapping (*ca.* 7 × 10^5^ s^--1^ and 40 s^--1^ respectively), *β* a stretching exponent and \[*A*\]~0~ a pre-exponential factor related to the initial yield of photoelectrons. Photoelectron trapping has been shown to occur on the ps--μs timescales and it is likely we have only fitted the tail of the trapping process here.[@cit39] {ref-type="fn"}](c5sc00423c-f6){#fig6} In previous studies the de-trapping rate of photoelectrons in extrinsically doped hematite correlated to the TPC decay rate as well as a distinct fast decay component in the hole population, leading to an assignment of both electron extraction and electron--hole recombination following de-trapping in the bulk of the electrode.[@cit40] In contrast, on α-Fe~2~O~3--*x*~ a sample with a high concentration of surface defect sites[@cit21],[@cit50] it is proposed that a significant level of electron trapping occurs at the surface, closer to the site of photohole accumulation leading to higher electron--hole recombination losses upon detrapping. We are able to assign electron--hole recombination as the fate of the de-trapped electrons due to the lack of correlation between the TPC decay rate and the rate of recovery of the 580 nm signal ([Fig. 4](#fig4){ref-type="fig"}, [6](#fig6){ref-type="fig"} and [7(b)](#fig7){ref-type="fig"}), which indicates that the electrons are unable to reach the external circuit. In the absence of photoelectron extraction it is expected that electron--hole recombination will occur. We can also conclude that a significant level of recombination with trap state electrons occurs close to the SCLJ as in the previous section TA experiments demonstrated that the a large portion of the photohole population reached the surface of α-Fe~2~O~3--*x*~ within the time resolution of our experiment (2 μs), *i.e.* prior to photoelectron detrapping. ![(a) Photocurrent of α-Fe~2~O~3--*x*~ before (black) and after deposition of an ALD Al~2~O~3~ layer (1 nm, red trace) recorded using a low power 75 W Xe lamp, dark traces are shown as dashed lines. (b) Overlay of the TA trace recorded at 580 nm, assigned to an interband trap state, and the normalized charge extracted for (b) α-Fe~2~O~3--*x*~ and (c) α-Fe~2~O~3--*x*~/Al~2~O~3~ following UV excitation (355 nm) at 1.4 V~RHE~. All traces are recorded in 1 M NaOH.[@cit51]](c5sc00423c-f7){#fig7} We further confirm the presence of a significant level of trap-mediated recombination in the oxygen deficient hematite through a detailed kinetic analysis of the photoholes. In the absence of a recombination process between surface trapped holes and bulk α-Fe~2~O~3~ electrons the photohole TA kinetics of α-Fe~2~O~3--*x*~ at 1.4 V~RHE~ would consist of three primary kinetic processes on the μs--s timescale corresponding to fast (μs) electron--hole recombination in the bulk, surface electron--hole recombination and hole transfer into solution. An excellent fit of the TA signal of α-Fe~2~O~3--*x*~ at 700 nm can indeed be achieved using a triphasic stretched exponential function, with the rate of bulk electron--hole recombination corresponding to the decay of the integrated TPC signal at the same potential (*k*~bulk~ ∼ 2 × 10^3^ s^--1^), the surface recombination rate constant matching the rate constant for photoelectron de-trapping (*k*~detrap~ ∼ 40 s^--1^), confirming the occurrence of trap-mediated recombination at the SCLJ, and a hole transfer rate constant into solution of *k*~WO~ ∼ 3.5 s^--1^ (full fitting parameters are in the ESI[†](#fn1){ref-type="fn"}), [Fig. 6(a)](#fig6){ref-type="fig"}. The assignment of the slowest kinetic phase of the hole decay is in line with previously measured rate constants for water oxidation on α-Fe~2~O~3~ [@cit40] which have ranged from ∼0.2--6 s^--1^ [@cit42] and is further supported by a plot of the yield of very long-lived photoholes (at 200 ms) *versus* the applied bias which strongly correlates with the measured photocurrent response of α-Fe~2~O~3--*x*~, (Fig. S9[†](#fn1){ref-type="fn"}). Our analysis is also further supported by fitting of the hole trace in [Fig. 5](#fig5){ref-type="fig"}, which is well fitted with a biphasic stretched exponential function, with *k*~bulk~ and *k*~detrap~ present, but *k*~WO~ absent, confirming the nature of this slowest kinetic component, (ESI Fig. S12[†](#fn1){ref-type="fn"}). We are able to now construct a full kinetic scheme for the key steps for water oxidation on α-Fe~2~O~3--*x*~, ([Scheme 1](#sch1){ref-type="fig"}). Interestingly the amplitude of the fitting components, *i.e.* the separate populations of holes that are decaying by each pathway, is approximately the same for trap-mediated recombination and hole transfer into solution even at 1.4 V~RHE~ demonstrating that a high level of trap mediated recombination is a critical factor limiting the efficiency of oxygen deficient hematite. {#sch1} In light of our TA studies that highlight the role of α-Fe~2~O~3--*x*~ surface states in mediating recombination, and show that the presence of oxygen vacancies also leads to a higher donor concentration and a narrower depletion layer suppressing the back electron injection, we are able to propose a route to both improve the activity of oxygen deficient hematite\'s and to test the mechanistic model proposed in [Scheme 1](#sch1){ref-type="fig"}. It is known that the preparation of α-Fe~2~O~3--*x*~, by thermal decomposition of β-FeOOH in an oxygen deficient atmosphere,[@cit19] leads to the formation of V~O~ both throughout α-Fe~2~O~3--*x*~ and at the surface.[@cit24] Removal of the states solely on the Fe~2~O~3--*x*~ surface by passivation, whilst maintaining a suitably high concentration of V~O~ both within the bulk and close to the SCLJ may be anticipated to be a route to obtaining both the desired improved *N*~d~ whilst lowering the level of trap-mediated recombination. Previously, passivation of surface states of α-Fe~2~O~3~ has been achieved through the use of high temperature annealing steps,[@cit29] the deposition of catalytic species including CoPi, IrO~*x*~, NiFeO~*x*~ [@cit32],[@cit37] and the use of inert metal oxides such as Al~2~O~3~ and Ga~2~O~3~ overlayers grown by atomic layer deposition (ALD).[@cit31],[@cit33],[@cit38] The deposition of thin Al~2~O~3~ layers appear particularly promising as the low temperatures (120 °C) required and the reduced oxygen pressures for ALD deposition are expected to limit the loss of bulk oxygen vacancies in α-Fe~2~O~3--*x*~. ALD Al~2~O~3~ layers have been shown to decrease the photocurrent onset potentials of Si-doped α-Fe~2~O~3~ electrodes by ∼100 mV,[@cit9] with a decrease in the surface capacitance of the Al~2~O~3~ coated α-Fe~2~O~3~ electrodes also reported,[@cit52],[@cit53] indicating the passivation of surface trap states. Here we use ALD to form a 1 nm Al~2~O~3~ layer. This thickness is chosen as previous experiments with Si--Fe~2~O~3~ [@cit9] have indicated that it provides reasonable coverage and stability for the duration of an experiment whilst remaining thin enough to allow for hole tunnelling transfer from the Fe~2~O~3~ to water. We find that the addition of an ALD layer on α-Fe~2~O~3--*x*~ leads to a large cathodic shift in the photocurrent onset potential (∼200 mV, 0.95 V~RHE~) and an overall increase in the magnitude of the photocurrent, [Fig. 7(a)](#fig7){ref-type="fig"} (150 W Xe lamp). Al~2~O~3~ was also deposited onto an air annealed α-Fe~2~O~3~ sample (Fig. S10[†](#fn1){ref-type="fn"}) however photocurrent measurements showed no improvement in activity confirming that sub-surface oxygen vacancies are required for photoelectrochemical activity in otherwise un-doped samples. To confirm that the improved photoelectrochemical activity of the α-Fe~2~O~3--*x*~/Al~2~O~3~ sample is related to the hypothesised decrease in trap mediated surface electron--hole recombination, we have also measured the TA kinetics of the 580 nm trap states at 1.4 V~RHE~ following surface treatment, [Fig. 7(b)](#fig7){ref-type="fig"}. To the best of our knowledge this report represents the first TA study of the role of an ALD Al~2~O~3~ passivation treatment on hematite. A large bleaching of the TA signal is observed at 580 nm even after the Al~2~O~3~ treatment showing that a significant concentration of trap states remains within the bulk of the α-Fe~2~O~3--*x*~/Al~2~O~3~ sample. Critically we also note a near four-fold increase in the rate of de-trapping with the α-Fe~2~O~3--*x*~/Al~2~O~3~ sample (*k*~detrap~ ∼ 150 s^--1^) indicating that photoelectrons in the ALD treated sample are trapped at energetically shallower sites. In contrast to the untreated α-Fe~2~O~3--*x*~ sample, where de-trapping leads to significant levels of recombination with surface trapped holes, we now see an excellent agreement between the recovery of the TA trap state band at 580 nm and the rate of charge extraction as measured by TPC for α-Fe~2~O~3--*x*~/Al~2~O~3~, [Fig. 7(b and c)](#fig7){ref-type="fig"}, further supported by the observation of an increase in the yield of very long-lived photoholes in the presence of the Al~2~O~3~ overlayer (Fig. S11[†](#fn1){ref-type="fn"}). This leads to the conclusion that detrapped electrons are reaching the external circuit in appreciable quantities. We assign the change in kinetics of the transient photocurrent and trap state occupancy to the passivation of solely the surface trap states. In the absence of a significant concentration of surface trap states, photoelectrons are able to be trapped at interband trap states in the α-Fe~2~O~3--*x*~ bulk, which are spatially separated from the population of surface accumulated holes, limiting trap-mediated recombination and enabling electron transport to the external circuit, thus lowering the photocurrent onset potential. Conclusions =========== The annealing of hematite in oxygen deficient atmospheres is being increasingly explored as an approach to improving the electrical properties of the photoelectrode, however conflicting reports exist regarding the mechanism of enhancement.[@cit40],[@cit52] In contrast to previously proposed mechanisms which have often indicated improved charge transport as being a significant factor,[@cit18] we find that the primary effect of the introduction of oxygen vacancies is to block the slow recombination of bulk electrons with surface accumulated holes, the so called "back electron recombination" pathway at even moderate applied biases (0.8 V~RHE~). A key target of mechanistic research is to provide design rules for rational material development and here we achieve that goal, by both proposing and verifying a modification to address the very anodic photocurrent onset potential of α-Fe~2~O~3--*x*~. We have investigated the effect of an ALD surface passivation treatment on both the photoelectrochemistry of α-Fe~2~O~3--*x*~ leading to validation of our mechanistic model and an improvement in the photocurrent onset potential of 0.2 V. Furthermore we report, to the best of our knowledge, the first TA measurements on ALD Al~2~O~3~ coated hematite providing important insights into this widely used surface treatment. Supplementary Material ====================== Supplementary information ###### Click here for additional data file. MF thanks the University of Liverpool for a GTA award. AJC gratefully acknowledges a fellowship from the EPSRC (EP/K006851/1). [^1]: †Electronic supplementary information (ESI) available: Full experimental details and sample characterization data can be found in the ESI. See DOI: [10.1039/c5sc00423c](10.1039/c5sc00423c)

{#sp1 .323}

The portals of entry for organisms responsible for most infections which dominate medicine in tropical countries (as elsewhere) are the skin, and respiratory and intestinal tracts. A very high proportion of infections of warm climes originates from ingestion of contaminated water and foodstuffs; many resultant diseases therefore fall into the subspecialty tropical gastroenterology.[@bib1], [@bib2], [@bib3] Most gastroenterological emergencies which occur in a temperate climate also occur in tropical and subtropical countries. However, there are notable differences in prevalence.[@bib4] Some are probably ethnically related (although elimination of environmental factors is often difficult), but the majority are superimposed upon an underlying communicable (infective) disease; important examples are ileal perforation or haemorrhage resulting from typhoid (enteric) fever, colonic perforation -- and far less often haemorrhage -- in amoebic colitis and shigellosis, and hepatic 'abscess' in invasive amoebiasis.[@bib4] MOUTH AND PHARYNX {#cesec1} ================= The mouth and rectum are the most accessible parts of the gastrointestinal tract from a clinical viewpoint;[@bib5] therefore, where endoscopic procedures are impossible (and that applies to many tropical and subtropical countries), as much information as possible should be derived from careful examination of these organs. Viral, bacterial, mycotic and parasitic infections all give rise to oropharyngeal pathology, which is frequently most pronounced in the presence of associated malnutrition (especially in infants and children). Herpes simplex virus, Epstein--Barr virus (EBV) (see Chapter 43) and many enteroviruses can produce a stomatitis; oral ulceration is also a frequent manifestation of Behçet\'s syndrome-common in the Middle East and Japan. Lassa fever (see Chapter 42) and diphtheria (see Chapter 67) are frequently characterized by severe pharyngeal involvement, and in rabies (see Chapter 44) dysphagia caused by spasm of the pharyngeal muscles is an important feature of the disease. In addition to acute bacterial infections, tuberculosis, leprosy, syphilis and yaws all exert oral manifestations. Candidiasis (exceedingly common in the acquired immune deficiency syndrome, AIDS) (Chapter 20), histoplasmosis, South American blastomycosis and coccidioidomycosis can also produce buccal lesions. Acute pharyngitis caused by infection with young adult *Fasciola hepatica* (ingested in raw sheep or goat liver -- reported from the Middle East and India -- and known locally as 'halzoun\'; Chapter 83) is caused by pentastomids.[@bib1] Therapeutic agents, such as sulfonamides (included in some antimalarial prophylactics, e.g. pyrimethamine + sulfamethoxazole, 'Fansidar\') can give rise to the Stevens--Johnson syndrome, in which oral ulceration is common. Manifestations of specific malnutrition states (vitamin B and C deficits, and iron deficiency anaemia) are usually obvious, whereas in kwashiorkor, these are frequently combined with infective complications. Cancrum oris is a gangrenous condition involving the gums and cheeks and is associated with *Borrelia vincentii* and *Fusiformis fusiformis* infection; it is especially common in malnourished children,[@bib1] especially in West Africa. Descriptions of the mouth, especially the tongue, in post-infective malabsorption (tropical sprue) (see below) were dominant in clinical accounts of this disease in the nineteenth century (i.e. before the advent of laboratory investigation). Periodontal disease and dental caries are also a major problem in tropical countries.[@bib1] Oral submucous fibrosis -- a chronic disease of unknown aetiology -- may affect any part of the oral cavity;[@bib1] most reports are from the Indian subcontinent and Southeast Asia. Fibroelastosis of the submucous tissues, accompanied by epithelial atrophy, are important sequelae and are probably premalignant. Of malignant disease(s), buccal carcinoma is pre-eminent;[@bib5] Burkitt\'s lymphoma, ameloblastoma and nasopharyngeal carcinoma (Chapter 35) are other malignancies that have important geographical distributions in tropical countries. Hypertrophy of the salivary glands is common in malnourished children; it can also be associated with *Ascaris lumbricoides* infection and chronic calcific pancreatitis (see below).[@bib1] Tumours of the salivary glands are probably no more common than in temperate regions. OESOPHAGUS {#cesec2} ========== The most important disease to involve this organ is oesophageal carcinoma[@bib6] ([Figure 10.1](#f1){ref-type="fig"} ) (Chapter 35); this malignancy possesses an enigmatic geographical distribution. It has a high prevalence in certain geographical locations:[@bib1], [@bib6] central and east-Africa (western Kenya, Malawi and eastern Zambia have the highest rates), the southern Caspian littoral (especially north-eastern Iran) and northern China (in and around the Taihang mountains). Various hypotheses have been advanced to explain the high incidence of this tumour in these areas (Chapter 35).[@bib1], [@bib6] Figure 10.1Barium swallow showing oesophageal carcinoma with gross mediastinal invasion. Megaoesophagus, a feature of chronic *Trypanosoma cruzi* infection (Chagas\' disease), is described in Chapter 76. [Table 10.1](#cetable1){ref-type="table"} lists some major causes of dysphagia in a tropical environment.Table 10.1Some causes of dysphagia in tropical countriesTraumaGastritisForeign bodiesCorrosive agentsInfectionSouth American trypanosomiasis (Chagas\' disease)Candidiasis (usually associated with AIDS)*Rhizopus, Absidia* (mucormycosis)NeoplasiaOesophageal carcinomaOesophagealMacronodular cirrhosis (usually varices, postviral)SchistosomiasisPortal vein thrombosisHyperreactive malarious splenomegalyOthersAchalasiaPeptic oesophagitisHiatus herniaExtrinsic pressureEndemic goitre Oesophageal varices ([Figure 10.2](#f2){ref-type="fig"} ) usually result from advanced macronodular cirrhosis (see below); however, hepatic schistosomiasis (caused by *Schistosoma mansoni*, *Schist. japonicum*, *Schist. intercalatum*, *Schist. matthei* and *Schist. mekongi*) are also important (Chapter 82). Portal vein obstruction (see below) is also common in some parts of Africa and Asia; this probably results in most cases (although more research is required) from umbilical sepsis in the neonatal period;[@bib1] it is occasionally a sequel to hepatocellular carcinoma. A very high splenic blood flow associated with hyperreactive malarious splenomegaly (HMS; tropical splenomegaly syndrome) can also give rise to oesophageal varices (see below).[@bib1] Where and when available, upper gastrointestinal endoscopic sclerotherapy is of enormous value in the management of oesophageal varices, but an ideal method of dealing with bleeding varices has yet to appear; in most tropical countries, older methods (see below) remain extant.Figure 10.2Advanced oesophageal varices in a Zambian woman with severe macronodular cirrhosis associated with HBV infection; barium swallow examination. Oesophageal trauma is a major problem in several African countries; foreign bodies (e.g. kola nuts and fish bones) and corrosive agents -- which give rise to strictures -- are also relatively common.[@bib1] Achalasia, peptic oesophagitis and hiatus hernia are all encountered, but are not unduly common. In HIV/AIDS infection, oesophageal candidiasis is a common manifestation; other systemic mycoses (Chapter 71) can also produce an oesophagitis. Emergencies {#cesec3} ----------- The most common oesophageal lesions in tropical countries are varices ([Table 10.1](#cetable1){ref-type="table"} summarizes the major causes) and carcinoma (see above);[@bib4] resultant acute complications are upper gastrointestinal haemorrhage and obstruction, respectively. Hookworm and *Ascaris lumbricoides* infections (Chapter 85) should not be neglected in this context.[@bib7] Of lesser importance, foreign bodies in the oesophagus (e.g. kola nuts) can cause dysphagia; corrosive lesions can result in stricture formation.[@bib1] ### Oesophageal varices {#cesec4} Reported prevalence of bleeding oesophageal varices in tropical countries is unreliable.[@bib4] Transport facilities are usually exceedingly unsatisfactory; therefore, the majority of those afflicted die before reaching medical care. Also, high technology (e.g. endoscopic sclerotherapy) and blood transfusion are less often available; outcome following medical intervention is therefore frequently less satisfactory than in a Western country.[@bib8] The cause of upper gastrointestinal bleeding in 131 successive patients admitted with haematemesis or melaena to a hospital at Harare, Zimbabwe, has been analysed;[@bib9] in 36 (27%) admissions (mean age 42 years) oesophageal varices were responsible. In 21, conservative management was followed by cessation of bleeding; however, nine suffered continuous bleeding, and six re-bleeding; five patients died (four within 24 h of admission) from haemorrhagic shock. Vasopressin infusions were used in four with the addition of oesophageal tamponade in two. The pathophysiological mechanisms underlying oesophageal bleeding have been addressed on numerous occasions.[@bib10] Both erosive and eruptive bases seem the most likely explanations; in addition, pressure and variceal size are probably important. In Egypt, endoscopic biopsies obtained from intervariceal mucosa (within 5 cm of the cardia) in 20 individuals with, and 30 without, a history of variceal bleeding (most suffered from schistosomal liver disease) were examined histologically;[@bib11] they showed dilated intraepithelial blood-filled channels within the squamous epithelium and lamina propria in all of the 'bleeders' and in 15 (50%) of the 'non-bleeders\'. Furthermore, oesophagitis was more pronounced in the bleeders compared with the non-bleeders: 11 (55%) and 7 (23%), respectively. The role of upper gastrointestinal endoscopy in a developing country has been studied in Kuwait;[@bib12] 345 (4%) of 8680 patients examined successively using this technique had evidence of oesophageal varices, the usual cause being chronic schistosomal liver disease (usually in Egyptian labourers). By examining 718 successive patients who presented with upper gastrointestinal bleeding within 24 h of admission, the exact site of the haemorrhage was delineated in 651 (91%), and the responsible lesion detected in 685 (97%). At Ibadan, Nigeria, a recent study has indicated that endoscopy gives a superior result to radiology in the diagnosis of variceal disease, resulting in upper gastrointestinal haemorrhage;[@bib13] endoscopy was successful in 64 (85%), but a barium meal correctly located the source of bleeding in only 38 (51%) of 75 patients. Three reports from New Delhi, India, focused on the role of endoscopic sclerotherapy in the management of bleeding oesophageal varices.[@bib14], [@bib15], [@bib16] A total of 79 patients underwent treatment (with either absolute or 50% alcohol) every 3 weeks, for oesophageal varices; active bleeding was controlled in 14 of 15 (93%) and 5 of 13 (54%) using the two fluids, respectively (*p*\< 0.05); the sole disadvantage of absolute alcohol was that it produced a higher incidence of retrosternal pain. In another study, using a similar regimen, 5% ethanolamine oleate was compared with absolute alcohol in 47 randomly allocated patients; the latter solution eradicated oesophageal varices earlier (12.9 vs 8.2 weeks, respectively) (*p*\< 0.001); the mean number of injection courses and necessary amount of sclerosant were also lower in the alcohol-treated group (*p*\< 0.001), but the frequency of re-bleeding did not differ significantly (*p*\> 0.05). A total of 31 children with variceal bleeding caused by extrahepatic portal vein obstruction (19), non-cirrhotic portal fibrosis (5) or cirrhosis (7) were treated by sclerotherapy using absolute alcohol; arrest of acute bleeding was achieved in 10 by emergency sclerotherapy, and a 3-week schedule was able to achieve variceal obliteration in all of them. During a 23-month follow-up period, recurrent varices occurred in three (two with cirrhosis and one with non-cirrhotic portal fibrosis) patients; a re-bleed was successfully controlled with emergency sclerotherapy in five, and an oesophageal stricture in four of them (which was easily dilated) which were the only significant complications. Although now rarely used in the Western world, oesophageal compression using a Sengstaken tube is often the only technique available. Intravenous pitressin is of limited value in acute bleeding. In long-term management, propranolol undoubtedly has a place in a developing country scenario. In an attempt to provide clinical guidelines for the prediction of outcome of upper gastrointestinal bleeding in a developing country, Clamp et al.[@bib8] carried out a multicentre study based on two centres, in Sikkim and China; in the former country, 60 (69%) of the patients put into the 'high-risk' group (by applying Bayes\' theorem using a computer system) for re-bleeding experienced this event (27 (54%) died), whereas this complication occurred in only six (2%) in the 'low-risk' group; furthermore, a simplified scoring system (little computer technology was available at Sikkim) gave almost exactly the same predictive accuracy. The authors suggest that, by using one of these systems, patients in remote areas can be categorized in order that scarce resources (which are available there) can be put to the best use. The optimal means of managing haemorrhage resulting from extrahepatic portal venous obstruction is summarized in the section on liver disease (see below). ### Oesophageal carcinoma {#cesec5} Presence of histologically diagnosed chronic oesophagitis (using upper gastrointestinal endoscopy) has been shown to be common in a high-risk population (15--26 years) in China.[@bib17] This lesion was significantly associated with: (1) consumption of 'burning hot' beverages; (2) a family history of oesophageal carcinoma (including second-degree relatives); (3) infrequent consumption of fresh fruit; and (4) infrequent consumption of dietary staples, other than maize. Associated factors which have been recorded in that population include: (1) positive cytological smears (568 individuals \>30 years of 42 190 had a positive result); and (2) a high prevalence of pharyngeal carcinoma in free-range chickens, which lived off domestic scraps[@bib18] in the local environment. This tumour often presents late in its clinical course in the heavily affected areas; in fact, complete luminal obstruction (accompanied by inability to swallow saliva) is not uncommon at presentation. Passage of a Celestin latex rubber tube (a palliative technique) is often the only available procedure;[@bib6] however, blockage is a frequent problem resulting largely from the bulky African (or other) diet. Chemotherapy and radiotherapy (when available) are of very limited value. STOMACH AND DUODENUM {#cesec6} ==================== Peptic ulceration was at one time considered an unusual cause of abdominal pain in tropical countries; it was felt by many physicians to be a rare disease.[@bib1] It is now clear, however, that this is not the case; many difficulties facing the clinical epidemiologist in a developing country are highlighted by studies of the geographical distribution of this disease. Because sophisticated methods of diagnosis, including barium meal and upper gastrointestinal endoscopy, have not until relatively recently been widely used in developing countries, diagnosis and attempts at establishing accurate prevalence rates have depended upon recording incidence rates of complications, especially pyloric stenosis; upper gastrointestinal haemorrhage seems an unusual presentation overall, but this probably results from the fact that such patients do not reach hospital before exsanguinations occur. Therefore, serious deficiencies exist in knowledge regarding the true prevalence of peptic ulceration, and it is currently impossible to draw accurate conclusions on regional and rural/urban patterns, and also on variations with time, i.e. during the course of 'westernization\'. As recently as the 1950s, duodenal ulcer (DU) was considered a rare disease in Africa;[@bib1] this is not so, because satisfactory radiological, and more recently endoscopic, investigations have yielded accurate facts on true prevalence rate(s). Prevalence of DU in Africa has been reviewed using the available literature;[@bib1], [@bib19] high-prevalence areas seem to exist in parts of West Africa, Rwanda, Burundi, eastern Zaire, western Tanzania, south-western Uganda and the Ethiopian highlands. In southern India[@bib19] (and Fijians descended from this population[@bib20]) and Papua New Guinea, the disease also seems relatively common. It has a marked male predominance; it is frequently post-bulbar, and presentation with pyloric obstruction is relatively usual. Genetic factors might be important;[@bib19] the role of diet remains difficult to assess. Whether low rates of presentation resulting from haemorrhage and/or perforation accurately reflect incidence, or are biased by the inability to transport a sick patient to hospital, is also impossible to evaluate. In Lima, Peru there is evidence that the prevalence of peptic ulcer (and also gastric adenocarcinoma) has declined; between 1985 and 2002 a reduction from 3.15% to 5.05% was documented.[@bib21] Evidence for a causative role for *Helicobacter pylori* in chronic active gastritis, peptic ulceration and possibly gastric malignancy has escalated during the last decade;[@bib19], [@bib22] however, Koch\'s postulates have not all been satisfied, and infection rate with this organism frequently approaches 100% at an early age in an affected population. In a study carried out in Belgium, South Africa, China and North America, a significantly higher rate was found in the presence of gastric carcinoma but not duodenal ulceration.[@bib23] Overall, gastric ulcer (GU) is uncommon in developing countries.[@bib1] In a study carried out at Kumasi, Ghana, however, perforated duodenal ulcer was less common than perforated gastric ulcer; the latter was related to the widespread use of NSAIDs and herbal medicines.[@bib24] An overall decline in ulcer mortality might be associated with a worldwide reduction in the occurrence of *H. pylori* infection.[@bib25], [@bib26] When it occurs, it usually has a male predominance, is most common in the fifth and sixth decades, and afflicts predominantly the lower social strata. Pyloric obstruction is a common presentation, due frequently to late-stage disease at presentation. Management of a bleeding peptic ulcer has been reviewed.[@bib1], [@bib4] Gastritis, often resulting from alcohol and spicy foods, is a major cause of abdominal pain/discomfort[@bib22] ([Table 10.2](#cetable2){ref-type="table"} ). Infective causes (including tuberculosis) are overall rare, although occasionally encountered; infections which involve predominantly lower sections of the gastrointestinal tract (e.g. *Salmonella typhi* and *Shigella* spp.) occasionally produce significant gastric pathology. A heavy infection with hookworm and/or *Ascaris lumbricoides* can also account for epigastric discomfort (see below) and should be differentiated from peptic ulceration.Table 10.2Some causes of severe abdominal pain (without features of intestinal obstruction) in relation to tropical exposureSite of painCauseEpigastriumHeavy nematode infection (e.g. *Ascaris lumbricoides*, hookworm)Mesenteric adenitis (helminthic eggs or tuberculosis)Acute pancreatitis (helminth related)GeneralizedPeritonitisTyphoid perforationAmoebic colitis with perforation (appendix, perforated peptic ulcer or diverticulitis)Abdominal tuberculosisRuptured hydatid cystSickle cell crisisRecurrent familial polyserositis (familial Mediterranean fever) (Chapter 34)Hyperinfective syndrome caused by *Strongyloides stercoralisAngiostrongylus costaricensis*Right upper quadrantHelminthic infection involving biliary systemLeft upper quadrantSplenomegaly (e.g. hyperreactive malarious splenomegaly \[HMS\])Splenic ruptureSolitary splenic abscessRight iliac fossaAppendicitis*Anisakis* spp. infectionIleocaecal tuberculosis When H~2~-receptor antagonists (e.g. cimetidine and ranitidine) are used in developing countries, a possibility exists that they will encourage proliferation of intestinal pathogen(s) -- bacterial and parasitic -- for the gastric acid defence mechanism is largely removed;[@bib27] available data are, however, presently inadequate for assessing the practical importance of this. Several studies of gastric acid production indicate that mean acid production probably varies little in different ethnic groups. Hypochlorhydria is relatively common in the tropics;[@bib1] whether it is the cause or consequence of intestinal infection (of bacterial, including *S. typhi*, and/or parasitic origin) remains far from clear. Gastric carcinoma is overall an uncommon malignancy in tropical countries (Chapter 35). At Sura, Fiji, gastric ulcer and carcinoma have been shown to be more common in Fijians than Indians. Emergencies {#cesec7} ----------- Many facts remain unclear regarding upper gastrointestinal haemorrhage in tropical countries. For example, DU is apparently common in descendants of southern Indians in Fiji (see above); however, haematemesis from a chronic DU is more common in Fijians. Many data suggest that pyloric obstruction is the most common complication of DU in developing countries. A report from Zaria, northern Nigeria, indicates that at that location perforation is by no means uncommon;[@bib28] between 1971 and 1983, 74 (24%) of 302 patients operated for DU, and 29 (58%) of 50 for GU, presented with perforation; furthermore, there was a progressive increase in the years 1971--1974 to 1979--1983 of from 16% to 45%, respectively. A rare case report from India has recorded massive haematemesis and melaena from a cholecystoduodenal fistula secondary to DU in a 24-year-old man;[@bib29] he was successfully managed surgically. Ideally, management of the complications of gastritis and peptic ulceration is exactly the same as in a Western country. In a study carried out at Ankara, Turkey, age, delayed surgery, presence of shock, status of the anaesthetist, and 'definitive surgery' were significantly associated with a fatal outcome in patients undergoing emergence surgery for perforated peptic ulcer.[@bib30] Although usually associated with oesophageal varices, gastric varices also occur alone. In New Delhi, India, 48 (16%) out of 309 patients with portal hypertension were shown to have gastric varices;[@bib31] in six (12%) there was no evidence of associated oesophageal varices. In 11 (28%) of 40 patients who completed endoscopic sclerotherapy for oesophageal varices, gastric varices disappeared concurrently with the former, or during the following 6 months. In the light of their experience, these authors considered that 'if they persist for 6 months after eradication of oesophageal varices, a combination of paravariceal and intravariceal sclerotherapy should be attempted for their obliteration\'. ABDOMINAL PAIN {#cesec8} ============== Epigastric pain/discomfort is a common presenting symptom in medical practice in tropical countries (see above);[@bib1], [@bib32] this frequently results from heavy small-intestinal helminthic infections, especially with *A. lumbricoides* and hookworm. Mesenteric adenitis as a sequel to the presence of helminthic ova, and tuberculosis, are further causes. Helminth-related acute pancreatitis is another possibility. [Table 10.2](#cetable2){ref-type="table"} summarizes some causes of severe generalized abdominal pain. This most commonly results from peritonitis, which has numerous aetiologies. Right upper quadrant pain is less likely to result from biliary tract disease than in a 'temperate' area of the world (see below); nevertheless, helminthic infections of the biliary system are occasionally encountered. Left upper quadrant pain can result from splenomegaly (following numerous 'tropical' infections; see below); an extreme example (HMS) occurs in most areas which are endemic for human *Plasmodium* spp. Ruptured spleen is a further cause of left hypochondrial pain; this event usually presents acutely. Solitary splenic abscess is by no means an uncommon event in West and Central Africa; the aetiology remains unclear. Right iliac fossa pain is less likely to be caused by appendicitis (see below) than in most Western countries. However, an appendix-like syndrome has been recorded in *Yersinia* spp., and *Anisakis* spp. infections and ileocaecal tuberculosis (see below). *Enterobius vermicularis* is not infrequently detected in an appendicectomy specimen; whether there is a cause--effect relationship to acute appendicitis is frequently unclear. Less common parasites involving the Appendix include *Taenia* species, *Trichuris trichiura* and *Angiostrongylus costaricensis* (see below). A peripheral blood eosinophilia is often (but by no means always) present when a helminthiasis is causatively related to appendicitis. Ileocaecal tuberculosis can account for chronic right iliac fossa pain; an ileocaecal mass is often palpable clinically (this can be confirmed by ultrasonography when this technique is available). A colonic amoeboma represents a possible source of diagnostic confusion. SMALL INTESTINE {#cesec9} =============== Tropical enteropathy, subclinical malabsorption and mechanism of diarrhoea {#cesec10} -------------------------------------------------------------------------- The small-intestinal mucosa of an individual living in a developing country possesses minor structural differences compared with that in one always resident in a temperate zone.[@bib1], [@bib33], [@bib34] Changes are not related to the clinical syndrome: post-infective malabsorption (tropical sprue; see below). Although the cause of these changes is not entirely clear, they seem to result from repeated low-grade viral and bacterial infection(s). Similarly, marginal xylose and glucose malabsorption has been demonstrated in large numbers of people indigenous to tropical countries; these abnormalities are certainly greater in lower socioeconomic groups. Using a breath-hydrogen test, bacterial overgrowth in the small-intestine was demonstrated in 37.5% of children living in slum conditions compared with only 2.1% (*p*\< 0.001) controls in urban Brazil.[@bib35] Subclinical malabsorption exists in many people in developing countries;[@bib1] xylose and B~12~ malabsorption have been demonstrated in 39% and 52%, respectively, of Peace Corps workers living under rural conditions in Pakistan. Apart from repeated small-intestinal infections, other factors are probably also important.[@bib36] Xylose, glucose and folic acid absorption have been shown to be impaired in individuals with systemic bacterial infections, e.g. pulmonary tuberculosis and pneumococcal pneumonia. Dietary folate depletion also results in xylose malabsorption. Marginal malnutrition and pellagra have both been suggested as causing subclinical malabsorption, but evidence is contradictory. The practical importance of subclinical malabsorption is unclear.[@bib1], [@bib33], [@bib34] It seems likely that it significantly contributes to malnutrition in people in developing countries who subsist on a marginally adequate dietary intake consisting largely of carbohydrate. Before any rigid conclusions are drawn, however, it should be appreciated that the small intestine has a very substantial functional reserve, and that the role of the colon in absorption of carbohydrate (and other substances) (see above) remains unclear. Diarrhoea resulting from small-intestinal disease consists of two main types;[@bib1], [@bib33] (1) profuse watery (e.g. cholera), and (2) steatorrhoeic (exemplified by post-infective tropical malabsorption (tropical sprue)). [Table 10.3](#cetable3){ref-type="table"} summarizes the most important causes; several of those responsible for the former type are infective, and then exert their pathogenic effect via an enterotoxin (either heat stable or heat labile); invasive disease involving the enterocyte is less important. The role of intestinal hormones -- especially vasoactive intestinal peptide -- in the production of watery diarrhoea has become clearer.[@bib34] The pathogenesis of diarrhoea in AIDS has a multifactorial basis, and is often by no means clear;[@bib37] some but not all cases are associated with an opportunistic infection(s), especially *Cryptosporidium parvum* (Chapter 79).[@bib34] The bacteria *Escherichia coli*, fungi *Candida albicans* and Histop*lasma capsulatum*, and the astroviruses and caliciviruses are also relevant. Other opportunistic infections in this syndrome include cytomegalovirus, *Mycobacterium avium intracellulare*, *Salmonella* species, and the protozoa *Isospora belli*, *Cyclospora cayatenensis, Sarcocystis hominis* and *Microsporidium* species infections; in addition, Kaposi\'s sarcoma (Chapter 35) causes severe small-intestinal disease.Table 10.3Small-intestinal diarrhoeaWatery diarrhoea (large volume, fluid stool(s)):Traveller\'s diarrhoea (TD) (turista)*Vibrio cholerae* (and other vibrios)*Escherichia coli* (enterotoxigenic)*Salmonella* spp.*Campylobacter jejuni*Rotavirus (and other enteric viruses)*Cryptosporidium* spp.(Food poisoning - *Staphylococcus, Clostridium perfringens)*Hypolactasia:Primary - genetically determinedSecondary - resulting from enterocyte damageSteatorrhoeic diarrhoea (malabsorption) (characteristically large pale, fatty, offensive stools; microscopy often shows fat globules in faecal smear):Post-infective tropical malabsorption ('tropical sprue')Intestinal parasites  *Giardia lamblia*  *Strongyloides stercoralis*  *Capillaria philippinensis* Coccidia:*Cryptosporidium parvumIsospora belliSarcocystis hominisMicrosporidium spp.Cyclospora cayetanensis*HIV enteropathyTrauma - short bowel syndrome (e.g. recovered pigbel disease)Lymphoma - Burkitt\'s, Mediterranean lymphomasIleocaecal tuberculosisChronic calcific pancreatitisAcute and chronic liver disease(Gluten-induced enteropathy (coeliac disease) seems to be uncommon or even rare in most tropical populations.Occasionally it can become clinically obvious in visitors fromWestern countries to the tropics) Many of the problems encountered in management, including chemoprophylaxis and chemotherapy, are exemplified by traveller\'s diarrhoea (see below). Traveller\'s diarrhoea {#cesec11} ---------------------- The clinical syndrome traveller\'s diarrhoea (TD)[@bib1], [@bib38], [@bib39], [@bib40], [@bib41], [@bib42] is arguably the world\'s most common disease entity; only rarely is it associated with mortality (usually in the presence of debility, or at the extremes of life), but the significant morbidity with which it is associated not infrequently interferes with a crowded schedule or a leisure or sporting activity. Numerous titles have been applied, including 'turista\', 'Montezuma\'s revenge\', 'Hong Kong dog' and 'Delhi belly\'. One estimate is that 12 million individuals travel annually from an industrialized (Western) country to one in the tropics or subtropics;[@bib43] in this group incidence of TD varies from around 20--50%. There is a highly significant geographical variation in prevalence; high-risk areas include North Africa, sub-Saharan Africa, the Indian subcontinent, South-east Asia, South America, Mexico and the Middle East; intermediate ones include the north Mediterranean, Canary Islands and the Caribbean islands; low-risk ones include North America, Western Europe and Australia. In a retrospective study carried out in Switzerland, a large group of travellers were asked to complete a questionnaire after travelling abroad; incidence of the disease varied greatly, the highest figure (50%) being associated with travel to Tunisia. (No detailed study exists of TD acquired in a European country.[@bib44]) The disease tends to become less common with advancing years; it is unclear whether this is due to the fact that older travellers (≥60 years) have a more discerning lifestyle, or whether relative immunity increases with advancing age.[@bib38] Individuals resident for substantial periods in areas where TD is common seem to experience it less frequently than those not previously exposed.[@bib38], [@bib39] ### Clinical features {#cesec12} TD is contracted by ingestion of contaminated water/food; it is characterized by acute-onset watery diarrhoea (usually of small-intestinal origin);[@bib38], [@bib39], [@bib40], [@bib41], [@bib42] when colorectal involvement exists, diarrhoea is often bloody (see below). Abdominal colic, nausea and vomiting may be present; fever is unusual, being recorded in 1--10% of infected individuals. Prostration and resultant dehydration (with electrolyte imbalance) cause major problems in a severe case. Rarely, symptoms become chronic, and it seems likely that a small proportion of cases of TD proceeds to post-infective malabsorption (see below).[@bib33] Unfortunately, for the investigator, by the time disease has become clinically overt, the initiating infection(s) has invariably been cleared. Chronic diarrhoea of lesser severity is a relatively common problem following recovery from acute disease; this can usually be attributed to (1) tropical enteropathy (in which there is major derangement of enterocyte structure and function) (see above) or (2) the irritable bowel syndrome (see below). On clinical grounds, an important differential diagnosis is inflammatory bowel disease presenting for the first time during, or immediately after, tropical exposure.[@bib45], [@bib46] In a retrospective review of UK residents presenting at the Hospital for Tropical Diseases, London, with acute onset/bloody diarrhoea, the majority had inflammatory bowel disease (usually ulcerative colitis); it was numerically more important than shigellosis and amoebic colitis.[@bib45] Acute disease pursues an especially virulent course in certain high-risk groups,[@bib38], [@bib39] e.g. those suffering from achlorhydria (*Salmonella* species and *Vibrio* species infections are known to be significantly more common in this group), known inflammatory bowel disease (see below), previous gastrointestinal tract surgery, a malignancy involving the gastrointestinal tract, and acquired or congenital immunodeficiency (including immunosuppressive therapy and HIV/AIDS). In addition, individuals on diuretic therapy (in whom maintenance of electrolyte balance is precarious) and others at the extremes of life also fall within the high-risk group. It is important to recognize these factors when advising chemoprophylaxis (see below). ### Aetiology {#cesec13} In 1970, Rowe et al.[@bib47] recorded results of a study involving British soldiers newly arrived in Aden; in 19 (54.3%) of 33 cases in which a recognized pathogen was not apparent, a 'new' serotype of *Escherichia coli* was isolated in the acute phase of TD; in a further 14 (40%), several different *E. coli* serotypes were also isolated. (B. H. Keane had suggested in the 1950s (on circumstantial evidence) that bacterial pathogens were implicated.[@bib33], [@bib34]) Sack[@bib48] recorded the identity of *E. coli* serotypes isolated from US Peace Corps volunteers serving in various countries: Kenya 06:H16, 06H^−^, 027:H7, 0159:H4 and 0159:H34; Morocco 06:H16, 0128:H12, 027:H20 and 0169:H^−^; Honduras 08:H9, 015:H49, 015:H^−^ and 027:H20. Therefore, many common strains of enterotoxigenic *E. coli* (ETEC) are relevant. Many other microorganisms are also involved. *Salmonella* species, *Shigella* species, *Campylobacter jejuni*, enteroadherent *E. coli* (EAEC) and *Vibrio* spp. (see Chapter 51); rotavirus and norovirus (Chapter 45), and *Giardia lamblia*, *Coccidia* species (including *Cryptosporidium* species, *I. belli* and *Blastocystis hominis*) and *Entamoeba histolytica* (Chapter 79). Other bacteria which have been implicated include *Aeromonas hydrophila*, *Plesiomonas shigelloides* and *Yersinia enterocolitica*. The causative agent(s) vary significantly in different locations, e.g. in an affected individual in Asia, Central America or Africa the likely organism is different on statistical grounds, although not relevant to a specific case. Furthermore, more than one organism is frequently present; in a study involving US Peace Corps workers in Thailand, 33% were infected by two to four different pathogens.[@bib38] Although protozoan parasites are usually incorporated in the list of aetiological agents, the incubation period is usually somewhat longer than is usual in TD; this applies especially to *G. lamblia*. When the colorectum is predominantly involved, *Shigella* species, enteroinvasive *E. coli* (EIEC), enterohaemorrhagic *E. coli* (EHEC) and *Ent. histolytica* may be responsible. Rarely, herpes simplex virus and *Chlamydia trachomatis* have been implicated. New pathogens will doubtless emerge in future years. ### Pathophysiology {#cesec14} The pathophysiology varies and depends on the site within the gastrointestinal tract to be involved.[@bib1], [@bib38], [@bib39], [@bib40], [@bib41] Whereas in the small intestine toxigenic diarrhoeas predominate (see above), in the colorectum (see below) invasive disease is more common. ETEC are characterized by both toxin production and mucosal adherence (via specific fimbriae); the latter property is required for disease production, for toxin-producing non-adherent mutants do not cause disease. Enteropathogenic *E. coli* (EPEC) (probably not a major cause of TD) adhere to intestinal mucosal cells and although they do not invade, destroy microvilli. EAEC (detected in up to 15% of patients suffering from TD) do not belong to classical serotypes of EPEC, but adhere to Hep-2 cells in culture; they neither produce a toxin nor invade.[@bib49] EIEC behave similarly to *Shigella* species and account for up to 5% of cases; the main site of action is the colorectum, and the major clinical manifestation is therefore dysentery resulting from epithelial cell invasion and intracellular multiplication; there is resultant mucosal inflammation and ulceration.[@bib49] EHEC (an uncommon cause of TD) produces disease via verotoxin production. ### Prophylaxis {#cesec15} Travellers should take maximal care to avoid water/food likely to be contaminated; common sense is of paramount importance! Use of prophylactic agents is controversial. Many chemoprophylactics have been used: doxycycline, co-trimoxazole, trimethoprim, mecillinam, bicozamycin and the fluroquinolone compounds (norfloxacin and ciprofloxacin). High protection rates (≥90%) have been claimed for co-trimoxazole and the fluoroquinolones; for trimethoprim a rate of around 50% has been recorded. Most cases of TD therefore possess a bacterial aetiology. The major problem with antibiotic chemoprophylaxis, however, is the risk of significant side-effects, dominated by dermatological reactions (including Stevens--Johnson syndrome) and pseudomembranous colitis (see below); using co-trimoxazole, a rate of up to 20% of significant skin reactions, necessitating discontinuation of prophylaxis, has been recorded. Also, the acquisition of resistant faecal *E. coli* during chemoprophylaxis has been recorded in several studies; an increase from 21% to 100% has been recorded using doxycycline in Kenya, and one of 3% to 100% with co-trimoxazole in Mexico. When chemoprophylaxis is used, either norfloxacin or ciprofloxacin seems to be the most appropriate, although strains of *Campylobacter jejuni* rapidly acquire resistance.[@bib44] In a recent study in Egypt, two of 105 individuals on norfloxacin developed TD, compared with 30 (26%) of 117 given a placebo.[@bib49] (Ciprofloxacin should be avoided in children because of experimental evidence indicating cartilaginous damage in young experimental animals; there is no evidence in *Homo sapiens*.) Should chemoprophylaxis be recommended widely in this essentially benign clinical syndrome? In addition to the objections so far outlined (see above), there is a possibility of inducing a false sense of security, resulting in increased exposure to other infections, e.g. viral hepatitis.[@bib49] The following groups should be seriously considered for chemoprophylaxis (for \<3 weeks):•Travellers with a bad 'history' of TD[@bib38], [@bib39], [@bib40], [@bib41]•Those in whom hypochlorhydria is proven (or a possibility)•Individuals suffering from inflammatory bowel disease•HIV-infected patients•Those in whom electrolyte balance is precarious (e.g. those receiving diuretic therapy) and others with chronic renal failure•The 'elderly' (not easily defined)•A nebulous group in whom TD is professionally embarrassing (e.g. members of the armed services, airline pilots, athletes, politicians, businessmen and other professionals on tight schedules, etc.). The role of prophylactic antiperistaltic agents is likewise controversial: action is unphysiological. It has been suggested that they can mask a more serious infection, e.g. *S. typhi*, although in this disease diarrhoea is an unusual presenting symptom (Chapter 52). By delaying excretion of pathogen(s) it is also possible that clinical disease is prolonged. In children, paralytic ileus is a major complication and has occasionally precipitated mortality.[@bib50] Bismuth subsalicylate has a role in prophylaxis; the bismuth moiety possesses antimicrobial activity and salicylate antisecretory properties.[@bib39] Early studies in Mexico by DuPont et al.[@bib51] showed that, given as a suspension (the sheer bulk required precluded its use by travellers), this agent significantly reduced TD; the same group, also working in Mexico, has demonstrated that, when given in tablet form (2 tablets 4 × daily for 3 weeks, i.e. 2.1 g daily), a 65% protection rate can be achieved;[@bib51] at half that dose, efficacy was greatly reduced. The number(s) of pathogen-positive TD cases in a group of treated patients was seven of 29, compared with 35 of 59 in a placebo group; ETEC was present in 3 and 22, respectively, and *Shigella* species in two and eight, respectively.[@bib51] A B-subunit/whole-cell (BS-WC) cholera vaccine has been shown to produce relative protection.[@bib52] In a study involving Finnish tourists to Morocco, BS-WC induced 52% protection against diarrhoea caused by ETEC, 65% with mixed infection, 71% when ETEC was present with another pathogen, and 82% when ETEC and *S. enterica* were present concurrently. (Sack[@bib48] has concluded that 'any advances in prevention and treatment of diarrhoea in travellers will be directly applicable to the worldwide problem of diarrhoea in children, which is far more important on a global scale\'. This statement does not apply to this BS-WC vaccine, because protection only lasts for about 3 months.) A further approach under consideration consists of oral administration of colostrum-derived antibodies against ETEC.[@bib39] A recent experimental investigation indicates that lactobacilli, which have the ability to adhere to the intestinal mucosa, can prevent *E. coli* colonization. In a limited clinical study, *Lactobacillus GG* reduced prevalence of TD by up to 40%.[@bib39] ### Management {#cesec16} Treatment (as in cholera, see below) devolves around oral rehydration (see below); all travellers should carry suitable preparations.[@bib1], [@bib41], [@bib53] When properly constituted, Dioralyte (Rhône-Poulenc Rorer) solution contains glucose 90, Na^+^ 60, K^+^ 25, Cl^−^ 45 and citrate 20 mmol/L. Corresponding concentrations for another proprietary preparation, Rapolyte (Janssen), are 111, 60, 20, 50 and 10 mmol/L. WHO/UNICEF rehydration fluid contains glucose 111, Na^+^ 90, K^+^ 20, Cl^−^ 80 and citrate 10 mmol/L. In a mild case adequate rehydration can usually be achieved using ordinary mineral water. The role of chemotherapy in established TD remains controversial. Early work carried out by DuPont et al.[@bib38] in Mexico showed that both co-trimoxazole and trimethoprim reduced the length of symptoms. Recent trials, using antibiotics which have been given for chemoprophylaxis (see above), have also indicated that the length of symptoms can be shortened; in Mexico, ofloxacin (600 mg daily for 3 days) produced cure in 77 (95%) of 81, compared with 56 (71%) patients who received placebo (*p*= 0.0001).[@bib54] In a study carried out in Thailand where the causative organism is usually *Campylobacter jejuni*, cure rate at 72 h was highest with single-dose azithromycin (96%), compared with lower rates with 3-day azithromycin and levofloxacin.[@bib55] Short-course chemotherapy can only be justified in a severe case; this applies at the extremes of life and in high-risk groups (see above), especially HIV-infected individuals.[@bib56] Cholera {#cesec17} ------- Cholera (see also Chapter 51) represents the archetypal disease in the context of small-intestinal secretory (watery) diarrhoea.[@bib34], [@bib36], [@bib57] The causative organism, *Vibrio cholerae*, is not invasive and exerts its effect by means of an enterotoxin.[@bib58] If untreated, the disease has a 20--80% mortality; with modern oral rehydration regimens that figure should be \<1%. Death results from dehydration, vascular collapse and renal failure. Historically, cholera was not confined to tropical countries and involved many temperate areas, including much of northern Europe. An epidemic in 1854 in London was traced to contaminated water supplied from the Broad Street pump in Soho. According to legend, when the handle of the pump was tied down by Dr John Snow, the London anaesthetist, a rapid decline in the incidence of new cases was recorded.[@bib59] ### Epidemiology {#cesec18} Cholera is endemic in India, Pakistan, Bangladesh, Afghanistan and many other countries of South-east Asia. Nosocomial transmission is reported. In recent years, epidemics have occurred in the Middle East, South America and Africa;[@bib58] most have been localized. Cholera is endemic along the Gulf Coast of the USA. The disease is closely associated with poverty, overcrowding and low socioeconomic status. In former times cholera was spread by population movements such as the annual 'hajj' to Mecca; outbreaks involving air travellers have been recorded. Overall, however, the disease is rare in British travellers.[@bib60] It tends to affect young people more often than the elderly. ### Aetiology and pathogenesis {#cesec19} There is probably a genetic predisposition: blood group O is associated with a higher infection rate than group A.[@bib34] Classical cholera is caused by *V. cholerae*, and is now localized to the Indian subcontinent, particularly the deltas of the Ganges and Brahmaputra rivers. Elsewhere, the El Tor biotype, which originated in Indonesia around 1960, and the 0139 strain have been responsible for most epidemics. Vibrio species are curved, Gram-negative, flagellated rods approximately 2 mm in length. Each biotype of cholera contains three serotypes: Inaba, Ogawa and Hikojima. For details of the organism and its pathophysiological effects, see Chapter 51. ### Pathology {#cesec20} Histologically, the small-intestinal mucosa is intact. Light and electron microscopical appearances are normal. Following circulatory collapse following gross dehydration, renal tubular necrosis can be demonstrated. ### Clinical features {#cesec21} There are no prodromal symptoms. The incubation period varies from a few hours to 5 days. The disease is similar whichever biotype is involved, but there is a wide spectrum of severity. When the El Tor biotype is responsible, a higher proportion of patients are asymptomatic. Onset is sudden, and mild diarrhoea rapidly gives way to the passage of a large volume of opalescent fluid -- the classic 'rice-water' stools. Up to 30 L of fluid, containing a high concentration of Vibrio spp. organisms, may be passed in 24 h.[@bib57] Vomiting of fluid of a similar composition is a later feature. Thirst, muscle cramps, hoarseness and anuria follow. Clinical signs of severe dehydration may be present by 24 h after onset in an untreated case. The body temperature is normal or mildly elevated. Circulatory failure and acute renal failure follow. Confusion, disorientation and hypoglycaemic convulsions may occur. Mortality rate is directly related to the degree of dehydration. Relative immunity is short lived. A carrier state, which lasts a few weeks, may occur, and gallbladder foci have been identified. ### Investigations {#cesec22} Vibrio spp. organisms are easily identified in a faecal specimen; material should be transported to the laboratory in alkaline peptone water (pH 9.0). A rapid diagnostic technique for field use has been described. For accurate serological identification of *V. cholerae*, rigid criteria are necessary. With classic cholera, organisms are present during the incubation period and up to 5 days after an attack; in the El Tor variety, Vibrio spp. can persist for weeks or months. Faecal samples are isotonic, with a protein concentration of approximately 10 g/L; pH is about 7.5; typical electrolyte concentrations are: sodium 139 mmol/L, potassium 23 mmol/L, chloride 106 mmol/L and bicarbonate 48 mmol/L. Specimens contain a high concentration of IgA. Serum IgA and IgM are elevated, the former most markedly in patients with an El Tor infection. In vitro animal studies, indicate that cholera toxin enhances IgA secretion from crypt epithelium to ileal lumen.[@bib56] Serum electrolyte, urea and creatinine concentrations vary with the stage and severity of the disease. Excessive potassium loss exacerbates metabolic acidosis. Urine is concentrated; its composition depends on the severity of the disease. ### Differential diagnosis {#cesec23} Diagnosis is usually straightforward; however, all other causes of small-intestinal diarrhoea (with and without vomiting) of acute onset (see below) should be considered. These include traveller\'s diarrhoea, *E. coli, Staphylococcus species, Clostridium perfringens, Cl. botulinum*, *Campylobacter jejuni* and viral causes (e.g. rotavirus, norovirus). *Salmonella* and *Shigella* spp. should also be considered. *Vibrio parahaemolyticus* (conveyed by infected raw seafood) and other non-cholera *Vibrio* species can produce a similar disease. Very occasionally, *Plasmodium falciparum* malaria presents with severe watery diarrhoea, especially in infants and children. Food poisoning, caused by toxic agents, should be added to the list of differential diagnoses. ### Prevention {#cesec24} Basic sanitation and public health procedures should be improved.[@bib61] Sterility of water supplies is of paramount importance. Contacts of proven cases should be vaccinated; all faeces and bed linen should be destroyed. Vaccination with inactivated (dead) *Vibrio* species organisms gives only limited protection;[@bib62] 0.5 mL and 1.0 mL vaccine should be given at an interval of 1 week, and a 0.5 mL booster every 6 months. The 26th Assembly of the WHO recommended, in 1973, that cholera vaccination should not be compulsory, due to its limited public health value. Despite this, a few countries continue to demand vaccination before entry. Important progress is being made towards an effective oral bivalent cholera--typhoid vaccine.[@bib63] ### Management {#cesec25} #### Rehydration regimens {#cesec26} Treatment was revolutionized by the introduction of oral rehydration regimens.[@bib64], [@bib65], [@bib66] The enterocyte sodium-glucose carrier system is not affected by cyclic AMP, and thus glucose (and glycine)-stimulated membrane transport takes place normally. It is impossible to overload the circulation by the oral route in a previously fit person. Quantity of ingested fluid should be regulated by faecal loss, best measured 2-hourly. Rehydration should be accomplished within 48 h. In an unsophisticated situation, sucrose is often more easily obtainable than glucose; results are usually good, although if severe mucosal damage pre-exists, sucrase concentration is lowered and satisfactory rehydration is less readily achieved. Cereal-based electrolyte solutions have also given satisfactory results.[@bib66] In a severe case, intravenous fluids may be necessary for initial rehydration.[@bib65] A widely-used formula consists of; sodium chloride 5.0 g, sodium bicarbonate 4.0 g, potassium chloride 1.0 g, made up to 1 L. Severity of dehydration should be assessed on clinical grounds; in a case of average severity, 5 L should be given (the first litre within 10 min) to a 50 kg subject. #### Drug treatment {#cesec27} Analgesics may be necessary for severe muscle cramps. Intravenous calcium gluconate is of value for tetany. Tetracycline hydrochloride, 1 g/day for 5 days, shortens duration of diarrhoea and clears the luminal content of *Vibrio* spp. organisms in the case of the El Tor biotype.[@bib66] A single dose (1 g or 2 g) has also been shown to be effective in *V. cholerae* infection, but is associated with asymptomatic bacteriological relapse.[@bib67], [@bib68] Tetracycline should be started several hours after rehydration therapy has begun. Single-dose doxycycline (300 mg) is probably as effective as tetracycline.[@bib69] There is clear evidence that in epidemics the El Tor biotype rapidly develops resistance not only to tetracycline, but also to several other antibiotics (including trimethoprim plus sulfamethoxazole), and is therefore of very limited value. Recently, *Vibrio cholerae* 01 biotype El Tor strains have proved resistant to furazolidone and co-trimoxazole.[@bib70] ### Prognosis {#cesec28} If cholera is adequately treated, there should be zero mortality, and complete recovery. A suggestion has been made that individuals who have suffered from cholera might be predisposed to α-chain disease (see below). Malabsorption in the tropics {#cesec29} ---------------------------- Apart from infective causes, primary hypolactasia (lactase deficiency)[@bib1], [@bib71] accounts for watery small-intestinal diarrhoea in some people indigenous to tropical countries. A low concentration of this enzyme in the enterocyte brush border is normal for adult *Homo sapiens* (as for other species within the mammalian kingdom); the enzyme is under genetic control. In a minority of the world\'s population, i.e. northern Europeans, Africans with an Hamitic ancestry, certain Middle Eastern populations (e.g. Saudi Arabians) and others in northern parts of the Indian subcontinent, a high concentration continues into adult life. Secondary hypolactasia results from brush border damage;[@bib1], [@bib71] concentration of all disaccharidases (and other digestive enzymes) is reduced, and slow recovery occurs after the initiating insult has disappeared. Thus, whenever there is enterocyte destruction (this includes post-infective malabsorption, see below) hypolactasia develops. Following ingestion of milk or another milk produce, in which lactose is incompletely hydrolysed, osmotic diarrhoea results; this is accompanied by abdominal colic, distension and flatulence ('lactose intolerance\'). In a study carried out at Penang, Malaysia, hypolactasia was demonstrated in all ethic groups, and although there was no clear association with gastrointestinal symptoms, the authors recommend a low lactose diet in all Asian countries.[@bib72] Yoghurt contains adequate bacterial lactase to hydrolyse the lactose component and is usually well tolerated. Lactic acid production (derived from hydrolysis of lactose by colonic bacteria) produces irritative diarrhoea, which contributes to the symptoms. The precise role of the colon in adaptation remains unclear; carbohydrate, in the form of free fatty acid(s) (and also nitrogen and electrolytes), can be absorbed from this organ. Investigation of hypolactasia most often utilizes the breath hydrogen test; lactose 'tolerance' test and lactase assay in a jejunal biopsy specimen are alternatives. In management, milk and all lactose-containing dairy products should be eliminated from the diet;[@bib1], [@bib71] individuals in countries with a high prevalence of primary hypolactasia can regulate bowel function by varying lactose ingestion. Post-infective malabsorption (PIM) (tropical sprue) {#cesec30} --------------------------------------------------- Relatively little is known about the prevalence and severity of malabsorption in acute infective conditions of the small intestine (viral, bacterial and parasitic) and the duration for which it can continue after the specific organism(s) has been eliminated.[@bib73] In some cases, malabsorption persists in the presence of mixed luminal flora, and a single infective agent cannot be detected. In others the recognizable initiating infective cause (or causes) may continue, culminating in a chronic form; a more precise term is therefore 'postacute infective' malabsorption. As with all infective diseases, the clinical spectrum of disease varies from subclinical to gross pathology (malabsorption). PIM is of particular clinical significance in tropical countries, where small (and large) intestinal infections are exceedingly common. PIM related to tropical exposure has been reviewed by Cook,[@bib1], [@bib33], [@bib74] Tomkins,[@bib75] Baker[@bib76] and Mathan.[@bib77] ### History and definition {#cesec31} Confusion has existed between PIM and tropical sprue; however, in tropical and subtropical countries, these entities are synonymous, and the difficulty is primarily one of semantics.[@bib1], [@bib33] Patrick Manson first coined the term tropical sprue (derived from a word used by Dutch workers in the East Indies) in 1880.[@bib78] The term was rapidly applied to all cases of malabsorption in tropical countries, undoubtedly including some resulting from tuberculosis and various parasitoses (both protozoan and helminthic). Historically, chronic diarrhoea accompanied by wasting was recognized in India before 600 BC; although the Englishman William Hillary is often credited with the first precise description of tropical sprue at Barbados,[@bib79] it now seems likely that he described either epidemic *G. lamblia* infection, or possibly strongyloidiasis. The clinical syndrome was well known to British physicians in India during the eighteenth and nineteenth centuries; most descriptions were made in British expatriate populations. It was in the early 1960s that reports of a high prevalence of epidemic PIM in indigenous Indians became available.[@bib1], [@bib76], [@bib77] Despite early suggestions that chronic tropical diarrhoea had an insidious onset, it is clear (after careful assessment) that the vast majority of cases always presented acutely. Confusion has been compounded further when acute epidemic cases of small-intestinal infection, associated with gross dehydration (in addition to xylose and fat malabsorption) and acute mortality, have been designated tropical sprue, as in numerous reports from southern India.[@bib77] It is essential to include a time factor in the definition of this clinical syndrome, e.g. chronic diarrhoea and malabsorption, with weight loss, of at least 3--4 months duration. The term tropical sprue (if used at all) would be better reserved for a condition where malabsorption of nutrients is quantitatively more important than that of water and electrolytes. Although the aetiology of PIM is not yet completely clear (see below), in most cases it undoubtedly follows an acute small-intestinal insult by either a bacterial, viral or parasitic (or mixed) infection. Overall, evidence for PIM following a small-intestinal insult is most complete for bacterial and parasitic infections; those of viral origin might, however, be more important numerically. Lack of precise data can be largely attributed to the fact that virology remains a relatively neglected discipline in most developing countries, where infections of all types are far more common than in the Western world. The effect of malabsorption on overall nutritional status is largely unknown (see above); children are especially at risk. The magnitude of energy loss is unclear; a deficit of 10% of dietary energy (one estimate) is substantial in tropical populations subsisting on a 'marginal' diet. The importance of anorexia in exacerbating associated malnutrition is also underexplored. ### Geographical distribution {#cesec32} [Figure 10.3](#f3){ref-type="fig"} summarizes the geographical localities where PIM has been reported either commonly or less frequently;[@bib33], [@bib34] the map does not include areas where sporadic cases have been rarely recorded. Although the disease is common (and endemic) in Asia and the northern part of South America, it is a very unusual condition in tropical Africa. It remains a problem in travellers to many tropical locations.[@bib80], [@bib81], [@bib82] Until recently, it was a common entity in overland travellers from the UK to Asia; the fact that it is now rarely seen is probably associated with early antibiotic administration. In the Middle East and Mediterranean littoral PIM is unusual, but undoubtedly occurs.[@bib74] Figure 10.3World map showing areas where post-infective tropical malabsorption is a significant problem. ### Aetiology {#cesec33} There can now be no reasonable doubt that PIM has an infective basis (see above): it is (1) more common in geographical areas where enteric infection abounds; (2) epidemic in certain areas, including southern India; (3) the small-intestinal lumen is colonized by aerobic enterobacteria; and (4) recovery usually occurs rapidly (and dramatically) following initiation of broad-spectrum antibiotic treatment. Despite this, however, Mathan[@bib77] is of the opinion that in southern India the primary lesion is enterocyte damage resulting from a 'persistent' lesion of the stem cell compartment on a 'background of tropical enteropathy\'. He further considers that 'an immunity-conferring agent may be responsible for the initiating damage\'. The widely used definition for this clinical syndrome in southern India, 'intestinal malabsorption of at lease two nutrients and the exclusion of diseases that give rise to secondary malabsorption in a tropical environment\', is inadequate; it does not exclude tropical enteropathy (see above), nor does it introduce a time (chronicity) factor. #### Genetic predisposition {#cesec34} All infective diseases, without exception, have a genetic background. In a limited study at Puerto Rico, 25 of 27 patients with PIM (not well defined) had at least one antigen of the HLA-Aw19 series;[@bib83] the strongest associated link was with Aw31. In India, a high frequency of HLA-B8 was documented;[@bib84] HLA-A1, A28 and Bw35 were significantly decreased in the affected group. More data are undoubtedly required on genetic markers in PIM. #### Infection {#cesec35} In severe PIM (in the absence of parasites) bacterial colonization has been demonstrated both within the jejunal lumen and in biopsy specimens. The importance of adhesive properties of bacteria in pathogenesis is unclear; many bacteria, including *E. coli, S. typhimurium and V. cholerae*, possess such properties, mediated by a transmissible plasmid. In tropical PIM, several groups have demonstrated a higher concentration of aerobic enterobacteria in relation to the enterocyte compared with luminal fluid. (In the normal individual, anaerobes outnumber aerobes by about 1000-fold.) It seems likely that a variety of toxins released by these enterobacteria induce net water secretion and malabsorption. In the blind-loop syndrome, enterobacteria (which are invariably obligate anaerobes) do not produce toxins. Several months after tropical exposure the upper small-intestinal intraluminal bacterial flora (mucosal biopsy or luminal fluid) remains abnormal;[@bib85] seven of 11 patients studied had enterobacteria in numbers ranging from 10^3^ to 10^8^/g or mL. The most common organisms were *Klebsiella pneumoniae, Enterobacter cloacae* and *E. coli; Citrobacter feundii, Serratia marcescens* and *Pseudomonas* spp. have also been detected. It seems highly likely, therefore, that these organisms were present since the onset of disease.[@bib86] In southern India, a viral aetiology has been sought, but there is little evidence for this. The origin of continuing overgrowth has not been adequately studied in tropical PIM; in patients in England with small-intestinal bacterial overgrowth, faecal flora account for most of the organisms, but salivary flora are probably important in some cases. #### Jejunal morphology {#cesec36} Morphological changes are non-specific and range in severity.[@bib74] Blunting of villi ('partial villous atrophy\') with increased lymphocyte and plasma cell infiltration (not a feature of tropical enteropathy) are present to a variable degree; a 'flat' mucosa is exceedingly unusual. Although the number of plasma cells is increased, distribution of IgA-, IgM- and IgG-containing cells is normal.[@bib87] In untreated gluten-induced enteropathy, T cells expressing T cell receptor *g*/*d* heterodimers are disproportionately raised; this is not so in PIM.[@bib87] The significance of elevated jejunal surface pH (demonstrated in southern India) is unclear, but is probably merely an indicator of enterocyte damage. Crypt hyperplasia has been demonstrated. Although a predisposing immunological deficit has been postulated in tropical PIM, there is no good evidence for this; immunological changes (increased IgG, IgE, C4 and orosomucoid, gastric parietal cell antibodies, and lymphopenia with a low peripheral blood T cell count) seem to be sequelae of mucosal damage, and are not causally related. #### Small-intestine stasis {#cesec37} In southern India whole-gut transit time (using a radio-opaque marker technique) has been shown to be unaltered in tropical PIM, despite a striking increase in faecal weight. Small-intestinal stasis has, however, been well documented in tropical PIM and might result from excessive enteroglucagon production in response to ileal (and colonic) mucosal injury (see below).[@bib88] However, many patients with PIM have received diphenoxylate or loperamide for acute diarrhoea; both agents produce relative small-intestinal stasis. Both of these agents interfere with peristalsis and prevent prostaglandin-induced diarrhoea; inhibition of small-intestinal secretion also occurs. Such stasis is of particular interest because peristalsis is usually increased in the presence of intraluminal bacteria. #### Gut hormones {#cesec38} Gut hormones have been studied in tropical PIM in the fasting state and following a standard meal.[@bib88] Fasting and postprandial plasma enteroglucagon concentrations (produced by cells in the distal ileum and colon) and motilin were markedly elevated; furthermore, the elevated enteroglucagon concentration is significantly correlated with a reduction in small-intestinal transit (using the H~2~ breath test). Both enteroglucagon and motilin concentrations fall after treatment. Concentration of another gut hormone, plasma peptide YY (also produced by endocrine cells in the ileum and colon and known to delay gastric emptying and small-intestinal transit, and to reduce gastric and pancreatic secretion) has been shown to be grossly elevated in PIM;[@bib89] it seems possible that this results from a change in peptide YY secretion, resulting from malabsorption, and is a compensatory mechanism in diarrhoea. Patients with PIM also have a reduced post-prandial rise in gastric-inhibiting polypeptide; gastrin and pancreatic polypeptide are normal. #### Role of the colon {#cesec39} The colonic mucosa, in addition to that of the small intestine, is abnormal in tropical PIM ('tropical colonopathy\').[@bib90] Few causes of diarrhoea are strictly confined to one or other of these organs; for example, *shigellosis* frequently involves the small intestine, and *salmonellosis* and *Campylobacter jejuni* infection of the colon. The normal colon is able to absorb 4--7 L of water/24 h,[@bib91] together with 100--160 mmol carbohydrate (as volatile fatty acid(s)). Failure of the diseased colon to 'salvage' the increased ileal effluent must increase the intensity of diarrhoea. Colonic abnormalities have been reported in tropical sprue; using a colonic perfusion system, impaired water and sodium absorption was demonstrated.[@bib92] Colonic function has not been investigated in tropical PIM investigated and treated in London. #### Animal model {#cesec40} A clinical syndrome which exhibits very close similarities to PIM has been described in the German shepherd dog.[@bib93] Jejunal biopsy specimens show villous atrophy with a variable infiltration of lymphocytes and plasma cells in the lamina propria. Aerobic bacteria are involved; both clinical and laboratory recovery take place after broad-spectrum antibiotic therapy. ### Clinical aspects {#cesec41} This is dominated by chronic diarrhoea with large, pale, fatty stools, and sometime excessive flatulence, usually following an acute intestinal infection.[@bib1], [@bib33], [@bib34], [@bib74] Weight loss is sometimes gross and is probably related to anorexia as much as to intestinal disease. [Figure 10.4](#f4){ref-type="fig"} shows an affected patient before and after chemotherapy. A wide range of clinical presentations exists, however, varying from the acute onset type (not strictly post-infective), described by Baker[@bib76] and Mathan[@bib77] as occurring in epidemics (with vomiting and pyrexia in up to 50%) at Vellore, India, to a far more chronic entity. Other clinical features, such as glossitis (aphthous ulceration was common in nineteenth-century reports), megaloblastic anaemia, fluid retention, depression, apathy, amenorrhoea and infertility, occur only after several months duration.Figure 10.4(A and C) A 19-year-old Englishman presented in London with post-infective tropical malabsorption (tropical sprue). Acute diarrhoea started soon after his arrival in Nepal and he lost approximately 12 kg in weight during the subsequent 2 months. The total urinary xylose excretion after a 25 g oral load was 2.5 mmol/5 h (normal range 8.0--16.0 mmol/5 h); the 24-h faecal fat was 83 mmol (normal range 11--18 mmol); the Schilling test result was 0.16% urinary excretion at 24 h (normal \>10%) and the 8-h serum concentration was 0% (normal \>0.6%) of the loading dose. Jejunal biopsy histology showed marked villous blunting with increased lymphocytes in the lamina propria. Parasites were not found in several faecal samples. Serum albumin 36 g/L; haemoglobin 13.2 g/dL; mean corpuscular volume 102.9; red blood cell folate 113 ng/L (normal \>150 ng/L); serum vitamin B~12~ 322 pg/L (normal \>150 pg/L). He responded rapidly to treatment with oral tetracycline and folic acid. (B and D) The same man 4 weeks after initiation of treatment when all investigations were normal. [Table 10.3](#cetable3){ref-type="table"} summarizes the more important differential diagnoses of chronic malabsorption in relation to tropical exposure (see below).[@bib80] There are also many non-infective causes of malabsorption in the tropics and subtropics; these should be excluded systematically.[@bib94] During, and immediately after, an acute small-intestinal infection, xylose, glucose, fat, B~12~ and folate malabsorption frequently occur (see above). After 4 months or so, moderate/severe morphological change occurs in the jejunal mucosa; serum folate and later B~12~ concentrations decline -- often to very low concentrations. Hypoalbuminaemia and oedema are late signs. Gastric acid secretion is often depressed, but whether this precedes, or is a sequel to, the initiating infection is unknown. The role of hypochlorhydria in the production of small-intestinal infection remains unclear. In a small proportion of cases in southern India, B~12~ absorption either improved or became normal with addition of intrinsic factor.[@bib90] Secondary hypolactasia may be present (see above).[@bib71] There is no good evidence that PIM predisposes to any gastrointestinal malignancy. ### Investigations {#cesec42} Investigations should include urinary D-xylose excretion, 72-h faecal fat estimation, a Schilling test and jejunal biopsy; faecal parasites should be excluded (1-h blood xylose concentration is in practice probably superior to a 5-h urinary collection in a tropical environment[@bib95]): serum B~12~ and red blood cell folate concentrations should be estimated; after 4 months of illness most patients have a low folate concentration. Serum albumin and globulin concentrations are often depressed. Monosaccharide absorption is impaired to a greater extent than that of amino acids.[@bib74] Barium meal and follow-through examination show dilated loops of jejunum with clumping of barium, in addition to reduced transit rate. Jejunal mucosal changes are variable, depending on the duration of the disease. By 3 or 4 months, most biopsies are ridged and/or convoluted; a flat mucosa is extremely unusual and, if present, gluten-induced enteropathy[@bib94] should be suspected. Submucosal invasion with lymphocytes (predominantly T cells) and plasma cells is usual. Ultrastructural changes in jejunal biopsy specimens have been studied;[@bib96] although lysosomes, peroxisomes and mitochondrial enzymes are not depressed, the organelles are more fragile. Endoplasmic reticulum is unchanged. A significant reduction in 5-nucleotidase in the basolateral (plasma) membrane persists after recovery. The latter finding might reflect an underlying abnormality in the enterocyte of individuals susceptible to PIM. Intestinal permeability has also been investigated;[@bib36], [@bib97] abnormalities in urinary excretion of lactulose and rhamnose following an oral load are similar to results obtained in gluten-induced enteropathy. ### Aetiology and treatment {#cesec43} A hypothesis to account for the aetiology of tropical PIM is summarized in [Figure 10.5](#f5){ref-type="fig"} .[@bib98] The 'vicious cycle' can be broken by (1) eliminating bacterial overgrowth, and (2) aiding mucosal recovery (with folic acid supplements). While this hypothesis has been challenged,[@bib99] a satisfactory alternative has not been produced. An adequate diet should be combined with tetracycline (250 mg three times a day for at least 2 weeks) and folic acid (5 mg three times a day for 1 month). Evidence of susceptibility of the responsible flora to antibiotics other than tetracycline is limited. Symptomatic treatment may be necessary in the acute stage of the disease; codeine phosphate (30 mg three times a day), diphenoxylate (2.5--5 mg four times daily), or loperamide (5--10 mg four times daily) are of value if stool frequency is excessive. Mild cases respond without treatment, but this may take several months. Recovery is usually rapid and straightforward;[@bib1], [@bib74], [@bib98] in the pre-antibiotic era a mortality rate of 10--20% was usual.Figure 10.5Hypothetical scheme to illustrate the pathogenesis of post-infective malabsorption. The open arrows indicate the vicious cycle which, once set in motion, is only broken by elimination of the abnormal intraluminal flora (†), and hastening of enterocyte recovery (\*). Evidence from south India suggests that response to antibiotics is less satisfactory;[@bib76], [@bib77] this has been used as evidence to support a viral rather than a bacterial aetiology being causative in that locality. ### Conclusion {#cesec44} The aetiology of PIM -- especially that presenting in association with tropical exposure -- is becoming clearer.[@bib99] It is probable that several primary insults to the enterocyte (of an infective nature) are involved. Whereas PIM resulting from most viral, bacterial and parasitic causes is usually self-limiting, this does not apply to the 'tropical sprue' syndrome, when well established. The reason why only a minority of affected individuals who suffer an acute small-intestinal infection are susceptible to PIM is unknown; a genetic (or ethnic) basis for susceptibility seems likely. Other causes of malabsorption in the tropics {#cesec45} -------------------------------------------- [Table 10.3](#cetable3){ref-type="table"} summarizes some of these. The role of parasitic infection has been highlighted by AIDS, in which prolonged diarrhoea accompanied by malabsorption and weight loss can be very troublesome.[@bib37] Incontrovertible evidence exists that HIV itself causes chronic enteropathy with villous blunting; crypt hypoplasia results from a direct effect of the viruses on cell replication, or by an unknown immunological reaction. This is a very common cause of persisting malabsorption in Africa. In this context, *Cryptosporidium parvum* and *Isospora belli* have recently come to the fore and it is now also clear that these organisms can produce a self-limiting illness simulating TD in immunocompetent adults and children (see below). *G. lamblia* (see below) is undoubtedly the most common cause of parasitic malabsorption.[@bib74], [@bib100], [@bib101] *Strongyloides stercoralis* (see below), which is widespread in tropical countries, was until very recently still present in approximately 15--30% of former prisoners of war in South-east Asia during World War II; it is an underdiagnosed cause.[@bib1], [@bib102] Of all causes of malabsorption related to tropical exposure, intestinal tuberculosis -- usually involving the ileocaecal region -- is probably that with the lowest index of suspicion among medical personnel.[@bib74], [@bib103] Abdominal tuberculosis can assume several clinical forms: apart from the hypertrophic ileocaecal form, glandular (involving the mesenteric glands), peritoneal (sometimes with ascites) and hepatic involvement (with granulomatous disease) are relatively common. With the first of these presentations, weight loss and diarrhoea are often accompanied by a low-grade febrile illness; in severe cases stools are large, pale and bulky. Examination reveals an ileocaecal mass in 35--50% of cases,[@bib103] and occasionally enlargement of one or more lymph glands; however, there is often no clinical abnormality. Late presentation can be as adult kwashiorkor. Anaemia and hypoalbuminaemia are common.[@bib103] Chest radiography is usually normal. Absorption tests are frequently abnormal; fat and B~12~ absorption are affected most severely. A protein-losing enteropathy may be present. Pathologically, the disease results either from miliary dissemination, or follows ileal ulceration. Malabsorption is caused by chronic bile salt loss; unabsorbed bile salts (normally re-absorbed in the terminal ileum) in turn interfere with colonic absorption. Barium meal and follow-through examination show ileal strictures,[@bib103] frequently multiple, in a high percentage of cases; the ascending colon may also be shortened. The major differential is Crohn\'s disease, which is statistically much less common in people indigenous to the tropics. *Yersinia* infection should also be considered. Chest radiography is usually normal. The tuberculin test is positive in 70--90% of cases.[@bib103] A needle liver biopsy specimen occasionally shows hepatic granulomas with caseation. Diagnostic laparotomy or peritoneoscopy (and peritoneal biopsy) is sometimes necessary in order to obtain a tissue diagnosis.[@bib103] Treatment is with an antituberculosis regimen (Chapter 56). Resection of stricture(s) and occasionally hemicolectomy are sometimes necessary; chemotherapy should be initiated before surgical intervention. A further cause of malabsorption in a tropical environment consists of the Mediterranean (α-chain) lymphoma,[@bib104], [@bib105] which occurs sporadically in many parts of the tropics. If started early, tetracycline usually produces a good result, but not always so. Although it seems overall uncommon in most indigenous populations in tropical countries recent reports of coeliac disease have been made from India[@bib106] and Turkey[@bib107] and some evidence exists that it might be increasing in prevalence.[@bib108] Other small-intestinal infections {#cesec46} --------------------------------- ### Viral infections {#cesec47} Significant intestinal protein loss (mean 1.7 g daily) and xylose malabsorption have been demonstrated in northern Nigerian children with measles (see Chapter 47); approximately 25% also had lactose malabsorption.[@bib109] Other infections in children caused by enteroviruses and herpes simplex viruses are also associated with diarrhoea and weight loss; malnutrition may result; the mechanism(s) (involving enterocyte damage) is probably similar to that in measles. Volunteers infected with enteric viruses develop small-intestinal morphological lesions which are not always associated with symptoms. Jejunal mucosal changes giving rise to severe malabsorption have been well documented in viral hepatitis;[@bib110] these may persist for a considerable time after resolution of the hepatic abnormalities. The norovirus (a 27 nm piconavirus) can also produce mucosal damage and malabsorption.[@bib111] Rotavirus infections give rise to morphological abnormalities and (especially in children) malabsorption.[@bib112], [@bib113] These viral infections are invasive, and the resulting diarrhoea and malabsorption are caused by enterocyte destruction. Malabsorption usually occurs after the virus has been shed into the intestinal lumen; the villi contain immature crypt-type enterocytes. In coronavirus infection(s) in piglets, which resemble human rotavirus infections, glucose absorption is significantly impaired.[@bib114] This has practical importance in management because sodium and water secretion cannot be reversed by glucose; oral rehydration fluids, commonly used in small-intestinal (including travellers\') diarrhoea (see above), contain a high glucose concentration which overwhelms the limited absorptive capacity. Baker[@bib76] and Mathan[@bib77] have suggested that coronavirus infections are responsible for at least some cases of 'tropical sprue' in southern India (see above); this might be the case, but asymptomatic individuals often excrete these viruses and this does not therefore indicate a cause-effect relationship. Also at *Vellore*, a search for evidence of *Berne* virus infection in 'epidemic tropical sprue' proved negative.[@bib115] ### Bacterial infections {#cesec48} Moderate to severe malabsorption is commonplace during acute intestinal infections of bacterial origin; subnormal absorptive capacity persists for variable periods after termination of the diarrhoea and apparent clinical recovery. In a study in Bangladesh, approximately 70% of patients had evidence of xylose malabsorption 1 week after the diarrhoea had ceased; this was less common after cholera than *Shigella* species, *Salmonella* species and/or *Staphylococcus* species infections; xylose and B~12~ malabsorption persisted for up to 378 and 196 days, respectively, after the diarrhoea had cleared. Although many different infective insults to the enterocyte are probably important in PIM (see above), evidence for bacteria being responsible currently has more solid support than that involving other agents. #### Escherichia coli {#cesec49} These organisms (with varying modes of pathogenicity) produce a spectrum of disease from TD to malabsorption by enterotoxin production and mucosal invasion-similar to that caused by Shigella species (Chapter 51). They are frequently food- or water-borne, and may cause outbreaks of gastroenteritis. Heat-labile enterotoxins exert an effect by activating adenylcyclase by a mechanism(s) similar to *V. cholerae*. Both heat-labile and heat-stable enterotoxins are probably important in TD (see above). A large pool of resistant *E. coli* (often showing resistance to multiple antimicrobials) now exists in the community. Enterotoxin production by *E. coli* may be transferred simultaneously with antibiotic resistance (Chapter 53); in a study, 72% and 44% of ETEC isolated in South-east Asia were resistant to one or more, and four or more antibiotics, respectively.[@bib116] Enterocyte adhesiveness of *E. coli* is also a property of some strains and that might be important in continuing colonization and subsequent malabsorption. The relationship between adherence and verotoxin production remains unclear.[@bib106] Attachment of microorganisms to the enterocyte prevents clearance by peristaltic activity; such mucosal receptors may be determined genetically.[@bib117] Ultrastructural studies have shown *E. coli* adherent to mucosal cells, with flattering of the microvilli, loss of the cellular terminal web and cupping of the plasma membrane around individual bacteria; intracellular damage was marked in the most heavily colonized cells. Histological improvement was demonstrated following clearing of *E. coli* with neomycin and nutritional support. This mechanism can lead to protracted diarrhoea in infants. In most cases, resultant malabsorption is short lived. #### Salmonellosis {#cesec50} Malabsorption occasionally follows infection with *Salmonella* species (Chapter 52),[@bib118] but the frequency is unknown. #### Campylobacter jejuni {#cesec51} Although unusual, dysenteric disease (bloody diarrhoea) has for long been known to predispose to tropical PIM;[@bib74] in addition to shigellosis it is clear that some cases are caused by *E. coli* (see above) and others by *Campylobacter jejuni* (Chapter 51). Although most cases of *Campylobacter jejuni* infection are acute, present with gastroenteritis and are self-limiting, initial symptoms can be prolonged.[@bib119] The disease is a zoonosis; poultry are frequently contaminated. Many outbreaks have been traced to infected cow\'s milk. Dogs also constitute a reservoir of infection. Although the infection is self-limiting, erythromycin probably hastens recovery when given early in a severe case. The carrier state is common. #### Enteritis necroticans (pigbel disease) {#cesec52} Although described in Germany at the end of World War II (1939--1945), and named Darmbrand,[@bib4], [@bib34] this acute infection ([Figure 10.6](#f6){ref-type="fig"} ), which is more common in children than adults, occurs in several tropical countries, notably the highlands of Papua New Guinea (where it is endemic),[@bib120] Thailand and Uganda. Recently, enteritis necroticans has been recorded in Khmer children at an evacuation site on the Thai-Kampuchean border of Thailand; in the former report 36 (58%) out of 62 affected children (10 months to 10 (mean 4) years) died.[@bib4] It seems likely that a disease termed 'necrotizing jejunitis' in rural areas of Bihar, India -- which also affects children -- represents the same entity; this condition ('segmental necrotizing enteritis\') has also been recorded in Jaipur, India, and in Sri Lanka.[@bib4] Scanty reports of a similar condition have also been made from northern Europe, which suggests that the disease exists worldwide, but only reaches epidemic proportions when suitable conditions exist, most importantly for the β-toxin of *Clostridium perfringens* type C (ingested in contaminated foodstuffs) to take its toll. Murrell[@bib121] has suggested (in the light of historical evidence) that the disease was widespread in medieval Europe when 'human habitats, food hygiene, protein deficiency and periodic meat feasting formed the basics of village life as they do in many Third World cultures today\'. Enteritis necroticans is now known to be caused by the ingestion (often at pig feasts or 'mumus\') of food contaminated by *Cl. perfringens* type C.[@bib120] The pathophysiology of the disease is complex, but the presence of a low concentration of trypsin (resulting from trypsin inhibitors in foodstuffs and chronic protein-energy malnutrition) allows the β-toxin of *Cl. perfringens* to survive and produce mucosal injury.[@bib34] It is sometimes associated with persisting structural changes in the small intestine; malabsorption may be a sequel.Figure 10.6Gangrenous small intestine at post-mortem in a Papua New Guinean child who had died from necrotizing enteritis (pigbel disease). Fluid and electrolyte replacement are essential (see below). Tetracycline or chloramphenicol, and type C gas gangrene antisera are of value; laparotomy is often indicated. In Papua New Guinea, immunization against *Cl. perfringens* type C has given good results;[@bib34] in a controlled trial, marked reduction in incidence and mortality was demonstrated in the treatment group. A management strategy has been outlined.[@bib120] ### Parasitic infections {#cesec53} A study carried out in Sierra Leone has indicated that both protozoan and helminthic infections are particularly common in displacement camps.[@bib122] #### Giardiasis {#cesec54} The spectrum of disease caused by this flagellated protozoan is broad.[@bib1], [@bib74], [@bib100], [@bib101] Symptoms vary from subclinical cases to those with severe malabsorption and malnutrition. The reason why some individuals are prone to symptomatic giardiasis is not clear; size of infecting dose, strain variability, genetic predisposition, acquired immunity factors, achlorhydria, a local secretory IgA deficiency and the presence of blood group A phenotype have all been considered. An increase in IgE and IgD cell numbers has been reported in the jejunal mucosa of 20 affected patients;[@bib123] the former reversed after treatment, when an increase in IgA cell numbers was also recorded. Genetic characterization has recently been reported from Ethiopia.[@bib124] The actual mechanism by which the trophozoites cause an absorptive defect is also unclear. Mucosal injury, with or without invasion, bacterial overgrowth in association with parasitization, and bile salt deconjugation by bacteria and/or parasites have all been considered. The extent of jejunal morphological abnormality varies widely. Clinical presentation is usually between 1 and 3 weeks after infection; contaminated water and, less commonly, food are the usual sources of infection. Infection occurs both endemically and epidemically. The disease can probably be contracted from domestic animals.[@bib125] It is more common in male homosexuals, but is not an opportunistic infection in AIDS sufferers. Diarrhoea of acute onset, flatus and weight loss may all be present; the stools have the characteristics of malabsorption. The disease is clinically indistinguishable from PIM; investigations also give similar results. A full-blown case has all of the clinical and laboratory features of the classical (historical) reports of 'tropical sprue' (see above). Cysts may be found in a faecal specimen; trophozoites can be detected in either a jejunal biopsy or jejunal fluid, or with the string test ('Enterotest\'). If mucosal changes and malabsorption exist, circulating antibodies to *G. lamblia* cysts can often be detected. Treatment is with metronizadole (2 g on three consecutive days); alcohol should be avoided during the treatment period. A single dose of tinidazole (2 g orally) has been used with success. Two 5-nitroimidazoles -- ornidazole and tinidazole (as a single 1.5 g dose) -- have been compared;[@bib126] recurrence of infection during the subsequent 2 months was similar in each case (about 10%). Nimorazole has also been used. An alternative is mepacrine (100 mg three times daily for 10 days), which is less often used. #### Cryptosporidium parvum {#cesec55} The importance of farms as a source of infection has been emphasized in a study from Zambia.[@bib127] Importance of domestic pets as a source of infection has also recently been emphasized in a study carried out in Peru.[@bib128] Like *G. lamblia*, this organism produces a broad spectrum of disease; prolonged infection usually, but not always, occurs in the immunosuppressed (including AIDS) sufferer where the organism is opportunistic. Diagnosis is similar to that for *G. lamblia* infection; oocysts are usually detectable in a faecal sample. Treatment (rarely indicated in the immunointact) is with spiramycin, but is usually ineffective in the immunosuppressed; although at least 70 other compounds have been tested, none, including spiramycin, has proven efficiency in vitro. #### Other parasites {#cesec56} The vast majority of small-intestinal parasitic infections do not result in signs/symptoms unless present at a high concentration.[@bib32] In a heavy infection, hookworm is responsible for hypochromic anaemia; *A. lumbricoides* rarely accounts for obstruction in the small intestine and biliary and pancreatic ducts (Chapter 85). The major clinical sequel of tapeworm infection is neurocysticercosis (*Taenia solium*) (Chapter 87) -- a complication unrelated to the intestinal tract. Although *A. lumbricoides, Ancylostoma duodenale* and *Necator americanus* have at various times been implicated in malabsorption, there is no clear evidence except in rare or anecdotal case reports.[@bib129] *Diphyllobothrium latum* infections are occasionally associated with a low serum B~12~ concentration; however, this is caused by B~12~ uptake within the small-intestinal lumen, and is not an example of true malabsorption. Clear evidence exists that *Strongyloides stercoralis* is causally related to malabsorption.[@bib1], [@bib34], [@bib74], [@bib102] This helminth can survive in the human host for several decades; some 10--20% of exprisoners of war in South-east Asia during World War II (1939--1945) remained infected until recently. Onset of diarrhoea is less acute than with *G. lamblia*. Larvae can be demonstrated by the 'Enterotest\', and less often by jejunal biopsy. Ova and larvae can occasionally be detected in faecal specimens. Eosinophilia may be gross; however, it is often absent. The immunofluorescent antibody test (IFAT) is positive in approximately 70% of cases; however, cross-reaction with filaria is common. The enzyme-linked immunosorbent assay (ELISA) test, when available, is more specific. A negative serological result is common in the immunosuppressed patient. Treatment is with thiabendazole (1.5 g twice daily on three successive days); repeated courses may be required. Albendazole (400 mg daily for 3 days) seems less effective. In animal experiments, cambendazole has given encouraging results; this has also been the case in limited clinical studies, but the compound has not been officially released for human use. Other *Strongyloides* species are important, especially in children. *Stongyloides fülleborni* has been implicated in the pathogenesis of severe PIM (see above) in Zambia and Papua New Guinea, where a significant mortality rate has been recorded.[@bib74] In the northern Philippines and Thailand, *Capillaria philippinensis* has been causally associated with PIM.[@bib1] It can occur in epidemics. Diarrhoea of acute onset is followed by malabsorption and, if untreated, infection carries a substantial mortality rate. Protein-losing enteropathy may also be present. Treatment with one of the benzimidazole compounds has given good results. The protozoa *Isospora belli* and *Sarcocystis hominis* (usually conveyed by undercooked pork and beef)[@bib130] also cause malabsorption. These organisms replicate within the enterocyte. *I. belli*, like *Crytosporidium parvum*, causes a spectrum of disease, from TD to PIM, and is more common in the immunosuppressed individual. Pyrimethamine + sulfadiazine, and co-trimoxazole + nitrofurantoin, have been used with some success. Other protozoan parasites, such as *P. falciparum* (in an acute infection) and visceral leishmaniasis (kala azar), can also produce significant malabsorption. Other protozoa which have assumed practical importance in the wake of the HIV/AIDS pandemic are *Cyclospora cayetanensis*,[@bib131], [@bib132], [@bib133] *microsporidiosis*,[@bib134], [@bib135] and *Blastocystis hominis*.[@bib136] All can be implicated in a wide range of small-intestinal problems ranging from traveller\'s diarrhoea to malabsorption. Emergencies {#cesec57} ----------- Severe dehydration consequent upon secretory watery diarrhoea accounts for enormous amounts of acute morbidity throughout the tropics; this applies especially to infants and children. Intravenous replacement therapy has been in use for more than 150 years; Dr Robert Lewins MD FRCP, of Leith, recorded that he had witnessed Dr Thomas Latta inject saline intravenously into a patient suffering from cholera (see above) in 1832,[@bib65] and George Leith Roupell,[@bib137] a physician at St Bartholomew\'s Hospital, London, seems to have been an early user of this technique. It is unlikely, however, that these were the first attempts at intravenous rehydration (in fact, Sir Christopher Wren, better known for his architectural achievements, had used the technique experimentally in 1657). Nearly three-quarters of a century passed before Sir Leonard Rogers, working at Calcutta, demonstrated a reduction in the mortality rate in cholera patients from 70% to 20% by use of this technique. Introduction of oral rehydration regimens had to wait much later, in fact until the latter half of the twentieth century. Introduction of this form of management, which followed upon important basic applied physiological observations, was, in a world context, one of the most important medical advances during the twentieth century.[@bib1] In many acute medical conditions, gastric emptying is delayed; however, this is not the case in cholera (and presumably other acute small-intestinal infections) and does not constitute a barrier to oral rehydration, even when fluid and electrolyte loss (in the stool) is severe.[@bib138] Oral rehydration therapy remains grossly underused,[@bib139] however, and infants and children in developing countries with acute gastroenteritis continue to die unnecessarily because this simple technique is not readily applied. The authors of this latter article have concluded: 'the impediment to its wide acceptance may be that it is counterintuitive for a simpler and much less expensive treatment to be an improvement over an effective but more complicated technology\'! ### Enteritis necroticans (pigbel disease) {#cesec58} This acute small-intestinal emergency (see above), which usually affects infants and children (see above) is characterized by gangrenous changes in the small-intestinal wall (in patchy distribution); the jejunum is most markedly affected, but the ileum is also involved. Presentation is usually as an acute abdominal (surgical) emergency, with abdominal pain, fever and bloody diarrhoea (see above). A chronic stage of the disease may ensue in which there is narrowing of the small-intestinal lumen (in one or more places) by a fibrotic stenosis or adhesion; clinical presentation is with subacute obstruction, often accompanied by malabsorption and malnutrition. Fluid and electrolyte replacement are vitally important in management; gastric suction is also required. Penicillin or another antibiotic should be given (see above). Laparotomy is frequently indicated to confirm the diagnosis and to resect the necrotized, haemorrhagic, segment(s) of small intestine. Fortunately, active immunization against the β-toxin has proved effective prophylaxis in Papua New Guinea; hospital admissions for pigbel in one area of the country fell to less than one-fifth of the previous figure (*p*\< 0.001) when a vaccination programme was introduced.[@bib140] Morbidity due to this acute abdominal emergency (with a very high mortality rate) should eventually fall in the seriously affected countries. ### Paralytic ileus and acute obstruction {#cesec59} In Pakistan, paralytic ileus has been recorded as a late complication of acute diarrhoeal disease in infants;[@bib141] despite rehydration and total parenteral nutrition, the mortality rate was 25%. When compared with others who did not develop ileus (following acute diarrhoeal disease), these infants were shown to have had significantly more antimotility agents preceding the ileus; furthermore, many had a depressed serum potassium concentration. The potential dangers associated with antiperi-staltic agents, especially in infancy and childhood, are thus re-emphasized. Acute intestinal obstruction constitutes a common surgical emergency in both children and adults in many parts of the tropics, including Africa. Strangulated hernia (usually of inguinal origin) is usually the most common cause; volvulus and intussusception are relatively common in tropical Africa; tuberculosis is a further cause due either to stenosis or to pressure on the third part of the duodenum or jejunum. A heavy *A. lumbricoides* infection (especially in children) can also produce small-intestinal obstruction;[@bib142] when diagnosed clinically, laparotomy can usually be avoided. Management consists of intravenous hydration, nasogastric suction and appropriate anthelmintic chemotherapy. Strangulated hernia, volvulus and intussusception nearly always require laparotomy.[@bib142] In a report from southern India, 904 children presented with intestinal obstruction;[@bib143] the most common causes in order of frequency were necrotizing enteritis (see above), acute intussusception, band obstruction, subacute obstruction, and remnants of the vitello-intestinal duct. Rare causes of smallintestinal obstruction include: Burkitt\'s lymphoma, Mediterranean lymphoma (α-chain disease) (see above) and intestinal schistosomiasis. Small-intestinal trauma -- caused by a road accident or knife, arrow or gunshot wound -- is also important in a tropical context. ### Typhoid (enteric) fever {#cesec60} In most areas within the developing world, typhoid (see also Chapters 52 and 53 (and to a lesser extent tuberculosis) accounts for much small-intestinal disease encountered in surgical practice;[@bib144] perforation, obstruction and less often haemorrhage constitute acute surgical emergencies. This seems especially important in West Africa. *S. typhi* infection is also an increasing problem in travellers from industrialized countries to the tropics;[@bib145] in the USA, 2666 cases (fatality rate 1--3%) of acute enteric fever were officially notified between 1975 and 1984; 62% of them were imported, the majority of infections having originated in either Mexico or India. Statistically, surgical complications are unusual; thus in a series of 82 culture-positive cases in The Gambia there were no surgical complications;[@bib146] this was also the case in a series of 192 cases of enteric fever -- most caused by *S. typhi* -- in Thailand.[@bib147] Despite its relative rarity, however, (perhaps 2--4% of cases worldwide), typhoid perforation is an extremely serious event, accounting for 20--60% of deaths in this disease (a statistic which is increased by late presentation, female sex, age ≥40 years and the presence of multiple perforations). Late perforation is often indistinguishable from a perforated appendix, amoebic liver abscess, tuberculous peritonitis, an infected ruptured ectopic pregnancy or intestinal strangulation. The optimal form of management seems to be surgical, provided the patient is not too shocked to endure such a procedure (a prolonged period of preoperative resuscitation is often required). There is as yet no general agreement, however, regarding the ideal type of operative intervention;[@bib148] simple closure, ulcer excision and closure, wedge excision and closure, ileal resection and anastomosis, resection and transverse ileotransverse colostomy, and right hemicolectomy have all found favour. When the perforation is single, simple closure (with or without excision) is the procedure of choice; an area(s) of impending perforation should not be oversewn; closure should always be in two layers: an inner one of chromic catgut and an outer of silk. When there are three or more perforations, bowel resection is probably advisable. Peritoneal lavage with a copious amount of washing with normal saline should be carried out. The incidence of postoperative complications is high, and includes peripheral vascular failure, respiratory infections, anaemia, sepsis, abscess formation, burst abdomen and intestinal obstruction.[@bib148] Re-perforation or a new perforation is possible. In a series of 108 consecutive cases of perforated typhoid enteritis managed in western Nigeria, 100 (93%) underwent 'debridement of the perforation and two-layer bowel closure\';[@bib149] 35 patients died, usually from overwhelming sepsis. In addition to specific chemotherapy, although chloramphenicol (1 g four times daily in an average adult, reduced to 1 g twice daily when body temperature is normal) remains the agent of choice, increasing numbers of reports of multiple-antibiotic-resistant strains of *S. typhi* are being reported (especially from India), metronidazole, and possibly corticosteroids, seem to improve the prognosis. Alternative chemotherapeutic agents include amoxicillin, co-trimoxazole, trimethoprim and ciprofloxacin; the last agent is indicated when there are serious doubts about sensitivity to the other compounds, as is frequently the case when infection has resulted in Asia. Despite these advances therefore, ileal perforation in enteric fever remains a potentially lethal complication, especially in children.[@bib150] Haemorrhage is rarely life-threatening, although recorded;[@bib151] whereas the majority of cases can be treated conservatively (blood transfusion when indicated), when selective angiography, fibreoptic endoscopy and high-resolution radionuclide imaging are available, localization of the bleeding site can be delineated and appropriate surgery instituted. ### Emergencies associated with helminthiases {#cesec61} Abdominal discomfort (and pain) are common sequelae to heavy small-intestinal nematode infections (see above), especially ancylostomiasis and *A. lumbricoides* (see above), but serious acute complications (see above) are fortunately rare.[@bib152] Anisakiasis, for example -- usually acquired from ingestion of undercooked or raw infected fish (sushi and sashimi) -- can present with an acute appendicitis-like illness.[@bib153], [@bib154], [@bib155] Invasive disease caused by this organism is usually localized to the ileocaecal region; there is no satisfactory parasitological or serological test, and chemotherapy is not effective. A diagnostic laparotomy is often necessary. Eosinophilic enteritis is an entity of multiple aetiology.[@bib156] A report from Townsville, Australia, suggested that *Ancylostoma caninum* (the dog hookworm) was responsible for an epidemic (93 cases) encountered there;[@bib157] nine were subjected to diagnostic laparotomy: eosinophilic infiltration involving a segment of ileum with indurated thickening of the distal small intestine and proximal dilatation was the usual underlying pathology. A rare case of acute mesenteric ischaemia (accompanied by segmental small-intestinal infarction and gangrene) caused by *Schist. mansoni* has been reported from Baghdad, Iraq.[@bib158] The small intestine can also be involved in *Schist. japonicum* infection; intestinal obstruction resulting from mesenteric ischaemia, an intussuscepting polypoid mass or fibrotic stenosis are possible sequelae. Intestinal perforation resulting from infection with the acanthocephalan helminth *Macracanthorhynchus hirudinaceus*, a natural intestinal parasite of the pig, has been described in Bangkok, Thailand[@bib159] (eight other cases are on record); this infection has also been reported from several other parts of the world, including China and southern Europe. Fatal gastrointestinal haemorrhage (associated with fluctuating jaundice, a tender liver, palpable gallbladder and an eosinophilia) has been attributed to *Fasciola hepatica* (liver fluke) infection in Harare, Zimbabwe;[@bib160] the site of bleeding was probably the biliary tree. COLORECTUM {#cesec62} ========== Most cases of colorectal disease occurring in a tropical environment have an infective basis ([Table 10.4](#cetable4){ref-type="table"} ); they are dominated by bacterial (*Shigella* species[@bib161] (Chapter 51) ([Figure 10.7](#f7){ref-type="fig"} ), *Campylobacter jejuni* and invasive *E. coli*) and protozoan (*Ent. histolytica* (Chapter 79) and *Balantidium coli*) infections. *Amoebic colitis* [@bib162] and *shigellosis* present classically with bloody diarrhoea; this should be differentiated from carcinoma, necrotizing colitis, antibioticassociated colitis and inflammatory bowel disease (which is overall not very common in tropical countries). Whether or not amoebic colitis can proceed to inflammatory bowel disease is debatable; however, misdiagnosis of amoebic colitis as inflammatory bowel disease (with subsequent corticosteroid therapy) can result in fatality. In AIDS, cytomegalovirus colitis is common; *Cryptosporidium* is usually a small-intestinal parasite, but colonic involvement can also occur. In addition, megacolon resulting from South American trypanosomiasis (Chagas\' disease) (Chapter 76) is another cause of colonic pathology. Of diseases localized to the anal region, lymphogranuloma is perhaps the most important although bacterial (including donovanosis, syphilis and gonorrhoea (Chapter 21)) and parasitic (including *Ent. histolytica, Schistosoma* species and *Enterobius vermicularis*) infections constitute differential diagnoses.Table 10.4Colorectal diarrhoea[a](#cetablefn1){ref-type="table-fn"}Bacterial infectionShigellosis*Campylobacter jejuniEscherichia coli* (enteroinvasive)Protozoan infection*Entamoeba histolyticaBalantidium coli*Schistosomiasis (usually *Schistosoma mansoni* and *Schist. japonicum*)Unusual causesNon-specific ulcerative colitis -- inflammatory bowel disease[b](#cetablefn2){ref-type="table-fn"}Crohn\'s disease[b](#cetablefn2){ref-type="table-fn"}AppendicitisDiverticulitisHaemorrhoidsColonic carcinomaIrritable bowel syndrome[^1][^2]Figure 10.7Severe amoebic colitis: operative specimen obtained from an Australian nurse misdiagnosed as having non-specific ulcerative colitis (inflammatory bowel disease) while working in Papua New Guinea. Overall, diseases of the colorectum are far less common in indigenous people in developing countries compared with individuals in industrialized ones;[@bib1], [@bib163] colonic carcinoma seems, for example, to be an unusual lesion in rural communities. Good evidence now exists that frequency of these diseases is increasing as urbanization advances, in Africa especially. Hypotheses to account for these differences include high dietary fibre consumption in most tropical countries; however, such associations rarely have a proven cause--effect relationship. Many data have been collected on colonic function in indigenous inhabitants of developing countries;[@bib1] it seems likely that mean 24-h faecal weight and volume is higher in Africa, and constipation unusual. Overall, intestinal transit rate also seems more rapid. Limited evidence indicates that colorectal histology is mildly different in indigenous people in developing countries, and is comparable to tropical enteropathy (see above). In PIM in India (see above) in vivo colonic functional abnormalities have been demonstrated. Whether colonic pathology is important in a nutritional context remains difficult to evaluate (see above): evidence now exists that this organ is important in the absorption of nitrogen and free (volatile) fatty acids. Inflammatory bowel disease (non-specific ulcerative colitis and Crohn\'s disease)[@bib164] is probably less common overall in indigenous people in developing countries compared with the UK and other Western countries.[@bib165], [@bib166] However, a recent study from Lebanon demonstrated a high prevalence of the disease there.[@bib167] The aetiology of this disease is unknown, although an infective basis has frequently been suggested; satisfactory evidence for a viral or bacterial (possibly mycobacterial) origin is at present lacking. A handful of reports of ulcerative colitis have been made from African countries, and a few more from Asia.[@bib164] In individuals in the UK with an ancestry in the Caribbean or Indian subcontinent this disease clearly exists but is unusual. Such differences also apply to Crohn\'s disease, although this disease also is well recognized in Caribbean people in the UK. Although Crohn\'s disease behaves very much like intestinal tuberculosis in clinical practice, response to antituberculous therapy is disappointing. When inflammatory bowel disease occurs, it seems to behave similarly to that in the indigenous population of the UK. It is a common cause of bloody diarrhoea in travellers who have returned to temperate from tropical countries ([Figure 10.8](#f8){ref-type="fig"} ).[@bib39], [@bib40], [@bib41] Similarly, appendicitis, diverticular disease and haemorrhoids are overall less common in a developing country population, where a high-fibre intake has been implicated in their prevention; a causative association has not, however, been proved.Figure 10.8Barium enema in a 35-year-old English woman who experienced bloody diarrhoea during a visit to Africa; she had not previously had significant gastrointestinal problems. Colonic biopsy specimen obtained at colonoscopy confirmed inflammatory bowel disease. Although irritable bowel (IBD) syndrome (spastic colon)[@bib168] is extremely common in UK residents (and others) following an intestinal infection acquired in a tropical country, it seems to be far less significant in indigenous peoples in Africa[@bib169] and Asia. Whether this constitutes a genuine difference is unclear because so many of the latter have more severe symptoms of different origin(s), which might mask symptoms resulting from IBD. This syndrome does not constitute a single entity;[@bib170], [@bib171] although some cases respond to mebeverine or peppermint oil, many do not. There is no doubt that recognition of the syndrome in developing countries leaves much to be desired.[@bib172] More studies are required. *Enterobius vermicularis* infection (Chapter 85) is arguably the most common gastrointestinal infection in the world;[@bib173] it exists in both tropical and temperate areas. Colonoscopy is an endoscopic technique which is now available in some, but by no means all, developing countries; frequently, it is available only at the teaching hospital and/or other (tertiary referral) centre(s). Emergencies {#cesec63} ----------- ### Invasive amoebic colitis {#cesec64} Perforation, although a rare event, can complicate this disease, with the production of amoebic peritonitis;[@bib1] there may be diffusion of *Ent. histolytica* from a 'blotting-paper\'-like colon, and perforation (especially in the rectosigmoid or caecal regions or to the retroperitoneal tissues) or leakage into a confined space (resulting in a pericolic abscess or internal intestinal fistula). Management consists of gastric suction and intravenous fluid replacement; metronizadole, 500 mg 8-hourly (preferably by the intravenous route), and a broad-spectrum antibiotic should immediately be given. The colon is extremely fragile; laparotomy is usually best avoided;[@bib174] overall, mortality is of the order of 50% and after surgery close on 100%. Two reports have recorded results of surgical intervention in 15 patients with fulminant amoebic colitis.[@bib175], [@bib176] In the first, five out of six patients (four had a subtotal colectomy with ileostomy, and two a right hemicolectomy and ileostomy) subsequently died (none was diagnosed either preoperatively or during surgery); in the second, three out of nine died, all of whom had exteriorization of the cut ends of the bowel following resection of the necrotic segment (four of those who died had end-to-end anastomoses, and two peritoneal drainage). ### Shigellosis {#cesec65} A recent study in China has drawn attention to a significant climatic factor in prevalence.[@bib177] Although perforation is less common in shigellosis compared with amoebic colitis, haemorrhage is well documented. The most recent pandemic of this disease in the Western Hemisphere began in Guatemala in 1969 and ended in 1973. It spread rapidly to Nicaragua, Belize, Honduras, Costa Rica, Panama and Mexico; with an estimated 500 000 affected, of whom 20 000 died.[@bib178] ### Appendicitis {#cesec66} Overall, this entity is less common in developing compared with 'westernized' countries. Nevertheless it certainly exists, and a predominance of appendicectomies in women has been recorded.[@bib179] Confusion with an acute gynaecological condition is a real problem and more widespread use of ultrasound and laparoscopy might be the solution.[@bib180] In Calabar, Nigeria, 603 consecutive cases were investigated prospectively during a 5-year period;[@bib181] there were no major differences from this disease in industrialized countries, and it constituted the second most common abdominal emergency during the study period, being less common than acute intestinal obstruction. Many causative agents have been implicated; in a retrospective review of 2921 appendicectomies carried out at Allahabad, India, during a 25-year period, 153 produced histological evidence of a specific infection:[@bib182] tuberculosis (70), *Ent. histolytica* (17), *A. lumbricoides* (13), *A. lumbricoides* and *Trichuris trichiura* (2), *Enterobius vermicularis* (41), and *Taenia* species (2). This acute disease should be differentiated from pelvic inflammatory disease, typhoid enteritis, ruptured ectopic pregnancy, psoas abscess, acute amoebic colitis, and *Schist. mansoni* colitis.[@bib183] Although the vast majority of cases of appendicitis in developing countries result from a bacterial cause, helminths, including *Schist. mansoni*, *Strongyloides stercoralis*, *Trichuris trichiura* and *E. vermicularis* have also been implicated.[@bib1], [@bib184] ### Volvulus of the colon {#cesec67} This is a disease with clear geographical differences; it is common in much of Central and East Africa, India and South America;[@bib1] numerous reports have been made from Uganda and Zimbabwe. Although genetic factors have been suggested for these high rates; a high-fibre diet, common in most of Africa, has also been implicated. The major complication is strangulation, and gangrene of a colonic segment; this should be differentiated from primary volvulus of the small intestine, compound volvulus (usually ileosigmoid) and internal herniae. Distension can be relieved with a flatus tube; at laparotomy the nature of the operation, and extent of resection, depends on the length of gangrenous colon. With simple volvulus, mortality rate should be low. Zimmerman et al.[@bib185] have emphasized the value of emergency colonoscopy in the diagnosis of colonic volvulus; when the mucosa is ischaemic or necrotic, emergency laparotomy is indicated, but when appearances are normal, relief of flatus (with a flatus tube passed per rectum) together with medical management followed by elective surgery (resection and anastomosis) 10 days later is recommended. ### Colonic intussusception {#cesec68} The common variety, especially in West Africa, is the caecocolic one; although children may be afflicted, the vast majority are in adults.[@bib1] The condition has also been reported to be by no means uncommon in Ethiopian adults.[@bib186] Aetiology, as with that of volvulus, is conjectural; while an intestinal polyp or amoeboma accounts for some, there is no obvious clue in most cases. Gangrene is about three times more common with the ileoileal and ileocaecal varieties compared with the caecocolic type. ### Acute colonic dilatation {#cesec69} Several gastrointestinal infections can cause toxic megacolon. These include: *Salmonella species, Campylobacter* species and *Y. enterocolitica* infection; however, there has been a growing recognition of *Shigella* species in this potentially lethal condition.[@bib187] Correct diagnosis is essential; an unnecessary laparotomy can thus usually be avoided. If the condition is misdiagnosed as ulcerative colitis, and corticosteroids administered, potentially fatal consequences can ensue. Diagnostically, the causative organism can usually be identified in a faecal sample. Choice of an appropriate antibiotic is often difficult; in *Shigella* species infection, a fluoroquinolone, e.g. ciprofloxacin (200 mg intravenously 12-hourly for 10 days), seems most appropriate. Toxic dilatation of the colon has also been reported, albeit rarely, in *Ent. histolytica* infection;[@bib188] these authors recorded a single case (in which total colectomy, and administration of metronidazole and emetine, was followed by recovery); they were able to detect seven cases in the world literature. ### Other colorectal lesions {#cesec70} Anorectal infections in relation to tropical exposure have been reviewed.[@bib189] Trauma to the colon, often resulting from road accidence, constitutes a medical emergency in most tropical countries.[@bib1] Necrotizing colitis (the pathology is similar to that of enteritis necroticans; see above) is rarely encountered. Colonic obstruction is rarely caused by carcinoma (a rare tumour in the rural tropics[@bib190]) but is recorded following introduction of a foreign body per rectum. Colorectal tuberculosis is an unusual cause of stricture formation, which occasionally requires surgical intervention.[@bib191] While the true prevalence of *Clostridium difficile* infection in developing countries is considered to be low, many more studies are required.[@bib192] LIVER AND BILIARY SYSTEM {#cesec71} ======================== Liver histology in an individual indigenous to a tropical country differs from that in one who has spent his/her life in a temperate region of the world.[@bib1] This organ is subjected to numerous systemic infections -- viral, bacterial and parasitic -- and it lies at the distal end of the portal circulation; it is therefore bathed with portal blood containing viruses, bacteria, parasites, ova, products of digestion and other antigens. Thus, Kupffer cell hyperplasia and periportal infiltration (with lymphocytes, plasma cells and eosinophils) are more common, and stellate fibrosis occurs more frequently. Also, nuclear pleomorphism in hepatocytes and sinusoidal lymphocytes are frequently prominent; these appearances are unusual in biopsies obtained in a temperate country. Malaria and schistosomal pigment are often also present. Granulomas are common ([Figure 10.9](#f9){ref-type="fig"} ) and a large number of differential diagnoses exist; [Table 10.5](#cetable5){ref-type="table"} lists some of them.Figure 10.9Liver biopsy specimen from a 30-year-old Zambian woman. A degenerating *Schistosoma mansoni* egg is surrounded by a well-formed granuloma.Table 10.5Some causes of hepatic granulomas in tropical countriesInfectionViral cytomegalovirus, Epstein--Barr virusBacterialTuberculosis and atypical mycobacteria, leprosy, syphilis, Q fever, brucellosisParasiticSchistosomiasis, ascariasis, strongyloidiasis, toxocariasis, filariasis, enterobiasis, visceral leishmaniasisFungiHistoplasmosis, coccidioidomycosis, aspergillosis, actinomycosis, candidiasisNeoplasmsLymphomas -- especially intra-abdominal Hodgkin\'s diseaseOthers(sarcoidosis) therapeutic agents -- especially sulfonamides Acute liver infections {#cesec72} ---------------------- Jaundice in a tropical context ([Table 10.6](#cetable6){ref-type="table"} ) is most commonly a result of viral hepatitis (types A,[@bib193], [@bib194] B (sometimes a combined infection with D), C,[@bib195] E[@bib196], [@bib197], [@bib198], [@bib199], [@bib200] and F) (Chapter 39), but other causes should also be considered; [Table 10.6](#cetable6){ref-type="table"} summarizes some of them. An important cause is the jaundice of acute bacterial infection -- most commonly caused by pneumococcal lobar pneumonia or pyomyositis.[@bib1] The mechanism of this form of jaundice is complex and consists of hepatocellular, cholestatic and haemolytic elements; the importance of the latter depends on the underlying prevalence of glucose-6-phosphate dehydrogenase (G6PD) deficiency in the population under consideration (Chapter 13). It is important to differentiate this form of jaundice from viral hepatitis, otherwise the appropriate antibiotic will not be administered for an underlying bacterial infection. In addition to yellow fever, several other viruses are implicated;[@bib195] dengue fever, Kyasanur Forest disease, herpes simplex and Coxsackie virus should also be considered.Table 10.6Some causes of jaundice in the tropicsJaundice of acute bacterial infection: pneumococcal lobar pneumonia, pyomyositisVirusesHepatitis (A-F)Yellow feverEpstein-Barr virusCytomegalovirusMarburg and Ebola diseasesLassa feverBacteriaLeptospirosisTyphoid feverSyphilisGonococcal diseaseBartonellosisParasitesMalaria (acute *Plasmodium falciparum* and *P. vivax)*SchistosomiasisAmoebiasis (rarely)ToxoplasmosisTrichinellosisFascioliasispredominantly large-duct obstructive jaundiceClonorchiasisOpisthorchiasisAscariasisHydatidosis (rarely)GeneticSickle cell diseaseGlucose-6-phosphate dehydrogenase deficiencyDubin-Johnson syndrome In AIDS, the liver is affected by many opportunistic organisms. These include viruses; hepatitis B (HBV) and C (HCV)[@bib201] infections can be especially virulent. A liver biopsy specimen may also yield evidence of cytomegalovirus, *Mycobacterium tuberculosis*, *M. avium intracellulare*, atypical mycobacteria, *Cryptosporidium parvum, Pneumocystis carinii, Cryptococcus* species and/or Kaposi\'s sarcoma. Cholestatic features are common. The co-existence of HIV and HBV should not be underestimated.[@bib202] In addition to septicaemia, several other infections can produce jaundice;[@bib1], [@bib203], [@bib204] leptospirosis is frequently accompanied by renal involvement, while overt jaundice in typhoid fever 'hepatitis' is unusual.[@bib205], [@bib206] Melioidosis, plague, tularaemia and relapsing fever can also produce hepatitis. Of parasitic causes, acute *P. falciparum* infection is probably the most important. In acute (Katayama syndrome) and severe chronic schistosomiasis jaundice may be present, but is rare in invasive hepatic amoebiasis. Most parasitic infections, including African trypanosomiasis (Chapter 75) and visceral leishmaniasis (Chapter 77), can produce significant hepatitis and deranged hepatocellular function -- often in the absence of clinical jaundice. Several parasites produce large duct biliary obstruction; for practical purposes, *A. lumbricoides* is the most important to recognize and treat urgently.[@bib207] Sickle cell disease and haemoglobinopathies (Chapter 13) are important causes of haemolytic jaundice; they possess a genetic basis. Jaundice in the presence of G6PD deficiency is frequently precipitated (or worsened) by therapeutic agents and/or toxins. In some parts of the tropics, especially Indonesia and Papua New Guinea, the Dubin-Johnson syndrome seems unusually common. Chronic liver disease {#cesec73} --------------------- Most cases of chronic active hepatitis in tropical countries result from HBV and HCV infections;[@bib208] corticosteroids should not be administered for they exacerbate hepatocyte viral infection; interferon-γ and adenine arabinoside have given encouraging results, but ethnic factors are probably important. There is no reliable evidence that either malnutrition (including kwashiorkor) or *Plasmodium* species infection are aetiologically important, although such beliefs linger.[@bib1] In tropical countries most cases of macronodular cirrhosis result from viral hepatitis, most commonly HBV, and to a lesser extent HCV hepatitis.[@bib201] The sequence of events is: acute hepatitis → chronic active hepatitis → macronodular cirrhosis → and, ultimately, hepatocellular carcinoma[@bib209], [@bib210], [@bib211], [@bib212] (hepatoma) (acute viral hepatitis is covered in Chapter 39 and hepatoma in Chapter 35). HBV and HCV are undoubtedly the most important aetiological factors in hepatoma, but the role of aflatoxin[@bib1] should not be totally disregarded. The true prevalence of autoimmune hepatitis, which has been studied in Brazil, is unknown.[@bib213] An important and probably underrated cause of chronic liver disease in a tropical context is schistosomiasis (Chapter 82).[@bib214], [@bib215] Although hepatocellular function is preserved until late in the disease, portal hypertension and its various complications (see below) are as important as in the various forms of cirrhosis. Clinically, cutaneous stigmata of chronic hepatocellular disease are difficult to detect in brown or black skins;[@bib1] similarly, other cutaneous stigmata of chronic liver disease may be absent. Diagnosis is often first suspected by abnormal liver function tests; a needle liver biopsy specimen is usually diagnostic. Peritoneoscopy is relatively simple and underused in developing countries; refined diagnostic techniques are rarely available. No treatment is of any avail in established cirrhosis, but some of the chromolytics in chronic schistosomal disease of the liver are reversible after treatment (Chapter 82). Major complications (see below) resulting from portal hypertension are: (1) haemorrhage, from oesophageal varices (see below); (2) fluid retention, including ascites; and (3) hepatic encephalopathy. Fluid retention is a major long-term problem, largely the result of a very low serum albumin concentration. This complication is often difficult to manage, largely because salt restriction is virtually impossible to impose in a tropical setting; diuretics, e.g. furosemide (Lasix) (40--120 mg daily) and spironolactone (Aldactone) (100 mg daily), usually achieve success. Paracentesis abdominis should rarely be undertaken; this procedure depletes albumin stores further, and electrolyte balance can be seriously disturbed; tapping ascitic fluid should be reserved for: (1) diagnostic purposes, to understand whether a bacterial infection, tuberculous peritonitis or hepatocellular carcinoma is present concurrently; and (2) management of tense ascites, accompanied by respiratory embarrassment. Hepatic encephalopathy is managed by accepted methods: oral neomycin (6 g daily) and/or lactulose (20--35 g three times daily); in the presence of hypolactasia, lactose can be substituted for lactulose. Other forms of chronic liver disease (with subsequent decompensation) (see below) include those resulting from excessive alcohol ingestion, Indian childhood cirrhosis, haemosiderosis and veno-occlusive disease. Wilson\'s disease (hepatolenticular degeneration) and other genetically determined forms of cirrhosis are of limited importance numerically in the tropics, although they too should enter the list of differential diagnoses. ### Alcoholic liver disease {#cesec74} Alcohol-related disease (including cirrhosis) is common in both indigenous and expatriate populations in tropical countries.[@bib1], [@bib216] Genetic factors are undoubtedly involved; HBsAg carriers are especially vulnerable. The liver in chronic alcoholic disease is classically micronodular, but not always so; liver biopsy histology sometimes shows characteristic Mallory\'s hyaline deposits, and haemosiderin may be present in excess. There are no major differences from the disease in temperate climates. The quantity of daily alcohol required to produce this disease is not known with accuracy, and estimates differ widely; an individual variation exists, and women seem to tolerate chronic alcohol ingestion less well than men. Acute alcoholic hepatitis is underdiagnosed and possesses a high mortality rate; the role of corticosteroids continues to be disputed;[@bib1], [@bib216] any beneficial effect is at best marginal and administration should probably be confined to severe and advanced cases. ### Indian childhood cirrhosis {#cesec75} Indian childhood cirrhosis[@bib217] is largely confined to India (especially south India, Calcutta and the Punjab) and surrounding countries; it is frequently familial. Diagnosis is usually made between 1.5 and 3 years of age; members of the upper strata of Hindu society are often affected. The disease may pursue fulminant, acute or subacute courses, and carries a high mortality rate. The clinical course therefore varies widely and is comparable to viral hepatitis (see above), with acute fulminant hepatitis at one extreme of the spectrum and cirrhosis (with one or all of its classic complications) ([Figure 10.10](#f10){ref-type="fig"} ) at the other. Histologically, there is usually progressive fibrosis, with absence of regeneration; macronodular and micronodular cirrhosis result. Hepatocellular carcinoma is an uncommon complication. The disease is associated with a high copper intake; epidemiological evidence suggests that early weaning followed by milk-feeding from copper vessels imparts an excessive copper intake.[@bib218] However, the possibility of an inherited defect resulting in excess copper absorption and/or metabolism has not been eliminated. There is no adequate treatment; in prevention, non-human milk for infant and childhood consumption should not be stored in coppercontaining vessels.Figure 10.10Indian child suffering from decompensated chronic liver disease -- Indian childhood cirrhosis. ### Haemosiderosis {#cesec76} Haemosiderosis (African or Bantu siderosis) is a disease of southern, and to a lesser extent other parts of (tropical) East and West Africa.[@bib219], [@bib220] Whether it can proceed to clear-cut cirrhosis is arguable; heavy alcohol intake is commonplace in many geographical areas where the disease is common; it is frequently impossible to exclude this as an aetiological factor (as with haemochromatosis). Iron-containing pots for cooking are commonly used in most areas, such as Zimbabwe, where haemosiderosis is common, but other factors also seem relevant. Also, chronic pancreatitis is relatively common in these areas; evidence exists that an excess of iron (and fat) is common. ### Veno-occlusive disease {#cesec77} Although first described in Jamaica, distribution of veno-occlusive disease is now known to be much wider.[@bib224] Bush-teas, which contain pyrrolizidine alkaloids (*Heliotropium, Crotalaria* and *Senecio*) are important aetiologically. Veno-occlusive disease occurs in many localized areas of the tropics, and is certainly not confined to the Caribbean. ### Other chronic liver diseases {#cesec78} The liver is involved in most chronic infective diseases; tuberculosis, leprosy, syphilis, actinomycosis, visceral leishmaniasis and African histoplasmosis are examples. It is, however, unusual for decompensation (and liver failure) to result. Major space-occupying lesions involving the liver are amoebic abscess (see below), pyogenic abscess and hydatid disease; tuberculomas, cysticercosis and melioidosis are of lesser importance. Of non-infective diseases, sickle cell disease, β-thalassaemia, haemoglobin-H disease, porphyria and α~1~-antitrypsin deficiency produce significant hepatic pathology. A change in disease profile of the Budd-Chiari Syndrome has been recorded over the past three decades in India.[@bib222] Portal hypertension {#cesec79} ------------------- Portal hypertension[@bib1], [@bib223] is a sequel to any form of chronic liver disease; [Table 10.7](#cetable7){ref-type="table"} summarizes some causes in a tropical country. Cirrhosis and schistosomal liver disease (Chapter 82) are numerically very important; however, in the latter entity hepatocellular function is preserved to a greater extent, and for longer in the course of disease than in cirrhosis; therefore, fluid retention and more importantly encephalopathy are less common. A form of non-cirrhotic chronic liver disease, sometimes associated with portal hypertension, exists in India; despite various suggestions (including arsenic poisoning), the aetiology remains unclear. Of pre-hepatic causes, HMS (see [Table 10.7](#cetable7){ref-type="table"}) is the most common; portal hypertension results from an increased splenic blood flow. Portal/splenic vein obstructions, probably resulting from neonatal umbilical sepsis, are important causes throughout tropical countries, and are undoubtedly underdiagnosed;[@bib223] hepatocellular function is usually intact. Posthepatic causes of portal hypertension include ([Table 10.7](#cetable7){ref-type="table"}) cardiac failure (usually resulting from chronic rheumatic cardiac disease), right-sided endomyocardial fibrosis (Chapter 12) and constrictive pericarditis, usually but not always resulting from tuberculosis. Other causes of portal hypertension are hepatocellular carcinoma (see above) and various dehydrating diseases, including dysentery and cholera. Splenomegaly is present whatever the cause of portal hypertension (which should be distinguished from other causes of enlargement of this organ in a tropical country). Barium swallow or upper gastrointestinal endoscopy usually confirms the presence of oesophageal varices. When available, ultrasonography is valuable in assessing portal vein patency.Table 10.7Causes of portal hypertension and oesophageal (and gastric) varices, showing those which are more common in developing countriesLevel of obstructionCausePre-hepaticHyper-reactive malarious splenomegaly (HMS) (increased portal blood flow)[a](#cetablefn3){ref-type="table-fn"}Portal vein occlusion[a](#cetablefn3){ref-type="table-fn"}Splenic vein occlusionHepatic macronodular cirrhosis[a](#cetablefn3){ref-type="table-fn"}Hepatosplenic schistosomiasis[a](#cetablefn3){ref-type="table-fn"}Veno-occlusive disease[a](#cetablefn3){ref-type="table-fn"}Congenital hepatic fibrosisPost-hepaticCardiac failure (secondary to chronic rheumatic disease)[a](#cetablefn3){ref-type="table-fn"}Endomyocardial fibrosis[a](#cetablefn3){ref-type="table-fn"}Constructive pericarditis[a](#cetablefn3){ref-type="table-fn"}Inferior vena caval obstructionHepatic vein thrombosis (Budd-Chiari syndrome)[^3] Biliary tract disease {#cesec80} --------------------- In tropical countries biliary pathology is largely attributable to parasites,[@bib1], [@bib207], [@bib224] ascariasis (Chapter 85), clonorchiasis and opisthorchiasis (Chapter 83); pigment stones (often intrahepatic) occasionally complicate sickle cell disease. *A. lumbricoides* infection (Chapter 85) is an underdiagnosed cause of large-duct obstruction. It should always be considered in this clinical situation, for it may be confused with pancreatic carcinoma. Endoscopy, if available, is of value; medical treatment is usually successful. Clonorchiasis and opisthorchiasis (Chapter 83), acquired from ingestion of raw fresh-water fish, may result in cholangiohepatitis and biliary obstruction; cholangiocarcinoma is a late complication of both infections. *F. hepatica* infection (Chapter 83) can give rise to tender hepatomegaly accompanied by jaundice; difficulty in diagnosis from viral hepatitis may be a problem; an eosinophilia is, however, common with this and all biliary trematode infections. Praziquantel is of no value in treatment; triclabendazole has now replaced it.[@bib225], [@bib226], [@bib227] Overall, cholesterol stones (and associated secondary infection) are uncommon in rural populations, especially in Africa. Gallbladder infection by *S. typhi* can result in the typhoid carrier state (Chapter 52); the focus of infection is usually intrahepatic. Gallbladder carcinoma is unusual. Emergencies {#cesec81} ----------- ### Acute hepatocellular failure {#cesec82} Acute liver failure (acute hepatic necrosis) is a major clinical problem in all developing countries (see above);[@bib4], [@bib228] various hepatitis viruses (most commonly B, C, D and E, and to lesser extent A) are all involved (see above), but some cases are caused by other viruses, bacteria or toxins. Although acute hepatocellular failure has been recorded in severe acute *P. falciparum* infection, this is of very limited clinical importance; it occurs as a terminal event but is of far lesser importance than other major organ failure.[@bib229] The role of several viruses involved in the production of acute liver injury has been summarized.[@bib201] Several reports highlight the aetiological basis of hepatitis in tropical countries; in Egypt, HBV and hepatitis A virus (HAV) accounted for 47% and 0.7% of cases of acute hepatitis (there was serological evidence of both viral infections in a further 1.4%), whereas 14.2% of cases were HBsAg carriers, 31% 'non-A, non-B' hepatitis and 6% were drug-induced.[@bib230] In other locations, however, hepatitis D virus (HDV) is important, especially in southern America, South-east Asia (and probably India) and northern Africa. Thus in Thailand, HDV is frequently present in drug abusers; it is also endemic in Chandigarh, India,[@bib231] and has been described in an epidemic of acute hepatitis in the Himalayan foothills in south Kashmir.[@bib232] In India and South-east Asia, hepatitis E virus (HEV) (see above) is responsible for most cases of the entity previously termed 'non-A, non-B' hepatitis; a similar situation probably pertains in Africa and South America. This virus is transmitted by the faecal--oral route and is transmitted in contaminated drinking water; the major importance of this infection is that it produces a high incidence of hepatocellular failure in pregnant women. HCV also causes severe disease -- including acute hepatic failure -- similar to that produced by HBV (Chapter 39). #### Differential diagnosis {#cesec83} Many other viruses present in tropical and subtropical regions may also produce acute hepatic necrosis; these include herpes simplex type 1, herpes virus 6,[@bib233] Epstein--Barr virus, cytomegalovirus, yellow fever[@bib234] and the haemorrhagic fever viruses, which include the Lassa fever virus, the Marburg virus, Ebola virus and Rift Valley fever virus (see above).[@bib235], [@bib236] Of bacterial causes of hepatitis, enteric fever is common, but rarely (if ever) proceeds to hepatocellular necrosis (see above). The jaundice of systemic bacterial infection[@bib1] often follows pyomyositis, especially in Africa. *P. falciparum* malaria causes deranged liver function tests resulting from centrilobular necrosis (see above). Hepatotoxicity resulting from herbal remedies is not confined to tropical countries.[@bib237] Alcoholic hepatitis is a significant clinical problem in both indigenous and expatriate populations. #### Management {#cesec84} Tandon et al.[@bib238] have outlined their experience of acute hepatic failure (resulting from viral hepatitis) in 145 (\>12 years old) patients managed by them using a 'simple supportive therapeutic regimen' during a 5.5-year period at New Delhi, India. Criteria for inclusion were:•Development of hepatic encephalopathy within 4 weeks of onset of symptoms and signs of acute hepatitis; and•Absence of evidence of pre-existent liver disease. There were 65 men and 80 women; 46 of them were pregnant and presumably infected by HEV. They used a simple intensive support mechanism; this consisted of:1.Isolation in an intensive care room.2.Attention to general hygiene and care of a comatose patient.3.Intravenous fluid to provide 1000--1500 calories daily using 10% dextrose, supplemented if necessary, by 20% dextrose.4.Nasogastric tube for aspiration of gastric contents and instillation of drugs.5.Gut sterilization by ampicillin (1.5 g 6-hourly via nasogastric tube); colonic washes twice daily.6.Liquid antacids (30 mL 2-hourly).7.'Lactisyn\' (1 ampoule = Lactobacillus lactus 490 million, L. acidophilus 490 million, Streptococcus lactus 10 million) three times daily.8.Condom or catheter drainage of the urinary bladder.9.Maintenance of electrolyte and fluid balance by intravenous supplementation. Complications were managed as follows:•Infection (diagnosis was based on clinical findings, leukocyte count \>15 × 10^9^/L, and/or chest radiograph abnormality): gentamicin 3.5 mg/kg body weight (as three divided doses), and/or cephalexin (2 g daily as four divided doses)•Cerebral oedema (criteria for diagnosis were: focal or generalized seizures, abnormal reactive or unequal pupils, decerebrate posture of the body after minor stimuli, and/or sudden deterioration of vital signs): intravenous mannitol (200 mL administered during 30 min and repeated three or four times per 24 h).•Gastrointestinal bleeding (diagnosed by aspiration of fresh or altered blood via nasogastric tube): liquid antacid (30--45 mL every 2 h), gastric lavage (with 100 mL cold saline containing 8 mg noradrenaline every 30 min) and occasionally cimetidine. (When the prothrombin time was \>7 s compared with a control, fresh frozen plasma was administered.)•Renal failure (the criterion used was: oliguria (urine output \<400 mg/24 h, and rising blood urea) despite adequate hydration): diuretics (judiciously used). Overall, 42 (28.9%) survived; of those ≤40 years old, 41 (33%) recovered, compared with only one (4.8%) of those ≥40 years; survival was not affected by pregnancy. Indicators of poor prognosis were: grade IV coma, presence of HBsAg, serum bilirubin concentration \>20 mg/100 mL and sodium \<119 mmol/L. In fatal cases the immediate complications resulting in death were cerebral oedema (65), bleeding (31), renal failure (11) and infection (8). The authors concluded that these results were comparable with results from centres using a variety of complex therapeutic regimens (e.g. exchange blood transfusion, charcoal perfusion and haemodialysis). ### Chronic hepatocellular failure and hepatoma {#cesec85} Cirrhosis, generally resulting from one of the hepatitis viruses (see above), is a very common problem throughout tropical and subtropical countries. A study carried out at New Delhi, India, has addressed the problem of survival in young (\<35 years old) and older patients with cirrhosis;[@bib239] numbers in the two groups were 63 and 106, respectively. Aetiology of cirrhosis in the young and adult groups was: HBV-related (32 and 51), alcohol-related (10 and 28), while 19 and 21, respectively, were labelled 'cryptogenic\'; in the former group, one had Wilson\'s disease and another α~1~-antitrypsin deficiency. During the surveillance period 27 and 47 deaths occurred: 40% and 64% from hepatic failure, and 52% and 26% from variceal bleeding. The 5-year survival (62% and 56%) and probability of survival within a similar grade of liver disease (Child\'s classification) were comparable. As anticipated, probability of survival was significantly higher in grade A and lowest in C. Aetiology of cirrhosis did not significantly influence prognosis in this study. Hepatocellular carcinoma usually presents as a rapidly progressive malignancy; however, an acute or chronic presentation can occur due to internal necrosis and haemorrhage.[@bib142] Such a lesion can in fact rupture into the peritoneal cavity, posing problems in differential diagnosis. In a patient with actively bleeding oesophageal varices, differentiation of the aetiology of underlying liver disease (from postviral (or another aetiology) cirrhosis and chronic schistosomal disease) is usually impossible on clinical grounds alone. In a study carried out at Cairo, Egypt, liver ultrasonography was undertaken in 50 patients who were undergoing an operation for bleeding oesophageal varices;[@bib240] ultrasonographic diagnosis was compared with a surgically obtained wedge biopsy specimen. The authors concluded that ultrasonography gave the more accurate diagnosis; the findings in schistosomal periportal (pipe-stem) fibrosis were characteristic and were not mimicked by other liver diseases (including cirrhosis); ultrasonography agreed with the histological diagnosis in 44 cases. #### Role of ultrasonography in management {#cesec86} The overall value of ultrasonographic scanning and scintigraphy in the diagnosis of chronic liver disease in developing countries has been addressed.[@bib241] Needle biopsy is frequently necessary to diagnose diffuse disease, but a high degree of specificity can be anticipated with a space-occupying lesion.[@bib155] A further problem surrounding ultrasonography has been highlighted:[@bib241] in Africa and other developing countries, focal lesions 'often present so late that lesions revealed by ultrasound are huge and bizarre\', and the inexperienced radiologist may therefore be baffled. ### Portal hypertension and its complications {#cesec87} The major causes of portal hypertension (and oesophageal varices) are summarized in [Table 10.7](#cetable7){ref-type="table"}. Some geographical variations have been reviewed.[@bib1], [@bib9] While in many parts of the world cirrhosis is the most common cause, in India non-cirrhotic portal fibrosis is relatively common.[@bib9] Indian childhood cirrhosis (see above) also accounts for cases in the younger age group(s). Extrahepatic portal vein obstruction is common in some countries (including India);[@bib223], [@bib242] however, in Egypt, Africa, the Middle East, South America and China, *Schist. mansoni* and *Schist. japonicum*, respectively, are frequently responsible. In Jamaica, South Africa, central Asia and the south-western USA, epidemic veno-occlusive disease (see above) (caused by *Heliotropium, Crotalaria, Senecio* and other alkaloids; see above) is important. Pitressin (vasopressin) forms the basis of management of variceal haemorrhage; if and where available, upper gastrointestinal endoscopic sclerotherapy is of value, but this technique usually has to be repeated at 6-monthly intervals. The Sengstaken tube (for variceal compression) still has a place in developing countries. Haemorrhage is not a major presenting feature at most tropical hospitals (see above). #### Bleeding varices resulting from extrahepatic portal obstruction {#cesec88} The cause of portal vein thrombosis in developing countries remains unclear; it is, however, a relatively common condition, and neonatal umbilical sepsis is usually cited as the likely aetiological factor.[@bib1] During an 8.5-year period, 136 patients with extrahepatic portal hypertension were treated surgically at New Delhi, India;[@bib242] in 22 it was carried out as an emergency (for variceal bleeding), and in 114 as an elective procedure (in 104 for a past haematemesis and in 10 for massive splenomegaly). The emergency strategy consisted of: splenectomy and splenorenal shunt (14), transoesophageal variceal ligation (4), splenectomy and gastro-oesophageal devascularization (3) and mesocaval shunt (1). Elective procedures were: splenectomy and splenorenal shunt (94), mesocaval shunt (8) and splenectomy and gastro-oesophageal devascularization (12). Operative mortality was 2 (9%) and 1 (1%), respectively; none of the survivors developed encephalopathy or postsplenectomy sepsis. One hundred and seventeen (86%) were followed up for 2--10 years; 17 had a further haematemesis, but 90% and 75% were alive at 5 and 10 years, respectively. Patients experiencing haematemesis are often far from medical facilities in a developing country; the authors therefore considered that in this setting operative intervention was more satisfactory than endoscopic sclerotherapy or management with propranolol (variceal compression was not considered). ### Space-occupying hepatic lesions {#cesec89} #### Invasive hepatic amoebiasis {#cesec90} Amoebic liver abscess is a cause of right upper quadrant pain (and hepatomegaly); this is usually accompanied by fever, and not infrequently right shoulder-tip pain. Travellers to infected areas as well as the indigenous population(s) of the tropics may be affected.[@bib1], [@bib243] Pathogenesis is dependent on an oral infection with a potentially invasive strain (zymodeme) of *Ent. histolytica*.[@bib244] The mode of evolution remains unclear.[@bib245] Diagnosis is based on an appropriate serological technique (IFAT, cellulose acetate or countercurrent immunoelectrophoresis) and hepatic ultrasonography or computed tomography. Clinical characteristics in a group of 52 patients suffering from amoebic liver abscesses have been recorded at Cairo, Egypt;[@bib246] while 22 (42%) presented with an acute illness (see above), 30 (58%) had a more chronic illness with dull aching in the right hypochondria, weight loss, fatigue, moderate to low-grade pyrexia and anaemia. A right-sided pleural effusion, emphysema, ascites and jaundice were present in three (6%), four (8%), seven (13%) and seven (13%), respectively. Forty-two (81%) abscesses were solitary and in the right lobe; 29 (43%) were initially solid or heterogeneous. Response to metronidazole (750 mg three times daily for 10 days) was described as good in 50; in four aspiration was carried out on account of the large abscess size. Whether needle aspiration of an amoebic abscess (in addition to satisfactory chemotherapy) is indicated remains controversial. A prospective, randomized controlled study carried out at New Delhi, India, has addressed this issue;[@bib247] in 17 of 37 patients (all received appropriate chemotherapy, 2--4 g metronizadole for 10 days) who completed the study, aspiration was carried out on the day of hospital admission; clinical improvement (and cure) was similar to that in 20 controls. 'Abscess' diameter was slightly lower in those who underwent aspiration (54 vs 72 mm). However, at Benin, Nigeria, needle aspiration was considered to 'enhance clinical recovery\';[@bib248] in a non-randomized trial, 19 patients were managed by needle aspiration in addition to chemotherapy, and 17 were given chemotherapy (metronidazole, diloxanide and chloroquine) alone; 18 and 10, respectively, experienced complete resolution (as shown by ultrasonography) after 21 days (*p*\< 0.021), and clinical response was also considered more rapid (*p*\< 0.01), especially when the abscess was \>6 cm in diameter. Delay in ultrasonographic 'recovery' is not important, there being good evidence that a residual abnormality after a year or more is compatible with complete, uncomplicated resolution. Although no in vitro evidence of *Ent. histolytica* resistance to the 5-nitroimidazole compounds exists, reports continue to be made from India of drug-resistant cases. The main problem with such reports is that, in few (if any) has diloxanide furoate (500 mg three times daily for 10 days) been administered; this is essential for a definitive cure because it is a far superior luminal amoebicide compared with the 5-nitroimidazole compounds -- and therefore kills the cysts (which could belong to invasive zymodemes). In a prospective randomized study of 50 such 'resistant' cases at New Delhi, four management regimens were used:[@bib249] (1) a repeat course of conservative therapy (with 1.25 mg/kg dehydroemetine given intramuscularly daily for 10 days); (2) needle aspiration (under ultrasonographic guidance); (3) percutaneous catheter drainage (under ultrasonographic guidance); and (4) open surgical drainage with catheter insertion. The authors concluded that 'the most impressive results' were obtained with regimen 3. To summarize, in the uncomplicated case, needle aspiration (under cover of a 5-nitroimidazole compound) is indicated when: (1) the abscess(es) cavity is large and the patient seriously ill; and (2) the site of the lesion is such that perforation into a nearby viscus (most importantly the pericardium) seems probable. All cases of invasive amoebiasis should receive a course of the luminal amoebicide, diloxanide furoate (500 mg three times daily for 10 days) after metronidazole (800 mg three times daily for 10 days) or tinidazole (2 g daily for 3 days). If this regimen is omitted, *Ent. histolytica* cysts remain in the colonic lumen and, in the event of their being of a pathogenic zymodeme, further tissue invasion (including liver abscess) might occur. Spontaneous perforation of an amoebic liver abscess is a serious complication which is associated with high morbidity and mortality rates;[@bib243] this applies especially when perforation takes place into the pericardial cavity. Successful percutaneous drainage (for 7--34 days) of a perforated abscess in five 'severely ill' patients (with a total of 11 lesions) under metronidazole cover has been recorded;[@bib250] there were resultant abscesses in the subhepatic space, pelvis, chest, right and left paracolic gutters, lesser sac, retroperitoneum and flank, and associated fistulas were demonstrated with the bile duct, duodenum and the colon; all healed completely. No patient required a laparotomy. These authors recommend wider use of catheter drainage for this serious complication of hepatic amoebiasis. #### Pyogenic liver abscess {#cesec91} Although in a tropical context it is far less common than invasive amoebiasis (see above), pyogenic abscess is a serious disease with high morbidity and mortality, even when managed in experienced hands.[@bib1] In most cases, a primary intra-abdominal focus of infection can be detected. Differentiation from invasive hepatic amoebiasis is usually straightforward, the patient being more severely and acutely ill; jaundice, septicaemia and renal impairment are common accompaniments. Ultrasonography is usually diagnostic. In Kuala Lumpur, 25 pyogenic abscesses were encountered between 1970 and 1985;[@bib251] during the same period, there were 90 amoebic and one tuberculous abscesses, while in 89 others the cause of the abscess was not discovered. At Kingston, Jamaica, fever and abdominal pain were present in 21 (80%) out of 24 cases of pyogenic abscess encountered between 1977 and 1986;[@bib252] the most common signs were right upper quadrant tenderness and hepatomegaly; leukocytosis, elevated alkaline phosphatase and hypoalbuminaemia were common. Reports from London[@bib253] and California[@bib254] have given encouraging reports of management by needle aspiration under antibiotic (usually gentamicin and metronidazole or clindamycin) cover. Another study has also recorded satisfactory results in 18 of 21 patients using this form of percutaneous drainage. Other authors have intimated, however, that this form of management should be reserved for selected patients.[@bib255] A report from Riyadh, Saudi Arabia, has provided results which were less encouraging. In Jamaica surgical drainage using a guided percutaneous technique gave comparable results.[@bib256] Taking all reports into account, it seems wise to perform a laparotomy and to institute surgical drainage as soon as possible after diagnosis. Using ultrasonographic control, a pyogenic abscess can be seen to 'resolve' significantly more rapidly than an amoebic abscess. It should be appreciated, however, that this disease carries a significant mortality rate; between 1975 and 1986, these authors treated 109 children with pyogenic liver abscess; the mortality rate was 15%.[@bib257] There is limited (suggestive) evidence that the overall prognosis is improving. #### Hydatid disease and schistosomiasis involving the liver {#cesec92} Only rarely, usually following trauma, does hydatidosis[@bib207], [@bib258], [@bib259] present as an abdominal emergency. Perforation into the peritoneal cavity may produce an anaphylactoid reaction with hypotension, and/or seeding of daughter hydatid cysts within the peritoneal cavity. A relatively high prevalence of alveolar echinococcosis has been recorded in China.[@bib260] Secondary bacterial infection is an unusual event. Chemotherapy is with albendazole and/or praziquantel (Chapter 86). Hepatic schistosomiasis[@bib261] is complicated by portal hypertension and oesophageal varices in an advanced case; however, hepatocellular function is maintained late into the course of disease and hepatic encephalopathy and ascites occur as advanced (usually terminal) signs. Praziquantel is the chemotherapeutic agent of choice; evidence of reversal of fibrotic changes is now available. PANCREAS {#cesec93} ======== The two major diseases involving this organ encountered in tropical countries, and which differ from those in temperate ones, are (1) 'J-type' diabetes, first reported in Jamaica (Chapter 36) and (2) chronic calcific pancreatitis. Diabetes, which is *not* associated with pancreatic calcification in young people, is encountered throughout tropical countries; those affected are usually thin, and require high doses of insulin; however, they do not rapidly develop ketosis when insulin is discontinued. J-type diabetes might have a viral aetiology, a Coxsackie virus being involved; a raised incidence of antibody to Coxsackie B~4~ has been demonstrated in affected patients in India. A suggestion has been made that these patients, especially those in Africa, are less susceptible to chronic diabetic complications than Europeans; this now seems unlikely. A popular Indian and Chinese vegetable, karela (*Momordica charantia*) possesses hypoglycaemic properties; these are enhanced by chlorpropamide, a fact that should be taken into account in the management of diabetes in a number of Asian countries. A syndrome consisting of pancreatic calcification associated with both exocrine and endocrine impairment is common in many tropical countries ([Figure 10.11](#f11){ref-type="fig"} );[@bib1], [@bib262], [@bib263] most observations have been made in Africa (East and West), southern India and Indonesia. The aetiology of *chronic calcific pancreatitis* remains unknown. Pancreatic disruption in childhood kwashiorkor can be severe and might be relevant. Cassava (*Manihot esculenta*) has also been implicated. Long-standing pancreatic damage can also follow viral hepatitis. A further hypothesis is that pancreatic ducts blocked by secretions and inspissated mucous plugs later calcify; this might be more common after starvation, gastroenteritis and dehydration. Presentation is with weight loss and malabsorption (in some parts of Africa, this is the most common cause of overt malabsorption); diabetes mellitus and pancreatic pain are important features. Management consists of providing pancreatic supplements (e.g. pancreatin BP, 6 g orally with meals) together with diabetic control.[@bib1] Pain is often difficult to manage and may be so severe that suicide is a sequel.Figure 10.11Abdominal radiograph showing calcified pancreas in the chronic calcific pancreatitis syndrome. There was no history of alcohol excess or infant malnutrition; aetiology was therefore undetermined. The pancreas can also be involved in many infections including *Schist. mansoni* and *Schist. japonicum*, trichinellosis, cysticercosis and hydatid disease. Pancreatic duct obstruction, complicated by acute pancreatitis, is most commonly a sequel to *A. lumbricoides* infection (see below); tapeworms are rarely implicated. Clonorchiasis and opisthorchiasis may involve the pancreatic duct system. Emergencies: pancreas, and biliary system {#cesec94} ----------------------------------------- One of the most widely distributed nematodes in tropical and subtropical countries is *A. lumbricoides*. By entering the biliary system (from the duodenum) this parasite can cause several acute medical and surgical conditions. Reporting from Kashmir, India, Khuroo et al.[@bib264] collected 500 cases in which *A. lumbricoides* involved the liver, biliary tract and pancreas; biliary ascariasis was present in 171 cases, and in 140 there was hepatic, in eight gallbladder and in seven pancreatic involvement. These authors recognized five clinical presentations: acute cholecystitis (64), acute cholangitis (121), biliary colic (280), acute pancreatitis (31) and hepatic abscess (4). Twenty-seven had a pyogenic cholangitis, which was treated by decompression and drainage, surgically in two and endoscopically in 25; removal of adult worms from the ampullary orifice (with extraction per os) led to rapid relief of biliary colic in 214, and acute pancreatitis in 16; four patients died, from acute pancreatitis (2), pyogenic cholangitis (1) and hepatic abscess (1). Worms persisted at 3 weeks in the biliary tree in 12 patients; dead worms were removed either by surgery (5) or by using an endoscopic basket (7). *A. lumbricoides* moved out of the ductal system in 211 cases. The patients were followed-up for a mean of 48 months; 76 became re-infected and had re-invasion of the biliary tree; in seven cases intrahepatic duct and bile duct calculi (superimposed on dead worms) were present. In South-east Asia, the two most common biliary parasites are *Clonorchis sinensis* and *Opisthorchis* spp. Although these cause chronic problems, notably secondary bacterial cholangitis[@bib142] and adenocarcinoma of the biliary system, an acute presentation[@bib1] is unusual. In most indigenous people of developing countries, gallstones are unusual; when they occur they are usually of the pigment variety, and often associated with haemolysis. A report from Saudi Arabia, where the average lifestyle has rapidly become westernized (with striking changes in diet) over the last few decades, indicates that cholecystectomy for cholelithiasis is now one of the most common major abdominal operations to be carried out;[@bib265] between 1977 and 1986, for example, 2854 individuals (most of them young Saudis) underwent this operation at 14 hospitals in the Eastern Province of the country. Acute pancreatitis is uncommon overall in developing countries, although severe abdominal pain caused by chronic calcific pancreatitis[@bib1] can give rise to problems in differential diagnosis. The pain may be severe. Biliary involvement by *A. lumbricoides* can result in acute pancreatitis.[@bib1], [@bib142] Other helminths, including *Clonorchis sinensis*, *Opisthorchis* and *Anisakis* species have also been associated with this condition. SPLEEN {#cesec95} ====== [Table 10.8](#cetable8){ref-type="table"} summarizes some causes of splenomegaly in the tropics.[@bib1] Most of these receive attention in other chapters. The most extreme form of splenomegaly (HMS) ([Figure 10.12](#f12){ref-type="fig"} ) is covered in Chapters 13 and 72; those caused by various viral, bacterial and parasitic infections are dealt with under these respective headings.Table 10.8Some causes of splenomegaly in the tropicsInfections ViralEpstein-Barr virus, cytomegalovirus, viral hepatitis and other virus diseases Bacterialtyphoid fever, brucellosis, tuberculosis Parasiticmalaria (especially hyper-reactive malarious splenomegaly (HMS)), schistosomiasis, visceral leishmaniasis, African trypanosomiasisPortal hypertensionHaemopoietic diseasesSickle cell disease, thalassaemiaReticuloendothelial diseasesBurkitt\'s lymphoma, leukaemia, reticulosesCystic lesionsHydatid diseaseAbscessAmoebic; unknown aetiologySpontaneous haemorrhage and ruptureMetabolicAmyloidosisFigure 10.12Papua New Guinea man suffering from hyperreactive malarious splenomegaly (HMS); all of the features of this syndrome were present. (B) Liver biopsy specimen showing severe sinusoidal lymphocytosis, a component of the syndrome. The spleen is an extremely important line of defence against many infections, especially pneumococcal and *Plasmodium* species infections. Splenectomized individuals in tropical countries should receive pneumococcal vaccine; prudent advice regarding malaria prophylaxis is mandatory. Splenic abscess is a well-documented tropical disease.[@bib1] Aetiology is usually unknown; underlying viral and parasitic diseases have been suggested, but not proved. A connection with carriage of the sickle cell gene has also been suggested, but this has also not been proved. Most reports have been made in West Africa and Zimbabwe. In most, the aetiology is unknown, but some undoubtedly result from a *S. typhi* infection. The clinical history is usually one of 2--3 weeks duration, and consists of pain/swelling in the left hypochondrium, associated with pyrexia. The splenic swelling is tender, often exquisitely so, and fluctuant. A radiograph may show gas within the abscess. Untreated, the abscess can rupture into the peritoneal cavity; splenectomy therefore has an important role in management. Should the condition become chronic -- an unusual event -- splenectomy is also the correct course of management. [^1]: Characteristically, numerous small stools containing mucus, pus and blood; microscopy shows pus cells and/or red blood cells in a faecal smear. [^2]: Although these diseases are uncommon, or even rare, in most tropical populations, they can become clinically overt for the first time in visitors from Western countries to the tropics. [^3]: More common in a developing than a developed country.

Introduction {#Sec1} ============ Relapsing--remitting multiple sclerosis (RRMS) is characterized by defined attacks separated by periods of stability. Over time, attacks become less frequent, while disability accumulates. Although the majority of patients with MS present with the relapsing form of the disease, relapses can continue to occur during the gradual transition to the progressive form of the disease, secondary progressive MS (SPMS) \[[@CR1]\]. Disease severity is assessed using the Expanded Disability Status Scale (EDSS) score, which ranges from 0 (normal) to 10 (death due to MS) and is based on assessment of clinical deficits in various central nervous system functions. Patients with MS who have EDSS scores 4.0--6.0, while not limited to wheelchair or bed, have moderate disability indicative of disability progression \[[@CR2]\]. Although there is no one agreed-upon definition of SPMS, it is usually defined as an initial relapsing--remitting disease course followed by progression with or without occasional relapses, minor remissions, and plateaus \[[@CR3]\]. Patients with SPMS usually have an EDSS score 5.0--9.5 with impaired ambulation. Patients with an EDSS score 4.0--6.0 may be transitioning to SPMS; however, the disease course varies between patients \[[@CR2], [@CR4]\]. PRISMS (Prevention of Relapses and disability by Interferon beta-1a Subcutaneously in Multiple Sclerosis), a double-blind, placebo-controlled study, demonstrated that subcutaneous (sc) interferon beta-1a (IFN β-1a) three times weekly (tiw) significantly reduced relapses and active T2 lesions over 2 years in patients with active (with relapses and/or evidence of new magnetic resonance imaging \[MRI\] activity \[[@CR5]\]) RRMS \[[@CR6]\]. Disability progression was significantly delayed by sc IFN β-1a 44 µg tiw in the overall population, and in the prespecified subgroup with baseline EDSS score \> 3.5 \[[@CR6]\]. SPECTRIMS (Secondary Progressive Efficacy Clinical Trial of Recombinant Interferon beta-1a in MS), a double-blind, placebo-controlled study, demonstrated that sc IFN β-1a tiw reduced relapses and active T2 lesions over 3 years among patients with SPMS \[[@CR7]\]. Disability progression was not significantly delayed in the overall population, although a greater, non-significant effect was seen in post hoc analyses of patients who had experienced a relapse ≤ 2 years before the study \[[@CR7]\]. Both PRISMS and SPECTRIMS included patients with advanced disease at baseline, as well as patients experiencing ongoing disease activity. While the past two decades have seen numerous effective therapies developed to reduce disease activity in RRMS, most therapies have not been evaluated specifically in patients with confirmed SPMS or in patients who are in the loosely defined transition period between RRMS and SPMS, and effective treatment and clinical management are still lacking \[[@CR4], [@CR8]-[@CR9]\]. Given the positive results seen with sc IFN β-1a tiw in the high-EDSS population of PRISMS and the subgroup of patients in SPECTRIMS with recent relapses, patients from these two studies with similar disease characteristics were pooled to evaluate the effects of sc IFN β-1a tiw in this unique cohort. Methods {#Sec2} ======= Study design {#Sec3} ------------ In the PRISMS trial, patients with RRMS were randomly assigned to sc IFN β-1a tiw or placebo for 2 years \[[@CR6]\]. A total of 560 patients between 18 and 50 years of age, with a history of \> 2 relapses in the previous 2 years and an EDSS score of 0--5.0, were randomized and received treatment. Diagnosis of RRMS was based on the Poser criteria \[[@CR14]\]. The primary endpoint was the number of relapses over 2 years. All patients had proton density (PD)/T2-weighted scans at baseline and twice yearly \[[@CR15]\]. MRI endpoints in the overall PRISMS population included burden of disease (total area of MS lesions identified on a PD/T2 scan) and active (new, recurrent, and enlarging) T2 lesions. Other efficacy measures included disability progression, defined as an increase in EDSS score of ≥ 1 point sustained over at least 3 months \[[@CR16]\]. In SPECTRIMS, 618 patients with SPMS (EDSS score increase of ≥ 1 point within the last 2 years \[≥ 0.5 points if baseline EDSS score was 6.0--6.5\]) and baseline EDSS score 3.0--6.5 were randomly assigned to receive sc IFN β-1a tiw or placebo for 3 years \[[@CR7]\]. Cranial MRI scans were performed at baseline and twice yearly \[[@CR17]\]. The primary endpoint was time to confirmed disability progression, defined as an increase from baseline in EDSS score of at least 1 point (or 0.5 points if baseline EDSS score was ≥ 6.0), confirmed 3 months later with no intervening score lower than the minimum required level. Additional clinical endpoints included relapse count and time to first relapse. MRI endpoints in the entire SPECTRIMS population included burden of disease and number of active T2 lesions \[[@CR17]\]. Exploratory analysis of PRISMS high-EDSS subgroup {#Sec4} ------------------------------------------------- A predefined subgroup of PRISMS included patients with active but advanced disease, characterized by EDSS 4.0--5.0 at baseline and \> 2 relapses in the previous 2 years (defined as PRISMS subgroup). Exploratory analysis of this subgroup over 2 years included assessment of the number of relapses, patients free of relapse, time to first relapse, time to 3-month confirmed disability progression (increase of ≥ 1 point in EDSS score), T2 burden of disease, and active T2 lesions. Post hoc analyses of pooled subgroups from PRISMS and SPECTRIMS {#Sec5} --------------------------------------------------------------- Post hoc analyses examined the treatment effect of sc IFN β-1a 44 μg tiw versus placebo in a pooled subgroup of patients from PRISMS and SPECTRIMS with baseline EDSS scores 4.0--6.0 (defined as PRISMS/SPECTRIMS subgroup). To identify a subset of patients with advanced but active disease, the PRISMS/SPECTRIMS subgroup was then refined to include patients within this disability range who had either ≥ 1 relapse within 2 years before baseline or ≥ 1 gadolinium-enhancing (Gd) lesion at baseline, referred to as the PRISMS/SPECTRIMS with baseline disease activity subgroup; patients without active disease are referred to as the PRISMS/SPECTRIMS without baseline disease activity subgroup. Post hoc analyses were also conducted for 3-month confirmed EDSS progression on a small subset of patients who had disease activity (≥1 relapse) during the study (defined as PRISMS/SPECTRIMS with disease activity during the study subgroup, regardless of baseline activity) to examine the pattern of progression that may be due to relapse activity. Both trials included sc IFN β-1a 44 and 22 μg tiw treatment arms; as 44 μg is most commonly used, the analyses presented here compare only this dose to placebo. The following endpoints were investigated in all three subgroups: annualized relapse rate (ARR) over year 1 (0--1) and year 2 (0--2), time to relapse over 2 years, risk of relapse at 1 and 2 years, 3- and 6-month confirmed disability progression (EDSS score increase of ≥ 1 point \[≥ 0.5 points if baseline EDSS score was ≥ 6.0\]) at 1 and 2 years, mean number of active T2 lesions over 2 years (new, recurring, and newly enlarging T2 lesions), and burden of disease (total T2 lesion area) at 1 and 2 years. Statistical analyses {#Sec6} -------------------- For the exploratory analysis of the PRISMS subgroup, comparisons were made between the subgroup who received sc IFN β-1a 44 μg tiw and those who received placebo. Independent sample *t* test was used to compare the number of relapses over time. Cochran--Mantel--Haenszel chi-square test was used to compare the percentage of patients who were relapse free. Between-treatment differences for time to first relapse and time to confirmed disability by 1 point on EDSS were compared using log-rank tests. Analysis of variance on the ranks with effects for baseline EDSS subgroup, center, and interaction between treatment and baseline EDSS subgroup was used to compare treatment groups for T2 burden of disease and the number of active T2 lesions. For the post hoc analyses of pooled PRISMS/SPECTRIMS patients, sc IFN β-1a 44 μg tiw was compared with placebo in each subgroup (overall, patients with baseline disease activity, and patients without baseline disease activity). Hazard ratios (HRs), confidence intervals (CIs), and *p* values based on Cox proportional hazards model were used to compare between-treatment differences for risk of relapse, time to first relapse, and time to 3-month confirmed EDSS progression over 1 and 2 years. For ARR comparisons, *p* values were based on negative binomial regression. All comparisons were adjusted for the number of relapses within 2 years prior, age group (\< 40 vs ≥ 40 years), and baseline burden of disease (with adjustment for baseline EDSS). T2 burden of disease *p* values at 6, 12, and 24 months were based on ranked analysis of covariance by adjusting for number of relapses within prior 2 years, age group (\< 40 vs ≥ 40 years), baseline EDSS, baseline burden of disease, and derived using the Wilcoxon rank-sum test. The *t* test was used to compare treatment difference in the number of active T2 lesions. Number of T2 lesions was not measured at baseline in the SPECTRIMS study and thus not analyzed in all three subgroups. Results {#Sec7} ======= PRISMS subgroup {#Sec8} --------------- In the PRISMS trial (*n* = 371), 59 patients had a high EDSS score (4.0--5.0; Tables [1](#Tab1){ref-type="table"} and [2](#Tab2){ref-type="table"}). As in the overall trial population \[[@CR6]\], PRISMS patients with EDSS 4.0--5.0 treated with sc IFN β-1a (*n* = 31) had significantly reduced relapses, T2 burden of disease, number of active T2 lesions, and delayed time to confirmed 3-month disability progression versus placebo (*n* = 28) (Table [3](#Tab3){ref-type="table"}).Table 1Patients with high EDSS (4.0‒6.0) in the SPECTRIMS and PRISMS studiesTreatment receivedPRISMS (*n* = 371)SPECTRIMS (*n* = 409)PRISMS/SPECTRIMS pooled (*n* = 335)*N* per treatment*n* (% with high EDSS)*N* per treatment*n* (% with high EDSS)*N* per treatment*n* (% from PRISMS)*n* (% from SPECTRIMS)Placebo18728 (15.0)205136 (66.3)16428 (17.1)136 (82.9)sc IFN β-1a 44 µg tiw18431 (16.8)204140 (68.6)17131 (18.1)140 (81.9)*EDSS* Expanded Disability Status Scale, *IFN β-1a* interferon beta-1a, *sc* subcutaneous, *tiw* three times weeklyTable 2Baseline characteristics in the high-EDSS subgroupsPRISMS (*N* = 59)PRISMS/SPECTRIMSAll patients (*n* = 335)With baseline disease activity^a^\ (*n* = 195)CharacteristicPlacebo (*n* = 28)sc IFN β-1a 44 μg tiw (*n* = 31)Placebo (*n* = 164)sc IFN β-1a 44 μg tiw (*n* = 171)Placebo (*n* = 92)sc IFN β-1a 44 μg tiw (*n* = 103)Age, years Mean (SD)37.6 (8.0)36.6 (7.6)41.5 (7.3)41.2 (7.3)40.0 (7.4)39.4 (7.2)Female sex, *n* (%)24 (86)17 (55)108 (65.9)107 (62.6)62 (67.4)66 (64.1)Time since diagnosis, years^b^ Mean (SD)8.9 (6.4)9.2 (6.4)13.3 (7.3)12.4 (7.0)12.0 (7.5)10.8 (6.4)EDSS score at baseline^b^ Mean (SD)4.4 (0.5)4.5 (0.6)5.2 (0.8)5.2 (0.8)5.1 (0.8)5.0 (0.8)Relapses in previous 2 years^b^ Mean (SD)3.2 (1.4)2.8 (1.1)1.3 (1.5)^c^1.3 (1.5)^d^2.2 (1.4)2.1 (1.4)Burden of disease,^b^ mm^2^ Mean (SD)4124.7 (3973.1)4110.2 (3324.8)4459.6 (3775.5)4441.6 (4213.7)4601.8 (4082.3)4879.7 (4608.9)*EDSS* Expanded Disability Status Scale, *IFN β-1a* interferon beta-1a, *sc* subcutaneous, *SD* standard deviation, *tiw* three times weekly^a^Active disease defined as having ≥ 1 relapse within 2 years before baseline or ≥ 1 gadolinium-enhancing lesion at baseline^b^Equals the number of patients with available data^c^*n* = 163^d^*n* = 170Table 3Clinical and MRI endpoints: PRISMS subgroupPlacebo (*n* = 28)sc IFN β-1a 44 µg tiw (*n* = 31)*p*Number of relapses at year 2 Mean (SD)3.1 (1.84)1.2 (1.20) Median3.01.0 \< 0.0001^a^Patients relapse free at year 2, *n* (%)2 (7.1)10 (32.3)0.0177^b^Time to first relapse Median, days (months)84 (2.8)324 (10.6)0.0012^c^Time to 3-month confirmed disability progression First quartile, days (months)218 (7.2)638 (21.0)0.0481^c^T2 burden of disease, % change Median (mean)5.4 (12.2)--6.9 (0.7)0.0207^d^Active T2 lesions per patient per scan Median (mean)1.9 (2.6)0.5 (0.9)0.0002^d^*EDSS* Expanded Disability Status Scale, *IFN β-1a* interferon beta-1a, *MRI* magnetic resonance imaging, *sc* subcutaneous, *SD* standard deviation, *tiw* three times weekly^a^Independent sample *t* test^b^Cochran--Mantel--Haenszel chi-square test^c^Log-rank test^d^Analysis of variance on the ranks with effects for baseline EDSS subgroup, center, and interaction between treatment and baseline EDSS subgroups PRISMS/SPECTRIMS subgroup {#Sec9} ------------------------- A total of 335 patients with EDSS 4.0--6.0 were included in the pooled PRISMS/SPECTRIMS subgroup (PRISMS, *n* = 59; SPECTRIMS, *n* = 276; Table [1](#Tab1){ref-type="table"}). Patients in the PRISMS/SPECTRIMS subgroup were slightly older than those in the PRISMS subgroup, with longer duration of disease, higher burden of disease, and fewer relapses in the previous 2 years (Table [2](#Tab2){ref-type="table"}). Within the PRISMS/SPECTRIMS subgroup, outcomes for patients with active disease (≥ 1 relapse in prior 2 years or ≥ 1 Gd lesion at baseline; *n* = 195 \[58%\] patients) versus those with no disease activity at baseline (no Gd lesions and no relapse in prior 2 years) were also examined. ### Relapses {#Sec10} In PRISMS/SPECTRIMS patients with high EDSS (4.0--6.0), sc IFN β-1a significantly reduced ARR versus placebo at year 1 and year 2 (Fig. [1](#Fig1){ref-type="fig"}a). The reduction in ARR was significant in the subgroup with active disease at baseline (Fig. [1](#Fig1){ref-type="fig"}b), but not significant in the subgroup without baseline disease activity (Fig. [1](#Fig1){ref-type="fig"}c). Treatment with sc IFN β-1a significantly lowered the risk of relapse versus placebo over year 1 and year 2 in the PRISMS/SPECTRIMS subgroup and the PRISMS/SPECTRIMS with baseline disease activity subgroup (Table [4](#Tab4){ref-type="table"}). Subcutaneous IFN β-1a significantly delayed the time to first relapse over 2 years' treatment (*p* = 0.0043) in the PRISMS/SPECTRIMS with baseline disease activity subgroup (Fig. [2](#Fig2){ref-type="fig"}).Fig. 1ARR over 1 and 2 years (PRISMS/SPECTRIMS). ^a^Active disease defined as having either ≥ 1 relapse within 2 years before baseline or ≥ 1 gadolinium-enhancing lesion at baseline. *p* values based on negative binomial regression, adjusted for number of relapses within 2 years prior, age group (\< 40 vs ≥ 40 years), and baseline burden of disease (in the EDSS 4.0--6.0 subgroup, adjustment was also made for baseline EDSS). *ARR* annualized relapse rate, *CI* confidence interval, *EDSS* Expanded Disability Status Scale, *IFN β-1a* interferon beta-1a, *sc* subcutaneous, *tiw* three times weeklyTable 4Risk of relapse versus placebo over 1 and 2 years (PRISMS/SPECTRIMS)Risk of relapsePRISMS/SPECTRIMSPRISMS/SPECTRIMS with baseline disease activity^a^PRISMS/SPECTRIMS without baseline disease activityPlacebo (*n* = 164)sc IFN β-1a 44 µg tiw (*n* = 171)Placebo (*n* = 92)sc IFN β-1a 44 µg tiw (*n* = 103)Placebo (*n* = 72)sc IFN β-1a 44 µg tiw (*n* = 68)Year 1 Risk of relapse vs placebo^b^  Patients with relapse, *n* (%)90 (54.9)77 (45.0)66 (71.7)59 (57.3)24 (33.3)18 (26.5)  Patients without relapse, *n* (%)74 (45.1)94 (55.0)26 (28.3)44 (42.7)48 (66.7)50 (73.5)  HR vs placebo (95% CI)0.696 (0.511--0.947)0.659 (0.461--0.942)0.759 (0.411--1.402)*  p* value0.02130.02230.3789Year 2 Risk of relapse vs placebo^b^  Patients with relapse, *n* (%)106 (64.6)96 (56.1)76 (82.6)69 (67.0)30 (41.7)27 (39.7)  Patients without relapse, *n* (%)58 (35.4)75 (43.9)16 (17.4)34 (33.0)42 (58.3)41 (60.3)  HR vs placebo (95% CI)0.696 (0.525--0.923)0.613 (0.438--0.858)0.866 (0.511--1.466)*  p* value0.01190.00430.5917*CI* confidence interval, *EDSS* Expanded Disability Status Scale, *HR* hazard ratio, *IFN β-1a* interferon beta-1a, *sc* subcutaneous, *tiw* three times weekly^a^Active disease defined as having either ≥ 1 relapse within 2 years before baseline or ≥ 1 gadolinium-enhancing lesion at baseline^b^HR and *p* value based on Cox proportional hazards model, adjusted for number of relapses within 2 years prior, age group (\<40 vs ≥ 40 years), and baseline burden of disease (in the EDSS 4.0--6.0 subgroup, adjustment was also made for baseline EDSS)Fig. 2Time to first relapse over 2 years in the PRISMS/SPECTRIMS with baseline disease activity^a^ subgroup. ^a^Active disease defined as having either ≥ 1 relapse within 2 years before baseline or ≥ 1 gadolinium-enhancing lesion at baseline. Hazard ratio and *p* value based on Cox proportional hazards model, adjusted for number of relapses within 2 years prior, age group (\<40 vs ≥ 40 years), and baseline burden of disease. *CI* confidence interval, *EDSS* Expanded Disability Status Scale, *IFN β-1a* interferon beta-1a, *sc* subcutaneous, *tiw* three times weekly ### Disability progression {#Sec11} In the PRISMS/SPECTRIMS subgroup, sc IFN β-1a was associated with a lower risk of 3-month EDSS progression versus placebo over 1 year \[HR 0.654 (95% CI 0.429--0.997); *p* = 0.0486\] and over 2 years, although this did not achieve statistical significance regardless of baseline disease activity or activity during the study (Fig. [3](#Fig3){ref-type="fig"}). Numerically fewer patients treated with sc IFN β-1a versus placebo had 3-month EDSS progression (year 1, 23% vs 29%; year 2, 38% vs 48%). Over 2 years, the time to first EDSS progression was delayed with sc IFN β-1a treatment; however, the HR was similar between all three subgroups (Fig. [3](#Fig3){ref-type="fig"}). There were no differences in the time to 6-month confirmed disability progression for patients treated with sc IFN β-1a compared with placebo over 2 years in the PRISMS/SPECTRIMS with baseline disease activity subgroup \[HR 0.995 (95% CI 0.597--1.657); *p* = 0.9832\] or the PRISMS/SPECTRIMS with disease activity during the study subgroup \[HR 0.762 (95% CI 0.490--1.187); *p* = 0.2293\].Fig. 3Time to 3-month confirmed EDSS progression over 2 years (PRISMS/SPECTRIMS). ^a^Active disease at baseline defined as having either ≥ 1 relapse within 2 years before baseline or ≥ 1 gadolinium-enhancing lesion at baseline, and active disease during the study defined as ≥ 1 relapse during the study (regardless of baseline activity). Hazard ratio (vs placebo) and *p* value estimated from a Cox proportional hazards model, adjusted for number of relapses within 2 years prior, age group (\< 40 vs ≥ 40 years), and baseline burden of disease. *CI* confidence interval, *EDSS* Expanded Disability Status Scale, *IFN β-1a* interferon beta-1a, *sc* subcutaneous, *tiw* three times weekly ### MRI endpoints {#Sec12} Subcutaneous IFN β-1a significantly reduced the T2 burden of disease from baseline compared with placebo through year 2 in all PRISMS/SPECTRIMS subgroups with EDSS 4.0--6.0 (Fig. [4](#Fig4){ref-type="fig"}) Compared with placebo, sc IFN β-1a also significantly reduced the mean numbers of active T2 lesions at 6, 12, and 24 months in the overall PRISMS/SPECTRIMS subgroup and PRISMS/SPECTRIMS with baseline disease activity subgroup, but not in the PRISMS/SPECTRIMS without baseline disease activity subgroup (Supplementary Fig. 1).Fig. 4T2 burden of disease at 6, 12, and 24 months in the PRISMS/SPECTRIMS subgroup, PRISMS/SPECTRIMS with baseline disease activity^a^ subgroup, and PRISMS/SPECTRIMS without baseline disease activity subgroup. *p* values are for between-treatment differences in change from baseline. ^a^Active disease defined as having either ≥ 1 relapse within 2 years before baseline or ≥ 1 gadolinium-enhancing lesion at baseline. \**p* \< 0.05 based on ranked analysis of covariance by adjusting for number of relapse within prior 2 years, age group (\<40 vs ≥ 40 years), baseline EDSS, baseline burden of disease, and by Wilcoxon rank-sum test. \*\**p* \< 0.005 based on ranked analysis of covariance by adjusting for number of relapse within prior 2 years, age group (\< 40 vs ≥ 40 years), baseline EDSS, baseline burden of disease, and by Wilcoxon rank-sum test. \*\*\**p* ≤ 0.0001 based on ranked analysis of covariance by adjusting for number of relapse within prior 2 years, age group (\<40 vs ≥ 40 years), baseline EDSS, baseline burden of disease, and by Wilcoxon rank-sum test. *EDSS* Expanded Disability Status Scale, *IFN β-1a* interferon beta 1-a, *sc* subcutaneous, *tiw* three times weekly Discussion {#Sec13} ========== In this post hoc analysis of the pooled subgroup of patients with EDSS 4.0--6.0 from the PRISMS RRMS trial and the SPECTRIMS SPMS trial, sc IFN β-1a 44 µg tiw was effective at reducing relapses and T2 lesion activity versus placebo. Greater efficacy was seen in patients with active disease at baseline (≥ 1 relapse in prior 2 years or ≥ 1 Gd lesion). In patients with high EDSS from the PRISMS trial, sc IFN β-1a delayed disability progression; in the subgroup of patients from both trials, sc IFN β-1a significantly delayed disease progression over 1, but not 2 years. However, no significant effect on delaying further disease progression was seen in the PRISMS/SPECTRIMS with baseline disease activity subgroup. The HR for 3-month disability progression was similar between the PRISMS/SPECTRIMS subgroups. Taken together, these data suggest that baseline disease activity may help identify those patients who could have relapses or radiological progression without treatment. For the PRISMS/SPECTRIMS without baseline disease activity subgroup, no statistically significant effects of sc IFN β-1a were observed on ARR; however, treatment reduced T2 lesion activity and number in this subgroup, although the low patient number in this subgroup may have influenced the result. Separation between treated and untreated groups in terms of time to disability progression could be seen early in the treatment course for this subgroup, with continued separation over 2 years, although statistical significance was not shown. These results are in line with the overall SPECTRIMS study in which inflammatory and radiological components of MS were more affected by sc IFN β-1a treatment than was disability progression \[[@CR7]\]. The relationship between relapses and disability progression in RRMS has not been not fully elucidated. In patients with RRMS, relapses not only affect EDSS score in the short term \[[@CR18], [@CR19]\] but also have been shown to predict future confirmed disability progression \[[@CR20]\]. However, other research in patients with more advanced disease has shown a lack of association between relapses and disability \[[@CR21]\]. Some studies have suggested that once patients achieved a clinical threshold of disability (EDSS score of 4.0), disability progression was not significantly affected by relapses \[[@CR22]\]. The results for the PRISMS and PRISMS/SPECTRIMS subgroups from this study are consistent with relapses having a greater effect on disability. Some patients with MS may enter a period of fewer interactions with their healthcare provider or withdrawal of disease-modifying drugs (DMDs) as their disability accumulates and they transition to SPMS \[[@CR23]\]. These changes in care and treatment are sometimes due to the perception of providers that there are no effective treatment options for patients who appear to be transitioning to SPMS. However, as shown in this study, patients with moderate disability can still experience clinical and MRI benefits from treatment. Findings have been inconsistent regarding the ability of DMDs to delay disability progression in patients with RRMS with higher EDSS or in patients with SPMS regardless of relapse activity. Four large-scale studies assessed the effectiveness of IFN β in patients with SPMS \[[@CR3], [@CR24]\]. Among these IFN β studies, the European SPMS (EUSPMS) trial was the only trial to show a positive effect of treatment on the accumulation of irreversible disability progression \[[@CR3], [@CR24]\]. The differences in treatment benefit within these studies could be due to the different patient populations included. For example, placebo patients in the North American SPMS (NASPMS) trial progressed less than both placebo and active treatment groups in the EUSPMS trial, even though the inclusion criteria were comparable \[[@CR25], [@CR26]\]. Thus, patients participating in the EUSPMS trial were more likely closer to the relapsing phase of MS, while patients in the NASPMS trial were further along in the course of the disease \[[@CR3]\]. Evidence is also inconclusive for the effects of other DMDs in patients with high EDSS or SPMS. Natalizumab treatment effect seemed to favor patients with RRMS who have lower baseline EDSS scores (≤ 3.5) over those with higher scores \[[@CR10]\]; furthermore, natalizumab did not delay progression of ambulatory disability in patients with SPMS (in a cohort with baseline EDSS score 3.0--6.5 \[mean 5.6\], 29% of whom had relapses within the previous 2 years) \[[@CR11]\]. In a subgroup analysis of the FREEDOMS study, fingolimod showed a 68% reduction in the odds of disability progression in those with higher baseline EDSS scores (\> 3.5) versus a 23% reduction among those with lower scores; however, the relapse activity in the two subgroups was not described \[[@CR12]\]. In a subgroup analysis of the TEMSO trial, teriflunomide 14 mg showed a trend towards a greater effect on the risk of disability progression in patients with higher baseline EDSS scores (\> 3.5) compared with those with lower scores; ARR was reduced most in patients with lower EDSS at baseline \[[@CR13]\]. It is important to note that the PRISMS/SPECTRIMS subgroup described here included patients with SPMS from the SPECTRIMS trial, which failed to meet the primary endpoint of delaying disability progression. Most of the advances over the past two decades have been limited to patients with RRMS, with few treatments showing efficacy in slowing the rate of disability progression, specifically in patients with SPMS, whose disease has accumulated further. Examinations of treatment efficacy in patients with moderate disability are of interest in light of the developing treatment outlook for patients with progressive disease. Two drugs, the sphingosine-1-phosphate receptor modulator siponimod and a purine antimetabolite, Cladribine tablets, were recently approved by the FDA for the treatment of adults with relapsing forms of MS, including SPMS with active disease \[[@CR27], [@CR28]\]. In a phase III study, siponimod significantly reduced risk of 3-month confirmed disability progression by 21% in patients with SPMS and reduced the ARR (0.07 \[95% CI 0.06--0.09\]) compared with placebo (0.16 \[95% CI 0.12--0.21\]). Further subgroup analysis identified favorable effects of siponimod versus placebo on the HR of 3-month disease progression in patients who had superimposed relapses in the 2 years before the study (HR 0.67 \[95% CI 0.49--0.91\]), which suggests that patients with active SPMS received a greater benefit from treatment with siponimod compared with patients with lower activity (HR 0.87 \[95% CI 0.68--1.11\]) \[[@CR9]\]. In the phase III CLARITY trial, Cladribine tablets 3.5 mg/kg reduced ARR by 57.6% versus placebo (*p* \< 0.001) in patients with RRMS, and reduced risk of 3-month disability progression (HR 0.69 \[95% CI 0.49‒0.96\]) \[[@CR29]\]. In post hoc analyses of the CLARITY trial in which baseline EDSS score ≥ 3.5 was used as a proxy for active SPMS, Cladribine tablets reduced ARR versus placebo (relative risk 0.43 \[95% CI 0.30‒0.62; *p* \< 0.001), and 49% of patients treated with Cladribine tablets achieved no evidence of disease activity compared with 17% of patients who received placebo (odds ratio 4.51 \[95% CI 2.65‒7.69\]; *p* \< 0.0001), indicating efficacy in patients with more advanced disease \[[@CR30], [@CR31]\]. In addition, the approved indications for other DMDs have been recently updated to include clinically isolated syndrome and active SPMS, and additional updates are expected \[[@CR32], [@CR33]\]. These expanded indications may be due to the recognition by regulatory agencies that clinically isolated syndrome, RRMS, and SPMS with relapses are all part of a spectrum of active disease and treatment is warranted at each stage. The present research is limited by its post hoc nature. The selected patient subgroups having the characteristics of interest made up a small part of the populations from each of the source trials. Furthermore, our analysis did not include stratification of efficacy by patient factors, such as age and sex. Age may be an important predictor of efficacy, as demonstrated in a recent meta-analysis of randomized, blinded clinical trials of MS DMDs against placebo or active comparator, in which the efficacy of immunomodulatory DMDs was found to decrease with age \[[@CR34]\]. Although our analysis did not include analysis by sex, a treatment-by-sex interaction was observed in female patients in the SPECTRIMS trial, showing a delay in progression compared with placebo with both sc IFN β-1a doses (*p* = 0.006 for 44 µg and *p* = 0.038 for 22 µg), whereas no difference was observed in male patients \[[@CR7]\]. An additional limitation is in the lack of a clear definition of "transition" from RRMS to SPMS, and the difficulty of making this assessment within the confines of clinical trials of relatively short duration. Overall, a similar magnitude of effect was observed for the overall PRISMS/SPECTRIMS subgroup and PRISMS/SPECTRIMS with baseline disease activity subgroup. While efforts were made to select a population consisting of patients from both trials with similar baseline characteristics, it should be noted that the trials had different entry criteria and reported discordant results of disability progression. There are also caveats while extrapolating these results to the modern MS patient population, as higher relapse rates were seen in placebo in PRISMS and SPECTRIMS than have been reported in more recent trials. These post hoc analyses suggest that treatment with sc IFN β-1a 44 µg tiw effectively reduced relapses, burden of disease, T2 lesions, and in some cases, delayed disability progression in a subgroup of MS patients appearing to transition from RRMS to SPMS. Such patients with active disease and continued disability worsening may still derive some benefit from continued treatment with sc IFN β-1a. Electronic supplementary material ================================= {#Sec14} Below is the link to the electronic supplementary material. Supplementary file1 (PDF 801 kb) Staley Brod: At time of research "Medical College of Wisconsin, Milwaukee, WI, USA". The authors thank Joanna Rakoczy, PhD, of Caudex, New York, NY, USA (supported by EMD Serono, Inc., Rockland, MA, USA \[a business of Merck KGaA, Darmstadt, Germany\]) for editorial assistance in drafting the manuscript, collating the comments of authors, and assembling tables and figures. The study was supported by EMD Serono, Inc., Rockland, MA, USA (a business of Merck KGaA, Darmstadt, Germany) and Pfizer Inc., New York, NY, USA. MSF has received honoraria or consulting fees from Actelion, Bayer HealthCare, Biogen, Chugai, EMD Canada, Genzyme, Hoffmann-La Roche, Merck Serono, Novartis, Sanofi-Aventis, and Teva Canada Innovation; has served on an advisory boards for Actelion, Bayer HealthCare, Biogen, Hoffmann-La Roche, Merck Serono, Novartis, Opexa, and Sanofi-Aventis; and has participated in a speakers' bureau for Genzyme. SB has consulting agreements or has received speaker honoraria from Acorda, Avanir, Bayer HealthCare, EMD Serono, Genzyme, Pfizer, Questcor, and Teva Neurosciences; has served on advisory boards for Bayer HealthCare, EMD Serono, Genzyme, Questcor, Sanofi, and Teva; and has received research or contractual support from the Clayton Foundation for Research, EMD Serono, Pfizer, and Questcor. BAS has received consulting and/or speaking fees from Acorda, Bayer, Biogen, EMD Serono, Genentech, Novartis, Roche, Sanofi Genzyme, and Teva; and has received research/grant support from Acorda, Biogen, Genzyme, MedImmune, Novartis, and Roche. BAC has received consulting fees from Biogen-Idec, Celgene, and EMD Serono; and contracted research support through Northwestern University from EMD Serono, Hoffman-La Roche/Genentech, MedDay, and Novartis. BH is an employee of EMD Serono, Inc., Rockland, MA, USA. FD is an employee of EMD Serono, Inc., Billerica, MA, USA. PKC has received consulting fees from Accordant, Bayer, Biogen, Celgene, EMD Serono, Genentech/Roche, Genzyme/Sanofi, and Novartis; and has received fees for contracted research with Actelion, Alkermes, Genzyme/Sanofi, MedDay, National Institute of Neurological Disorders and Stroke (NINDS), and Novartis. In the PRISMS and SPECTRIMS studies, all patients provided written informed consent prior to treatment. Both studies were conducted in accordance with the ethical principles set forth in the Declaration of Helsinki and standards for Good Clinical Practice.

1. Introduction {#sec1} =============== Southeast Asia is a rich source of natural gas and petroleum, especially Brunei, Vietnam, Malaysia, and Indonesia. From GlobalData statistic, Malaysia and Indonesia will contribute around 80% and 70% of the Southeast Asia\'s total crude oil and natural gas production from eight planned projects in 2025, respectively. In brief, once the well is successfully drilled and installation is completed, the product must be transported to a facility where it can be treated, stored, processed, and refined through the pipeline system. Therefore, the environmental conditions of the production and pipeline systems are required for the prediction of materials life cycle and its maintenance requirements. Thus, any corrosion that occurs inside oil and gas pipeline systems is a serious factor that could lead to system failure. Corrosion not only causes economic losses but also greatly affects the safety and protection of the oil and gas resource. Hence, in upstream oil and gas applications, the most desirable alloy should be made of a strong material with good localized corrosion resistance, low cost, and suitable mechanical characteristics. Austenitic stainless steel type 316L (UNS S31603) has excellent corrosion resistance to the electrochemical properties of the passive film that forms on its surface. The passive film mainly consists of iron (Fe), chromium (Cr), and nickel (Ni) oxides. Cr promotes the formation of protective surface oxide, and Ni enhances the stability of the oxide film. Thus, stainless steel alloys with a higher composition of Cr and Ni prevent the iron from rusting and provide heat resistant properties \[[@B1]\]. Besides, 316L steel is a low carbon alloy that contains molybdenum (Mo), which also makes it more corrosion resistant, especially in highly sour conditions \[[@B1]\]. The physicochemistry of passive films directly influences the film properties. Therefore, many studies have investigated the effect of hydrogen sulfide corrosion at different hydrogen sulfide partial pressures (pH~2~S) \[[@B2]\], pH \[[@B3]\], and temperature \[[@B4]\] on the material surface. However, this area remains underexplored due to the cross effects between the parameters of the study \[[@B3]\]. Ding et al. \[[@B1]\] investigated the electrochemical behavior of 316L steel in Cl^−^ solutions under different pH~2~S. Additionally, Gao et al. \[[@B4]\] found an increased presence of iron oxide (Fe~3~O~4~) and iron sulfide in the corrosion product layer on substrate after performing H~2~S corrosion tests involving increased exposure times at 120°C. When both H~2~S and Cl^−^ were present, the passive film would become more sensitive to temperature, therefore significantly affecting the corrosion rate, the corrosion mechanism, and the properties of the material surface. Although various corrosion reactions could occur, the corrosion rate generally increases in the presence of H~2~S until sulfide saturation was achieved \[[@B2], [@B5]\]. Therefore, the authors aim to determine the effect of different pH~2~S exposure on the formation of multiple corrosion product layer of the 316L steel surface. The surface analysis of 316L steel was performed using a multianalytical technique consisting of X-ray photoelectron spectroscopy (XPS), field emission scanning electron microscopy (FESEM), and energy dispersive X-ray spectrometry (EDS). The microstructure characterization and crystalline phase characterization of 316L steels were performed using transmission electron microscopy (TEM). The metal ions that had dissolved in the test solution after 7 days were characterized via inductively coupled plasma-mass spectrometry (ICP-MS). Finally, the reaction mechanism occurring on the surface of 316L steel was explained in terms of the dissolution of CrO~3~ and Cr~2~O~3~ on the steel protective film at 3 bar pH~2~S after 7 days of testing. 2. Experiment {#sec2} ============= 2.1. Materials and Method {#sec2.1} ------------------------- For the experiment, a commercial austenitic stainless steel type 316L with a chemical composition listed in [Table 1](#tab1){ref-type="table"} was used. Then, optical emission vacuum spectrometer analysis was performed on the stainless steel using the point-to-plane excitation technique, following ASTM E 1086 Standards \[[@B6]\]. The 316 L steel sample was machined into a dimension of 20 × 20 × 3 mm^3^, abraded with a series of silicon carbide papers (up to 1200 grit), polished with a diamond spray (up to 1.0 *μ*m), rinsed with distilled water, degreased with acetone, dried at room temperature, and then stored in a desiccator. Each procedure for preparing the 316L steels was performed following ASTM G1-03 Standards \[[@B7]\]. All experiments were carried out in NACE TM0177 Solution A \[[@B8]\]. The solutions were added to an autoclave and deoxygenated by purging with nitrogen gas for two hours before the beginning of each experiment. All experiments were conducted at 60°C and pH 3.0 ± 0.5 to simulate the crude oil conditions in Southeast Asia especially around Malaysian region (30°C--90°C and pH 2--6) \[[@B9]\]. 2.2. Experimental Procedure and Weight-Loss Measurement {#sec2.2} ------------------------------------------------------- Experiments were carried out in a 5 L autoclave using experimental conditions which are summarized in [Table 2](#tab2){ref-type="table"}. A schematic diagram of the corrosion test apparatus was set up for the 0 bar and 3 bar of pH~2~S and is shown in Figures [1(a)](#fig1){ref-type="fig"} and [1(b)](#fig1){ref-type="fig"}, respectively, in Sour lab, at DNVGL Singapore laboratory. The samples were weighed using an analytical balance for the initial weight (*W*~i~), before being immersed vertically in test solution. Then, the 316L steel sample with a holder was attached to the top lid and sealed in the autoclave before immersing it vertically in the test solution. Besides, 3.5 L of the NACE A solution and 1.5 L vapor were used to allow the water vapor to expand in high pressure and temperature conditions. After that, a deoxygenation process was conducted by purging the autoclave with nitrogen gas for two hours before each test. The samples were heated at 60°C and pressurized with premixed gases (3 bar of H~2~S and 27 bar of N~2~ gas) equaling a total pressure of 30 bar until a stable pressure was reached. The pH and temperature of the solution were measured and recorded at the start and end of each experiment as shown in [Table 2](#tab2){ref-type="table"}. After 7 days of immersion, the samples were taken out from the solution, cleaned, and preserved with nitrogen according to ASTM G1-03 Standards \[[@B7]\]. The final samples were weighed again to obtain the final weight (*W*~f~) based on which weight loss was determined. After that, the corrosion rate was calculated using the following equation:$$\begin{matrix} {\text{corrosion rate} = \,\frac{kW}{pAT\,},} \\ \end{matrix}$$where *K* = constant, *W* = weight loss in mg, *ρ* = density of the metal in g/cm^3^, *A* = area in cm^2^, and *T* = time of exposure in hours. Then, the procedure was repeated with H~2~S-free conditions for the control sample in this experiment. 2.3. Characterization of Corrosion Products {#sec2.3} ------------------------------------------- Morphological images characterization of all 316L steels were performed using a Zeiss SUPRA 55VP field emission scanning electron microscope (FESEM) and a high-efficiency in-lens detector for clear topographic imaging in high-vacuum mode, in conjunction with energy dispersive X-ray spectroscopy (EDS) to characterize the elemental composition of the samples. The surface analysis was performed using X-ray photoelectron spectroscopy (XPS) (X-ray microprobe PHI Quantera II, Ulvac-PHI, Inc.) with a monochromatic Al-K*α* (hv = 1486.6 eV) X-ray source. Before deconvolution, the charging effects were corrected using a Kratos charge neutralizer system in order to minimize the carbon charging effect. In this case, all spectrum is corrected and referred to adventitious carbon binding energy at 284.8 eV. A TEM thin foil was prepared using the focused ion beam (FIB) technique (Helios Nanolab 600i, FEI). Also, detailed subsurface microstructure characterization was performed on a transmission electron microscope (TEM) (Tecnai G2 F20, FEI) with an EDS detector operated at 200 kV. Finally, the elemental analyses of the corrosion product dissolution in the test solutions were carried out on an inductively coupled plasma-mass spectrometer (ICP-MS) PerkinElmer Sciex Elan 9000. 3. Results and Discussion {#sec3} ========================= 3.1. The Effect of H~2~S Pressure on the Corrosion Rate {#sec3.1} ------------------------------------------------------- The average corrosion rate of 316L steels after immersion in test solution at 0 bar and 3 bar pH~2~S were calculated using weight loss measurement. The 316L steel at 3 bar pH~2~S had higher corrosion rate compared to that in 0 bar pH~2~S of 316L steel. The corrosion rate increased by 90.1% from 0.07 mm/yr at 0 bar pH~2~S to 0.74 mm/yr at 3 bar pH~2~S due to aggressiveness of sulfide ions in test solution \[[@B1]\]. The data show that overall mass loss is symmetrical with H~2~S pressure, and sulfide ions play a significant role in determining the kind of corrosion scales and reducing the surface protection of 316L steel \[[@B1]\]. 3.2. FESEM-EDS {#sec3.2} -------------- The surface morphology of the austenitic 316L steels immersed in test solution without H~2~S at 60°C for 7 days is shown in [Figure 2(a)](#fig2){ref-type="fig"}. Under an environment free of H~2~S, the 316L steel surface showed scratch marks because of polishing, but no grains or pits are shown. Besides, the elemental composition of 316L steel free of H~2~S was done using EDS, as shown in [Figure 2(a)](#fig2){ref-type="fig"}. The surface of the 316L steel in the H~2~S-free conditions showed that the passive film was formed in the presence of Cr. However, when the 3 bar pH~2~S was applied, the surface of the 316L steel shows a cracked surface, which indicates local breakdown of the passive layer, as shown in [Figure 2(b)](#fig2){ref-type="fig"}. The sulfide compounds were detected on the 316L steel surface as a result of applying pH~2~S. Due to local breakdown of the protective layer, microcracking was observed, as shown in [Figure 2(b)](#fig2){ref-type="fig"}. Therefore, corrosion products are formed after 3 bar pH~2~S exposure, demonstrating that the protectiveness of the passive film had very much degraded. The EDS analysis in [Figure 2(b)](#fig2){ref-type="fig"} shows the presence of a sulfur element, suggesting the presence of metal sulfides layer on the 316L steel surface \[[@B10]\]. However, Cr signals were depleted in the protective layer after exposure with H~2~S due to the dissolution of Cr in the test solution, as reported by Beverskog and Puigdomenech (1997) \[[@B11]\]. 3.3. X-Ray Photoelectron Spectroscopy (XPS) {#sec3.3} ------------------------------------------- The chemical composition of the 316L steel passive films that formed under different pH~2~S conditions after 7 days of immersion in test solution was investigated using XPS measurements, as shown in [Figure 3](#fig3){ref-type="fig"}. The major peaks present in the spectra for both 316L steels corresponded to C, O, Fe, Cr, Ni, Mo, and S. Generally, austenitic stainless steel is self-passivated, as chromium oxide (CrO~3~) instantly forms when it is exposed to air, but the nature of the passive layer changes when exposed to an aqueous solution \[[@B12]\]. Apart from that, the element contribution analysis of Cr and Mo was performed on the sample without exposure to H~2~S as shown in [Figure 4](#fig4){ref-type="fig"}. The Cr element is known as a key element in passive film formation. However, [Figure 4(a)](#fig4){ref-type="fig"} shows only three constituent peaks in the Cr 2p signal, representing Cr(OH)~3~ (Cr^3+^ 2p~3/2~; 577.5 eV), Cr~2~O~3~ (Cr^6+^ 2p~1/2~; 578.8 eV), and CrO~3~ (Cr^3+^ 2p~1/2~; 587.1 eV) formed during pre-exposure (during immersion) and Cr~2~O~3~ (577.1 eV) formed via the anion-cation reaction during postexposure (after immersion). The absence of Cr(OH)~3~ in this analysis implies that Cr was easily oxidized at lower pH (∼3.5) in H~2~S-free conditions. Besides, [Figure 4(b)](#fig4){ref-type="fig"} shows the narrow scan spectra of Mo 3d obtained on the sample in 0 bar pH~2~S conditions. The spectra of the doublet Mo (3d~3/2~ and 3d~5/2~) can be split into two components, corresponding to the oxides formed of Mo/MoO~2~ (Mo^4+^ 3d~5/2~; 228.8 eV and Mo^4+^ 3d~3/2~; 232.6 eV) and MoO~4~/MoO~3~ (Mo^6+^ 3d~5/2~; 232.9 eV and Mo^6+^ 3d~3/2~; 235.8 eV), respectively. It could also be observed that MoO~3~ was the primary Mo species. Meanwhile, the elemental contribution analysis of the specimen after immersion in the test solution in H~2~S-containing conditions yielded Ni, Fe, Mo, and S, as shown in [Figure 5](#fig5){ref-type="fig"}. The Ni 2p narrow spectra consisted of three compounds, namely, Ni(OH)~2~ (857.4 eV), NiO (855.7 eV), and NiS (853.9 eV), as shown in [Figure 5(a)](#fig5){ref-type="fig"}. The dominant peaks of Ni 2p~3/2~ in the spectrum corresponded to NiO, indicating that NiO could stably exist in the passive film. However, [Figure 5(b)](#fig5){ref-type="fig"} shows the narrow scan spectra of Fe 2p~3/2~ obtained on the 316L steel sample, which was immersed in the H~2~S solution at 3 bar pH~2~S for 7 days. Two deconvolution signals were observed, attributed to Fe~3~O~4~ (710.1 eV) and FeS (712.8 eV). From the reactivity series, Fe was more active than Ni and could participate in the passive film formation in the form of Fe~3~O~4~, Fe~2~O~3~, Fe(OH)~2~, FeOOH, and Fe(OH)~3~ \[[@B13]\]. However, iron sulfide (FeS) also appeared in the film due to the high affinity of sulfides with iron in the H~2~S solution \[[@B14]\]. Moreover, the possibility of Fe species dissolved preferentially in the H~2~S-containing solution and affected the migration of S^2−^ into the passive film, preventing the formation of Fe~2~O~3~ on the passive film \[[@B13]\]. Apart from that, the Mo 3d~3/2~ spectra consisted of MoO~3~ (3d~3/2~; 235.5 eV and 3d~5/2~; 232.4 eV), MoO~2~ (3d~3/2~; 230.2 eV), and bulk Mo from metal (226.9 eV), as shown in [Figure 5(c)](#fig5){ref-type="fig"}. This result is similar to Wang et al. (2017), which reported a Mo 3d photoelectron peak binding energy position at 232.6 eV and 235.8 eV, corresponding to molybdenum oxides (MoO~3~) \[[@B15]\]. Although the H~2~S molecule has a chemical structure similar to H~2~O, the polarizability of the sulfides such as S^2−^ or HS^−^ is higher than that of halides (Cl^−^) and OH^−^ when adsorbed onto the passive film and could appear in that film \[[@B16]\]. Since chromium sulfides were not detected, S^2-^ could only unite with Fe^2+^ and Ni^2+^. Therefore, from the deconvolution of S 2p shown in [Figure 5(d)](#fig5){ref-type="fig"}, four peaks were fitted by the narrow spectra at the binding energy of FeS~2~ (2p~3/2~; 162.2 eV), NiS (2p~3/2~; 162.9 eV), FeS~2~ (2p~1/2~; 163.7 eV), and NiS (2p~1/2~; 164.5 eV) corresponding to disulfide (S~2~^2−^), S^2−^, and S, respectively \[[@B16]\]. Based on the reactivity series of metals, nickel is less reactive than Cr and Fe, but nickel oxide could be porous and could easily entrap S^2-^ within the porous oxide. However, traces of NiO and/or Ni(OH)~2~ were also found in the passive film, similar to that reported by Luo \[[@B17]\]. The O 1s narrow scan spectra for austenitic 316L steel for both 0 bar and 3 bar H~2~S-containing conditions are shown in [Figure 6](#fig6){ref-type="fig"}. From the deconvolution of O 1s, several oxide species were found on the samples with O^2-^ constituents in H~2~S free conditions attributed to Cr~2~O~3~ (531.9 eV) and MoO~3~ (530.6 eV), as shown in [Figure 6(a)](#fig6){ref-type="fig"}. Meanwhile, [Figure 6(b)](#fig6){ref-type="fig"} shows the O 1s spectra of the sample in the 3 bar pH~2~S-containing conditions being dominated by Ni(OH)~2~ (532.5 eV)~,~ Fe~3~O~4~ (531.6 eV), MoO~3~ (530.7 eV), and NiO (529.4 eV), formed via the lattice nonconserving reaction through diffusion vacancy or interstitial vacancy filling based on the cation-anion reaction \[[@B18]\]. Besides, the Cr content in the passive film was very low, thus having no corresponding XPS spectra and revealing a decreased stability of this species at high H~2~S-Cl^−^ concentration in the test solution. Based on the XPS deconvolution in the atomic concentration analysis of [Table 3](#tab3){ref-type="table"}, 316L with H~2~S-free conditions showed signals attributed to Mo, Cr, and O. Meanwhile, Cr and Mo compounds were found associated with O to form Cr~2~O~3~, CrO~3~, MoO~3,~ and MoO~2~. Besides, the sample in H~2~S conditions produced more signals corresponding to Mo, Ni, Fe, S, O, and C. The atomic concentration of Cr was reduced after exposure to 3 bar pH~2~S that could be due to the adsorption and dissociation of sulfide ions (HS^−^ and S^2-^) on 316L steel surface, and then react with chromium oxides (Cr~2~O~3~ and CrO~3~) \[[@B19]\]. The Mo compound showed MoO~3~ signals with a photoelectron signal of 232.41 eV and metallic Mo with a photoelectron signal of 226.91 eV. The presence of sulfide contributed to the signals of FeS~2~ and NiS on the sample surface. Aside from that, the Ni signals were attributed to NiO, Ni(OH)~2~, and NiS. The product of the final chemisorption process was derived from the Fe elements, namely, Fe~3~O~4~ and FeS~2~. The summary of corrosion product layer formed at each condition is shown in [Table 4](#tab4){ref-type="table"}. The corrosion product formed at 0 bar H~2~S consists of Cr~2~O~3~, CrO~3~, MoO~3,~ and MoO~2~. Whilst the corrosion product at 3 bar H~2~S consists of the multiple corrosion product layer including Mo, MoO~3~, NiS, Ni(OH)~2~, NiO, FeS, and Fe~3~O~4~. 3.4. Inductively Coupled Plasma-Mass Spectrometry (ICP-MS) {#sec3.4} ---------------------------------------------------------- Metal ions of 0 bar and 3 bar pH~2~S after immersion in the test solutions for 7 days were evaluated via ICP-MS, as shown in [Figure 7](#fig7){ref-type="fig"}. *Y*-axis in [Figure 7](#fig7){ref-type="fig"} shows metal ions concentration after 7 days immersion in the test solution, while *x*-axis shows the main elemental composition in the 316L steel with two plots for each element which represents a partial pressure of H~2~S. In brief, both solutions (0 and 3 bar pH~2~S) showed no Mo, but Fe and Cr were selectively dissolved. The highest concentration of Fe dissolved in the test solution suggests that Fe is the main component of the 316L stainless steel. The selective dissolution of Cr became more pronounced as 3 bar pH~2~S was exposed. One point to note is that metals that rely on Cr~2~O~3~ for corrosion protection may be at risk of having the Cr~2~O~3~ dissolve into the solution with increased pH~2~S. This issue is a considerable concern since an increase in the concentration of HS^−^, S^2-^, or Cl^−^ anions could accelerate the Cr~2~O~3~ film dissolution. In comparison to the 0 bar pH~2~S sample, the sample exposed to 3 bar pH~2~S had significantly increased the concentrations of Fe and Cr. Castle and Qiu (1990) \[[@B20]\] reported that Cr dissolves into solution as trivalent ions, while Fe dissolves as divalent ions. Additionally, the adsorbance of S on the metal surface produces metal sulfides that make the oxide film more defective \[[@B21]\]. Therefore, the values obtained from the ICP-MS results are generally in very good agreement with the XPS results when comparing the intensity of the composition of the compound on the sample surface to the dissolved metal ions in the test solution. The result showed that the depleted compound present in the corrosion product layer dissolved directly into the electrolyte during the experiment. Furthermore, the surface atomic percentage in [Table 3](#tab3){ref-type="table"} shows that the Cr content was very low for the sample exposed to 3 bar pH~2~S, indicating that Cr dissolution had indeed happened. Pourbaix \[[@B11]\] stated that Cr(II) ions are unstable and react very rapidly with oxygen in acidic solutions. Referring to the Pourbaix diagram, the formation of Cr, Cr(OH)~2~, Cr(OH)~3~, and HCrO~4~^−^ is related to the dissolution of Cr in H~2~O-Cl^−^ and H~2~O-S-Cl^−^ solutions, mostly in conditions close to pH 3.5 (blue dash line) with a potential range expected between 0 and 1 volt, as shown in Figures [8(a)](#fig8){ref-type="fig"} and [8(b)](#fig8){ref-type="fig"} \[[@B22], [@B23]\]. Overall, the dissolution process, which is the release of alloying elements in the electrolyte from the bulk or the film, was investigated using ICP-MS after 7 days of exposing the sample in NACE A solution. The main elements to dissolve were iron and chromium, which were detected in the electrolyte in the same proportion as the chemical composition of the 316L steel ([Table 1](#tab1){ref-type="table"}). 3.5. Transmission Electron Microscopy (TEM) {#sec3.5} ------------------------------------------- Transmission electron microscopy (TEM) was conducted to further investigate the structure of passive films. TEM micrograph with the point EDS of both 316L steels was recorded as shown in [Figure 9](#fig9){ref-type="fig"}. 316L steels were prepared using FIB with a platinum coating layer, an oxide layer, and a substrate of 316L steel. The layer formed in both samples showed different patterns and multilayered in the 3 bar pH~2~S 316L steel ([Figure 9(c)](#fig9){ref-type="fig"}) compared to that in the 0 bar pH~2~S 316L steel ([Figure 9(a)](#fig9){ref-type="fig"}). These films acted as layers that barred the reaction products from the corrosive environment and took the form of mono- or multilayer oxygen or other chemical species adsorbed onto the metal surface. The four-point EDS (marked as P1, P2, P3, and P4) was analyzed in both 0 bar and 3 bar pH~2~S conditions. P1 represents the outer layer/interface, P2 and P3 represent the inner layer, and P4 is the sample substrate. This analysis focused on elements that had a major contribution to the samples such as Fe, Ni, S, O, and Mo. Point 1 in the 0 bar pH~2~S sample comprised O, Cr, Mn, Fe, Ni, and Mo elements at the outer layer of the sample. However, Mn and Cr elements were not detected at P1 in 3 bar pH~2~S, but the S element was present in the outer layer of the 3 bar pH~2~S sample as shown in [Figure 9(d)](#fig9){ref-type="fig"}. The Cr signal disappeared at the outer layer, due to formation of Cr(OH)~3~, hence suggesting that more Cr(OH)~3~ was dissolved in the test solution \[[@B24]\]. Thus, one of the primary elements that may contribute to P1 in the 3 bar pH~2~S is MoO~3~ and a result that is in good agreement with the XPS results. Moreover, points P2 and P3 represent the inner part of the multilayer that gives strong signals of Fe, Cr, and Ni for 0 bar pH~2~S and Fe, Ni, and S for the EDS analysis of 3 bar pH~2~S. In addition, Ni enriched about 51.42% at P2 and 53.43% at P3 for the 3 bar pH~2~S sample ([Figure 9(d)](#fig9){ref-type="fig"}). As the EDS and XPS results of 0 bar pH~2~S are linked, the primary elements that contributed to P2 and P3 were MoO~3~, MoO~2~, Cr~2~O~3~, and CrO~3~. However, it has been proposed that Mo could thicken the passive film, increase the surface affinity towards oxygen, and decrease the propensity for Cl^−^ to adsorb or form an extra secondary protective film \[[@B18]\]. Nevertheless, the primary elements that contributed to P2 and P3 at 3 bar pH~2~S were NiO, Ni(OH)~2~, NiS, FeS, Fe~3~O~4~, and FeS~2~, consistent with the EDS and XPS results. Since Ni did not participate in the passive film formation, it would be segregated by the oxides, and it consequently enriched underneath the passive film in the form of NiO and Ni(OH)~2~, as also verified by the XPS results in [Figure 5(a)](#fig5){ref-type="fig"}. Finally, EDS P4 showed a strong signal attributed to the Fe element in both 0 bar and 3 bar pH~2~S conditions. The atomic composition showed Fe enriched by about 61.01% and 74.07% at 0 bar pH~2~S and 3 bar pH~2~S, respectively, both of which correspond to the bare metal. Meanwhile, the enlargement of TEM micrograph of 3 bar pH~2~S 316L steel in [Figure 9(c)](#fig9){ref-type="fig"} shows that the porousness of the oxide layer was precipitated at the upper layer of the film. Generally, passive films are formed with a highly disordered "barrier" layer adjacent to the substrate and film, comprising a precipitated phase that may incorporate anion and/or cation from the solution \[[@B25]\]. However, the contribution of ions in the NACE A solution, i.e., NaCl, CH~3~COOH, and H~2~O in the presence of H~2~S, affected the oxide layer of the 316L steel samples, forming multilayer corrosion products after 3 bar partial pressure H~2~S exposure as shown in [Figure 10](#fig10){ref-type="fig"}. 3.6. The Formation and Dissolution of 316L Corrosion Products {#sec3.6} ------------------------------------------------------------- A passive film is envisaged to be a bilayer or multilayer structure comprising a point defective, nanocrystalline barrier layer and a porous outer layer that is formed by the hydrolysis of cations, depending upon the local conditions \[[@B25]\]. Corrosion products formation and dissolution are a continuous process and, generally, the film will have defects (vacancies and interstitials). At 3 bar pH~2~S condition, a decrease in corrosion resistance due to the introduction of sulfide ions (H~2~S, HS^−^, and S^2-^) to 316L steel was associated with the increase in proton reduction, leading to the anodic dissolution. Sulfides in the passive film are reported to have more defects than oxides \[[@B16]\], and thus, the film is less protective and stable in a high sour environment. In brief, Cr and Mo oxides were first generated under H~2~S-free conditions following the reactions in equations ([2](#EEq2){ref-type="disp-formula"})-([3](#EEq3){ref-type="disp-formula"}) for Mo oxides and the reactions in equations ([4](#EEq4){ref-type="disp-formula"})--([6](#EEq6){ref-type="disp-formula"}) for the Cr oxides:$$\begin{matrix} \left. \text{Mo}^{+} + {4\text{OH}}^{-}\longrightarrow\text{MoO}_{2} + {2\text{H}}_{2}\text{O} \right. \\ \end{matrix}$$$$\begin{matrix} \left. \text{Mo}^{6 +} + {6\text{OH}}^{-}\longrightarrow\text{MoO}_{3} + {3\text{H}}_{2}\text{O} \right. \\ \end{matrix}$$$$\begin{matrix} \left. 2\text{Cr}^{3 +} + 6\text{OH}\longrightarrow 2\text{Cr}\left( \text{OH} \right)_{3} \right. \\ \end{matrix}$$$$\begin{matrix} \left. 2\text{Cr}\left( \text{OH} \right)_{3}\longrightarrow\text{Cr}_{2}\text{O}_{3} + 3\text{H}_{2}\text{O} \right. \\ \end{matrix}$$$$\begin{matrix} \left. \text{Cr}^{6 +} + 6\text{OH}^{-}\longrightarrow\text{Cr}_{2}\text{O}_{3} + 3\text{H}_{2}\text{O} \right. \\ \end{matrix}$$ After long-time passivation, a stable passive film forms on the 316L steel surface with a dual-layer structure, as shown in [Figure 9(a)](#fig9){ref-type="fig"}. The amorphous outer layer consists of CrO~3~, Cr~2~O~3~, MoO~3~, and MoO~2~ in H~2~S-free conditions, as proven by the XPS surface analysis ([Figure 4](#fig4){ref-type="fig"}). In addition, MoO~2~ and MoO~3~ metals are also formed after exposure to 3 bar pH~2~S via a reaction with hydroxyl, similar to the reactions in equations ([2](#EEq2){ref-type="disp-formula"})-([3](#EEq3){ref-type="disp-formula"}), respectively. The passivation of Mo at higher pH~2~S conditions could mainly be associated with the formation of MoS~2~ (equation ([7](#EEq7){ref-type="disp-formula"})) \[[@B26]\]. However, the replacement of MoS~2~ by MoO~2~ could be predicted by the reaction with water as shown in equation ([8](#EEq8){ref-type="disp-formula"}) \[[@B26], [@B27]\]:$$\begin{matrix} \left. \text{Mo}^{4 +} + 4\text{HS}^{-}\longrightarrow\text{MoS}_{2} + 2\text{H}_{2}\text{S} \right. \\ \end{matrix}$$$$\begin{matrix} \left. \text{MoS}_{2} + {2\text{H}}_{2}\text{O}\longrightarrow\text{MoO}_{2} + 2\text{H}_{2}\text{S} \right. \\ \end{matrix}$$ Since the corrosion product of Cr compound did not form after 3 bar pH~2~S exposure due to dissolution of Cr in the test solution to produce Cr^3+^ ([Figure 8](#fig8){ref-type="fig"}), as per the reactions are shown in equations ([9](#EEq9){ref-type="disp-formula"})--([11](#EEq11){ref-type="disp-formula"}):$$\begin{matrix} \left. 2\text{CrO}_{3} + 2\text{OH}^{-}\longrightarrow\text{Cr}_{2}\text{O}_{7}^{2 -} + \text{H}_{2}\text{O} \right. \\ \end{matrix}$$$$\begin{matrix} \left. \text{Cr}_{2}\text{O}_{7}^{2 -} + \text{H}_{2}{\text{O} + 3\text{H}}_{2}\text{S}\longrightarrow 2\text{Cr}^{3 +} + \text{H}_{2}\text{O} \right. \\ \end{matrix}$$$$\begin{matrix} \left. \text{Cr}_{2}\text{O}_{3} + \text{H}_{2}{\text{O} + 3\text{H}}_{2}\text{S}\longrightarrow 2\text{Cr}^{3 +} + {3\text{S}}^{2 -}{+ 4\text{H}}_{2}\text{O} \right. \\ \end{matrix}$$ The depletion of Cr in the passive film resulted in the creation of a porous-precipitated layer because of the vacancies of cations and anions, as shown in [Figure 10](#fig10){ref-type="fig"}. Thus, depletion of Cr induces a higher S content in the passive film and results in passive film degradation. Besides, Ni(OH)~2~, NiO, and NiS are formed after exposure to 3 bar pH~2~S, respectively, based on the reactions in equation ([12](#EEq12){ref-type="disp-formula"})--([14](#EEq14){ref-type="disp-formula"}):$$\begin{matrix} \left. \text{Ni}^{2 +} + 2\text{OH}^{-}\longrightarrow\text{Ni}\left( \text{OH} \right)^{2} \right. \\ \end{matrix}$$$$\begin{matrix} \left. \text{Ni}\left( \text{OH} \right)^{2}\longrightarrow\text{NiO} + \text{H}_{2}\text{O} \right. \\ \end{matrix}$$$$\begin{matrix} \left. \text{Ni}^{2 +} + 2\text{HS}^{-}\longrightarrow\text{NiS} + \text{H}_{2}\text{S} \right. \\ \end{matrix}$$ Aside from that, Fe~3~O~4~, FeS, and FeS~2~ formations were slightly different compared to other elements in 3 bar pH~2~S conditions. The OH^−^ from the solution reacted first with ionic Fe in the outer passive film layer to form iron hydroxide followed by oxidation to form Fe~3~O~4~. The main reactions are described by the following equations \[[@B28]\]:$$\begin{matrix} \left. \text{Fe}^{2 +} + {2\text{OH}}^{-}\longrightarrow\text{Fe}\left( \text{OH} \right)^{2} \right. \\ \end{matrix}$$$$\begin{matrix} \left. 4\text{Fe}\left( \text{OH} \right)^{2} + 2\text{H}_{2}\text{O} + \text{O}_{2}\longrightarrow 4\text{Fe}\left( \text{OH} \right)_{3} \right. \\ \end{matrix}$$$$\begin{matrix} \left. 2\text{Fe}\left( \text{OH} \right)^{3} + \text{Fe}\left( \text{OH} \right)_{2}\longrightarrow\text{Fe}_{3}\text{O}_{4} + 4\text{H}_{2}\text{O} \right. \\ \end{matrix}$$ In an H~2~S environment, HS^−^ or S^2-^ could also incorporate into the film to form FeS and FeS~2~ based on the reactions in equations ([18](#EEq18){ref-type="disp-formula"}) and ([19](#EEq19){ref-type="disp-formula"}) \[[@B29]\]:$$\begin{matrix} \left. \text{Fe}^{2 +} + \text{S}^{2 -}\longrightarrow\text{FeS} \right. \\ \end{matrix}$$$$\begin{matrix} \left. \text{Fe}^{2 +} + 2\text{HS}^{-}\longrightarrow\text{FeS}_{2} + \text{H}_{2}\text{Fe}^{2 +} + \text{S}^{2 -} \right. \\ \end{matrix}$$ Therefore, the protectiveness of the 316L passive film could decrease, and the corrosion resistance of the 316L also deteriorates at higher H~2~S conditions. Hence, different materials could be subjected to further study and are analyzed at a range of different scales for safety concern, relating to the oil and gas production system. 4. Conclusions {#sec4} ============== From this study, the effect of H~2~S pressure on the formation of corrosion products on the 316L steels was investigated. The corrosion rate was increased at 3 bar pH~2~S, and the morphology of corrosion product experienced cracked surface and local breakdown of barrier layer compared to H~2~S-free conditions. Similarly, the TEM results have shown that the 316L steel at 3 bar pH~2~S condition exhibits more porous in the corrosion product layer. XPS results indicated the signals (peaks) of passive films which are Mo, MoO~3~, NiO, Ni(OH)~2~, NiS, Fe~3~O~4~, and FeS~2~. However, in H~2~S-free conditions, the only passive film was observed at MoO~2~, MoO~3~, Cr~2~O~3~, and CrO~3~ signals. Finally, Cr signals in XPS analysis were decreased after exposure to 3 bar pH~2~S, and it has been confirmed by ICP-MS that the chromium was dissolved into solution. Therefore, 316L could not be sustained in high partial pressure of H~2~S due to local breakdown of the crystalline layer, and the protectiveness of the passive film had very much degraded. The authors would like to thank The National University of Malaysia (UKM) and DNV GL Malaysia Sdn. Bhd (grant numbers ST-2018-015, DIP-2018-032, and GUP-2017-075) for providing the experimental facilities and financial support to conduct the MOU research project. Data Availability ================= The Pourbaix diagrams of Cr with pH~2~S-free and pH~2~S-containing conditions in aqueous at *T* = 25°C \[[@B23], [@B24]\] and data supporting (inductively coupled plasma-mass spectrometry (ICP-MS) analysis) are from previously reported studies and datasets, which have been cited. The processed data are available from the corresponding author upon request. Conflicts of Interest ===================== The authors declare that there are no conflicts of interest regarding the publication of this paper. {#fig1} {#fig2} {#fig3} {#fig4} {#fig5} {#fig6} {#fig7} ![Pourbaix diagrams of Cr with (a) pH~2~S-free and (b) pH~2~S-containing conditions in aqueous at *T* = 25°C \[[@B22], [@B23]\].](TSWJ2020-3989563.008){#fig8} {#fig9} {#fig10} ###### The chemical composition of the tested 316L austenitic stainless steel. Composition C Si Mn P S Cr Mo Ni Cu Al Fe ------------- ------ ------ ------ ------ ------ ------- ------ ------- ------ ------ ------ Weight (%) 0.02 0.43 1.31 0.03 0.01 16.48 2.08 10.18 0.34 0.01 Bal. ###### The experimental parameters and weight loss of 316L steel exposed to two different environments. Test condition Without H~2~S With H~2~S ------------------------------ ---------------------------------------------------------------------------- ------------ Temperature (°C) 60 Partial pressure H~2~S (bar) 0 3 Total pressure (bar) 30 pH (before/after) 2.52/2.63 2.95/3.42 Weight loss (g) 0.01 0.086 Immersion time 7 days (128 h) Solution NACE TM0177 solution A (5wt% NaCl and 0.5wt% CH~3~COOH in distilled water) ###### The XPS deconvolution of percentage surface atomic concentration for 316L steels at different conditions. Condition Surface atomic concentration % -------------- -------------------------------- ------ ------ ------ ------ ------ ------ 0 bar pH~2~S 26.31 1.74 0.20 --- --- --- Bal. 3 bar pH~2~S 34.52 0.10 0.19 0.20 6.94 1.15 Bal. ###### The list of corrosion products from XPS deconvolution of percentage surface atomic concentration for 316L steels in different conditions. Corrosion product Binding energy (eV) ------------------- --------------------- ------- CrO~3~ 578.48 Nil Cr~2~O~3~ 577.03 Nil Mo Nil 232.4 MoO~2~ 228.85 Nil MoO~3~ 232.96 226.9 NiO Nil 855.7 Ni(OH)~2~ Nil 857.4 NiS Nil 853.9 FeS Nil 712.8 Fe~3~O~4~ Nil 710.1 [^1]: Academic Editor: Michael Pravica

1. Introduction {#sec1} =============== Yttrium-90 (^90^Y) radioembolization or selective internal radiation therapy (SIRT) with resin microspheres is an FDA approved therapy for patients with colorectal cancer liver metastases (CLM) who are not candidates for hepatic resection or ablation but have adequate liver function and a satisfactory performance status \[[@B1]\]. Often, patients present for SIRT in the salvage setting with progression of disease after several prior treatments, including surgery, ablation, and/or chemotherapy with cytotoxic and biologic agents \[[@B2]--[@B5]\]. We report a case of chemorefractory CLM progressing after hepatic resections, hepatic arterial pump chemotherapy (HAC), and multiple regimens of systemic chemotherapy that was successfully treated with SIRT after biliary stenting above the papilla to relieve obstructive hyperbilirubinemia. 2. Case Presentation {#sec2} ==================== A 75-year-old man presenting ten years earlier with synchronous metastatic colorectal cancer (T3N0M1) and multiple comorbidities was referred to our department for image-guided therapy due to progressive metastatic liver disease. After initial diagnosis, the patient underwent right hemicolectomy and concurrent wedge biopsy followed by intraoperative radiofrequency ablation (RFA) of a single hepatic lesion in segment 7. Thirteen months after surgery, due to multiple new enlarging hepatic metastases, a right hepatectomy was performed with subsequent placement of a hepatic arterial infusion pump (HAIP). Postoperatively, combination floxuridine (FUDR) HAC with six cycles of systemic FOLFOX chemotherapy was administered. Recurrence occurred approximately two years after discontinuation of chemotherapy leading to reinitiation of HAC and administration of irinotecan and cetuximab for a four-month period, which downsized the tumors and enabled a limited hepatic resection of segment 4A. After six months, systemic chemotherapy was resumed to address increasing CEA levels and infiltrative soft tissue at the previous resection margin. For two years, disease control was achieved with different chemotherapy regimens, including cetuximab, panitumumab/irinotecan, and capecitabine. Disease progression led to an additional surgical procedure, consisting of resection of the remainder of segment 4. After surgery the panitumumab/irinotecan regimen was reinitiated but was eventually discontinued due to poor tolerance and poor tumor response. In the face of progressing hepatic metastases the patient was referred for evaluation regarding appropriateness of SIRT. Bilirubin levels were slightly elevated at presentation (1.6 mg/dL) and for the past six months ranged between 1.3 and 2.3 mg/dL. Imaging revealed several lesions throughout the remnant left liver, including a central lesion causing partial biliary obstruction ([Figure 1](#fig1){ref-type="fig"}). Shortly after presentation, bilirubin levels were found to have rapidly increased to 7.5 mg/dL. To address this issue, a 10 × 68 mm Wallstent was primarily placed, under fluoroscopic guidance, across the left hepatic duct obstruction without violation of the papilla. No additional drainage catheters were required and total bilirubin levels following this intervention dropped significantly. With bilirubin levels continuing to drop, the patient was considered a candidate for SIRT and underwent standard evaluation with pre-SIRT arteriography and Tc-99 macroaggregate albumin (MAA) mapping. Angiographic evaluation revealed tumor arterial supply by the left hepatic artery and the right phrenic artery. The latter was embolized with microcoils to redistribute arterial flow to the tumors via the hepatic artery and thus optimize ^90^Y microsphere distribution. Tc-99 MAA hepatic scintigraphy demonstrated 6% lung shunting ([Figure 2(a)](#fig2){ref-type="fig"}), within acceptable values (\<20%). One month after biliary stent placement and two weeks after the mapping session, serum bilirubin levels were marginally above the upper limit of normal (1.4 mg/dL). SIR-spheres (^90^Y-resin microspheres; Sirtex Medical, Sydney, Australia) were injected selectively in the left hepatic artery through a microcatheter, delivering the entire dose of 52.6 mCi (1.95 GBq). SPECT/CT Bremsstrahlung liver imaging confirmed successful delivery of the full dose within the liver ([Figure 2(b)](#fig2){ref-type="fig"}), without any extrahepatic activity. The patient tolerated the procedure well, without any significant acute or delayed side effects/toxicities. Total bilirubin levels during this period did not exceed the levels before SIRT. Imaging revealed complete PET response to SIRT and resolution of SUV activity within the liver ([Figure 3](#fig3){ref-type="fig"}). Two months after the procedure, despite control of hepatic disease, progression in nodal sites outside the liver and enlarging pulmonary nodules mandated reinitiation of systemic chemotherapy with 5-FU/leucovorin and panitumumab, which temporarily controlled metastatic disease. The patient eventually succumbed to his illness at an outside hospital seven months after SIRT and 10.5 years after initial diagnosis. There was no evident progression of CLM on the latest available CT scan obtained 5 months after SIRT, despite extrahepatic nodal disease involvement. 3. Discussion {#sec3} ============= For patients with unresectable chemorefractory CLM, SIRT has established safety, is well tolerated, and provides promising survival and response rates \[[@B1], [@B6], [@B7]\]. Patient selection characteristics for SIRT include, among others, liver-dominant tumor burden, patients with a life expectancy greater than 3 months, adequate hepatic function (serum bilirubin levels \<2 mg/dL), and a satisfactory performance status \[[@B8]\]. In this case, the patient had a performance status within the inclusion criteria (ECOG status 0) despite being heavily pretreated for over 10 years with systemic chemotherapy and HAC, as well as with multiple hepatic resections. It has been shown that, in the salvage setting after different chemotherapy regimens or after systemic and HAIP chemotherapy, SIRT is safe provided that bilirubin levels remain below 1.5 mg/dL \[[@B2]--[@B5]\]. Biliary intervention (drainage, stenting, or bilioenteric anastomosis) prior to SIRT has been reported and in several institutions is not considered an exclusion criterion \[[@B9], [@B10]\]. However to our knowledge this is the first reported case of primary biliary stenting as a bridge to SIRT in a patient with coexisting limited hepatic reserve due to multiple resections and HAC. In our patient, it was contemplated that the rapid elevation in bilirubin was primarily due to a centrally located metastasis causing biliary obstruction rather than a consequence of HAC toxicity or of rapid deterioration of hepatic function. The latter is a significant factor to take into consideration when treating patients with hepatocellular carcinoma and underlying cirrhosis, as well as those with metastatic disease and chemotherapy induced steatohepatitis \[[@B12]\]. Whenever feasible, treatment of high level biliary obstruction with primary stent placement avoiding violation of the papilla is recommended. This provides optimal drainage and normalization of bilirubin levels and at the same time minimizes the contamination of the biliary tree by enteric contents and the risk of hepatic abscess formation, as the protective role of the papilla is maintained \[[@B13]\]. The presence of an incompetent sphincter of Oddi is a well-recognized risk factor for hepatic abscess formation after chemoembolization \[[@B14]\], but recent evidence suggests that this risk may be lower for SIRT \[[@B9]\]. To prevent the development of cholangitis and abscess formation in this patient, two intravenous doses of cefotetan were administered before and after biliary stenting. Prior to SIRT, an additional dose of cefotetan was administered and in the postprocedural setting the patient received prophylaxis with oral metronidazole and ciprofloxacin for five days. Despite the fact that the patient had limited hepatic reserve due to previous hepatectomies, the entire liver remnant was treated with administration of the full calculated ^90^Y dose within the left hepatic artery without stasis. This approach may be followed when treating patients with limited hepatic volume, although a conservative approach with sequential lobar treatment is preferred when treating patients with bilobar disease \[[@B2], [@B15]\]. After the procedure and during follow-up evaluation there was no evidence of significant toxicity, liver failure, or radioembolization induced liver disease (REILD), even though the patient had risk factors for the development of these complications (treatment of the entire liver remnant, elevated baseline bilirubin levels, and multiple chemotherapy regimens) \[[@B16], [@B17]\]. Although at least in theory patients who receive HAC should not require any arterial embolization for flow redistribution, it is noteworthy that extrahepatic collaterals or accessory gastric or pancreaticoduodenal branches can recanalize and supply the hepatic tumors. This was the case in this patient with phrenic arterial recruitment supplying tumor. Rarely after HAIP placement or coil embolization for prior SIRT sessions, collaterals can still develop or previously ligated/embolized vessels may recanalize, requiring additional embolization for safe and optimal delivery of the ^90^Y microspheres \[[@B18]\]. In conclusion, the current case demonstrates that, with appropriate evaluation and biliary intervention, SIRT can still be safe and effective in the most compromised patients, even in the face of biliary obstruction. Conflict of Interests ===================== Constantinos T. Sofocleous has received in the past research support from and is currently a consultant for Sirtex Medical, Inc. The other authors do not report any potential conflict of interests. {#fig1} {#fig2} {#fig3} [^1]: Academic Editor: Zu-Yau Lin

Introduction ============ Melanoma is one of the most severe forms of malignant diseases [@B1]. Although recent therapeutic agents for melanoma have prolonged patient survival, their prognosis remains poor. In recent years, some epochal strategies against melanoma have been proposed, focusing on immunostimulatory therapy [@B2], because cellular immunity, especially interferon (IFN)-γ released from cytotoxic T lymphocytes (CTLs) is essential to reject bulky melanoma tumors [@B3]. Serum osteocalcin (OC), a noncollagenous bone matrix protein secreted by osteoblasts, in serum has been correlated with bone remodeling under pathological conditions including cancer bone metastasis [@B4], and during normal bone turnover. OC in serum exits as two types, γ-carboxylated OC (GlaOC) and lower- (or un-)γ-carboxylated OC (GluOC) [@B5]-[@B8]. We have reported that GluOC, but not GlaOC, improves glucose metabolism [@B5]-[@B8], which is consistent with a previous report [@B9]. These results indicate that each OC type might play a different role in various pathophysiological conditions. In terms of cancer, our recent study revealed that GlaOC and GluOC have completely opposite effects on the growth of human prostate cancer cells*in vitro* [@B10], that is, GluOC suppresses cancer cell growth, whereas GlaOC accelerates it. In the present study, we investigated anticancer mechanisms of exogenous GluOC using a mouse melanoma cell line, B16. There is an intimate crosstalk between bone lineage cells and immune cells [@B11], and OC has been reported to be regulated by proinflammatory cytokines [@B12] and decreases in a weakened immune system [@B13]. Therefore, we investigated the immunological aspect of GluOC in cancer progression, as well as its direct antitumor effects. Materials and Methods ===================== Materials --------- Recombinant GluOC was prepared as described previously [@B6]. GlaOC was purchased from AnaSpec (Fremont, CA, USA). Cells ----- The mouse melanoma cell line, B16, was obtained from RIKEN BioResource Center. The cells were maintained in Dulbecco\'s modified Eagle\'s medium supplemented with 10% fetal bovine serum (Thermo Fisher Scientific, Waltham, MA, USA), penicillin (100 U/ml), and streptomycin (0.1 mg/ml) at 37°C in humidified air containing 5% CO~2~. Measurement of cell viability ----------------------------- Cell viability was determined with a WST8 assay (Cell Count Reagent SF, Nacalai Tesque, Kyoto, Japan) and a BrdU uptake assay (BrdU Cell Proliferation ELISA Kit, Exalpha Biologicals, Shirley, MA, USA), as reported previously [@B10]. Each cell seeding density was optimized before experiments (GluOC: 1 × 10^4^ cells/ well, GlaOC: 5 × 10^3^ cells/ well). Receptor tyrosine kinase (RTK) phosphorylation antibody array ------------------------------------------------------------- The RTK phosphorylation antibody array was emploved a Mouse Phospho-RTK Array Kit (R&D Systems, Minneapolis, MN, USA), as reported previously [@B10]. After precultured for 24 h, B16 cells were treated with GluOC (10 ng/ml) or GlaOC (10 ng/ml) for 6 h. Cells were lysed in lysis buffer 17 (R&D Systems) including 10 μg/ml aprotinin, 10 μg/ml leupeptin, and 10 μg/ml pepstatin. Cell lysate (57 μg protein) was applied to the array. Animals ------- Female C57Bl/6N mice (Charles River Laboratories, Yokohama, Japan) were maintained under specific pathogen-free conditions. All animal experiments and a part of experiments were approved by the Animal Ethics Committees of Kyushu University and Fukuoka University, respectively. Determination of tumor growth in B16 transplants of C57Bl/6N mice ----------------------------------------------------------------- A micro-osmotic pump (ALZET, model 1004; DURECT, Cupertino, CA, USA) was used to administer saline, GluOC at a high dose (7.5 µg/mouse/day), or GluOC at a low dose (1.5 µg/mouse/day) by sustained release into the left flank. After 1 week, B16 cells (1×10^6^cells) in 0.1 ml of PBS were injected subcutaneously into the right flank of the 9-week-old female C57Bl/6N mice (n=8-10) a week later. Tumor size was measured with calipers every 3 or 4 days. Tumor volume was calculated as *ab*^2^/2, where *a* and *b* are the largest and smallest central cross-sectional dimensions, respectively. At 20 days after tumor transplantation, tumors were extracted after euthanasia. Immunohistochemistry -------------------- Tissue was fixed with 4% paraformaldehyde, embedded in paraffin, and sectioned at a thickness of 4-6 μm. Following routine procedures including quenching of endogenous peroxidase and antigen retrieval, the sections were incubated with rabbit polyclonal antibodies against osteocalcin (1:100 dilution, ab93876; Abcam, Cambridge, UK), CD3 (1:100 dilution, 17A2; Biolegend, San Diego, CA, USA), or IFN-γ (1:100 dilution, bs-0480R; Bioss, Woburn, MA, USA). After washing, the sections were incubated for 30 min with biotin-conjugated goat anti-rabbit (1:500 dilution, BA-1000; Vector, Burlingame, CA, USA) or mouse IgG (1:500 dilution, BA-9200; Vector, Burlingame, CA, USA). Tissue sections were visualized using and avidin-horseradish peroxidase (HRP) and 3,3′-diaminobenzidine (DAB), followed by counterstaining with hematoxylin. Images were acquired under a microscope. Primary splenocyte culture -------------------------- Splenocytes were prepared as described previously [@B14]. Briefly, spleens were disrupted in RPMI-1640 (Sigma-Aldrich, St. Louis, MO, USA) with 10% fetal calf serum (Life Technologies Japan, Tokyo, Japan). After erythrocytes were lysed by treatment with 0.83% ammonium chloride, the remaining splenocytes were washed twice in RPMI-1640 with 10% fetal calf serum. Splenocytes were seeded at a cell density of 1 × 10^6^ cells/ well in a 96-well culture plate. The cells were stimulated with 5 μg/ml concanavalin A (ConA) (Sigma-Aldrich) in the presence or absence of GluOC or GlaOC. For mitogen-induced lymphocyte proliferation assays, we used Cell Count Reagent SF (Nacalai Tesque) was used. The culture supernatants of lymphocytes were collected at 15, 45, 65 h after stimulation to measure cytokine concentrations by ELISAs. Enzyme-immuno assay (EIA) and enzyme-linked immunosorbent assays (ELISAs) ------------------------------------------------------------------------- To measure IFN-γ and IL-6 concentrations in mouse serum or splenocyte culture supernatants, anti-IFN-γ antibody (Affymetrix, San Diego, CA, USA) or anti-IL-6 antibodies (Affymetrix) were pre-coated onto 96-well plates (Nunc-Immuno Module; Nunc A/S, Roskilde, Denmark) at 4°C overnight. After blocking with ELISA Diluent Solution (eBioscience, San Diego, CA, USA) for 1 h, samples were incubated in the wells at room temperature for 1 h. After washing three times, biotin-anti-mouse IFN-γ (Biolegend, San Diego, CA, USA) or biotin-anti-mouse IL-6 (Biolegend) antibodies were added, followed by incubation at room temperature for 1 h. HRP-streptavidin and tetramethylbenzidine substrate were applied each for 1 h each. Then, absorbance was measured using a microplate reader at 450 nm (655 nm reference). For measurement of serum GluOC concentrations, we used a mouse Glu-osteocalcin high sensitive EIA kit (Takara Bio, Shiga, Japan TAKARA BIO INC. TAKARA BIO INC. TAKARA BIO INC.) was used. Immunoblot analysis ------------------- B16 cells were lysed in 20 mM phosphate buffer, containing 5 mM EDTA, 1% Triton X-100, pepstatin A (2.5 μg/ml), leupeptin (5 μg/ml), 4-(2-aminoethyl) benzenesulfonyl fluoride hydrochloride (25 μg/ml), and aprotinin (1.7 μg/ml). The lysates were centrifuged at 12,000 ×*g* for 30 min at 4 °C. The protein concentration of the resulting supernatants was determined with a Protein Assay Rapid Kit (Wako, Osaka, Japan). Protein samples (10 μg) were fractionated by polyacrylamide gel electrophoresis on 8-15% polyacrylamide gels containing 0.1% sodium dodecyl sulfate. The separated proteins were transferred to a polyvinylidene difluoride membrane (Merck-Millipore, Darmstadt, Germany), that was then exposed to Blocking One (Nacalai Tesque) at 4°C overnight. After incubation with anti-cleaved Caspase-3 (Cell Signaling Technology, Danvers, MA, USA) or anti-β-actin (Sigma-Aldrich) antibodies at 4 °C overnight, immune complexes were detected with HRP-conjugated secondary antibodies and a chemiluminescence substrate kit (GE Healthcare, Buckinghamshire, England). Fluorimetric caspase 3/7 assay ------------------------------ An Amplite™ Fluorimetric Caspase 3/7 assay kit (AAT Bioquest^®^, Sunnyvale, CA, USA) was used to determine caspase 3/7 proteolytic activities according to the manufacturer\'s protocol. Precultured B16 cells for 24 h in 100 μl medium/well were treated with the vehicle, GluOC (10 ng/ml), or GlaOC (10 ng/ml) for 6 h in 96-well culture plates. Caspase assay solution (100 µl/well) was then applied at room temperature for 1 h in the presence or absence of Ac-DEVD-CHO, a caspase 3/7 inhibitor. For measurement, the fluorescence increase at 350/450 nm (Ex/Em) was monitored. Statistical analysis -------------------- The student\'s *t*-test or Dunnett\'s test were performed for statistical analysis as appropriate. *P* values of less than 0.05 were considered to be statistically significant. Results are expressed as the mean ± standard error of the mean (SEM). Results and Discussion ====================== Antitumor effects of GluOC *in vivo* ------------------------------------ We first investigated the effect of GluOC on B16 melanoma isografts *in vivo* using C57Bl/6N wild-type mice (Fig. [1](#F1){ref-type="fig"}A). As a result, B16 tumor growth was significantly attenuated by GluOC administration (7.5 μg/mouse/day) (Fig. [1](#F1){ref-type="fig"}B-C). GluOC administrated by a micro-osmotic pump was detected in the tumor mass and tumor-bearing mouse serum in a dose-dependent manner (Fig. [1](#F1){ref-type="fig"}D-E). These results indicated that GluOC administrated through a subcutaneous micro-osmotic pump was hematogenously delivered to the isografts to directly exert antitumor effects on B16 melanoma cells. We next examined serum IFN-γ levels in serum and tumors and T cell population in tumors, because recent studies focusing on biological therapy of malignant melanoma have shown the effectiveness of IFN-γ in cellular immunity, especially CTL-mediated immunity, as an adjuvant treatment [@B3], [@B15]-[@B25]. In addition, we observed that the growth of B16 melanoma cells was inhibited with recombinant IFN-γ *in vitro* as the previous study demonstrated (data not shown) [@B3]. As a result, GluOC administration at 7.5 μg/mouse/day significantly increased serum IFN-γ levels both in serum and tumor mass compared with saline administration (Fig. [1](#F1){ref-type="fig"}F-G), and CD3-popsitive T cell population was greater in GluOC-treated mouse tumors than in control ones (Fig. [1](#F1){ref-type="fig"}H). IFN-γ, a type II interferon and pleiotropic cytokine that has diverse biological functions [@B26], is secreted by CTLs and activated natural killer (NK) cells to increase antitumor activity by enhancing antigen presentation and promoting the proliferation, expansion and survival of CD8^+^T cells [@B27], [@B28]. It also binds to cognate receptors at the cell surface and activates the JAK (Janus kinase)-STAT (signal transducer and activator of transcription) pathway [@B29]. We observed that Stat1 was activated in response to recombinant IFN-γ*in vitro*(data not shown), which satisfied with the previous report [@B30]. Take these results into consideration, JAK-STAT pathway was potentially activated in B16 melanoma cells treated with GluOC. We also found that similar GluOC administration using athymic Balb/c nu/nu mice did not show an antitumor activity against human prostate cancer cells *in vivo* and serum IFN-γ concentrations in those mice were not increased (data not shown), even though GluOC has significantly suppressed the growth of the same cancer cells by reducing phosphorylation of RTKs *in vitro* [@B10]. Considering these data, we speculated that cell-mediated immunity, especially T cell-mediated production of IFN-γ, is essential for GluOC to show an antitumor activity*in vivo*. Effects of GluOC on ConA-specific lymphocyte proliferation and cytokine production ---------------------------------------------------------------------------------- We next tested the effects of GluOC on ConA-stimulated lymphocyte proliferative (blastogenesis) activation and cytokine production. Splenocytes from mice, which were administrated with GluOC (7.5 μg/mouse/day) via a subcutaneous micro-osmotic pump for 3 weeks, showed an increased proliferative response at 15, 45, and 65 h after ConA stimulation compared with those from saline-treated mice (Fig. [2](#F2){ref-type="fig"}A). In addition, splenocytes of GluOC-treated mice produced significantly higher amounts of IFN-γ following ConA stimulation compared with the control (Fig. [2](#F2){ref-type="fig"}B), while the amount of the interleukin (IL)-6, another proinflammatory cytokine [@B31], was unaffected by GluOC treatment (Fig. [2](#F2){ref-type="fig"}C). Next, we examined whether these results were reflected by the direct effects of GluOC on splenocytes. Splenocytes from C57Bl/6N mice were treated with ConA in the presence or absence of GluOC at 0, 12.5, 25, 50 ng/ml *in vitro*. ConA-stimulated proliferative response of T cells (Fig. [2](#F2){ref-type="fig"}D) and IFN-γ production (Fig. [2](#F2){ref-type="fig"}F) were increased by *in vitro* administration of GluOC at 65 h after stimulation, as well as *in vivo* GluOC administration. In addition, the level of IL-6 in the supernatant was unaltered (Fig. [2](#F2){ref-type="fig"}E). Both IFN-γ and IL-6 are regarded as proinflammatory cytokines that regulate immune responses, cell proliferation, and tumor development and progression [@B3], [@B31], [@B32]. However, these cytokines frequently have functionally different roles. IFN-γ is a major Th1 cytokine associated with the cellular response [@B3], [@B32], while IL-6 is a Th2 cytokines associated with the humoral response to extracellular pathogens [@B31]. Although it is currently unknown whether GluOC contributes to the predominance of the Th1 response in cancer, there is the possibility that GluOC might regulate the Th1/Th2 paradigm in several pathophysiological processes including cancer. It has been reported that GPRC6A, a receptor for GluOC, is highly expressed not only in organs where it is involved in glycolipid metabolism [@B33]-[@B35], but also in the spleen [@B34], [@B35]. Although its function in the spleen is unknown, recent reports suggested that the receptor is associated with immune responses [@B36], [@B37]. Our immunological experiments also showed the expression of GPRC6A in mouse spleen and splenocytes (data not shown), that include a variety of immune cell populations. It could not be concluded without further detailed analysis, but there is a possibility that GluOC increased IFN-γ secretion via GPRC6A expression in certain splenic immune cells. Because IFN-γ is produced by CTLs and NK cells [@B26]-[@B29], we examined whether NK cells were needed for GluOC to perform immunopotentiative functions under ConA stimulation. GluOC did not increase lymphocyte proliferation or IFN-γ production of NK cell-depleted splenocytes treated with ConA (Fig. [S1](#SM1){ref-type="supplementary-material"}A, B). This result indicates that not only T cell but also NK cells are essential for the strong GluOC-mediated immunostimulation. In addition, such immunostimulating effects were specific to GluOC, because no changes occurred in ConA-stimulated splenocytes treated with GlaOC (Fig. [S2](#SM1){ref-type="supplementary-material"}A, B). These results indicate that GluOC suppresses cancer cell growth*in vivo* by upregulating the anticancer effect of cellular immunity, especially increased secretion of IFN-γ. Direct effects of GluOC and GlaOC on B16 cell viability *in vitro* ------------------------------------------------------------------ We have reported that GluOC directly suppresses, while GlaOC promotes, human prostate cancer cell growth*in vitro*[@B10]. We therefore investigated the effect of each OC type on B16 cell viability *in vitro*. WST-8 and BrdU uptake assays showed that GluOC significantly suppressed B16 cell viability (Fig. [3](#F3){ref-type="fig"}A). In contrast, GlaOC increased B16 cell viability in the WST-8 assay, although there were no significant differences in the BrdU uptake assay (Fig. [3](#F3){ref-type="fig"}B). These results indicate a functional difference between GlaOC and GluOC in B16 mouse melanoma cell growth, and that GluOC or GlaOC has suppressing or promoting effects on these cells, respectively, which is consistent with our previous report [@B10]. We next subjected B16 mouse melanoma cells to phospho-RTK arrays. RTKs are a family of cell surface receptors that mediate key signaling pathways involved in cell proliferation [@B38]. As shown in Fig. [3](#F3){ref-type="fig"}C, GluOC reduced phosphorylation levels of many kinds of RTKs including epidermal growth factor receptor (EGFR), ErbB2, ErbB3, ErbB4, fibroblast growth factor receptor (FGFR) 3, FGFR4, InR (insulin-like receptor), insulin-like growth factor 1 receptor, Axl, platelet-derived growth factor receptor (PDGFR) α, PDGFRβ, Flt-3 (fms-like tyrosine kinase), c-Ret, tyrosine kinase with immunoglobulin-like and EGF-like domains (Tie)-1, Tie-2, tropomyosin receptor kinase C (TrkC), and vascular endothelial growth factor receptor 3 (VEGFR3). These results indicate that GluOC directly suppresses melanoma cell growth by reducing RTK activation as shown in human prostate cancer cells [@B10]. In contrast, GlaOC enhanced the phosphorylation levels of many kinds of RTKs, particularly EGFR, ErbB2, ErbB3, ErbB4, macrophage-stimulating protein receptor, PDGFRα, PDGFRβ, TrkC, and VEGFR3. These nine RTKs are growth factors belonging to the families of EGFR [@B39], MET proto-oncogene [@B40], PDGFR [@B41], [@B42], Trk [@B43], and VEGFR [@B44], which are closely related to tumor progression of melanoma [@B38], [@B43], [@B45]-[@B51]. These data were consistent with the results of proliferation assays, although whether the phosphorylation is a direct effect or a secondary effect has not been established. A multi-kinase inhibitor, which inhibits many growth factors, has been greatly focused on in clinical cancer therapy [@B45], and some RTK multi-inhibitors have already been used in clinical pathology [@B46]-[@B49], [@B52]-[@B54]. However, they also have the risk of side effects by downregulating kinase activity required in normal physiological functions. Considered that administration of GluOC did not result in any observable toxic effects on normal tissues or cells of the treated mice *in vivo* or *in vitro* [@B10], GluOC might be a promising multi-targeted inhibitor of RTKs in cancer therapy. We also analyzed another molecular mechanism of the inhibitory effect on cell growth by GluOC. In a caspase-3/7 fluorescence assay, GluOC, but not GlaOC, elevated the proteolytic activity of caspase-3 and -7, which was inhibited by Ac-DEVD-CHO, an inhibitor of caspase-3 and -7 (Fig. [3](#F3){ref-type="fig"}D). This result was also supported by immunoblot analysis in which cleaved caspase-3 was visualized in GluOC-treated B16 cell lysates in a dose-dependent manner, but not in GlaOC-treated B16 cell lysates (Fig. [3](#F3){ref-type="fig"}E). These results indicate that GluOC suppresses B16 cell growth *in vitro*, which directly involves downregulating phosphorylation of multiple RTKs as well as promotion of apoptosis. Conclusion ========== In summary, we propose a new therapeutic strategy using GluOC, which is not only a multi-kinase inhibitor and apoptosis-inducer, but also an adjuvant in tumor immunotherapy. Although further studies are needed to elucidate the potential molecular mechanism of GluOC in cancer progression, studies are now in progress to evaluate the applied dosage and an effective method for clinical application. Supplementary Material {#SM0} ====================== ###### Supplementary figures. ###### Click here for additional data file. This work was supported by the Japan Society for the Promotion of Science (KAKENHI grants 24229009 to M. Hirata, 26861554 and 16K11496 to T. Kawakubo-Yasukochi, and 26861553 and 16K20421 to A. Mizokami), the Central Research Institute of Fukuoka University (157103 to T. Kawakubo-Yasukochi. and M. Hazekawa), and Fukuoka Foundation for Sound Health Cancer Research Fund. The authors acknowledge useful discussion with Prof. M. Nakashima (Fukuoka University). Y. Hayashi was a recipient of The Iwadare Scholarship Foundation and Morita Scholarship Foundation. {#F1} {#F2} ###### GluOC, but not GlaOC, suppresses B16 cell growth *in vitro*. B16 cells were incubated with GluOC (A) or GlaOC (B) for 24 h, followed by WST-8 (left panels) and BrdU uptake (right panels) assays. Mean data are expressed as the ratio to the control. Each experiment was repeated three times. Data represent the mean ± SEM. \**P*\< 0.05 and \*\**P*\< 0.01 versus the control. (C) Phospho-RTK array. After B16 cells were treated with the vehicle, GluOC, or GlaOC for 6 h, the cell lysate was applied to the array. Each pair of kinase dots that increased (red) or decreased (blue) compared with the controls is enclosed by a square. Quantitation of the dot densities of phospho-RTKs was performed using scanned images and ImageQuant LAS 4000 software (GE Healthcare). (D) Fluorimetric assay of caspase-3 and -7 activities detected in B16 cells treated with the vehicle, GluOC, or GlaOC for 6 h. Fluorescence was monitored in the presence or absence of DEVD-CHO. Each experiment was repeated three times. (E) Immunoblot analysis of cleaved caspase-3 after stimulation with GluOC or GlaOC for 6 h in B16 cells. β-actin was used as an internal control. Quantitation of cleaved caspase-3 expression (normalized by the amount of β-actin) in an immunoblot is shown in the bottom panel. Each experiment was repeated three times. Data represent the mean ± SEM. \**P*\< 0.05 and \*\*\**P*\< 0.001 versus the control.    [^1]: Competing Interests: The authors have declared that no competing interest exists.

{#sp1 .320}

Introduction {#S0001} ============ Carcinomas of unknown primary (CUPs) are a heterogeneous group of cancers for which the metastases are clinically and histologically confirmed, but the primary tumor site remains occult after comprehensive workup.[@CIT0001] Patients with CUPs mainly received empirical chemotherapy (paclitaxel/platinum was the most commonly used regimen), and the median overall survival (OS) was within 1 year.[@CIT0002],[@CIT0003] In the era of precision medicine, the discovery of driver genes and the development of target agents enormously improve the survival and quality of life of patients with cancer. Although most malignant tumors are still treated according to the pathological classification of primary tumors, a growing number of studies have revealed that different malignant tumors with the same driver genes can be treated with the same targeted agents with good therapeutic effects.[@CIT0004] Therefore, the target treatment pattern according to driver genes but not primary tumor site may greatly improve the diagnosis and treatment of CUPs. In this study, the therapeutic efficiency of crizotinib was examined in a patient with CUP harboring mesenchymal-epithelial transition factor (MET) gene amplification and neurotrophic tyrosine receptor kinase 1 (NTRK1) gene co-occurring mutation and a history of long-term hypertension and renal cysts. The treatment strategy of CUPs based on molecular diagnosis, therapeutic effect of crizotinib, and management of related adverse events (AEs) was also discussed. Case report {#S0002} =========== A 67-year-old male patient had a history of smoking with a smoking index of 800. He had quit smoking for more than 1 year. He had a long-term history of hypertension and cerebral infarction. He regularly took his antihypertensive drugs, and his blood pressure was controlled. In January 2017, the patient accidentally found a mass in the right neck, accompanied by swelling and discomfort in the right upper extremity, which was not taken seriously at the beginning. Later, masses occurred in the right axillary fossa in succession. The patient was treated in a local hospital in April 2017. Enhanced neck and thoracic computed tomography (CT) showed 1) multiple enlarged lymph nodes (the maximum was 3.2×2.4 cm) in the right neck, right supraclavicular region, right axillary fossa, and right tracheal and esophageal sulcus and 2) nodule in the posterior segment of right upper lung (0.63×0.78 cm). Pathological puncture of the right supraclavicular lymph node showed metastatic, poorly differentiated adenocarcinoma. Positron emission tomography (PET)/CT did not find a clear primary tumor, and the clinical diagnosis was "metastatic poorly differentiated adenocarcinoma with unknown primary site." The patient then received 6 cycles of TP (taxanes/cisplatin) regimen in the hospital. Partial remission (PR) of target lesions was assessed after 3 cycles of chemotherapy, and no significant change in the nodules in the right upper lung was seen. One month after the last chemotherapy, the lymph nodes in the right neck and right axillary region showed a progressive increase, indicating progression of the disease. In October 2017, the patient was transferred to our department for treatment. The Eastern Cooperative Oncology Group (ECOG) score was 2 points. The patient was fatigued and had poor appetite. He lost 5 kg of weight. Physical examination revealed signs of anemia and multiple enlarged lymph nodes in the right neck and right axilla, with the largest one fused into a hard, painless mass, about 3 cm in diameter, with poor mobility. No obvious positive signs were found in other systems. PET/CT showed 1) multiple nodular soft tissue density shadows in the right neck, right upper and lower clavicular regions, right axilla and mediastinum, besides increased fluorodeoxyglucose (FDG) metabolism (SUVmax =10.8), which was considered as metastatic lymph nodes ([Figure 1A](#F0001){ref-type="fig"}); 2) nodular shadows in the posterior segment of the right upper lung, with not-so-high FDG metabolism ([Figure 1B](#F0001){ref-type="fig"}); and 3) multiple cerebral infarctions, splenomegaly, and multiple cysts in the left kidney. Laboratory tests indicated normal liver and kidney function; normal white blood cell, neutrophil, and platelet count; hemoglobin 87 g/L; and serum CEA 41.35 ng/mL (normal, \<5 ng/mL). Clinical diagnosis indicated 1) metastatic adenocarcinoma in the right neck, right clavicular region, right axillary fossa, and mediastinal lymph nodes with unknown primary site (stage IV); 2) anemia (moderate); 3) hypertension of grade 2 (very high-risk group); 4) multiple cerebral infarctions; and 5) multiple cysts in the left kidney. The patient was poorly tolerant to further chemotherapy and had a willingness to seek targeted therapy. Thus, a second biopsy of the lymph nodes in the right neck was performed to obtain real-time pathology of the tumor tissue. Moreover, peripheral blood and freshly punctured tumor tissues were taken to perform the next-generation sequencing (NGS)-based multigene panel analysis (Burning Rock Biotech, Guangzhou, China) to reveal possible drug targets. The results of histopathological and immunohistochemical (IHC) diagnoses were consistent with those of metastatic poorly differentiated adenocarcinoma ([Figure 1C](#F0001){ref-type="fig"}--[E](#F0001){ref-type="fig"}): CK7 (2+), TTF-1 (--), CK5/6 (--), and p63 (--). The targeted NGS analysis of neck metastatic carcinoma revealed EGFR mutation (--), ALK fusion (--), ROS1 fusion (--), BRAF mutation (--), RAS mutation (--), MET gene amplification (9.12 times) ([Figure 2A](#F0002){ref-type="fig"}), NTRK1 gene c.339-2A \> C mutation (+) \[at a mutant allele frequency (MAF) of 16.17%\] ([Figure 2B](#F0002){ref-type="fig"}), Cyclin D1 (CCND1) gene amplification (4.75 times) ([Figure 2A](#F0002){ref-type="fig"}), and TP53 gene R249S mutation (at a MAF of 11.54%). Tumor mutation burden was moderate. The tumor was microsatellite stable. However, targeted NGS of ctDNA only identified NTRK1 (at a MAF of 0.42%) ([Figure 2B](#F0002){ref-type="fig"}) and TP53 gene mutation (at a MAF of 0.39%). IHC was performed against MET to verify the findings of the gene test, and the results indicated that the protein staining was positive and the expression level was high (2+ to 3+) ([Figure 2C](#F0002){ref-type="fig"},[D](#F0002){ref-type="fig"}). Anti-MET-amplification therapy with crizotinib (250 mg, twice daily) was recommended in November 2017 on the basis of the molecular findings, accessible drugs in China, and treatment willingness of the patient. After 1 week of medication, the patient's clinical symptoms ameliorated significantly and promptly. Scans demonstrated a complete response (CR) after 1 month of treatment ([Figure 3](#F0003){ref-type="fig"}), and the tumor markers returned to normal levels. During the administration of crizotinib, the main AEs were grade II creatinine elevation, grade I localized edema, and a new cyst in the right kidney ([Figure 4](#F0004){ref-type="fig"}). Considering the decrease in creatinine clearance and long-term history of chronic hypertension, the dose was reduced by half (250 mg, once daily) for maintenance after 2 months of medication, and treatments for protecting the kidneys were combined. Tumor was excellently controlled until the patient died of tumor progression in September 2018, with a progression-free survival (PFS) and OS time of 8.5 months and 10.0 months, respectively, from the initiation of crizotinib. The treatment using half-dose crizotinib was well tolerated, with a good quality of life and without deterioration in renal complications.Figure 1Imaging and histopathological characteristics of the patient before crizotinib treatment.(**A** and **B**) PET/CT scans showed increased FDG uptake in multiple lymphatic metastases in the whole body and no FDG uptake in the small nodule located in the posterior segment of the right upper lung. (**C**) H&E staining confirmed the right neck lymph nodes as poorly differentiated metastatic carcinomas. (**D** and **E**) Immunohistochemical staining of the right neck lymph nodes showed positive staining of CK7 (2+) and negative staining of TTF-1 (×400).**Note:** The red arrow in figure B indicates the small nodule located in the right upper lung.**Abbrevations:** CK7, cytokeratin 7; FDG, fluorodeoxyglucose; H&E, hematoxylin & eosin; PET/CT, positron emission tomography/computed tomography; TTF-1, thyroid transcription factor-1.Figure 2Molecular analysis of the patient before crizotinib treatment.(**A**) NGS analysis based on tissue samples showed MET and CCND1 copy-number amplification. (**B**) NGS analysis based on plasma samples showed NTRK1c.339--2 A\>C mutation. (**C** and **D**) IHC staining of tissue samples showed positive staining of MET protein (2 to 3+, ×100, and 400, respectively) in metastatic lesions.**Abbrevations:** CCND1, cyclin D1; GATA3, GATA binding protein 3; HIST2H3D, histone cluster 2 H3D; IHC, immunohistochemistry; MCL1, myeloid cell leukemia 1; MET, mesenchymal-epithelial transition factor; NGS, next-generation sequencing; N.S., no significance; NTRK1, neurotrophic tyrosine receptor kinase 1; PI3KCG, phosphatidylinositol 3-kinase catalytic subunit gamma.Figure 3Chest CT scans showing dynamic changes in metastatic right axillary lymph nodes (**A**) before, (**B**) 1 month after, (**C**) 4 months after, and (**D**) 8.5 months after crizotinib treatment.**Note:** The red arrows in all figure parts indicate the metastatic lesions located in the right axillary fossa.**Abbrevation:** CT, computed tomography.Figure 4Abdominal CT scans showing dynamic changes in renal cysts (**A**) before, (**B**) 1 month after, and (**C**) 8.5 months after crizotinib treatment.**Note:** The red arrows in all figure parts indicate the cysts located in both kidneys.**Abbrevation:** CT, computed tomography. Discussion {#S0003} ========== Exploration of precise treatment mode for CUPs {#S0003-S2001} ---------------------------------------------- As one of the characteristics of malignant tumors, genomic instability and mutation were important links in the occurrence and development of tumors.[@CIT0005] The rapid development of molecular biology and translational medicine led to the discovery of different tumor-driven genes using high-throughput NGS to assist the judgment of primary tumors and guidance of treatment.[@CIT0006]--[@CIT0008] It may bring new enlightenment to the treatment of CUPs because if this treatment mode is verified, the CUPs may be the optimal indication for targeted therapy based on genetic information regardless of tumor location. For example, Ross et al found that gene mutation was most commonly observed in adenocarcinoma among the 200 samples of CUPs, and 72% of adenocarcinoma with an unknown primary site had mutations of RTK and Ras pathways.[@CIT0009] Gatalica et al also obtained similar conclusions by analyzing 1,086 CUPs and concluded that the expression profiling analysis of tumor genes could provide more treatment options for patients with CUPs.[@CIT0010] The present case was of an elderly male with a metastatic adenocarcinoma of unknown primary. He received the empirical chemotherapeutic TP regimen and obtained a first-line PFS of less than 6 months. His quality of life drastically deteriorated, although the target lesion was temporarily and partially relieved. After the patient was admitted to our department, we confirmed his diagnosis and revealed a MET gene amplification of the patient as evidenced by both NGS and IHC analyses of the tumor tissues. The patient gained his first CR of the target lesions after 1 month of treatment with anti-MET-amplification drug crizotinib, and the CR was maintained for another 6.5 months in the case of subsequent medication with half the dose. The good effect of crizotinib in treating this patient was speculated to have come from precise gene profiling and effective drug use, but the possibility that it might be related to the unique biological behavior of the CUPs themselves cannot be ruled out. In the meantime, it was unknown whether other composite mutations in this patient also influenced this process because tumors harboring NTRK1 fusion, such as MPRIP--NTRK1 and LMNA--NTRK1, might be sensitive to crizotinib treatment.[@CIT0011],[@CIT0012] Indeed, c.339-2A\>C variable splicing mutation on the sixth exon of NTRK1, which was located in the leucine-rich repeat region, was identified in this patient. Because this kind of variation was not reported in the COSMIC and cBioPortal databases, the significance was unknown for this novel case report, and hence further research is warranted in the future. Treatment of crizotinib-related renal AEs {#S0003-S2002} ----------------------------------------- According to the instructions for crizotinib, 4% of patients may have complicated renal cysts during medication,[@CIT0013] which has now increased to more than 22%.[@CIT0014],[@CIT0015] Related studies found that the use of crizotinib was associated with the growth of existing renal cysts and the development of new renal cysts.[@CIT0016] Renal cysts generally occur in the early stage of crizotinib treatment. The older the age, the higher the incidence, which can be complicated (Bosniak class III or higher) or be similar to malignancy. At the same time, previous studies also found that patients with significant changes in renal cysts usually received more antitumor treatment before crizotinib was used and also received crizotinib for a longer duration. The current consensus was that the occurrence of renal cysts after crizotinib treatment might be a benign process that could be spontaneously resolved by discontinuing the drug, and that it would not hinder the use of crizotinib.[@CIT0014],[@CIT0015],[@CIT0017] Although the mechanism of action of crizotinib in the formation of renal cysts was not fully understood, studies suggested that it might be associated with the inhibition of c-MET and activation of the NF-κB pathway, leading to renal injury and fibrosis.[@CIT0018] The present case was of an elderly patient with underlying hypertension. Unlike previous crizotinib-associated renal cysts in clinical trials, this patient not only experienced an increase in the number of left renal cysts but also had a new cyst in his right kidney, along with grade II creatinine elevation and grade I localized edema. The emergence and management of renal-related AEs in this patient would provide new insights into the use of crizotinib. Recent studies on crizotinib-related renal cysts were based mostly on retrospective studies of clinical trials for crizotinib, and in such population, patients often had no history of hypertension or kidney disease. Due to the strong renal compensatory capacity, the single population may mask the damage of renal cysts to renal function, resulting in research bias. The medical history and imaging data of this patient revealed that long-term chronic hypertension caused the renal cortex to become thinner, which was consistent with the imaging findings of hypertensive contracted kidney. Crizotinib might aggravate renal inadequacy that already existed and was unrecognized in the patient, causing deterioration of renal function, thereby eliciting the elevation of creatinine and occurrence of edema. Therefore, it is necessary for the clinicians to be aware of the potential complication of renal cysts caused by the use of crizotinib. Especially for patients with a history of long-term hypertension or kidney disease, regular monitoring of renal function, and reexamination of renal CT should be performed to avoid more severe renal impairments. Conclusion {#S0004} ========== In summary, this study reported the clinical course of an elderly man with a metastatic adenocarcinoma. MET amplification and unclear NTRK1 mutations were identified using targeted NGS analysis upon the progression of empirical chemotherapy. The treatment with half the standard dose of crizotinib resulted in not only a satisfactory CR but also an effective management of drug-related renal AEs, in this CUP patient harboring MET amplification and NTRK1 co-occurring mutation. Therefore, gene profiling may contribute to the precise diagnosis and treatment of CUPs. The clinical management of targeted drugs is critically important in promoting the precision medicine, especially in drug-related AEs. The authors sincerely thank the patient for his contribution to the publication of this case report. This work was supported by the Excellent Young Talents Fund Program of Higher Education Institutions of Anhui Province (No. gxfx2017066) and an internal grant from the First Affiliated Hospital of Bengbu Medical College (No. Byyfykj201801). Ethics approval and informed consent {#S0005} ==================================== This study was approved by the ethics committee of the First Affiliated Hospital of Bengbu Medical College, and written informed consent to participate was obtained from the family member of the patient. Consent for publication {#S0006} ======================= Written informed consent was obtained from the family member of the patient for the publication of this case report and any accompanying images. Abbreviations {#S0007} ============= AE, adverse event; CCND1, cyclin D1; CR, complete response; CT, computed tomography; CUP, carcinoma of unknown primary; ECOG, Eastern Cooperative Oncology Group; FDG, fluorodeoxyglucose; IHC, immunohistochemistry; MAF, mutant allele frequency; MET, mesenchymal-epithelial transition factor; NGS, next-generation sequencing; NTRK1, neurotrophic tyrosine receptor kinase 1; OS, overall survival; PET, positron emission tomography; PFS, progression-free survival; PR, partial remission; TP, taxanes/cisplatin. Disclosure {#S0008} ========== The authors report no conflicts of interest in this work.

{#sp1 .317} {#sp2 .318}

Background {#Sec1} ========== Asthma is a chronic global health problem having a significantly detrimental effect on quality of life and is often associated with significant healthcare utilisation costs. It is estimated that asthma affects 30 million people in Europe and up to 300 million people worldwide, with prevalence expected to rise \[[@CR1], [@CR2]\]. In Germany, asthma has been estimated to affect approximately 7 % of the population \[[@CR3]\]. Most asthma cases are due to allergic conditions with previous research showing that over 90 % of all allergic asthma patients display symptoms that correlate with allergic rhinitis \[[@CR4]\]. Although the disease does not exhibit high mortality rates, the loss in quality of life is significant, with both physical and psychological dimensions found to be affected \[[@CR5], [@CR6]\]. It is also associated with substantial costs to the healthcare system and wider society. One European-wide cost-of-illness study found that the mean annual costs of asthma in 2010, including both direct and indirect costs, were €509 and €2281 for controlled and uncontrolled asthma patients respectively \[[@CR7]\]. The total asthma burden in Germany was estimated to be €3.3 billion in 2008 \[[@CR3]\]. The SQ^®^ HDM SLIT-tablet (ACARIZAX^®^, ALK, Hørsholm, Denmark) is a recently developed allergy immunotherapy (AIT) tablet, targeted specifically at house dust mite (HDM) allergens. It is a sublingual treatment option for HDM allergic asthma patients already taking pharmacotherapy whose symptoms are not well controlled, and is a 1:1 mixture of allergen extract from *Dermatophagoides pteronyssinus* and *Dermatophagoides farinae* \[[@CR8]\]. The objective of this analysis was to assess the cost-effectiveness of ACARIZAX from a German societal perspective. Methods {#Sec2} ======= Design {#Sec3} ------ A cost-utility analysis was undertaken based on the results of a double-blinded phase III randomised controlled trial. The MT-04 trial was conducted across 13 European countries, with the primary objective of comparing the efficacy of ACARIZAX with placebo in patients with HDM allergic asthma, as measured by a reduction in the risk of asthma exacerbation. Eligible patients were adults, sensitised to HDM and not well controlled by inhaled corticosteroids (ICS; equivalent to budesonide, 400--1200 μg) at inclusion. Furthermore, they could have multiple sensitizations but a relevant clinical history of perennial allergic asthma or rhinitis caused by other allergens to which the patients were regularly exposed led to exclusion as this may have caused symptoms to arise in the efficacy period that would interfere with the trial results \[[@CR9]\]. Patient characteristics are summarised in Table [1](#Tab1){ref-type="table"}. Two treatment groups (6 SQ-HDM and 12 SQ-HDM), and one control group (placebo) were included in the trial to investigate different therapy doses, with the 12 SQ-HDM group considered here. Patient diagnosis took place during the trial screening period. Following screening there was a 7--13 month treatment maintenance period, in which patients in both arms were the given their allocated treatment plus pharmacotherapy in the form of ICS and short-acting β2-agonist (SABA). Patients could also be given oral steroids to treat severe acute asthma symptoms or to restrict the deterioration of asthma symptoms. Following the treatment maintenance period, the daily ICS dose was reduced by 50 % for 3 months and then removed for patients who did not experience an asthma exacerbation. The treatment maintenance period was adopted for the analysis as it was judged to be better aligned with present clinical practice. A more comprehensive description of the trial design has been published previously \[[@CR9]\]. Based on the setup of MT-04, two treatment options have been incorporated into the cost-utility analysis: ACARIZAX plus pharmacotherapy (ACARIZAX henceforth) and placebo plus pharmacotherapy (pharmacotherapy henceforth).Table 1Summary of patient characteristics from MT-04PlaceboACARIZAXNo. of subjects277282Sex (%) Male151 (55 %)147 (52 %) Female126 (45 %)135 (48 %)Mean age (SD)33.0 (12.2)37 (11.6)Ethnic origin (%) Caucasian273 (99 %)277 (98 %) Other4 (1 %)5 (2 %)Mean years with HDM AA (SD)13.3 (10.6)12.9 (11.5)Control level at randomisation (%)^a^Controlled0 (0 %)0 (0 %) Partly controlled200 (72 %)200 (71 %) Uncontrolled77 (28 %)82 (29 %)*AA* allergic asthma, *HDM* house dust mite, *SD* standard deviation^a^Classification system based on GINA, Masoli et al. \[[@CR2]\] Healthcare resource use {#Sec4} ----------------------- Within MT-04, patients recorded medication use using electronic diaries during the last 4 weeks of the treatment maintenance period. Physician and emergency room visits were also recorded by trial investigators at each visit. In the analysis, all resources recorded within MT-04 were combined with relevant unit costs from a German perspective, to estimate mean patient costs over a one year time horizon. The cost of ACARIZAX was also included, with a patient requiring one tablet per day. ACARIZAX is suitable for home treatment, although the first tablet should be taken under the surveillance of a physician. This additional visit was also incorporated into the model. Healthcare resource use values implemented in the analysis are summarised in Table [2](#Tab2){ref-type="table"}.Table 2Summary of cost and resource use inputs incorporated in the analysisResourceUnit priceAnnual resource useTotal costACARIZAXPharmacotherapyACARIZAXPharmacotherapyACARIZAX tablet^a^€2.533650€923€0GP visits^b^€29.350.1750.105€5.13€3.07Emergency room visits^b^€74.960.0100.025€0.75€1.89ICS daily dose (μg)^c^€18.14563555€373€363SABA intake (doses)^c^€22.15266297€9.82€10.96Three drugs and two other medical resources were included as parameters in the analysis. Resource use was based on data recorded in MT-04 and, therefore, they relate specifically to allergic asthma patients. The values have been multiplied by the unit price of each resource to generate total costs. These costs were applied to a one year period, and applied equally across all years in the analysis (with costs also discounted at 3 % per year in line with German guidelines)^a^Source: <http://www2.lauer-fischer.de/>^b^Source: <http://www.kbv.de/html/>^c^Source: <https://www.gkv-spitzenverband.de> The analysis was also run with sick days considered, to capture the impact of indirect costs. Within MT-04, the impact of asthma on productivity was captured via the administration of the work productivity and activity impairment (WPAI) questionnaire. The WPAI is a well-validated instrument that measures absenteeism, presenteeism and impairments in unpaid activity over the previous seven day period \[[@CR10]\]. Patient quality of life {#Sec5} ----------------------- The impact of allergic asthma on patient health-related quality of life (HRQoL) was captured. Patient reported health outcomes were used to elicit utility scores; which are the valuing of health on a scale of 0--1, with 0 representing health states equivalent to death and 1 representing full health. Quality-adjusted life years (QALYs) combine utility values and time to determine HRQoL over time. A patient who experiences one year in full health, (i.e. with a utility value of 1), would gain 1 QALY. Similarly, a patient who experiences 2 years with health valued at 0.5 would gain 1 QALY \[[@CR11]\]. Utilities have been used to estimate QALYs for patients in both treatment groups. Utility values used in the model were taken from the end of the treatment maintenance period in MT-04 (i.e. before ICS reduction and removal). Within the trial the SF-36 health survey was used to measure patient utility. For values used within the analysis, the data was corrected for baseline to determine between group differences at the end of the treatment maintenance period. These utility scores are summarised in Table [3](#Tab3){ref-type="table"}.Table 3Summary of utility values applied to patients in the analysisUtility at baseline---all patientsChange from baselineUtility adopted in analysis*Placebo*ACARIZAX*p* value*Placebo*ACARIZAX0.7360.00590.03150.03180.7420.768Within MT-04 patient utility was measured using the SF-36 survey instrument. There was a statistically significant difference in utility change (i.e. a measurement of patient quality of life) from baseline to the end of the treatment maintenance period in MT-04 between ACARIZAX and placebo (*p* \< 0.05). These values were applied to the mean utility score at baseline for all patients, to estimate the values that were applied at baseline in the analysis. Overall, ACARIZAX patients had a greater quality of life compared to placebo patients, at the end of the treatment maintenance period Pharmacoeconomic analysis {#Sec6} ------------------------- To assess the long-term impact of ACARIZAX on the healthcare system and patient HRQoL, costs and QALYs from MT-04 have been extrapolated over a nine-year time horizon. For this extrapolation, costs that occurred during year one are applied equally across all years. To examine the impact of treatment on patient health, QALY scores were altered using an annual rate of change in utility (i.e. quality of life). There is evidence that AIT may have both a curative and preventative impact on respiratory allergies, equating to a long-term treatment effect \[[@CR12]--[@CR14]\]. Evidence specific to GRAZAX^®^ therapy, shows that the treatment has a disease-modifying effect, as patients receive clinical benefit for at least 5 years despite treatment only lasting 3 years \[[@CR15]\]. In the analysis it has been assumed that there will be a 5 % increase in utility for ACARIZAX patients during years two and three of treatment, based on the assumption patients will continue to receive a clinical benefit from treatment. Alternatively, for pharmacotherapy patients it is assumed that patient health remains stable, based on the assumption that the improvement that was observed during the trial for pharmacotherapy patients will remain throughout this period. Following the treatment period it is assumed that both patient groups remain stable for years four and five (i.e. 2 years after discontinuation of treatment), followed by a 5 % decline in health during years 6--9. Based on these assumptions, costs and QALYs have been estimated for both treatment groups over the nine-year time horizon and an assessment of the cost-effectiveness undertaken. Cost-effectiveness was defined using the incremental cost-effectiveness ratio (ICER) of ACARIZAX in addition to pharmacotherapy versus pharmacotherapy alone. To estimate the cost-effectiveness of an intervention to society as a whole, it is necessary to compare the ICER against a willingness-to-pay threshold. This allows the value of one QALY to society to be defined in monetary terms. In Germany, rationing decisions are made by Gemeinsamer Bundesausschuss (G-BA) on a case-by-case basis and, therefore, there is no defined threshold \[[@CR16]\]. However in the United Kingdom, the National Institute for Health and Care Excellence (NICE), a world leader in national health technology assessments, uses a threshold of £20,000 to £30,000 per QALY (€27,473 to €41,209) \[[@CR17]\]. For an intervention to be cost-effective, the ICER should fall within or below this threshold. For this analysis, a threshold value of €40,000 (\~£30,000) has been used. This means that compared with pharmacotherapy only, ACARIZAX should generate one QALY at a cost less than or equal to €40,000.$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\text{ICER}} = \frac{{Cost_{\text{Treatment}} - Cost_{\text{Comparator}} }}{{{\text{QALY}}_{\text{Treatment}} - {\text{QALY}}_{\text{Comparator}} }} = \frac{{\Delta {\text{Cost}}}}{{\Delta {\text{QALY}}}}$$\end{document}$$ Because a nine-year time horizon has been adopted, cost and QALY values were discounted using an annual discount rate of 3 %, in line with German guidelines \[[@CR18]\]. Uncertainty analysis {#Sec7} -------------------- To investigate the impact of changes in long term patient utility, two scenarios have been tested alongside the base case analysis. The rate of utility change for ACARIZAX and pharmacotherapy in the two scenarios are depicted graphically in Fig. [1](#Fig1){ref-type="fig"}. In short, within scenario one it is assumed that utilities (i.e. quality of life) remain stable during the treatment period (i.e. years 2--3), followed by a 5 % decline for all subsequent years, for both treatment groups. This is to test the impact of a decline in health as soon as treatment is stopped, an unfavourable scenario. In scenario two it is assumed that ACARIZAX patients have a small 5 % increase in utility for years two and three, followed by stability for all remaining years, whilst the utility for pharmacotherapy remains stable for the full time horizon. This scenario has been implemented to test the impact of a lasting disease-modifying effect for ACARIZAX that leads to long-term improvements in patient health.Fig. 1Graphical representation of change in patient utility over time, for each of the three scenarios assessed. Within all three scenarios, utility scores were based on results from the end of the treatment maintenance period in MT-04. For all remaining years, utility scores were altered via an annual rate of change, to see the impact long-term changes in patient outcomes had on the results. The rates of change in each scenario are given here in table In order to account for first-order uncertainty around the data used for input parameter values, one-way deterministic sensitivity has also been undertaken. This involves altering the value used for individual parameters within realistic ranges, to assess the impact on the model's results. Results {#Sec8} ======= The overall setup of the analysis and key results are presented in Fig. [2](#Fig2){ref-type="fig"}. The results of the base case analysis indicate that, over the nine year time horizon, ACARIZAX in addition to pharmacotherapy leads to a total of 6.16 QALYs per patient at a cost of €5658, compare with 5.50 QALYs at a cost of €2985 for pharmacotherapy alone. Therefore, ACARIZAX produced an extra 0.66 QALYs at an incremental cost of €2673, which equates to an ICER of €4041. This is substantially lower than the €40,000 threshold adopted for the analysis.Fig. 2Overview of the structure of the analysis. A cost utility analysis was undertaken to assess the impact of ACARIZAX on allergic asthma patients taking pharmacotherapy. Two treatment options were included; ACARIZAX plus pharmacotherapy, and placebo plus pharmacotherapy. A nine year time horizon was used, with ACARIZAX patients given treatment for 3 years. Over the nine year time horizon ACARIZAX patients accumulated 6.16 QALYs on average, at a mean cost of €5658, compared to an average of 5.50 QALYs at a mean cost of €2985 for pharmacotherapy only patients. This equates to an incremental cost-effectiveness ration (ICER) of €4041, substantially lower than the threshold value of €40,000 adopted here. This indicates that ACARIZAX is a cost-effective treatment option Indirect costs were also incorporated based on the administration of the WPAI during the final month of the MT-04 treatment maintenance period. Following this assessment it was found that asthma caused 0.8 and 1.6 % of work time to be missed for ACARIZAX and pharmacotherapy respectively. Based on an average of 1371 h worked per year in Germany \[[@CR19]\] and an average work day of 7.5 h, it was estimated that 1.51 days per year would be missed with ACARIZAX compared with 3.02 days with pharmacotherapy. Klussman and colleagues have previously estimated the cost per sick day in Germany is €93.69 \[[@CR20]\]. Therefore, the total annual indirect costs were estimated to be €141 for ACARIZAX and €283 for pharmacotherapy. When these costs were included the total per patient costs rose to €6760 for ACARIZAX and €5188 for pharmacotherapy respectively, with overall incremental costs reducing from €2673 to €1572. Within the two additional scenarios, the ICER remained below €40,000. For scenario one (i.e. the unfavourable scenario) the ICER increased from the base case value to €14,091, as the QALY gain reduced to 0.19. Alternatively, for scenario two (i.e. scenario testing the disease-modifying effect) the ICER was lower than that produced in the base case, being as it was €3832, and this was due to a QALY gain of 0.70. Within both scenarios, the inclusion of indirect costs had no impact on the direction of the results but did affect the magnitude, with the incremental costs reducing. The results of the sensitivity analyses indicate that the model's results are sensitive to changes in two key input parameters. If, during years two and three of treatment, ACARIZAX patients have a decline in health of 2 % or more, or if pharmacotherapy patients have an improvement in health of 5 % or more during years 6--9, then the ICER increases above the threshold value of €40,000. For all remaining parameters, the changes had no impact on the model's conclusions. Discussion {#Sec9} ========== The cost-utility analysis illustrates that, over a nine year time horizon, ACARIZAX plus pharmacotherapy is cost-effective compared with placebo plus pharmacotherapy, in HDM allergic asthma patients not well controlled by ICS and associated with mild to severe allergic rhinitis in a German setting. The deterministic sensitivity analyses show that there is a degree of uncertainty regarding the results of the analysis, given modest changes in two input parameters impacted on the cost-effectiveness of ACARIZAX (as measured by the ICER). If patient health declines at a rate of 2 % or higher during the 3 years of treatment with ACARIZAX, then it can no longer be considered a cost-effective treatment option at the given threshold value. However, evidence regarding AITs indicates that the efficacy of these treatments may be sustained for the duration of treatment, and beyond \[[@CR12]--[@CR15]\]. Therefore, it is unlikely that the health of ACARIZAX patients will decline during the treatment period. Similarly, if patient health with pharmacotherapy improves during years 6--9 then ACARIZAX is no longer cost effective. However, this improvement is unlikely given the applicable patient population, as they are patients whose symptoms are uncontrolled by pharmacotherapy. The analysis was conducted based on the results of a single RCT and there are certain limitations with this approach. In particular, healthcare utilisation is based on resource use data collected solely within MT-04, with costs specific to Germany applied to this data. This means the values used in the analysis may not be reflective of clinical practice, as the resource use within the trial was protocol driven, and certain resources required by allergic asthma patients may not have been captured as they were not recorded in the trial. For example, pharmacotherapy patients received more clinical supervision than can be expected in clinical practice. This may have reduced the total number of contacts with the healthcare system outside of the trial protocol (e.g. emergency room visits). Related to this, better overall supervision and the Hawthorne effect (i.e. patients modifying their behaviour as they were being monitored) may have led to a clinical gain that will not be found in practice, and patient health may have improved regardless of treatment as outcome measures naturally tended towards the population mean (i.e. regression to the mean). Despite the limitations of undertaking economic evaluations alongside a single RCT such an approach is becoming more common, with some funders now specifically requesting these evaluations alongside RCTs, as they allow for an early stage estimate of cost-effectiveness (i.e. before use has become widespread in clinical practice) \[[@CR22]\]. Furthermore, MT-04 was a large-scale trial conducted across several countries with multiple investigators. Therefore, the treatment effect of ACARIZAX should have been appropriately captured across a wide population group. The analysis covers a nine year time horizon with the assumption that the health (i.e. utility) of ACARIZAX patients remains stable in the 2 years post-treatment, which implies some form of disease-modifying effect. However, there are no long-term efficacy data to prove this effect, with the assumption based on the findings of other AITs. Therefore, to ensure that ACARIZAX does not gain an unfair advantage, it was also assumed that pharmacotherapy patients remained stable during this period and in the longer term that both ACARIZAX and pharmacotherapy patients have a decline in health during years 6--9. Furthermore, at the willingness-to-pay threshold of €40,000, ACARIZAX was cost-effective compared with pharmacotherapy at each of the 9 years included in the analysis indicating it is cost-effective even when long-term extrapolation of the data is not undertaken. In the future, further health economic evaluations could be undertaken should long-term, real world data become available (e.g. registry studies). This evaluation could estimate the true benefits of ACARIZAX, particularly if a disease-modifying effect has been proven in practice, and the impact of the placebo effect observed in MT-04 could also be assessed. If a disease-modifying effect can be proven in practice then this analysis may prove to be conservative estimation of the cost-effectiveness of ACARIZAX. AIT options, such as ACARIZAX, can be associated with adverse events that will impact on patient quality of life and overall treatment costs. Such events were not formally included in this analysis as the rate of serious adverse events was very low (less than 1 % for 12 SQ HDM patients) in MT-04 and because there was no clear difference in the rate of serious adverse events between ACARIZAX and placebo patients. It should be noted that the cost of asthma exacerbations were also not included in the analysis. However, ACARIZAX has a positive impact on exacerbations as shown via the quantification of number needed to treat (NNT) to avoid any moderate to severe exacerbation, which was measured during the ICS reduction period of trial, and found to only be 10 for ACARIZAX \[[@CR21]\]. This beneficial impact of ACARIZAX was not quantified in the analysis. In the future it may be pertinent to compare ACARIZAX with other AITs, which are considered standard care in Germany. It was, however, not feasible to compare ACARIZAX with all potential AITs in a double-blinded, controlled trial. Furthermore, currently, there are no other AITs for which the efficacy in (HDM) allergic asthma has been demonstrated using similar design; hence comparison between ACARIZAX and other AIT is not possible \[[@CR23]\]. Conclusion {#Sec10} ========== The analysis presented here indicates that ACARIZAX in addition to pharmacotherapy is a cost-effective treatment option compared to pharmacotherapy alone in HDM allergic asthma patients not well controlled by ICS and associated with mild to severe allergic rhinitis. If a disease-modifying effect can be proven the results of this analysis may underestimate the true benefits of ACARIZAX as conservative assumptions were used to predict long-term patient outcomes. AIT : allergy immunotherapy G-BA : Gemeinsamer Bundesausschuss GP : general practice HDM : house dust mite HRQoL : health-related quality of life ICER : incremental cost-effectiveness ratio ICS : inhaled corticosteroids NICE : national institute for health and care excellence QALY : quality-adjusted life year SABA : short-acting β2-agonist SF-36 : 36 item short form health survey WPAI : work productivity and activity impairment MW reviewed drafts of the manuscript and gave advice to the structure and content of the analysis. She was a clinical investigator within the clinical trial (MT-04) that the economic analysis is based upon. WG completed the initial drafting of the manuscript and oversaw all major revisions. He also developed the economic model that is the basis of the economic analysis discussed in the manuscript. JHP & JNA were involved in both the process of economic model development and revising the initial manuscript drafts. Both are employed by ALK-Abelló, with Julie Hahn-Pedersen leading the project for the company. MT oversaw both the development of the manuscript and the economic model upon which the manuscript is based, providing editorial input throughout the project. All authors read and approved the final manuscript. Acknowledgements {#FPar1} ================ None. Competing interests {#FPar2} =================== The authors declare that they have no competing interests. Availability of data and materials {#FPar4} ================================== The analysis undertaken here is based on a clinical trial that has been described in a previous publication \[[@CR7]\]. Therefore, the data and materials will not be shared here. Funding {#FPar3} ======= ACARIZAX, the drug considered in this manuscript, is owned by ALK Abelló. JHP and JNA were both employed by ALK Abelló at the time of manuscript preparation. WG and MT are employed by York Health Economics Consortium (YHEC) who were sponsored by ALK Abelló to undertake the cost-utility analysis of ACARIZAX, which this manuscript is based on. A separate consultancy fee was also provided for to YHEC the preparation of the manuscript. MW was the principal investigator in MT-04 and received honoraria for consultancy and speaker activity.

Introduction {#S0001} ============ Chaque année, près de 25 millions de femmes enceintes, dont 20% de primipares, sont confrontées aux conséquences du paludisme en Afrique subsaharienne \[[@CIT0001]\]. Le placenta en effet est le site préférentiel de séquestration et de développement des plasmodiums pendant la grossesse. La femme enceinte a ainsi une forte susceptibilité d'être infestée par *Plasmodium falciparum*. Ceci se traduit par une forte fréquente d'épisodes palustres avec une forte densité parasitaire par rapport aux femmes non enceintes \[[@CIT0002]\]. L\'infection palustre chez la femme enceinte se caractérise par une anémie maternelle. Les carences nutritives qui en résultent pour le foetus sont parfois source d\'avortements, de retard intra utérine, de mort in utero et de prématurité \[[@CIT0003], [@CIT0004]\]. Au regard des conséquences du paludisme durant la grossesse, il est important que les femmes enceintes vivant dans les zones d\'endémie palustre soient sous prophylaxie antipalustre \[[@CIT0005]\]. En Afrique subsaharienne, l\'OMS recommande des stratégies de prévention pendant la grossesse, basées sur l′administration d\'un traitement préventif intermittent (TPI) et sur l′utilisation de moustiquaires imprégnées d′insecticide (MII) \[[@CIT0006]\]. L\'efficacité du TPI à la sulfadoxine-pyriméthamine (SP) a été démontrée en Afrique de l\'Ouest \[[@CIT0007]--[@CIT0011]\] et en Afrique de l\'Est \[[@CIT0012], [@CIT0013]\]. Au Burkina Faso, la prévention du paludisme chez la femme enceinte constitue une priorité pour les autorités sanitaires nationales. C\'est ainsi qu\'avec la chloroquino-résistance ayant atteint 26,9% à 63,3% en 2003 \[[@CIT0014]\], une nouvelle stratégie de prévention du paludisme chez la femme enceinte a été mise en place à partir de Février 2005 \[[@CIT0015]\]. Conformément aux recommandations de l\'Organisation Mondiale de la Santé (OMS), cette stratégie préconise l\'administration de la sulfadoxine-pyriméthamine (SP) en traitement préventif intermittent (TPI) chez la femme enceinte \[[@CIT0015]--[@CIT0017]\]. La thérapie intermittente pour le paludisme avec la SP est recommandée pour les femmes enceintes vivant en zone d\'endémie palustre où *Plasmodium falciparum* est résistant à la chloroquine (CQ) et sensible à la SP \[[@CIT0018], [@CIT0019]\]. Le TPI avec au moins de deux doses de sulfadoxine - pyrimethamine administrées pendant le second et le troisième trimestre de grossesse représente une stratégie de prévention alternative dont l\'efficacité a été démontrée par plusieurs études conduites en Afrique dans la réduction du taux de l\'infestation placentaire par *Plasmodium falciparum* \[[@CIT0007], [@CIT0010], [@CIT0018], [@CIT0020]--[@CIT0022]\], le faible poids de naissance \[[@CIT0004], [@CIT0019]\] et l\'anémie sévère au cours de la grossesse \[[@CIT0004], [@CIT0023]\]. La prise de SP doit être observée par un agent qualifié au niveau de la formation sanitaire. Toutefois, le succès d\'un tel programme de prévention dépend non seulement de l\'efficacité du médicament utilisé et de sa disponibilité mais également de l\'observance des directives par les femmes enceintes et les agents de santé. Depuis la mise en place de cette nouvelle politique au Burkina Faso, aucune donnée relative à l\'efficacité du TPI à la SP chez la femme enceinte en milieu urbain n\'est disponible \[[@CIT0015], [@CIT0016]\]. C\'est dans ce contexte que la présente étude se propose trois années après l\'adoption de cette stratégie, d'évaluer son efficacité thérapeutique et son observance chez la femme enceinte dans deux hôpitaux urbains du Burkina Faso. Méthodes {#S0002} ======== Site d'étude {#S20003} ------------ L'étude s\'est déroulée dans les services de gynécologie-obstétrique du Centre Hospitalo-universitaire (CHU) de Ouagadougou et de Bobo - Dioulasso, les deux principales villes du Burkina Faso. Le Burkina Faso est un pays sub-saharien situé dans la boucle du fleuve Niger et couvre une superficie de 274 200 Km^2^. Il possède un climat de type soudano-sahélien caractérisé par l′alternance d′une saison des pluies (mai-octobre) et d′une saison sèche (novembre-avril). Il existe une recrudescence de la transmission palustre pendant la saison des pluies à Ouagadougou et Bobo - Dioulasso. Population d'étude {#S20004} ------------------ Nous avons conduit une étude transversale prospective recrutant des femmes enceintes venues accoucher dans les services de gynécologie-obstétrique du CHU Yalgado Ouédraogo (CHUYO) de Ouagadougou et du CHU Sourô Sanou (CHUSS) de Bobo - Dioulasso entre février et novembre 2008. Ont été inclues dans l'étude les femmes enceintes résidant dans les villes de Ouagadougou et de Bobo- Dioulasso et consentantes librement de participer à l'étude. Ces femmes ont au préalable bénéficié de consultations prénatales (CPN) trimestrielles dans les centres de santé et de promotion sociale (CSPS) ou dans les services de santé maternelle et infantile (SMI) des deux villes. Les critères de non-inclusion dans l'étude ont été les suivants: les femmes en mauvais état général; celles porteuses de grossesses pathologiques pouvant altérer le placenta (hématome rétro-placentaire, placenta *praevia*) et les femmes qui ont affirmé avoir présenté une allergie à la SP et donc n\'ayant pas respecté le TPI. Chimioprophylaxie {#S20005} ----------------- Comme le préconise le programme national de lutte contre le paludisme (PNLP) au Burkina Faso \[[@CIT0016]\], la SP a été administrée à la posologie de 3 comprimés en prise unique (soit 1,5 g de sulfadoxine et 0,75 g de pyriméthamine). Cette administration a été faite lors des CPN conjointement à du fer et à de l\'acide folique. Deux doses ont été en principe délivrées, l\'une lors du second trimestre de la grossesse et la deuxième au troisième trimestre selon les recommandations de l\'OMS \[[@CIT0001]\]. Collecte des données {#S20006} -------------------- Chaque femme répondant aux critères d\'inclusion de l'étude a été soumise à un questionnaire après l\'obtention de son consentement éclairé. Ce questionnaire a pris en compte les variables suivantes: l\'identité et l'état civil de la femme, les antécédents obstétricaux (gestité), les conditions socio-économiques, les conditions socio-culturelles (religion, profession), le niveau d\'instruction (analphabète, niveau d\'instruction primaire, secondaire, ou supérieur), les particularités de la grossesse actuelle et les conditions d\'observance de la chimioprophylaxie (prescription du TPI, nombre de prises, effets secondaires du TPI, autre chimioprophylaxie prescrite). Les carnets de santé des femmes ont été consultés afin de s\'assurer des dates et lieux des CPN fréquentés, de la prescription de la SP ou d\'une autre alternative, de la prescription d\'autres traitements éventuels. Le degré de concordance entre les recommandations du médecin et les comportements de la femme a été vérifié. Les réponses de la femme ont été également comparées à celles mentionnées dans le carnet (chimioprophylaxie prescrite, nombre de prise, les antécédents obstétricaux). Diagnostic parasitologique {#S20007} -------------------------- Lors de l\'accouchement, l\'apposition placentaire a été systématiquement pratiquée après obtention de l\'autorisation de la femme ou des accompagnants. Des morceaux de la face maternelle du placenta ont été déposés sur des lames de façon à confectionner des frottis minces. La lecture a été faite au microscope à l\'objectif 100 à immersion. Au total, 100 champs microscopiques ont été lus à la recherche de trophozoïtes ou de schizontes de plasmodiums témoins d\'une infection palustre placentaire. Il a été estimé à 200 en moyenne le nombre de globules rouges parasités par champ examiné, et à 4.000.000 le nombre de globules rouges par microlitre (µl) de sang. La densité parasitaire a été exprimée en nombre de parasites par microlitre de sang \[[@CIT0024]\]. Critères de jugement de l\'efficacité et de l\'observance {#S20008} --------------------------------------------------------- Le traitement à la SP a été considéré comme efficace en l\'absence de *Plasmodium sp* à l\'examen du placenta \[[@CIT0001]\]. Il a été considéré qu\'une femme a veillé à la bonne observance du traitement quand elle a reçu 3 comprimés de SP en une prise orale au deuxième trimestre et au troisième trimestre \[[@CIT0016]\]. Considérations éthiques {#S20009} ----------------------- Avant le début de l'étude, son protocole a été soumis et approuvé par le comité d'éthique institutionnel du Centre Muraz à Bobo-Dioulasso au Burkina Faso. Analyse des données {#S20010} ------------------- Les données ont été analysées sur le logiciel EPI-INFO version 6.04. Le test statistique de Chi carré au seuil de signification de 5% a été utilisé pour la comparaison des proportions. Résultats {#S0011} ========= Au total, 542 femmes ont participé à l'étude. Leur âge moyen a été de 26,0 ± 6,45 ans (extrêmes 13-43 ans). Les femmes de moins de 18 ans ont représenté 26,6% (n = 144) vs 73,4% (n = 398) pour celles de plus de 18 ans ([Tableau 1](#T0001){ref-type="table"}), (p\<,05). Les femmes analphabètes ont représenté 41,5% (n= 225) vs 58,5% (n= 159) de scolarisées ([Tableau 1](#T0001){ref-type="table"}), (p \>0,05). Une analyse des caractéristiques culturelle et professionnelle de la population d'étude indique que parmi les 542 femmes, les ménagères ont constitué 50,5%( n= 273), ont suivi ensuite les commerçantes (24,3%, n= 132 /542) puis les élèves-étudiantes (17,3% n= 94) et enfin les fonctionnaires (8%, n= 43) ([Tableau 1](#T0001){ref-type="table"}), (p \> 0,05). La religion musulmane a été pratiquée par 56,7% (n= 307) des femmes; vs 44,3% (n= 235) de chrétiennes ([Tableau 1](#T0001){ref-type="table"}), (p \> 0,05). ###### Caractéristiques socio -- démographiques et taux de bonne observance du TPI à la SP Caractéristiques Effectifs global (N = 542) Effectif sous TPI à la SP (N = 433) Taux de bonne Observance (N= 238) ---------------------------------- ---------------------------- ------------------------------------- ----------------------------------- **Age**  \< 18 ans 144 (26,6%) 89 (61,8%) 19 (21,3%)  \>18 ans 398 (73,4%) 354 (90%) 219 (61,9%) **Niveau d\'instruction**  Analphabète 225 (41,5%) 171 (76%) 52(30,4%)  Primaire 157 (29%) 132 (84%) 59 (44,7%)  Secondaire 111 (20,5%) 102 (91,9%) 91 (89,2%)  Supérieur 49 (9%) 38 (77,5%) 36 (94,7%) **Profession /occupation**  Ménagère 273 (50,5%) 240 147 (61,1%)  Commerçante 132 (24,3%) 92 48 (52,2%)  élèves-étudiante 94 (17,3%) 83 35 (42,1%)  Fonctionnaires 43 (8%) 28 8 (28,5%) **Religion**  Musulmane 307 (56,7%) 237 (77,2%) 129 (54,3%)  Chrétienne 235 (44,3%) 196 (83,4%) 109 (55,6%) **Antécédents obstétricaux**  Gestité  Primigeste 220 (40,6%) 183 (42,3%) 81 (43,3%)  Secondigestes 135 (24,4%) 98 (22,6%) 53 (51,4%)  3 et plus 187 (34,5%) 152 (35,1%) 104 (68%) **Gratitude de la SP**  SP gratuite \- 96 (21,7%) 51 (53,1%)  SP achetée \- 347(78,3%) 187 (54%) **Effets secondaires à la SP**  Effets secondaires notifiés \- 88 (19,9%) 46 (52,3%)  Effets secondaires non notifiés \- 355 (80,1%) 192 (54,1%) TPI : Traitement préventif intermittent ; SP : sulfadoxine -- pyriméthamine Antécédents obstétricaux et type de chimioprophylaxie {#S20012} ----------------------------------------------------- Le nombre moyen de grossesses de la population d'étude a été de 2,7 avec un maximum de 11 grossesses. La proportion des primigestes, secondigestes et de multigestes ont été respectivement de 40,6% (n= 220), 24,9%(n= 135) et de 34,5% (n= 187) ([Tableau 1](#T0001){ref-type="table"}), (p \> 0,05). Sur 542 femmes, la chimioprophylaxie prescrite au cours des CPN a été la SP pour 80% (n= 433) vs 8,7% (n= 47) pour la CQ en chimioprophylaxie hebdomadaire (25 mg/kg sur 3 jours à la première CPN, puis 300 mg/semaine) ([Tableau 2](#T0002){ref-type="table"}). Et, 11,4% (n= 62) des femmes n\'ont reçu aucune prophylaxie médicamenteuse antipalustre ([Tableau 2](#T0002){ref-type="table"}). Parmi les 433 femmes qui ont été sous la chimioprophylaxie à la SP, la majorité ont reçu les deux doses (55% n= 238); vs 29,5% (n= 128); 13,1% (n= 57) et 1,2% (n= 5) pour respectivement une, trois et quatre doses. Les femmes ignorant le nombre de doses de SP reçues ont été au nombre de 5 (1,2%) ([Tableau 2](#T0002){ref-type="table"}). ###### Répartition des femmes selon la chimioprophylaxie et le nombre de dose de SP Nombre de prise de SP prescrites Effectif \% ---------------------------------- ---------- ------- **Chimioprophylaxie**  SP 433 80%  Chloroquine 47 8,7%  Aucune chimioprophylaxie 62 11,4% **Nombre de prise de SP**  1 128 29,5%  2 238 55%  3 57 13,2%  4 5 1,2% **Inconnu** 5 1,2 **Total** 433 100 SP: sulfadoxine -- pyriméthamine L\'analyse des antécédents obstétricaux indique que parmi les 433 femmes ayant reçu le TPI à la SP, 42,3% (n= 183), 22,6% (n= 98), et 35,1% (n= 152) ont été respectivement primigestes, secondigestes et multigestes ([Tableau 1](#T0001){ref-type="table"}). Selon la gratuité de la SP, 78,3% (n= 347) des femmes ont affirmé avoir acheté la SP, vs 21,7% (n= 96) qui l\'ont reçu gratuitement ([Tableau 1](#T0001){ref-type="table"}). Influence de la chimioprophylaxie à la SP sur l\'infestation palustre placentaire {#S20013} --------------------------------------------------------------------------------- De façon globale, sur 443 appositions placentaires analysées, 4,8% (n = 21) ont eu des trophozoïtes de *Plasmodium falciparum*. Le taux d\'infestation placentaire à *Plasmodium falciparum* a été de 7% (9/128), 4,2% (10/238) et 3,5% (2/57) pour une, deux et trois doses respectivement ([Tableau 3](#T0003){ref-type="table"}). Aucune relation entre le nombre de dose et le taux d\'infestation placentaire n\'a été établie. (p \> 0,05). ###### Efficacité du TPI à la SP selon le nombre de grossesse et de dose de SP Variables liées à l\'infestation placentaire Placenta ---------------------------------------------- --------------- --------- --------- **Nombre de geste**  Primigestes 7 (3,7%) 180 187  Secondigestes 9 (8,7%) 94 103  Multigestes 5 (3,5%) 148 153  Total **21(4,7%)** **422** **443** **Nombre de dose**  1 9 (7%) 119 128  2 10 (4,2%) 228 238  3 2 (3,5%) 54 57 **Autre** 0 10 10 **Total** **21 (4,7%)** **401** **433** TPI: Traitement préventif intermittent; SP : sulfadoxine -- pyriméthamine Relation entre le nombre de grossesses et l\'infestation placentaire {#S20014} -------------------------------------------------------------------- Parmi les 21 (4,8%) de frottis placentaires qui ont eu des plasmodiums, les primigestes et les secondigestes ont représenté 76,1% (16/21). Le taux d\'infestation placentaire a été plus élevé chez les secondigestes (8,7; 9/103) que chez les primigestes (3,7%; 7/187) ([Tableau 3](#T0003){ref-type="table"}). Toutefois, ces différences n\'ont pas été statistiquement significatives (p \> 0,05). En analysant la répartition temporelle de l\'infection placentaire, il ressort qu\'aucune infestation palustre placentaire n\'a été observée durant les 5 premiers mois de l'étude (février à juin). Le taux d\'infection placentaire a augmenté de juillet à octobre (42,9%, 9/21) et a diminué significativement à 9,5% (2/21) en novembre. Nous avons pu établir une relation entre la répartition temporelle de l\'infection placentaire (p \< 0,05) ([Figure 1](#F0001){ref-type="fig"}). {#F0001} Observance du TPI à la SP {#S20015} ------------------------- Le taux global de bonne observance du TPI à la SP a été de 55% (238/433). Nous avons pu établir une relation entre la bonne observance du TPI et l'âge des femmes. En effet, le taux de bonne observance a été de 21,3% (19/89) chez les femmes de moins de 18 ans et de 61,9% (219/354) chez les plus de 18 ans ([Tableau 1](#T0001){ref-type="table"}), (p 0 \< 0,05). En analysant la relation entre cette bonne observance et le niveau d\'instruction, il ressort que le taux de bonne observance a été de 94,7% (36/38), 89,2% (91/102), 44,7% (59/132) et 30,4% (52/171) pour les femmes du niveau d\'instruction supérieur, secondaire, primaire et les analphabètes, respectivement ([Tableau 1](#T0001){ref-type="table"}). Nous n\'avons pas pu établir une relation entre la bonne observance et le niveau d\'instruction (p \> 0,05). Nous n\'avons pas noté de relation entre l\'observance et la confession religieuse ([Tableau 1](#T0001){ref-type="table"}). En effet, le traitement a été bien observé aussi bien chez les femmes chrétiennes (55,6%; 109/196) que chez les femmes musulmanes (54,3%; 129/237), (p \> 0,05). La meilleure observance de 61,1% (147/240) a été notée chez les ménagères. Les niveaux d\'observance ont diminué ensuite, avec des taux de 52,2% (48/92 pour les commerçantes, 42,1% (35/83) pour les élèves - étudiants et 28,5% (8/28) pour les fonctionnaires ([Tableau 1](#T0001){ref-type="table"}). Cependant nous n\'avons pas trouvé de relation entre l\'observance et la profession des femmes (p \>0,05). Le taux de bonne observance a été de 36,6% (67/183) chez les primigestes, 44,3% (81/183) vs 54,1% (53/98) et 68,1% (104/152) pour les secondigestes et les multigestes ([Tableau 1](#T0001){ref-type="table"}). Cependant nous n\'avons pas trouvé de relation entre l\'observance et le nombre de grossesses (p \> 0,05). L\'observance a été similaire entre les femmes qui ont reçu la SP gratuitement (53,1% (51/96) et celles qui l\'ont achetée (54%; 187/347). Cependant nous n\'avons pas trouvé de relation entre l\'observance et la religion des femmes (p \>0,05) ([Tableau 1](#T0001){ref-type="table"}). L\'observance a été bonne chez 52,3% (46/88) des femmes ayant déclaré avoir présenté un effet secondaire et de 54,1% (192/355) chez celles qui ne l\'ont pas signalé (P \> 0,05) ([Tableau 1](#T0001){ref-type="table"}). Une analyse de l\'incidence de ces effets secondaires indique que 20,3% (88 /433) des femmes ont déclaré avoir présenté au moins un effet secondaire à la suite de la prise de la SP ([Tableau 1](#T0001){ref-type="table"}). Les effets secondaires les plus fréquemment rapportés ont été l\'asthénie (30,7%; 27/88); les vomissements (27,2%; 24/88); les nausées (239%; 21/88) et les sensations de vertiges (18,2%; 16/88) ([Tableau 1](#T0001){ref-type="table"}). Tous ces effets ont survenu en moyenne dans les 4 heures suivant la prise médicamenteuse. Discussion {#S0016} ========== L\'efficacité du TPI à la SP a été démontrée en milieu rural du Burkina Faso à travers trois études \[[@CIT0007], [@CIT0020], [@CIT0021]\]. Ces études ont rapporté une bonne efficacité de cette stratégie à travers une réduction de la charge parasitaire placentaire palustre chez la femme enceinte. Cependant il n\'existe aucune donnée en ce qui concerne le milieu urbain, qui pourtant présente un faciès complètement différent du point de vue épidémiologique et socio- culturel. La présente étude qui est la première réalisée en milieu urbain permettra de fournir au PNLP des données complémentaires sur les deux villes (Ouagadougou et Bobo -Dioulasso) les plus peuplées du Burkina Faso. L'âge moyen des parturientes est de 26 ± 6,45 ans, chiffre voisin de celui d\'une étude menée en milieu rural au Burkina Faso \[[@CIT0020]\], au Bénin \[[@CIT0025]\] et au Nigeria \[[@CIT0011]\]. Dans la présente étude, plus de la moitié des femmes (80%) ont bénéficié du traitement prophylactique tel que préconisé par le PNLP, la majorité d\'entre elles (55%), ont reçu deux doses de SP ([Tableau 2](#T0002){ref-type="table"}). Le même constat a été fait par les travaux antérieurs conduits au Burkina Faso (96,2%) \[[@CIT0007]\]. Au vu de cette bonne proportion, nous pensons que la sensibilisation des femmes à la problématique du paludisme gestationnel initiée à travers les campagnes d\'information, d'éducation et de communication (IEC) du PNLP a un impact sur l\'adhésion des femmes enceintes au TPI à la SP au Burkina Faso \[[@CIT0016]\]. Des taux similaires au nôtre ont été rapportés ailleurs au Ghana (77%) \[[@CIT0010]\]. Par ailleurs, dans la présente étude, les femmes ayant bénéficié d\'une chimioprophylaxie par la chloroquine ont représenté 8,7% de notre population d'étude. Bien que l\'objectif de notre étude n'était pas de comparer les deux médicaments (SP et CQ), nos résultats indiquent que trois ans après la mise en place de la nouvelle politique de prévention du paludisme, et malgré le retrait de la chloroquine, celle-ci est toujours utilisée pour la prévention du paludisme chez la femme enceinte. Bien que cette utilisation ne concerne qu\'une faible proportion de notre population d'étude (8,7%), il soulève néanmoins la question de la problématique du changement de comportement et de l\'appropriation par les praticiens et les populations des nouvelles stratégies d\'intervention en matière de santé publique. Toutefois, les résultats obtenus ont le bénéfice de confirmer que le TPI à la SP est efficace en chimioprophylaxie antipalustre chez la femme enceinte, comme cela avait été précédemment démontré par des études antérieures conduites au Burkina Faso \[[@CIT0007], [@CIT0020], [@CIT0021]\], et suggéré par le Ministère de la santé du pays \[[@CIT0016]\]. Le taux d\'infestation placentaire palustre global de la présente étude est de 4,7% ([Tableau 3](#T0003){ref-type="table"}) chez des femmes ayant bénéficié du TPI à la SP. Des taux similaires au nôtre ont été enregistrés au Bénin (4,1%) \[[@CIT0025]\]. Par contre des taux supérieurs au nôtre (4,7%) ont été rapportés par d\'autres auteurs. Ces taux ont été en effet de 19,2% au Burkina Faso selon une étude antérieure \[[@CIT0021]\]. Ailleurs au Mali, des taux de 16,7% \[[@CIT0026]\], à 24,5% \[[@CIT0008]\] ont été notifiés par des études. Les mêmes tendances de 13,8% et de 22,8% ont été respectivement rapportées au Kenya \[[@CIT0013]\] et au Malawi \[[@CIT0027]\]. Le faible taux observé dans notre étude comparé à celui des études antérieures menées au Burkina Faso \[[@CIT0021]\] pourrait s\'expliquer par la bonne observance globale observée dans notre étude. Toutefois de façon générale, ces constats renforcent l\'efficacité de la SP en chimioprophylaxie antipalustre pendant la grossesse et justifie de ce fait son administration en traitement préventif intermittent (TPI) depuis février 2005 au Burkina Faso \[[@CIT0016]\]. La thérapie intermittente à la SP est efficace et peut être utilisée comme une stratégie pratique pour la réduction du risque de l\'infestation placentaire par des plasmodiums en zone d\'endémie palustre. Les mêmes conclusions ont été faites par des auteurs au Burkina Faso \[[@CIT0007], [@CIT0020], [@CIT0021]\], ailleurs en Afrique de l\'Ouest au Mali \[[@CIT0008], [@CIT0026]\], au Bénin \[[@CIT0025]\], au Nigéria \[[@CIT0011]\], au Ghana \[[@CIT0010]\], et d\'autre part en Afrique de l\'Est au Kenya \[[@CIT0014]\] et au Malawi \[[@CIT0018], [@CIT0027]\]. Par ailleurs, les primigestes et les secondigestes sous TPI à la SP demeurent les plus exposées au paludisme, 76,1% d\'entre elles ont eu, en effet, un frottis placentaire positif ([Tableau 3](#T0003){ref-type="table"}). Les mêmes tendances ont été observées au Burkina Faso \[[@CIT0021]\], au Bénin \[[@CIT0025]\] et au Malawi \[[@CIT0018], [@CIT0027]\]. Cela pourrait s\'expliquer par le fait que les multigestes, développent avec les grossesses antérieures une immunité protectrice contre l\'infection palustre placentaire \[[@CIT0028], [@CIT0029]\]. Le fait qu\'aucun placenta n\'est été trouvé parasité durant les 5 premiers mois de l'étude, puis que la fréquence d\'infestation ait augmenté de juillet à octobre, est lié à la recrudescence de la transmission palustre pendant la saison des pluies (mai-octobre) ([Figure 1](#F0001){ref-type="fig"}). Le même constat a été fait en milieu rural au Burkina Faso par des auteurs qui ont noté que 90,6% de la prévalence du paludisme placentaire était notée pendant la saison pluvieuse \[[@CIT0021]\]. Cela devrait interpeller les autorités sanitaires sur la nécessité de développer des stratégies de surveillance rapprochée pour une meilleure observance du TPI pendant la saison de transmission où le risque est plus important. Le taux de couverture du TPI à la SP des femmes de notre étude a été de 80%. Ce résultat est en conformité avec les objectifs du plan stratégique 2006-2010 de lutte contre le paludisme au Burkina Faso, qui préconisait un taux de couverture en SP de 80% pour 2010 \[[@CIT0016]\]. Le taux d\'ensemble de bonne observance du traitement à la SP, de 55%, est bien inférieur à celui de 99,1% dont fait état une étude menée au Mali \[[@CIT0008]\]. Si dans notre étude, la gratuité de la SP à 78,3% ne semble avoir eu d\'influence sur l\'observance, dans l'étude réalisée au Mali \[[@CIT0008]\], chaque femme recevait gratuitement la SP lors des consultations prénatales et cela a semblé jouer positivement sur le taux de couverture. Des études complémentaires devraient être conduites dans notre cas pour comprendre les raisons du taux de 55% malgré la gratuité chez 78,3% des femmes. Des raisons de nature socioculturelle, économique et environnementale pourraient en être les causes. L\'analyse des facteurs socio-démographiques fait apparaître que l'âge, et non la gestité, constitue un déterminant majeur de l\'adhésion au TPI. Nous avons, en effet, observé une amélioration de l\'observance avec l'âge des femmes. Cela pourrait être lié à une meilleure prise de conscience des complications du paludisme au cours de la grossesse. Conclusion {#S0017} ========== Notre étude, qui est la première réalisée sur l\'efficacité du TPI en milieu urbain au Burkina Faso, confirme l\'efficacité plusieurs fois rapportée de cette stratégie lorsqu\'elle est bien suivie. Cette étude qui confirme également l\'efficacité du TPI à la SP et aussi la prescription de la chloroquine en chimioprophylaxie, soulève cependant la question de l\'appropriation des nouvelles stratégies d\'intervention en santé publique. En effet, 3 ans après le changement de la politique et le retrait de la chloroquine, cette dernière est toujours prescrite et utilisée. Avec un taux de couverture du TPI à la SP de 80%, les résultats de notre étude confirment l\'atteinte en milieu urbain des objectifs du plan stratégique 2006-2010 de lutte contre le paludisme au Burkina Faso, qui préconisait un taux de couverture en SP de 80% pour 2010. Cependant les raisons du taux d\'observance de 55%, malgré la gratuité de la SP chez 78,3% des femmes de notre série dans certains cas méritent d'être explorées en tenant compte des facteurs socioculturels, économiques et environnementaux de notre contexte d'étude. Les auteurs adressent leurs sincères remerciements à tout le personnel du service de gynéco-obstétrique du CHU de Ouagadougou et de Bobo - Dioulasso et ainsi qu'à toutes les femmes qui ont participé à cette étude. Les auteurs remercient sincèrement le Dr Liance Martine du Laboratoire de Parasitologie-Mycologie de la faculté de Médecine de Créteil pour sa contribution inconditionnelle à la relecture du document. Conflit d\'intérêts {#S0018} =================== Les auteurs ne déclarent aucun conflit d\'intérêt. Contribution des auteurs {#S0019} ======================== Adama Séré et Rodrigues Nikiéma ont contribué à la collecte des données. Sanata Bamba, Adama Séré, Rodrigues Nikiéma, Halidou Tinto, Blandine Thiéba, Blami Dao, Tinga Robert Guiguemdé ont tous contribué à la rédaction, lecture et conception finale du document.

The prevalence of type 2 diabetes is increasing rapidly. In the U.S., \>13% of adults have been diagnosed with type 2 diabetes ([@B1]), and a similar prevalence has been reported in Asia ([@B2]). Up to 25% of newly diagnosed diabetic patients already had microvascular complications, which suggests that there is a 6- to 7-year time lag between the onset and the diagnosis of type 2 diabetes ([@B3]). When considering the clinical implications of diabetes and its complications, it is important to identify individuals with undiagnosed diabetes or those who are prone to diabetes in the near future. The American Diabetes Association (ADA) recommends screening asymptomatic people at 3-year intervals using a fasting plasma glucose (FPG) test or 2-h oral glucose tolerance test (OGTT) ([@B4]). However, it is not easy to perform the OGTT in primary practice, and it is debatable whether the FPG concentration alone provides an accurate diagnosis of diabetes, as indicated by the estimated 40% of people who have undiagnosed diabetes ([@B1]). The hemoglobin A~1c~ (A1C) level is measured in a standardized test that produces data consistent with those of the international A1C-derived average glucose and the Diabetes Control and Complications Trial ([@B5],[@B6]). The A1C level provides a reliable measure of chronic glycemic control without the need for a fasting or timed sample, and it correlates well with the risk of long-term diabetes complications and mortality ([@B7],[@B8]). Several population-based studies have investigated the utility of the A1C level for detecting undiagnosed diabetes and the potential to use the A1C level as a good screening tool for type 2 diabetes ([@B9],[@B10]). However, the recent ADA redefinition of the diagnosis of diabetes using an A1C level ≥6.5%, which considers many aspects of diagnostic testing and the economic burden, raises concerns about the possible delay in diagnosing diabetes ([@B11],[@B12]). Thus, there is widespread debate about the appropriate A1C cutoff value for diagnosing diabetes. To evaluate the predictive value of the A1C level and to find the appropriate A1C cutoff for identifying undiagnosed diabetes and new-onset diabetes over a 6-year follow-up, we analyzed the data from a large-scale, prospective cohort study of people from a homogeneous ethnic background. RESEARCH DESIGN AND METHODS {#s5} =========================== The design and baseline characteristics of the Ansung-Ansan cohort study have been published by our group ([@B13]). Briefly, it is an ongoing prospective, community-based cohort study that is part of the Korean Health and Genome Study, a community-based epidemiological survey to investigate the trends in diabetes and associated risk factors. The baseline examination was performed in 2001--2002, and biennial follow-up examinations will continue through 2010. The eligibility criteria included an age of 40--69 years, residence within the borders of the survey area for at least 6 months before testing, and sufficient mental and physical ability to participate. Participants were recruited from the residents of two Korean communities within 60 km of Seoul. Ansung is a representative rural farming community that had a population of 132,906 in 2000 ([@B14]). Of 7,192 eligible individuals in Ansung, 5,018 were surveyed (70% response rate) using a cluster-sampling method stratified by age, sex, and residential district. Ansan is a representative urban community that had a population of 554,998 in 2000 ([@B14]). We successfully recruited 5,020 subjects from 124,775 eligible subjects (4.0%) using a random-sampling method of the local telephone directory. At baseline, we excluded 572 (5.7%) individuals with known type 2 diabetes and 91 who had an unknown glucose status. Among 9,375 (4,415 men and 4,960 women) participants without a previous history of diabetes, 635 (6.8%) were newly diagnosed with type 2 diabetes at the baseline examination. Of 8,740 remaining nondiabetic subjects, 5,945 (3,022 from Ansung and 2,923 from Ansan) were included and underwent repeated examinations during the 6-year follow-up period. The recall rate was 85.7% at the 2-year follow-up examination, 74.9% at year 4, and 66.7% at year 6. To deal with the bias arising from missing data, we used a data-deletion method and found that there was no significant bias caused by loss to follow-up. Informed written consent was obtained from all participants. The study protocol was approved by the ethics committee of the Korean Center for Disease Control and the Ajou University School of Medicine Institutional Review Board. Throughout the study, the same trained researchers and instruments were used to collect the data. Anthropometric parameters and blood pressure were measured by standard methods. The fasting plasma concentrations of glucose, insulin, total cholesterol, triglycerides, HDL cholesterol, and high-sensitivity C-reactive protein (hsCRP) were measured in a central laboratory. After an 8- to 14-h overnight fast, all subjects underwent a 2-h 75-g OGTT at inclusion and biennially. A1C level was measured using high-performance liquid chromatography (Variant II; BioRad Laboratories, Hercules, CA). Pancreatic β-cell function and insulin resistance were calculated by the homeostasis model assessment (HOMA-β and HOMA-IR, respectively). Definition {#s6} ---------- For both baseline and during follow-up, the definition of diabetes was based on plasma glucose results during the 75-g OGTT, defined according to the 1997 ADA criteria: FPG concentration ≥7.0 mmol/L (126 mg/dL) or 2-h plasma glucose ≥11.1 mmol/L (200 mg/dL) or current treatment by oral antidiabetes drugs or insulin ([@B15]). A family history of diabetes was coded if there was at least one diabetic first-degree relative. Hypertension was defined as systolic blood pressure ≥140 mmHg or diastolic blood pressure ≥90 mmHg or taking antihypertensive medication. Statistical analysis {#s7} -------------------- The data are presented as means ± SD, as numbers and percentages, or as a relative risk (RR) with 95% CIs. Fasting insulin, triglycerides, and hsCRP concentrations and HOMA-β and HOMA-IR were normalized by logarithmic transformation. The means were compared by Student *t* tests or by ANCOVA. For qualitative variables, the results are expressed as percentages and were compared by c^2^ or by logistic regression. Pearson correlation analysis was used to determine the relationships between A1C level and plasma glucose concentration. The diagnostic properties of the specific threshold levels of A1C were evaluated by calculating the sensitivity, specificity, and positive and negative predictive value by the receiver operating characteristic (ROC) curve. To decide optimal-cutoff A1C, we were referencing the Youden-Index (*J =* max*~c~*{*sensitivity(c)* + *specificity(c)* − 1}, for all possible cutoff values *c*) ([@B16]). Risk of new-onset diabetes according to the A1C cutoff was modeled using the Cox proportional hazards model, after adjusting for age, and using those variables with *P* ≤ 0.25 in the age-adjusted comparison between the diabetic and nondiabetic groups. We first examined the age-adjusted effects of the A1C cutoff on the 6-year incidence of diabetes (model A). Model B comprised model A with additional adjustment for anthropometric and social parameters. Model C was the adjusted model B plus triglyceride, HDL cholesterol, and hsCRP concentrations and HOMA-β and HOMA-IR. The final Cox models fulfilled the proportional hazards assumption. We compared the predictive performance of A1C level and FPG concentration as continuous variables using the ROC curves and by calculating the area under the curves. For detecting undiagnosed diabetes at baseline by ROC curve analysis, we used the baseline data of participants without a previous history of diabetes. For incident diabetes after the 6-year follow-up, we used the 6-year follow-up data of participants who were nondiabetic at baseline and who had completed the 6-year follow-up. MedCalc software was used to calculate the ROC curves; the significance of differences between areas under these curves was calculated as shown elsewhere ([@B17]). All other analyses were performed using SPSS software (version 12.0; SPSS, Chicago, IL). Significance was defined as *P* \< 0.05 for two-sided tests. RESULTS {#s8} ======= Baseline characteristics {#s9} ------------------------ Of 9,375 participants without a previous history of diabetes, 635 (6.8%) subjects revealed previously undiagnosed diabetes at the baseline 75-g OGTT test ([Table 1](#T1){ref-type="table"}). The clinical characteristics of participants with and without undiagnosed diabetes at baseline are shown in [Supplementary Table 1](http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc10-0644/-/DC1). At baseline, the Pearson correlation coefficients were 0.759 between A1C level and FPG and 0.673 between A1C level and 2-h plasma glucose (all *P* \< 0.001). ###### Baseline characteristics of men and women who developed or did not develop diabetes at 6 years Men Women --------------------------------------- ----------------- ---------------- --------- ----------------- ----------------- --------- *n* 2,328 478 2,722 417 Age (years) 51.1 ± 8.4 52.5 ± 8.7 51.6 ± 8.7 54.1 ± 8.8 BMI (kg/m^2^) 24.1 ± 2.8 24.8 ± 3.1 \<0.001 24.6 ± 3.1 26.0 ± 3.3 \<0.001 Waist circumference (cm) 83 ± 7 85 ± 8 \<0.001 81 ± 9 85 ± 10 \<0.001 Systolic blood pressure (mmHg) 116 ± 16 121 ± 17 \<0.001 115 ± 18 123 ± 20 \<0.001 Diastolic blood pressure (mmHg) 76 ± 11 78 ± 11 \<0.001 73 ± 11 77 ± 12 \<0.001 FPG (mmol/L) 4.7 ± 0.5 5.1 ± 0.6 \<0.001 4.6 ± 0.4 4.9 ± 0.6 \<0.001 2-h glucose (mmol/L) 6.1 ± 1.6 8.0 ± 1.9 \<0.001 6.6 ± 1.5 8.4 ± 1.6 \<0.001 A1C (%) 5.3 ± 0.3 5.6 ± 0.5 \<0.001 5.3 ± 0.3 5.8 ± 0.5 \<0.001 Fasting insulin (pmol/L) 35.8 ×/÷ 25.4 38.3 ×/÷ 28.1 0.016 41.1 ×/÷ 30.6 46.1 ×/÷ 27.2 \<0.001 HOMA-IR 1.2 ×/÷ 0.9 1.4 ×/÷ 1.1 \<0.001 1.4 ×/÷ 1.1 1.7 ×/÷ 1.0 \<0.001 HOMA-β 105.3 ×/÷ 123.4 84.5 ×/÷ 223.2 \<0.001 139.6 ×/÷ 142.2 120.9 ×/÷ 150.0 \<0.001 Total cholesterol (mmol/L) 5.0 ± 0.9 5.1 ± 1.0 \<0.001 4.9 ± 0.9 5.2 ± 0.9 \<0.001 HDL cholesterol (mmol/L) 1.2 ± 0.3 1.1 ± 0.3 0.026 1.2 ± 0.3 1.2 ± 0.3 \<0.001 Triglycerides (mmol/L) 1.6 ×/÷ 1.2 1.9 ×/÷ 1.2 \<0.001 1.3 ×/÷ 0.8 1.8 ×/÷ 1.1 \<0.001 hsCRP (mg/dL) 0.12 ×/÷ 0.57 0.14 ×/÷ 0.42 0.005 0.10 ×/÷ 0.60 0.15 ×/÷ 0.25 \<0.001 Serum creatinine (mg/dL) 1.0 ± 0.2 1.0 ± 0.2 0.072 0.73 ± 0.16 0.73 ± 0.13 0.898 Hypertension (%) 8.6 14.9 \<0.001 10.9 25.2 \<0.001 Family history of diabetes (%) 9.2 14.0 \<0.001 10.9 17.5 \<0.001 Smoker (%) 46.1 48.2 0.319 2.4 5.6 0.001 Living in urban area (Ansan) (%) 50.1 61.7 \<0.001 45.2 51.3 \<0.001 Sporting activity (≥1 per week) (%) 39.2 38.7 0.929 32.9 34.3 0.150 Alcohol intake (≥60 Kcal per day) (%) 45.2 49.3 0.053 2.8 4.1 0.042 Child with birth weight \>4 kg (%) --- --- --- 11.2 12.3 0.338 Data are means ± SD, geometric mean ×/÷ SD, or column percentage. Comparisons are adjusted for age. [Table 1](#T1){ref-type="table"} showed the different characteristics between diabetic converters versus nondiabetic subjects. Over 6 years, 895 (10.2%) subjects developed new type 2 diabetes, and the mean follow-up periods were 5.68 ± 0.99 years. After adjusting for age, BMI, waist circumference, blood pressure, FPG, and 2-h plasma glucose, A1C level, fasting insulin, HOMA-IR, total cholesterol, triglycerides, and hsCRP concentrations were higher in those who developed diabetes. In both sexes, a family history of diabetes, hypertension, and urban residence (Ansan) were more frequent in the incident diabetic group. A1C cutoff for detecting undiagnosed diabetes and predicting progression to diabetes {#s10} ------------------------------------------------------------------------------------ [Table 2](#T2){ref-type="table"} shows the sensitivity, specificity, and positive and negative predictive values of A1C level for detecting undiagnosed diabetes and predicting 6-year incident diabetes at A1C cutoff values of 5.0--6.6%. For detecting undiagnosed diabetes, an A1C cutoff of 5.9% produced the maximum sum of sensitivity (68%) and specificity (91%) by ROC analysis. The positive and negative predictive values of this cut point were 34 and 98%, respectively. ###### Sensitivity, specificity, and positive and negative predictive value of increasing A1C cutoff levels for detecting undiagnosed diabetes and for predicting the incidence of type 2 diabetes at the 6-year follow-up A1C cutoff (%) Baseline undiagnosed diabetes Incident diabetes after 6 years of follow-up -------------------------------------- ------------------------------- ---------------------------------------------- ----------- ----------- ----------- ----------- ----------- ----------- 5.0 (−1.00 SDs above normal mean) 0.972 0.115 0.074 0.982 0.962 0.121 0.162 0.947 5.1 (−0.75 SDs above normal mean) 0.956 0.185 0.079 0.98 0.935 0.198 0.171 0.945 5.2 (−0.50 SDs above normal mean) 0.945 0.279 0.087 0.986 0.886 0.302 0.184 0.937 5.3 (−0.25 SDs above normal mean) 0.915 0.390 0.098 0.984 0.827 0.419 0.201 0.932 5.4 (0.00 SDs above normal mean) 0.887 0.506 0.115 0.984 0.768 0.547 0.231 0.930 5.5 (0.25 SDs above normal mean) 0.866 0.616 0.141 0.984 0.682 0.665 0.265 0.922 **5.6 (0.50 SDs above normal mean)** 0.822 0.717 0.174 0.982 **0.594** **0.769** **0.313** **0.914** 5.7 (0.75 SDs above normal mean) 0.770 0.797 0.216 0.979 0.508 0.847 0.370 0.907 5.8 (1.00 SDs above normal mean) 0.720 0.862 0.274 0.977 0.420 0.908 0.448 0.898 **5.9 (1.25 SDs above normal mean)** **0.676** **0.907** **0.344** **0.975** 0.333 0.947 0.527 0.889 6.0 (1.50 SDs above normal mean) 0.619 0.935 0.411 0.971 0.263 0.967 0.586 0.881 6.2 (2.00 SDs above normal mean) 0.523 0.968 0.544 0.965 0.152 0.987 0.677 0.868 6.6 (3.00 SDs above normal mean) 0.372 0.992 0.771 0.956 0.051 0.999 0.885 0.856 For predicting incident diabetes at 6 years, an A1C level of 5.6% was the optimal cutoff; the sensitivity, specificity, and positive and negative predictive values of this cut point were 59, 77, 31, and 91%, respectively. To test the A1C cut points to predict future diabetes, we tested reliability by randomly dividing our cohort into the two groups. Half of the cohort was used to define the cut point and the other half to test reliability by calculating the incidence and adjusted RRs. The incidences and RRs were 31.7, 37.6, 46.5, 53.3, 58.9, 67.6, and 89.7% and 4.9, 5.9, 8, 9.4, 10.8, 13.8, and 51.5 at the A1C cutoff of 5.6, 5.7, 5.8, 5.9, 6, 6.2, and 6.6%, respectively. A1C level and prediction of new-onset diabetes {#s11} ---------------------------------------------- The RR of new-onset diabetes in subjects whose A1C levels were above or below 5.6% was assessed using the Cox proportional hazards model ([Table 3](#T3){ref-type="table"}). In the age-adjusted model (model A), an A1C cutoff ≥5.6% predicted incident diabetes in both men and women with a RR of 3.4 (95% CI 2.9--4.1) in men and 4.6 (3.7--5.7) in women (both *P* \< 0.001). This increased risk remained after additional adjusting for other confounding factors (model C). ###### The RR of incident type 2 diabetes at the 6-year follow-up in Cox proportional hazards models based on A1C status at baseline Men Women ------------------------------------------------------- ------------------- --------- -------------------- --------- A1C ≥5.6% (vs. \<5.6%) in the entire study population  Model A[\*](#t3n1){ref-type="table-fn"} 3.44 (2.87--4.13) \<0.001 4.60 (3.75--5.66) \<0.001  Model B[†](#t3n2){ref-type="table-fn"} 3.17 (2.62--3.84) \<0.001 4.00 (3.24--4.95) \<0.001  Model C[‡](#t3n3){ref-type="table-fn"} 2.41 (1.98--2.93) \<0.001 3.06 (2.46--3.81) \<0.001 A1C ≥5.8% (vs. \<5.8%) in subjects with IFG  Model A[\*](#t3n1){ref-type="table-fn"} 3.15 (2.13--4.64) \<0.001 6.29 (3.03--13.05) \<0.001  Model B[†](#t3n2){ref-type="table-fn"} 3.57 (2.36--5.41) \<0.001 5.99 (2.83--12.66) \<0.001  Model C[‡](#t3n3){ref-type="table-fn"} 3.47 (2.27--5.29) \<0.001 5.15 (2.39--11.11) \<0.001 There was significant interaction between A1C and sex. The interaction between A1C and FPG also was significant. \*Age adjusted. †Model A and waist circumference, family history of diabetes, living in urban area, hypertension, smoking, and alcohol intake were adjusted. ‡Model B and triglycerides (log), HDL cholesterol, HOMA-IR (log), HOMA-β (log), and hsCRP (log) were adjusted. Because the A1C level displayed a significant interaction with sex and FPG concentration, we stratified by sex and performed subgroup analysis in the subjects with impaired fasting glucose (IFG) ([Table 3](#T3){ref-type="table"}; [Supplementary Table 2](http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc10-0644/-/DC1)). A total of 457 participants had IFG at baseline, and 138 subjects developed incident diabetes during the 6 years. In IFG group at baseline, an A1C level of 5.8% produced the highest sum of sensitivity and specificity for predicting new-onset diabetes at 6 years ([Supplementary Table 3](http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc10-0644/-/DC1)). After multivariate adjustment, those with an A1C level ≥5.8% had a 3.5-fold increased risk of incident diabetes in men and 5.2-fold increased risk in women. The RR of incident diabetes increased as baseline A1C level increased in both the group with normal glucose tolerance and with IFG, respectively ([Supplementary Table 2](http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc10-0644/-/DC1)). ROC curves {#s12} ---------- [Figure 1](#F1){ref-type="fig"} shows the ROC curves representing the sensitivity and specificity of the A1C levels in detecting undiagnosed diabetes ([Fig. 1*A*](#F1){ref-type="fig"}) and predicting new-onset diabetes ([Fig. 1*B*](#F1){ref-type="fig"}) at each possible A1C cutoff level. The analysis indicated a high predictive value for A1C level in screening for undiagnosed diabetes and in predicting future diabetes. {#F1} For the identification of undiagnosed diabetes in the entire study population of 9,375 subjects, the areas under the curve for A1C level were similar with that for FPG concentration (0.85 \[95% CI 0.84--0.87\] vs. 0.88 \[0.86--0.89\]; *P* = 0.14). The optimal FPG cutoff for predicting undiagnosed diabetes was 5.5 mmol/L (99 mg/dL), with 70% sensitivity and 94% specificity ([Fig. 1*A*](#F1){ref-type="fig"}, *dotted line*). For predicting new-onset diabetes after the 6-year follow-up, the area under the curve for A1C level was significantly greater than that for FPG concentration (0.74 \[95% CI 0.72--0.76\] vs. 0.69 \[0.67--0.71\]; *P* \< 0.001). For FPG concentrations, the cutoff value of 4.8 mmol/L (87 mg/dL) yielded the maximum sum of sensitivity (62%) and specificity (67%) in predicting new-onset diabetes ([Fig. 1*B*](#F1){ref-type="fig"}, *dotted line*). CONCLUSIONS {#s13} =========== The main finding of this study is that the A1C assay was useful as a screening test for type 2 diabetes and as a predictor of future diabetes. In our population, an A1C cutoff of 5.9% was able to identify people with undiagnosed diabetes, and individuals with an A1C ≥5.6% had an increased risk for progression to type 2 diabetes independent of other confounding factors. This was a large, prospective cohort study that used stringent criteria to diagnose diabetes and to evaluate the usefulness of A1C level in diabetes screening and in the prediction of new-onset diabetes. In this homogeneous population-based study, we applied the OGTT to all participants and used the same instruments and personnel for all clinical and biochemical assessments during the 6 years. Use of the A1C level in the diagnosis of or screening for diabetes has been debated for many years. Most A1C assays, such as the National Glycohemoglobin Standardization Program ([@B18]), are standardized, and recent expert committee reports suggest an A1C cutoff of 6.5% for diagnosing diabetes ([@B11]). For screening a general population, the A1C level has several advantages over the currently used FPG concentration or 2-h glucose concentration after an OGTT. The A1C assay does not need a fasting or timed sample. It is a better indicator of chronic glycemic level, has less preanalytic instability ([@B19]), and has a more consistent relationship with diabetic microvascular complications than does FPG concentration ([@B20],[@B21]). However, concerns remain about the risk of underdiagnosing people with overt diabetes when using an A1C cutoff of 6.5% ([@B12]). Several cross-sectional studies have evaluated the accuracy of the A1C cutoffs in screening for diabetes. In analysis of the National Health and Nutrition Examination Survey data, Buell et al. ([@B9]) reported that an A1C level of 5.8% showed the highest sensitivity (86%) and specificity (92%) in identifying undiagnosed diabetes when using FPG concentration as the diagnostic test for type 2 diabetes. In the current study, the definition of diabetes was based on plasma glucose results from the 75-g OGTT, and the A1C value of 5.9% was appropriate for detecting undiagnosed type 2 diabetes in this Korean cohort population. In a Japanese study of OGTT results in 1,904 people, an A1C cut point of 5.6% identified undiagnosed type 2 diabetes, and this value is used as a supplementary diagnostic criterion by the Japanese Diabetes Society ([@B10]). Only a few studies have investigated the utility of A1C level in predicting new-onset diabetes. Recent Japanese and French cohort studies reported that A1C level is effective in predicting type 2 diabetes ([@B22],[@B23]) but was less sensitive and specific than FPG concentration for predicting FPG-defined diabetes ([@B22]). This might be because many people with an abnormal 2-h glucose concentration after an OGTT have a normal FPG concentration ([@B24]). We used OGTT to define diabetes and found that A1C level was independently related to an increased risk of new-onset diabetes, even in those with IFG at baseline. The predictive value of the A1C level was greater than that of the FPG concentration. Determining the optimal A1C cutoff for diabetes screening is somewhat arbitrary because the risk of diabetes is continuous over a range of glycemic measures. To maximize the diagnostic efficiency, the optimal A1C cutoff should be considered in balancing both sensitivity and specificity. Despite the A1C cutoff value of 5.6% for identifying individuals with increased risk of future diabetes, as was chosen by the Youden Index, it showed only 31% of the positive predictive value. However, we considered the clinical situation because diabetes is a common disease and the action for prevention is highly beneficial and does relatively little harm to healthy subjects. Our study has some limitations. All participants were enrolled from Korean rural and urban communities of homogeneous ethnic background, and it is debatable whether these results can be generalized. Although racial differences in A1C level have been suggested ([@B25]), the significance of any differences is not clear, and the use of different A1C values according to ethnicity is not currently recommended. However, after multivariate adjustment of confounders, A1C level remained as an independent predictor of incident diabetes. In addition, the stringency of our study method and prospective follow-up of a large community-based cohort for 6 years make our results stronger than those of other studies. In conclusion, we found that A1C level was effective and convenient for diabetes screening. An A1C cutoff of 5.9% may identify a high proportion of people with undiagnosed type 2 diabetes. Individuals with A1C ≥5.6% have an increased risk for future diabetes, and early preventive intervention could be helpful. Supplementary Material ====================== ###### Supplementary Data This article contains Supplementary Data online at <http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc10-0644/-/DC1>. This study was supported by the National Genome Research Institute, Korean Center for Disease Control and Prevention (contract nos. 2001-2003-348-6111-221, 2004-347-6111-213, and 2005-347-2400-2440-215). The funding source had no role in the collection of data or in the decision to submit the manuscript for publication. No potential conflicts of interest relevant to this article were reported. S.H.C. wrote and edited the manuscript. T.H.K. performed the statistical analysis and wrote the manuscript. S.L. contributed to discussion and statistical analysis. K.S.P. and H.C.J. reviewed and edited the manuscript. N.H.C. was the principal investigator of the project, collected and researched data, designed and performed the cohort study, and wrote the manuscript. [^1]: S.H.C. and T.H.K. contributed equally to this work.

Background {#Sec1} ========== Interstitial microdeletions of the short arm of chromosome 1 are rare and to date at least 19 cases have been reported \[[@CR1]--[@CR17]\] (Table [1](#Tab1){ref-type="table"}). It is thought that many genes, which are critical for early development, reside within the short arm of chromosome 1 \[[@CR4]\]. These heterozygous deletions, in some cases, might have resulted in the eventual death of the affected individuals \[[@CR3], [@CR7], [@CR9], [@CR11]\]. Patients with 1p interstitial microdeletions display a number of clinical features including intellectual disability, craniofacial anomalies, seizures, tooth abnormalities, urinary tract anomalies, and skeletal, cardiac as well as limb defects \[[@CR3], [@CR4], [@CR7], [@CR12], [@CR16], [@CR17]\].Table 1Patients with 1p interstitial microdeletionsPatientKaryotypeSourcePhenotypeMethods and coordinates1del(1)(p21p32)\[[@CR1]\]^a^1979Severe psychomotor retardation, craniofacial and skeletal anomalies, short stature, overweight.Karyotyping2del(1)(p22.1p31.2)\[[@CR2]\]^a^1987Psychomotor retardation, craniofacial and limb anomalies.Karyotyping3del(1)(p22.1p31.2)\[[@CR3]\]^a^1991 Pt 1Developmental delay, intellectual disability, craniofacial anomalies.Karyotyping4del(1)(p22.3p31.3)\[[@CR3]\]^a^1991 Pt 2Developmental delay, seizures craniofacial and limb anomalies.Karyotyping5del(1)(p32.3p34.1)\[[@CR4]\]^a^1991Mental and motor developmental delay, craniofacial and limb anomalies, hypotonia.Karyotyping6del(1)(p36.1p36.2)\[[@CR5]\]^a^1993Developmental delay, craniofacial anomalies, neuroblastoma.Karyotyping7del(1)(p32.1p32.3)\[[@CR6]\]^a^1995Developmental delay, craniofacial and limb anomalies.Karyotyping8del(1)(p21p22.3)\[[@CR7]\]^a^1997Craniofacial and limb anomalies, congenital heart disease, rib abnormalities.Karyotyping9del(1)(p34.1p34.3)\[[@CR8]\]^a^1999 Pt 1Severe learning disability, attention deficit disorder (ADD), craniofacial anomalies.Karyotyping and FISH with probes specific for chromosome 1 (COATA-SOME^TM^1, p5205-DG.5)10del(1)(p34.1p34.3)\[[@CR8]\]^a^1999 Pt 2Attention deficit hyperactivity disorder (ADHD), craniofacial anomalies, disturbed behaviors.Karyotyping and FISH with probes specific for chromosome 1 (COATA SOME^TM^1, p5205-DG.5)11del(1)(p32.1p32.3)\[[@CR9]\]^a^ 2002 Pt 1Global developmental delay, craniofacial anomalies, absence of corpus callosum, type I Chiari malformation, tethered cord.Karyotyping and FISH with probes specific for whole chromosome painting12del(1)(p32.1p32.3)\[[@CR9]\]^a^2002 Pt 2Intraventricular hemorrhage, seizures, thin corpus callosum, limb anomaliesKaryotyping and FISH with probes specific for whole chromosome painting13del(1)(p32.1p32.3)\[[@CR10]\]^a^2003Delayed psychomotor development, craniofacial anomaliesKaryotyping and FISH using whole chromosome 1 painting probe (wcp1)14del(1)(p36)\[[@CR11]\]^a^2004Craniofacial anomalies, moderate intellectual disability, seizures.FISH with TelVysion 1p (Vysis), P5124 (Oncor), a YAC 273d11 (CEPH)15del(1)(p31.3p32.2)\[[@CR12]\] 2010Craniofacial anomalies, hypoplasia of corpus callosum, ventriculomegaly, hypotonia105K Oligoarray CGH & BAC array CGH, 4.93 Mb deletion, 1p32.2p31.3 (chr1:58,193,565-63,125,273)X1 *dn*, hg1916del(1)(p36.3)\[[@CR13]\]^a^ 2010Prader-Willi like features.MLPA, FISH with BAC probes, real-time q-PCR17del(1)(p31.1p32.2)\[[@CR14]\] 2014Craniofacial anomalies, partially hypoplastic corpus callosum, mild ventriculomegaly, intraparenchymal hemorrhages, cerebral palsy.SNP- microarray (SNP-CMA), 22.9 Mb deletion (chr1:55,113,975-77,992,492)X1, hg1918del(1)(p36.3)\[[@CR16]\] 2008Developmental delay, facial dysmorphisms, neuroblastomaMLPA, 244 K oligo microarray, a deletion of 1.59 Mb at 1p36.3 and a duplication of 3.26 Mb at 1p36.3 (1,741,058-5,004,693)X3, hg18, FISH with Vysis 1p36 and BACs19^b^del(1)(p31.3p32.1)\[[@CR17]\] 2007Abnormal corpus callosum, Ventriculomegaly, developmental tethered spinal cord, Chiari I malformationarray Comparative genomic hybridization (aCGH)20del(1)(p32.1p32.3)\[[@CR22]\] 2015Microcephaly, urogenital anomalies, hearing loss, choanal atresia.SNP microarray, 6.4 Mb deletion (chr1:54,668,618-61,113,264)X1, hg19, FISH with BACsDCP274057partial C-terminal duplication of *NFIA*DECIPHER90 kb duplication at 1p31.3, global developmental delayDCP260253intragenic deletion of *NFIA*DECIPHER117 kb deletion at 1p31.3, phenotype not availableDCP285169intragenic duplication of *NFIA*DECIPHER419 kb duplication paternally inherited at 1p31.3, additional 2.07 Mb maternal deletion at 15q13, delayed fine motor development, expressive language delay, impaired social interactions, receptive language delayDCP288170intragenic *NFIA* deletionDECIPHER229 kb deletion at 1p31.3, intellectual disability, macrocephalyDCP300407del(1)p32.1DECIPHER281 kb deletion at 1p32.1, cognitive impairmentDCP264827del(1)p31.3DECIPHER5.43 Mb deletion at 1p31.3, abnormally folded helix, ADHD, constipation, flat forehead, global developmental delay, hypothyroidism, malar flatteningDCP276512del(1)p31.3p32.1DECIPHER8.83 Mb deletion at 1p31.3-1p32.1, delayed speech and language^a^ Microdeletions without genomic coordinates reported in the pre-microarray era. Pt denotes patient^b^Pt 19 has additionally a chromosomal translocation 46,XY,t(1;3)(p22;q21)dn Revisiting CNV (copy number variation) cases reported during the pre-microarray era is uncommon, because the patients are deceased or lost to follow-up. Re-evaluation of earlier cases using microarray analysis, however, provides more accurate information on the candidate genes contained within the CNV regions. Furthermore, it also allows monitoring of any changes in the patient's clinical features over time \[[@CR18]\]. More than a decade ago, an apparent interstitial microdeletion at 1p32.1p32.3 was reported in a 10-year old girl with delayed psychomotor development and facial dysmorphism. This *de novo* structural chromosomal rearrangement was detected by karyotype analysis alone followed by fluorescent *in situ* hybrization (FISH) without defining the exact size and location of the microdeletion \[[@CR10]\]. Here we revisited this proband to revise the karyotype and to define the size of this microdeletion by molecular analysis. From molecular dissection of our case with six published cases of microdeletions, a balanced translocation, and seven unpublished DECIPHER cases \[[@CR23]\], we propose that there are at least six chromosomal segments within this region, and each segment harbors at least one candidate gene for intellectual disability. We also determined the transcript levels of six candidate genes (*DAB1*, *HOOK1*, *NFIA, DOCK7*, *DNAJC6* and *PDE4B*) for syndromic intellectual disability. Methods {#Sec2} ======= Case presentation {#Sec3} ----------------- The proband (DGDP005) is a 23-year-old female with a history of developmental delay and intellectual disability. In infancy, developmental milestones were delayed and she exhibited impaired motor skills (Table [2](#Tab2){ref-type="table"}). The patient displayed craniofacial anomalies including macrocephaly (Fig. [1a, b, d-g](#Fig1){ref-type="fig"}), frontal bossing (Fig. [1b](#Fig1){ref-type="fig"}) and low-set ears (Fig. [1d](#Fig1){ref-type="fig"}). She was diagnosed with attention deficit hyperactivity disorder (ADHD), obsessive compulsive disorder (OCD), ocular hypertension (OHT), and developmental encephalopathy with cognitive impairment (Table [2](#Tab2){ref-type="table"}).Table 2Timescale of DGDP005's medical evaluations as well as her clinical featuresAgeEvaluationClinical features\<5 yearsPediatrician- Crawling, sitting, walking and attaining language milestones delayed- Craniofacial anomalies including macrocephaly- Problematic motor skills.\~10 yearsDevelopmental pediatrician- Special education implemented in curriculum- Difficulty falling asleep and treatment with melatonin- ADHD- Hypertonia.- Possible sensory motor difficulties.14 yearsStandford Binet Intelligence Scales-Fifth edition (SB5)- Difficulty staying on the task and needed to be redirected from time to time- Overall thinking and reasoning at \~13 percentile.- Borderline nonverbal reasoning abilities- Verbal reasoning at \~12 percentile- Verbal and nonverbal problem solving at \~ 58 percentile- Ability to gather information at \~ 18 percentile- Numerical problem solving at \~7 percentile.- Borderline visual display abilities- Ability to maintain attention at \~13 percentile.- Full scale quotient of 83 (95 % confidence interval 79--87) suggesting intellectual disability.16 yearsIndividualized education program review team- OHT-ADHD- Language disability in written language expression- Disability in math calculation and reasoning- Difficulties in fine graph motor functions with soft neurological signs.- Continues to be a student with disability18 yearsChild neurology services- Developmental encephalopathy- Cognitive impairment- OCD as well as ADHD-Treatment with fluoxetine (aka Prozac10 mg) and Adderall (15 mg)- More pronounced facial dysmorphisms including prominent forehead, low-set ears, narrow nose thin lips22 yearsAdmitted to emergency- Diagnosed with intraventricular hemorrhage with layering in the posterior fossa.- Right anterior communicating artery aneurysm- Developmental delay, OCD and ADHDFig. 1Phenotypic features of DGDP005. **a** Facial and head appearance including macrocephaly and upper body picture at 4 years (**b**) Lateral facial feature of 4 years of age showing frontal bossing (**c**) Full body picture as a 7 years old (**d**) upper body and head appearance presenting with macrocephaly at 12 (**e**) 13 (**f**) 16 and (**g**) 18 years of age At age 10, her grades were Bs and Cs, but school challenges warranted special education. She also had difficulty falling asleep and she took nightly melatonin. She was diagnosed with ADHD, hypertonia, and possible sensory integration difficulties (further details on her phenotype up to age10 can be found in Zinner and Batanian (2003) \[[@CR10]\]). At 14 years of age, she took the Stanford Binet Intelligence Scales-Fifth edition (SB5). She had difficulty staying on a task requiring redirection. Her general cognitive ability was estimated to be in the low average range with a full scale intelligence quotient (FSIQ) score of 83 (95 % confidence interval 79--87), suggesting intellectual disability. She received a score of 79 in the nonverbal reasoning abilities, which is in the borderline range, and higher than only 8 % of her peers. In verbal reasoning abilities, she scored 88, which is in the low average, and just above \~12 % of her peers. Her ability to solve verbal and nonverbal problems using inductive or deductive reasoning was average. Her overall thinking and reasoning was worse than \~87 % of children of her age. She scored in the low average range for general information gathering, attention span, and concentration. Her ability to see patterns, relationships, and spatial orientations among diverse pieces of visual display, as well as her ability with numbers and numerical problem solving with word problems or with picture relationships were borderline. At age 16, she was found to display OHT and ADHD, language disability in written language expression as well as in math calculation and math reasoning. She passed the vision test scoring 20/10 and 20/20 for the right and left eye respectively. Her hearing was adequate for the classroom functioning. She showed significant difficulties in fine motor functions with soft neurological signs. Overall, the assessment showed that she continued to be a student with intellectual disability and ADHD. When she was 18, the patient was re-evaluated by the child neurology services. She was diagnosed with developmental encephalopathy with cognitive impairment, OCD and ADHD. She was treated with fluoxetine and Adderall for OCD and ADHD, respectively. Some of her facial dysmorphisms became more evident at age 18, such as her prominent forehead, narrow nose, and thin lips (Fig. [1](#Fig1){ref-type="fig"}). She still displayed low-set ears (Fig. [1d](#Fig1){ref-type="fig"}). At the age of 22, she was admitted to the hospital and diagnosed with subarachnoid as well as intraventricular hemorrhage with layering in the posterior fossa. She was also diagnosed with a right anterior communicating artery aneurysm and continued to show developmental delay, OCD and ADHD. Cell culture {#Sec4} ------------ Lymphocytes were isolated from the patient's blood samples by density gradient centrifugation as we have done previously \[[@CR19]\]. Lymphoblastoid cell line (LCL) was generated from the patient's blood in order to determine the transcript levels of genes of interests. Genomic DNA extraction and microarray {#Sec5} ------------------------------------- Genomic DNA was extracted from the patient's blood samples using a standard phenol-chloroform protocol with minor modifications. The human genome 244 K array (Agilent technologies, G4411B) was used to detect copy number variation as described previously in Miller et al. (2012) \[[@CR20]\]. Real-Time genomic qPCR and RT-qPCR {#Sec6} ---------------------------------- Primers targeting regions encompassing the proximal and distal deletion breakpoints were designed for qPCR. Primer pairs were first evaluated by determining the primer efficiency using a serial dilution of genomic DNA as template. Primers were also designed against exonic regions of genes of interests for RT-qPCR. Total RNA was isolated from cell lines established from patient DGDP005 using the RNeasy Plus Mini kit (Qiagen) following the manufacturer's instructions. cDNA was synthesized from 1 μg of total RNA using the RevertAid First cDNA Synthesis Kit (Thermo Scientific) according to the manufacturer's protocol. All Real-Time qPCR reactions were carried out using 2 μl of sample, 10 ml of Fast Essential DNA Green Master (Roche) and 2.5 μM primers in a total reaction volume of 20 μl. Results {#Sec7} ======= Microarray {#Sec8} ---------- Microarray performed on genomic DNA derived from patient DGDP005 revealed a 9.45 Mb microdeletion at 1p31.3p32.2 (chr1: 57,633,718- 67,087,056, GRCh38/hg38). The deleted chromosomal region contains at least 35 genes, including *NFIA* (Fig. [2](#Fig2){ref-type="fig"}).Fig. 2Comparative deletion mapping of patients with CNVs at 1p31.3p32.2. The genes located within this interval are displayed. Microdeletions are represented by solid red bars while microduplications are displayed in blue. Patients with CNVs from the DECIPHER database (Firth et al., 2009) are denoted by 'DCP' followed by the reference number. The sizes of the deleted regions in Rao 2014, Lu 2007 (DGAP174), Koehler 2010, Kehrer 2015, Campbell 2002 (DGAP205-1), Ji 2014 and seven DECIPHER cases relative to our patient (DGDP005) are displayed. Four CNVs (DCPP274057, DCP260253, DCP288170, Rao 2014) have only *NFIA* disrupted implying that it is causative gene in all four patients. Patient DCP285169 with a 419 kb duplication including the second half of *NFIA* presents with language delay, impaired social interactions, and delayed motor development. But he also has an additional 2.07 Mb microdeletion at 15q13. The microdeletion in patient DCP300407 does not include *NFIA*, yet the patient displays cognitive impairment, suggesting the likely contribution of *HOOK1*. Two microdeletions (Ji 2014 and Campbell 2002) extend beyond the 1p32.2 and 1p31.3 interval. Vertical lines in blue represent the proximal and distal boundaries of the microdeletion in DGDP005. Two orange lines are refined candidate gene region overlapping eight microdeletions including DCP300407. Two vertical black lines depict narrowed candidate region among 5 small CNVs \[4 DCP cases and patient of Rao et al. (2014)\]. Asterisk denotes candidate genes in this region Comparative deletion mapping {#Sec9} ---------------------------- We compared all clinical features displayed by DGDP005 with five previously reported microdeletion cases namely, Campbell et al. (2002) \[[@CR9]\], Koehler et al. (2010) \[[@CR12]\], Ji et al. (2014) \[[@CR14]\], Lu et al. (2007) \[[@CR17]\], Rao et al. (2014) \[[@CR21]\], Kehrer et al. (2015) \[[@CR22]\] at 1p31.1p32.2 overlapping our microdeletion at 1p31.3p32.2, as well as one translocation case involving the *NFIA* gene in Lu et al. (2007) \[[@CR17]\] (Table [3](#Tab3){ref-type="table"}). We also compared the microdeletion in DGDP005 to seven cases from the DECIPHER database \[[@CR24]\] (Fig. [2](#Fig2){ref-type="fig"}, Table [1](#Tab1){ref-type="table"}). The refined candidate gene region represented by DCP300407 and proximal breakpoint of DCP276512 was depicted in two orange lines (Fig. [2](#Fig2){ref-type="fig"}), while two vertical black lines depict narrowed candidate region among 5 small CNVs including 4 DCP cases and a microdeletion of Rao et al. (2014) \[[@CR21]\]. We examined the functions of all genes involved in the microdeletion by reviewing the literature to highlight candidate genes that could contribute to phenotypes observed in DGDP005 (Table [4](#Tab4){ref-type="table"}). Patient DGAP174 of Lu et al. (2007) was not included in determining genotype-phenotype relationships (Table [3](#Tab3){ref-type="table"}), because of an accompanying chromosomal translocation t(1;3) (p31.1; q25.1) \[[@CR17]\].Table 3Clinical features of DGDP005 along with five additional microdeletion cases and a balanced translocation t(1;20)(p31.3;q13.31)*dn* disrupting the *NFIA* geneClinical featuresDGDP005 del(1)p31.3 p32.2Koehler et al. 2010 del(1)p31.3 p32.2Campbell et al. 2002 DGAP205-1 & DGAP205-1S del(1)p31.3 p32.3Ji et al.2014 del(1)p31.1 p32.2Rao et. al 2014 del(1) p31.3Lu et al., 2007 DGAP174^a^del(1)(p31.3 p32.1)Lu et al., 2007 DGAP104 Balanced translocation t(1;20) (p31.3;q13.31)*dn*Developmental delay+++++++Intellectual disability+NS+NSNS++Macrocephaly+++++++Frontal bossing+-+NSNSNSNSDevelopmental encephalopathy+\-\-\-\-\--OHT+\-\-\-\-\--Intraventricular hemorrhage+-++\-\--Impaired motor skills+NS+NSNS++Attention deficit disorder+NS-NSNS++Hypertonia+NSNSNSNS-NSOCD+-+-NS\--Seizures\--++-NSNSAbnormal corpus callosumNS++++++Ventriculomegaly-+++++-Tethered spinal cords\--+\--++Chiari I malformation-NS+\--++Urinary tract defects\--+++NS-NS: Not Stated; Pt: Patient^a^DGAP174 also has an additional chromosome translocation, 46,XY, t(1:3)(p31.1;q25.1)*dn*Table 4Functions of candidate genes possibly contributing to clinical phenotypes observed in patient DGDP005Gene nameGene symbolAgeAssociation with neuro-developmental featuresRemarksNuclear factor I/A*NFIA*\[[@CR16]\] \~2 yrsAbnormal corpus callosum, ventriculomegaly, hydrocephalus, developmental delay, tethered spinal cord, chiari I malformation, and urinary track defectDisruption of *Nfia* in mice result in severe developmental defects including agenesis of the corpus callosum, severe communicating hydrocephalus, neurological defects, male sterility, female subfertility \[[@CR24]\]\[[@CR21]\] 8 yrsDab, reelin signal transducer homolog 1*DAB1*\[[@CR37]\] \~7.5 yrsAutismHook homolog 1(Drosophila)*HOOK1*\[[@CR23]\] 5 yrsCognitive impairment (DCP300407)Cytosolic protein possessing a conserved N-terminal domain that binds to microtubules. Interacts with CLN3, the causative gene for autosomal recessive Batten disease.Dedicator of cytokinesis 7*DOCK7*\[[@CR38]\] 5--7 yrsEpileptic encephalopathy, dysmorphic features and intellectual disability*DOCK7* plays a role in neural developmentDnaJ (Hsp40) homolog, C, member 6*DNAJC6*\[[@CR31]\] 17--44 yrsParkinson disease Subfamily-Phosphodiesterase 4B, cAMP-specific*PDE4B*\[[@CR35]\] NSSchizophrenia-NS; Not stated Real-Time qPCR {#Sec10} -------------- We performed qPCR to refine the flanking coordinates of the deletion breakpoints. qPCR assays using primers designed from the *DAB1* gene revealed that the distal deletion breakpoint is located between *DAB1* exon 1 and its 5′-UTR. The *SLC35D1* gene was found to be completely deleted as predicted by microarray (Figs. [2](#Fig2){ref-type="fig"} and [3](#Fig3){ref-type="fig"}). The proximal deletion breakpoint was found to reside in the intergenic region between gene *SLC35D1* and uncharacterized gene *C1orf141* (Figs. [2](#Fig2){ref-type="fig"} and [3](#Fig3){ref-type="fig"}).Fig. 3Refining the deletion breakpoints in DGDP005 by qPCR. The 5′-UTR of the *DAB1* gene is deleted while its first exon is intact indicating that the distal deletion breakpoint lies between *DAB1* exon 1 and the 5′-UTR. The *NFIA* gene and *SLC35D1* are both completely deleted in our patient. Two separate loci located in the proximal and distal intergenic region between *SLC35D1* and *C1orf141* were also assayed by qPCR. The distal intergenic region residing closer to *SLC35D1* was found to be contained in the microdeletion, while the proximal intergenic region was intact. A value close to 1 indicates that the locus is not deleted, while a value near 0.5 shows deletion on another allele RT-qPCR {#Sec11} ------- Expression analysis of six genes of interest by RT-qPCR revealed that the transcript levels of *NFIA*, *DAB1* and *DNAJC6* were markedly reduced in the patient relative to a healthy white female control (Fig. [4](#Fig4){ref-type="fig"}). Transcripts derived from *HOOK1* and *DOCK7* were reduced approximately by half, while *PDE4B* transcripts decreased moderately (Fig. [4](#Fig4){ref-type="fig"}).Fig. 4Transcript levels of six candidate genes involved in CNVs were determined by RT-qPCR. The level of transcripts of *DAB1*, *NFIA* and *DNAJC6* were markedly reduced in DGDP005 relative to the healthy white female control. Transcripts derived from *HOOK1* and *DOCK7* were reduced approximately by half, while *PDE4B* transcripts decreased moderately Discussion {#Sec12} ========== The female patient (DGDP005) examined in this study was first reported as a 10-year old girl possessing a rare 1p32.1p32.3 microdeletion \[[@CR10]\]. Characterization of the microdeletion in the patient by microarray allowed the refinement of the microdeletion from 1p32.1p32.3 to 1p31.3p32.2. The health condition of DGDP005 has been monitored regularly since the first few months of her life. Psychomotor delay and craniofacial anomalies were apparent since infancy \[[@CR10]\]. During the follow-up, she has continued to display developmental delay and as a result, special education has been implemented in her curriculum. Craniofacial anomalies such as macrocephaly and frontal bossing became more pronounced over time, suggesting the involvement of culprit gene(s) in continuous skull development since infancy \[[@CR10]\]. Additional features such as OCD and intraventricular hemorrhage were also observed after the age of ten, emphasizing the importance of follow-up. We have now determined the size of the deletion in DGDP005 by microarray and refined the locations of the flanking deletion breakpoints by qPCR (Figs. [2](#Fig2){ref-type="fig"} and [3](#Fig3){ref-type="fig"}). The deleted region is at least 9.45 Mb encompassing more than 35 annotated genes including *NFIA* \[[@CR17]\]. Only four smaller-sized heterozygous microdeletions (Rao 2014, Lu 2007, Koehler 2010 and Kehrer 2015) have been reported so far within this interval (Fig. [2](#Fig2){ref-type="fig"}). The patient of Koehler et al. (2010) \[[@CR12]\] had a 1p31.3p32.2 microdeletion while another patient DGAP174 of Lu et al. (2007) had a deletion at 1p31.3p32.1 with additional translocation, 46,XY, t(1;3)(p31.1;q25.1) \[[@CR17]\]. The patient of Rao et al. (2014) had a small microdeletion spanning exons 4--9 within *NFIA* \[[@CR21]\]. The distal part of the microdeletion of Kehrer overlaps with the proximal part of DGDP005 and the more refined candidate region is represented by DCP300407. Two larger microdeletions of two half siblings (DGAP205-1 and DGAP205-1S) (Table [3](#Tab3){ref-type="table"}) reported in Campbell et al. (2002) and a patient described in Ji et al. (2014) encompassing the whole chromosomal region deleted in DGDP005, have also previously been published \[[@CR9], [@CR14]\] (Fig. [2](#Fig2){ref-type="fig"}). Haploinsufficiency of the *NFIA* gene is expected to cause an abnormal corpus callosum, ventriculomegaly or hydrocephalus, developmental delay, tethered spinal cord, Chiari I malformation, seizures, urinary tract defects, and craniofacial anomalies \[[@CR17], [@CR21]\]. Furthermore, disruption of *nfia* in mouse results in severe developmental defects including agenesis of corpus callosum, severe communicating hydrocephalus, female subfertility, and male sterility \[[@CR24]\]. Five CNV cases (DCDP274057, DCP260253, DCP285169, DCP288170, Rao 2014) only had *NFIA* disrupted (Fig. [2](#Fig2){ref-type="fig"}). Among these five DECIPHER cases, DCP288170 with an intragenic deletion within *NFIA*, displayed intellectual disability. Although, this phenotype was not emphasized in Lu et al. (2007) \[[@CR17]\], DGAP104 with a balanced translocation of t(1;20)(p31.3;q13.31) disrupting *NFIA* at 1p31.3 breakpoint (Table [3](#Tab3){ref-type="table"}) also displayed intellectual disability (global IQ score of 52). Patient DCP274057 presents with global developmental delay while DCP285169 displays expressive and receptive language delay. The patient reported in Rao et al. (2014) with a deletion involving exons 4--9 of *NFIA* was only 8 years-old, and had not yet been evaluated intellectual disability \[[@CR21]\]. The aforementioned patients clearly suggest that disruption of *NFIA* is likely to cause intellectual disability, or cognitive deficits. Indeed, the *NFIA* transcript levels were markedly reduced in DGDP005 (Fig. [4](#Fig4){ref-type="fig"}) suggesting that *NFIA* may be contributing to syndromic intellectual disability. Notably, four microdeletion patients namely Campbell et al., 2002 \[[@CR9]\], Koehler et al., 2010 \[[@CR12]\], Ji et al., 2014 \[[@CR14]\] and DGDP005 had in common macrocephaly (Table [3](#Tab3){ref-type="table"}). Importantly, a balanced translocation patient DGAP104, in whom *NFIA* is disrupted at 1p31.3 \[[@CR17]\] also had macrocephaly. Furthermore, the patient reported in Rao et al. (2014) \[[@CR21]\] with a smaller microdeletion within *NFIA* also had macrocephaly (Fig. [2](#Fig2){ref-type="fig"} and Tables [1](#Tab1){ref-type="table"} and [3](#Tab3){ref-type="table"}). Collectively, these findings underscore that the haploinsufficiency of NFIA is sufficient to cause intellectual disability and macrocephaly, two of the core phenotypic features of del (1)(p31.3p32.2) microdeletion syndrome. While comparing the clinical features of our patient with the two half-siblings \[[@CR9]\] who have a larger microdeletion including *NFIA*, we found that DGDP005 does not manifest ventriculomegaly, tethered spinal cord, Chiari I malformation, or urinary tract defects. Similarly, the patient having a smaller microdeletion encompassing *NFIA* \[[@CR12]\] displayed only three of the clinical features as a result of *NFIA* deletion, namely abnormal corpus callosum, ventriculomegaly and developmental delay. Incomplete penetrance may partly explain why all phenotypes expected from *NIFA* disruption are not manifested in the patients with an *NFIA* deletion (Table [3](#Tab3){ref-type="table"} and Fig. [2](#Fig2){ref-type="fig"}) \[[@CR25]\]. We have not included DGAP174 \[[@CR17]\] in the comparison because he had a *de novo* translocation t(1;3) (p31.1; q25.1)*dn* in addition to a 2.2 Mb interstitial deletion at 1p31.3-p32.1 (Table [3](#Tab3){ref-type="table"}). Reviewing phenotypes displayed by patients with microdeletions at 1p31.3-1p32.2 enabled identification of at least 17 clinical features seen across all patients (Table [3](#Tab3){ref-type="table"}). Because of the putative chromosomal segments with corresponding candidate genes for cognitive impairment within this region, it is obvious that not all phenotypes observed can be attributed to *NFIA* alone. It is likely that genes including *NFIA* in this region are haploinsufficient thereby contributing to the 17 clinical features observed across all patients (Table [3](#Tab3){ref-type="table"}) \[[@CR17]\]. It supports the idea that the 1p31.3p32.2 microdeletion may constitute a "chromosome syndrome" postulated in Zinner and Batanian (2003) \[[@CR10]\], which meant to be contiguous gene deletion syndrome \[[@CR26]\]. Recently, Ji et al. (2014) \[[@CR14]\] also postulated the possible involvement of genes other than *NFIA*. We have reviewed the functions of all genes within the 1p31.3p32.2 microdeletion in DGDP005. Molecular dissection of this region suggests the presence of at least five chromosomal segments, with six candidate genes for intellectual disability. Each segment contains *DAB1*, *HOOK1*, *NFIA*, *DOCK7*, and *DNAJC6* as well as *PDE4B*, respectively. We propose six candidate genes (*DAB1*, *HOOK1*, *DOCK7*, *DNAJC6*, *PDE4B*, and *NFIA*) for the clinical features such as intellectual disability and OCD observed in DGDP005, because each of these genes is involved directly or indirectly in neurological disorders (Table [4](#Tab4){ref-type="table"}). While HOOK1 interacts with CLN3, the causative gene for the autosomal recessive Batten disease characterized by visual impairment, gait anomalies, seizures, dementia and sometimes mental deterioration \[[@CR27]--[@CR30]\], *DNAJC6* has been shown to be implicated in Parkinson disease \[[@CR31]\]. It is noteworthy that Batten disease is also characterized by progressive neurodegeneration and impaired motor skills \[[@CR32]\] like Parkinson disease \[[@CR33], [@CR34]\]. Furthermore, patient DCP300407 with only *HOOK1*, *CYP2J2* and *C1orf87* deleted (Fig. [2](#Fig2){ref-type="fig"}) displays cognitive impairment \[[@CR24]\]. Additionally, *PDE4B* is implicated in schizophrenia \[[@CR35]\], whereas *DAB1* truncated in our patient is required for synaptic function as well as associative learning in mice \[[@CR36]\], and has been shown to be associated with autism \[[@CR37]\]. Recently, mutation in *DOCK7* has been found to be associated with epileptic encephalopathies, dysmorphic features and intellectual disability \[[@CR38]\]. Interestingly, DGDP005 displays some of the phenotypes expected from *DOCK7* mutation indicating that *DOCK7* may also contribute to intellectual disability and craniofacial anomalies in our patient. As is the case for *DOCK7*, cognitive impairment is an overlapping clinical feature also produced by disruption in *NFIA*. This further supports the hypothesis that deletion at 1p31.3p32.2 constitutes a contiguous gene deletion syndrome. As expected, the transcript levels of *DOCK7* and *HOOK1* was reduced approximately by half. The level of transcripts of *DAB1*, *DNAJC6* and *PDE4B* were also reduced in our patient relative to a healthy white female control (Fig. [4](#Fig4){ref-type="fig"}). The significant reduction of transcripts of *NFIA* and two additional genes (*DAB1* and *DNAJC*6) suggests an allele-specific expression pattern (Fig. [4](#Fig4){ref-type="fig"}). This may be due to a loss of the allele actively transcribed in the microdeletion. It is unclear whether the causative effect of each candidate gene is stoichiometrically contributing to the manifestation of the phenotype or an effect of one gene is masked by the effect of other gene when more than one candidate gene is deleted in a microdeletion. If some of the candidate genes participate in overlapping molecular pathways leading to cognitive impairment, the effect of one gene may be masked by another. Conclusion {#Sec13} ========== Revisiting patient DGDP005 has allowed us to refine the flanking proximal and distal breakpoints, enabling us to compare the clinical phenotypes of DGDP005 to reported cases and unpublished DECIPHER cases. Comparative deletion mapping showed that there are at least six candidate genes including, *NFIA*, *HOOK1*, *DOCK7*, *DAB1*, *DNAJC6*, and *PDE4B*, that could contribute to the syndromic or non-syndromic intellectual disability. Most importantly, the genomic and clinical delineation result implies *NFIA* responsible for intellectual disability coupled with macrocephaly. Smaller-sized microdeletions across this interval and rare point mutations in the genes should reveal a better understanding of the pathological role of individual gene residing within this region. Additionally, expression studies including haploinsufficiency of aforementioned candidate genes in this interval will provide more insights both on the number of candidate genes and their impact on this contiguous gene syndrome. Consent {#Sec14} ======= This study was approved by the Institutional Review Board of Augusta University and written informed consent was obtained from the mother of DGDP005 for the publication of this report and accompanying images. qPCR : quantitative polymerase chain reaction DAB1 : DISABLED, DROSOPHILA, HOMOLOG OF, 1 HOOK1 : HOOK, DROSOPHILA, HOMOLOG OF, 1 NFIA : NUCLEAR FACTOR I/A DOCK7 : DEDICATOR OF CYTOKINESIS 7 DNAJC6 : DNAJ/HSP40 HOMOLOG, SUBFAMILY C, MEMBER 6 PDE4B : PHOSPHODIESTERASE 4B, cAMP-SPECIFIC CNV : copy number variation FISH : fluorescent in situ hybridization RT-qPCR : reverse-transcriptase quantitative polymerase chain reaction ADHD : attention deficit hyperactivity disorder ADD : attention deficit disorder OCD : obsessive compulsive disorder OHT : ocular hypertension SB5 : Stanford binet intelligence scales-fifth edition FSIQ : full-scale intelligence quotient DCP : DECIPHER SLC35D1 : SOLUTE CARRIER FAMILY 35 (UDP-GLUCURONIC ACID/UDP-N-ACETYLGALACTOSAMINE DUAL TRANSPORTER), MEMBER D1 C1orf141 : Homo sapiens chromosome 1 open reading frame 141 **Competing interest** The authors declare that they have no competing interests. **Authors' contributions** JL generated data and wrote the manuscript. HGK conceived the project, analyzed and interpreted the data of comparative deletion mapping, and wrote the manuscript. Microarray was performed by YS. LL, MD, IKK revised the manuscript. All authors read and approved the final manuscript. We would like to thank the patient who kindly consented to participate in our study. We wish to thank Lynn Chorich and Megan Sullivan. We also extend our thanks to Paul Browne. The present study makes use of data generated by the DECIPHER community. A full list of centres who contributed to the generation of the data is available from <http://decipher.sanger.ac.uk> and via email from decipher\@sanger.ac.uk. Funding for the DECIPHER project was provided by the Wellcome Trust. This study was undertaken following the guidelines established by the Institutional Review Board at AU.

Introduction {#section1-0898264315609907} ============ In the International Classification of Functioning, Disability, and Health model (ICF), pre-existing health conditions in an individual can give rise to functional impairments, which in turn cause activity limitation and restriction in social participation ([@bibr11-0898264315609907]; [@bibr32-0898264315609907]). Disability in turn results from the interaction between these factors in the individual and the surrounding environment. In general, no attempt is made by the ICF to distinguish between cognitive and physical causes of disability, which often interact. The World Mental Health Survey Initiative has demonstrated that mental disorders were more likely to be associated with severe disability than physical ones ([@bibr26-0898264315609907]). The key improvement from the earlier disability model, the International Classification of Impairments, Disabilities, and Handicaps (ICIDH), was a shift from a purely medical to a bio-psycho-social framework reflecting the interaction between an individual's health status and ambient factors, both personal and environmental. In addition, there was an emphasis on viewing disability as an expression of the aforementioned interaction, rather than as a characteristic of the individual ([@bibr32-0898264315609907]). Previously, a person with paralysis after a stroke was deemed to be disabled due to loss of limb function. When viewed through the lens of the ICF model, the stroke is a chronic disease that gives rise to functional impairment (limb paralysis), which limits activity performance (driving), which in turn restricts social participation (going to play poker). The stroke itself may also limit activity performance and social participation directly by making it harder to retain a driving license or getting fewer invitations to play poker. Personal factors such as motivation (loves to play poker), and environmental factors (poker buddy who can drive) enable the affected person to still participate in his favorite social activity in spite of significant functional impairment or activity limitation. The interaction of all these elements thus gives rise to a certain level of disability in that person. Although the above example was simplified for clarity, it can be observed that disability originates from the underlying chronic disease as it directly gives rise to functional impairment, and affects activity performance and social participation both directly and indirectly. In older people especially, multiple chronic diseases may co-exist contributing to disability, although not everyone with a high chronic disease burden is necessarily disabled. Hence, this relationship may not be as clear as one might assume. The next observation is that for a certain degree of functional impairment, the level of restriction in activity and participation is variable depending on the personal and environmental factors surrounding the individual. The last observation is that functional impairment directly impacts on activity and participation, and although the correlation between them may be lessened by the aforementioned personal and environmental factors, it is still likely to be fairly strong. Earlier studies on the ICF were restricted to examining disability in a particular disease-specific context or outcome, rather than looking at the general validity of the model as a whole ([@bibr6-0898264315609907]; [@bibr8-0898264315609907]; [@bibr16-0898264315609907]; [@bibr23-0898264315609907]). These disease-specific studies may also have limited generalizability to the elderly population, where there are often significant and concurrent co-morbidities contributing to disability. The use of a single large data set in examining the ICF model has the advantage of analyzing the connections between ICF constructs simultaneously, thus allowing a comparison of strengths between these inter-relationships. It would also be useful to determine how cognitive factors contribute to disability in the ICF model. Thus, our study aimed to examine in a cohort of community-dwelling older adults some of the observations or hypotheses stemming from the ICF model, which are that (a) chronic disease burden is strongly related to disability as a whole; (b) for a given level of functional impairment, individuals vary significantly in performance of activities; and (c) functional impairment is strongly related to activity performance, which in turn correlates with social participation ([@bibr32-0898264315609907]). We also studied where relevant, the impact of dementia status on disability as well as the relationships between the ICF constructs. Method {#section2-0898264315609907} ====== This study was based on the "Determinants of Wellness among Older Malaysians: A Health Promotion Perspective" survey conducted by trained interviewers from Universiti Putra Malaysia and the Malaysian Ministry of Health in 2008 to 2009. The questions were set in English and Malay, and translated into the primary language of the respondents by interviewers who were native speakers where necessary. It was not feasible to pre-translate into all the local languages due to the large number of tribes and ethnic groups especially in East Malaysia. The interviewers read out the individual questions and noted the responses in a set of printed study forms. The study was approved by the Malaysian Ministry of Health Ethics Review Board (approval reference P08-219) in accordance with current guidelines on Good Clinical Practice and the Declaration of Helsinki. Specific informed consent was not required by the Ethics Review Board as the data were analyzed anonymously. This study used a geographical cluster sampling design based on the World Health Organization STEPwise approach to Surveillance (STEPS), with clusters represented by enumeration blocks (EBs) determined by the Malaysian Department of Statistics ([@bibr7-0898264315609907]; [@bibr33-0898264315609907]). The inclusion criteria were above the age of 60 years, a Malaysian citizen, and willing to be interviewed. They were excluded from the study if mobility, mental functions, or hearing was sufficiently impaired to prevent them from completing the survey. Each EB had a target recruitment of eight people, but less could have been recruited if suitable residents were not found in the allocated houses. The World Health Organization Disability Assessment Schedule II (WHODAS-II) was developed to assess disability based on the ICF ([@bibr31-0898264315609907]). It has been validated across different cultures for use in subjects with chronic diseases, older people, and is literacy-independent ([@bibr10-0898264315609907]; [@bibr27-0898264315609907]). In our study, we used simple percentile sum scoring. Severe disability was defined as a WHODAS-II score above the 90th percentile for the population ([@bibr26-0898264315609907]). Dementia status was defined using the Malaysian version of the 30-item Folstein mini-mental state examination (MMSE-M). The sample was stratified into two groups (with and without dementia) based on locally validated cutoffs, with good sensitivity (88%-97%) and specificity (75%-93%) in classifying dementia against *Diagnostic and Statistical Manual of Mental Disorders* (4th ed.; *DSM-IV*; [@bibr1-0898264315609907]) criteria. The question testing attention and calculation was repeated using serial 7's, serial 3's, and spelling the word "WORLD" backwards, and was scored according to the best performance. The respective MMSE-M cutoffs for dementia classification were 21/22, 18/19, and 17/18, respectively, depending on the question used ([@bibr14-0898264315609907]). Respondents were asked whether they had any of the chronic non-communicable disease groups, which may contribute to disability: cancer (all types), respiratory (asthma, obstructive airways disease, pulmonary tuberculosis), uncorrected visual or hearing impairment, renal, gastrointestinal-urinary (dyspepsia, constipation, incontinence), musculoskeletal (joint pain, gout, arthritis), and cardiovascular (hypertension, diabetes mellitus, heart disease, stroke). They were also questioned whether the diagnoses had been medically confirmed. Although pulmonary tuberculosis is not strictly a chronic non-communicable disease, it was included in the list as it can cause long-term impairment of lung function, and is commonly found among the elderly in Malaysia. Each of the seven chronic disease groups and dementia were marked as either present (1) or absent (0). The Timed Get-Up and Go Test (GUGT), short International Physical Activity Questionnaire (IPAQ), and Lubben Social Network Scale (LSNS-6) are measures of basic functional mobility, physical activity, and social support ([@bibr4-0898264315609907]; [@bibr19-0898264315609907]; [@bibr22-0898264315609907]). They are feasible to perform, valid, and reliable in the elderly ([@bibr9-0898264315609907]; [@bibr13-0898264315609907]; [@bibr21-0898264315609907]; [@bibr29-0898264315609907]). Unless specified in the original definition, missing values were replaced by the mean of the remaining values in the scale provided internal consistency was demonstrated to be high (Cronbach's α \> .7). Up to one missing value could be replaced in this way and further missing values invalidated the score for that respondent. Education level, net monthly income, smoking, alcohol consumption, ethnicity, marital status, and religion were classified according to standard census criteria ([@bibr7-0898264315609907]). Fall status was determined as whether the respondent had any falls resulting in injury in the previous 6 months. Body mass index (BMI) and waist circumference were obtained using a stadiometer (206CM Bodymeter Measuring Tape, SECA, Hamburg, Germany) and electronic weight scale (HD-314 Digital Weight Scale, Tanita, Arlington Heights, Illinois, USA). Dietary adequacy was assessed using a Dietary Diversity Score (DDS; [@bibr17-0898264315609907]). Analysis {#section3-0898264315609907} -------- The overall contribution of chronic disease to severe disability was modeled using population attributable risk (PAR). PAR was calculated for each chronic disease group along with a total PAR for their combined effect ([@bibr28-0898264315609907]). The combined PAR for all (*n*) disease groups is shown below ([@bibr5-0898264315609907]). $$PAR\left( n \right) = 1 - {\prod\limits_{i = 1}^{n}\left( {1 - PAR_{i}} \right)}.$$ A composite chronic disease score (CDBS) for all (*n*) disease groups including dementia was derived for each respondent, weighted based on the PAR of each group as shown below. A composite chronic disease score without dementia (CDBS-ND) was also derived using the same method. $$CDBS\left( n \right) = \frac{\sum\limits_{i = 1}^{n}\left( {D_{i} \times PAR_{i}} \right)}{\sum\limits_{i = 1}^{n}\left( {PAR_{i}} \right)},$$ where CDBS is expressed as a percentage and *D~i~* represents the presence (1) or absence (0) of the chronic disease group "*i*." To test the first observation that chronic disease burden is strongly related to disability, a multivariate approach was used, including covariates that could potentially act as confounders. Subjects were divided into two groups based on the presence or absence of severe disability. The initial list of covariates screened were age, gender, marital status, race, religion, years of formal education, net income, smoking and alcohol use, DDS, fall status, BMI, and waist circumference. To reduce the number of covariates, they were tested against the grouped WHODAS-II score using a Pearson's chi-square test for categorical covariates, and either an unpaired *T* test or Mann--Whitney test for continuous covariates ([Table 1](#table1-0898264315609907){ref-type="table"}). ###### Baseline Characteristics of Respondents, Selected Summary Data From the Malaysia 2010 Census, and Key Outcome Measures.  Categorical covariates Distribution ----------------------------------------------------------------------------- --------------------------------------------------------------- Gender 45.0% males, 55.0% females Ethnicity 55.4% Malays, 13.9% Chinese, 5.3% Indians, 25.4% Indigenous and Others Gender (Malaysia 2010 census) 48.1% males, 51.9% females Ethnicity (Malaysia 2010 census) 65.9% Malays, 11.8% Chinese, 2.2% Indians, 20.1% Indigenous and Others Marital status 2.1% not married, 58.3% married, 2.2% divorced, 37.4% widowed Religion 71.1% Islam, 12.1% Christianity, 11.3% Buddhism, 4.4% Hinduism, 1.1% no religion and others Smoking 84% not current smoker, 16% current smoker Alcohol use 95.8% no current alcohol use, 4.2% current alcohol use Fall status 87% no falls, 13% fell with injury Continuous covariates *M* (*SD*)/distribution Age 63.2% (60-69 years), 29.9% (70-79 years), 7.0% (≥80 years) Years of formal education 3.4 years (3.7) Net income RM616 (1,329; approx. US\$170, early 2015 conversion rates) Dietary Diversity Score 4.3 (0.8) (max 5.0) Body mass index 24.2 (4.8) Waist circumference 83.9 (15.0) cm Key outcome measures^[a](#table-fn2-0898264315609907){ref-type="table-fn"}^ Median, IQR, non-response rate WHODAS-II Median 14.6, IQR 4.2-31.3, non-response 3.4% CDBS^[b](#table-fn3-0898264315609907){ref-type="table-fn"}^ Median 13.0, IQR 13.0-25.0, non-response \<0.1% CDBS-ND^[b](#table-fn3-0898264315609907){ref-type="table-fn"}^ Median 14.0, IQR 14.0-29.0, non-response \<0.1% GUGT time (s) Median 17.0, IQR 13.0-21.0, non-response 20.4% IPAQ energy expenditure (MET-min/week) Median 1233, IQR 0-3947, non-response \<0.1% LSNS-6 Median 15.0, IQR 10.0-19.0, non-response 0.4% *Note.* IQR = inter-quartile range; WHODAS-II = World Health Organization Disability Assessment Schedule II; CDBS = composite chronic disease score; CDBS-ND = composite chronic disease score without dementia; GUGT = Get-Up and Go Test; IPAQ = International Physical Activity Questionnaire; LSNS-6 = Lubben Social Network Scale; MET = Metabolic Equivalent Task. The key outcome measures were shown as medians and IQRs as most of them were non-normal in distribution. Pearson's correlation coefficient = .93 between CDBS and CDBS-ND. Significant covariates were then entered into binary logistic regression analysis, with the CDBS as the main independent variable and the grouped WHODAS-II score as the binary dependent variable. The same analysis was also repeated excluding subjects with dementia on the MMSE-M, utilizing CDBS-ND as the main independent variable. To test the second observation that individuals vary significantly in performance of activities for a given level of functional impairment, we plotted IPAQ activity levels (activity performance) for each GUGT functional status category (functional impairment) to illustrate the degree of variability in activity performance. As a comparison, LSNS-6 scores (social participation) were divided into quartiles and similarly charted against GUGT (functional impairment). The dispersion for IPAQ and LSNS-6 scores in each GUGT category was quantified using the inter-quartile range (IQR). For the third observation, we tested the relationship between functional impairment and activity performance by comparing IPAQ energy expenditure (activity performance) with GUGT (functional impairment) using Spearman's rank correlation coefficient, the Kruskal--Wallis non-parametric ANOVA, and also graphing the comparison. The relationship between activity performance and social participation was tested by comparing the LSNS-6 score (social participation) and IPAQ activity levels (activity performance) in a similar manner. This analysis was repeated stratifying by dementia status on the MMSE-M. As a comparison, we quantitatively tested the relationship between functional impairment and social participation in a similar manner. Sample size was calculated using the STEPS Sample Size Calculator from the [@bibr33-0898264315609907]. Based on a 95% confidence level, 5% margin of error, an estimated prevalence of risk factors of 50%, a design effect correction of 3, an expected response rate of 60%, the calculated sample size was 3,840 subjects, translating to 480 EBs of eight subjects each. All computations were performed using SPSS for Windows version 20.0 (IBM Corp., Armonk, New York, USA). Statistical tests were two-tailed and conducted at 5% level of significance. Results {#section4-0898264315609907} ======= The sample included a total of 2,563 respondents with an overall response rate of 66.7%, which was more than the expected response rate of 60% ([Figure 1](#fig1-0898264315609907){ref-type="fig"}). Baseline characteristics of respondents are detailed in [Table 1](#table1-0898264315609907){ref-type="table"}, along with selected equivalent summary data from the Malaysia 2010 census ([@bibr7-0898264315609907]). The gender and ethnic distribution approximates that from the census data, with the exception of having less Malays and more Indigenous people. {#fig1-0898264315609907} The responses for medically confirmed diagnoses essentially followed the same pattern as self-reported conditions (Pearson's correlation coefficient *r* = .68), and hence were not analyzed separately. Median values and non-response rates for the key outcome measures are summarized in [Table 1](#table1-0898264315609907){ref-type="table"}; 21.5% were classified as having dementia using the locally validated cutoffs for MMSE-M. The non-response rate for GUGT testing (20.4%) was disproportionately higher than for other outcome measures, and mostly attributable to lack of proper equipment or space (7.7%) and refusal (5.5%). There was a significant difference in WHODAS-II scores between respondents who refused GUGT testing and those who agreed, with those who refused having higher scores (mean 4.4 points, 95% CI = \[0.6, 8.1\], *p* = .02). Cronbach's alpha for WHODAS-II, DDS, LSNS-6, CDBS, and CDBS-ND were .92, .46, .83, .28, and .28, respectively. Replacement of missing values occurred in 8% of WHODAS-II scores but not for the DDS, LSNS-6, CDBS, or CDBS-ND. For the first observation, the 90th percentile for WHODAS-II scores used to define severe disability was 50 points. Significant covariates obtained after initial testing against WHODAS-II were gender, marital status, smoking, fall status, age, years of formal education, and net income ([Table 2](#table2-0898264315609907){ref-type="table"}). Binary logistic regression using these covariates gave a good model fit with a non-significant Hosmer and Lemeshow statistic (*p* = .26), −2 log-likelihood of 1,184, overall correct classification percentage of 92.4%, and Nagelkerke *R*-square of .78. The Wald statistic and significance level for the CDBS were 25.5 and *p* \< .001. The other significant covariates from regression analysis were gender, age, education status, and social support ([Table 3](#table3-0898264315609907){ref-type="table"}). When the analysis was repeated excluding subjects with dementia, the Wald statistic and significance level for the CDBS-ND were 28.5 and *p* \< .001, with a Nagelkerke *R*-square of .78. The combined PAR for chronic disease was 53.8%, whereas the PAR for dementia alone was 15.3%. ###### Covariates Tested Against WHODAS-II Scores Using Pearson's Chi-Square for Categorical Covariates and Unpaired *T* Test for Continuous Covariates.  Categorical covariates χ^2^ Odds ratio (95% CI) *p* --------------------------- ------------------------------------------------------------- ------------------------------------------------------------ ------------ Gender (M/F) 18.0 2.0 \[1.4, 2.7\] **\<.001** Marital status 25.0 n/a^[a](#table-fn5-0898264315609907){ref-type="table-fn"}^ **\<.001** Ethnicity 2.9 n/a^[a](#table-fn5-0898264315609907){ref-type="table-fn"}^ .407 Religion 4.2 n/a^[a](#table-fn5-0898264315609907){ref-type="table-fn"}^ .515 Smoking (Y/N) 4.5 1.7 \[1.0, 2.7\] **.033** Alcohol use (Y/N) 1.2 1.6 \[0.7, 4.1\] .281 Fall status (Y/N) 4.2 1.5 \[1.0, 2.2\] **.041** Continuous covariates *T* test/*z* score *p* Age −7.2^[b](#table-fn6-0898264315609907){ref-type="table-fn"}^ **\<.001** Years of formal education −6.3^[b](#table-fn6-0898264315609907){ref-type="table-fn"}^ **\<.001** Net income −5.1^[b](#table-fn6-0898264315609907){ref-type="table-fn"}^ **\<.001** Dietary Diversity Score −0.4^[b](#table-fn6-0898264315609907){ref-type="table-fn"}^ .705 Body mass index −0.6^[b](#table-fn6-0898264315609907){ref-type="table-fn"}^ .527 Waist circumference −0.6^[b](#table-fn6-0898264315609907){ref-type="table-fn"}^ .581 *Note.* WHODAS-II = World Health Organization Disability Assessment Schedule II; CI = confidence interval. Boldfaced values have p\<0.05. Categorical covariate with more than two categories. The *z* score for the Mann--Whitney Test was used where the variables were non-normal in distribution. ###### Relevant Statistics for the Logistic Regression Analysis From the First Observation (CDBS Only).  Covariates β Wald Significance Exp(β) (95% CI) --------------------------- ------- ------- -------------- ---------------------- Gender 0.431 6.22 .013 1.54 \[1.01, 2.16\] Age 0.024 58.06 **\<.001** 1.03 \[1.02, 1.03\] Years of formal education 0.118 18.27 **\<.001** 1.13 \[1.07, 1.19\] CDBS 1.874 25.48 **\<.001** 6.52 \[3.15, 13.49\] LSNS-6 0.070 35.04 **\<.001** 1.07 \[1.05, 1.10\] *Note.* CDBS = composite chronic disease score; CI = confidence interval; LSNS-6 = Lubben Social Network Scale. Boldfaced values have p\<0.05. For the second observation, bar charts of IPAQ activity levels for each GUGT functional status category clearly showed that at low levels of functional impairment, activity performance varied considerably with a large IQR (2.00; [Figure 2](#fig2-0898264315609907){ref-type="fig"}, right panel). However, at higher levels of functional impairment, activity performance was consistently low with a small IQR (0.00; [Figure 2](#fig2-0898264315609907){ref-type="fig"}, left panel). In contrast, significant variability was present in social participation throughout all levels of functional impairment (IQR = 9-10; [Figure 3](#fig3-0898264315609907){ref-type="fig"}). {#fig2-0898264315609907} {#fig3-0898264315609907} For the third observation, ANOVA showed a highly significant difference (χ^2^ = 253.3, *p* \< .001) in IPAQ energy expenditure between GUGT functional status groups, and the bar chart showed a clear rise in energy expenditure with improving functional status (Spearman's rho = 0.33, *p* \< 0.001). When analyzed according to dementia status, the trends were generally similar although it could be seen that those without dementia were able to maintain a higher level of activity even with moderately reduced ambulatory function ([Figure 4](#fig4-0898264315609907){ref-type="fig"}, bottom panel). {#fig4-0898264315609907} Similarly, ANOVA showed a highly significant difference (χ^2^ = 83.8, *p* \< .001) in LSNS-6 scores between IPAQ activity levels, whereas the bar chart demonstrated improving social participation with better activity performance (Spearman's rho = 0.18, *p* \< .001). The presence of dementia was shown to substantially reduce social participation, and those with concurrent low physical activity and dementia had the lowest scores (median LSNS-6 = 11.0; [Figure 5](#fig5-0898264315609907){ref-type="fig"}, middle panel). {#fig5-0898264315609907} The difference in LSNS-6 scores between GUGT groups was also significant (χ^2^ = 26.3, *p* \< .001) showing improving social participation with better functional status (Spearman's rho = 0.11, *p* \< .001). Discussion {#section5-0898264315609907} ========== For the first observation that examined the relationship between chronic disease burden and disability, the model used for regression analysis had a good fit and high correct classification percentage. Hence, the highly significant Wald statistic for CDBS indicates that chronic disease burden is strongly related to disability, even after accounting for significant confounding variables and dementia status. The large Nagelkerke *R*-square value shows that most of the variability in disability between individuals was accounted for by chronic disease and socio-demographic characteristics such as gender, age, education level, and social support. In addition, the PAR calculations indicate that chronic disease accounted for more than half of prevalent disability, with dementia causing a large proportion of it. These findings are all consistent with results from the multi-national 10/66 Dementia Research Group, which reported that the significant covariates were chronic disease, age, gender, and education level. Similar to our study, PAR for chronic disease accounted for two thirds of disability with dementia also being the largest contributor ([@bibr28-0898264315609907]). The second observation states that for a given level of functional impairment, activity performance varies significantly between individuals due to the effect of factors such as motivation, self-esteem, and a conducive environment. However, we found that this holds true only for individuals with relatively little functional impairment ([Figure 2](#fig2-0898264315609907){ref-type="fig"}, right panel). With worsening function, the role of environmental factors diminishes as the ability to perform activities becomes limited ([Figure 2](#fig2-0898264315609907){ref-type="fig"}, left panel). The reduction of variability in activity performance is evidenced by a corresponding diminution of the IQR. This means that once an individual's physical function deteriorates beyond a certain point, modification of the environment cannot meaningfully improve that person's disability. In contrast, functional impairment does not seem to be a major limiting factor for social participation, and it is likely that individuals retain their pre-existing social networks even after their functional status worsens ([Figure 3](#fig3-0898264315609907){ref-type="fig"}). This relates very well to Antonucci's support bank theory, which postulates that social capital built up when a person is younger carries over into old age in the form of social support ([@bibr2-0898264315609907]). [Figures 4](#fig4-0898264315609907){ref-type="fig"} and [5](#fig5-0898264315609907){ref-type="fig"} and the highly significant ANOVA tests all show a clear relationship between functional impairment, activity performance, and social participation, thus validating the third observation. The relative weakness of the relationship between functional impairment and social participation (χ^2^ = 26.3) compared with the other two relationships (χ^2^ = 253.3 and 83.8) suggests that activity performance is the intermediary construct in this model ([Figure 6](#fig6-0898264315609907){ref-type="fig"}). A path analysis could provide more clarity although this is out of the scope of the present study. {#fig6-0898264315609907} Although the cognitive aspect of disability was not studied in depth, it could clearly be seen that the presence of dementia significantly worsened both physical activity and social participation. Furthermore, cognitively intact individuals with a moderate level of functional impairment retained a significant level of activity performance, implying that this group may have the potential for further improvement with appropriate rehabilitation ([Figure 4](#fig4-0898264315609907){ref-type="fig"} and [5](#fig5-0898264315609907){ref-type="fig"}). In contrast, concurrent low physical activity and poor cognitive function identify individuals who are at risk of social isolation ([@bibr20-0898264315609907]). For the first observation, because chronic non-communicable disease is so strongly tied to disability, every effort should be made to identify the major conditions in each population, which contribute to it. These conditions should then be targeted with effective preventive care programs to reduce the eventual occurrence of end organ functional impairments in the community. Dementia is an especially good candidate to focus on due to its prevalence in community-dwelling elderly and the high PAR for severe disability ([@bibr12-0898264315609907]). In addition to these primary health conditions, attention should be paid to the secondary and co-morbid conditions that arise from them, the provision of adequate health care, and removing barriers to performance and participation ([@bibr32-0898264315609907]). For the second observation, it can be seen that modification of the environment will be most effective in increasing activity performance in individuals with moderate functional impairment. In Japan, the Heart Building Law (1994), Barrier Free Transport Law (2000), and the Barrier Free Law (2006) provide a formal legal framework to enhance accessibility for the aged and disabled. Although these regulations have greatly improved facilities for the physically disabled, this study suggests that cognitive disablement is also an important issue that needs to be considered. For example, signage needs to be written in clear simple language with unambiguous universal symbols to aid understanding. For those with high levels of functional impairment especially with concurrent cognition issues, this study shows that mobility is much harder to improve even with the best barrier-free environment. Resources would be better targeted in enhancing their support network, and hence social participation with interventions such as community social worker assistance and programs for organized home visits by school children and voluntary welfare organizations. For the third observation, the strong linkages between functional impairment, activity performance, and social participation, mean that success in improving any one of these will have secondary benefits for the downstream components, hence reducing overall disability. Logically this means that resources would be better spent on areas that have the most potential to benefit each individual, rather than dealing with all aspects of disability simultaneously. From this study, the implication is that younger people are best targeted with prevention programs to reduce the risk of developing disability, whereas older people with milder disability should focus on removing barriers to activity performance, and those with more severe disability should be provided more support to improve their social participation ([@bibr18-0898264315609907]). It can be seen from [Figures 4](#fig4-0898264315609907){ref-type="fig"} and [5](#fig5-0898264315609907){ref-type="fig"} that the presence of dementia has an important effect on the relationships between the ICF constructs. Moreover, data from this study and the 10/66 Dementia Research Group show that dementia contributes significantly to disability in the community-dwelling elderly ([@bibr28-0898264315609907]). Seen together, there would seem to be good reason to investigate the cognitive aspects of disability in greater detail that what was achieved in this study. Study Considerations and Limitations {#section6-0898264315609907} ------------------------------------ We opted to study disability in community-dwelling older people as very little work has been done locally in this population aside from the National Health and Morbidity Survey, which was based on the ICIDH ([@bibr15-0898264315609907]). Mindful of the need to establish the external validity of the model, we chose instruments that were developed separately from the ICF when testing the second and third observations. The relevant domains of the WHODAS-II (cognition, mobility, life activities, and participation) were unsuitable for this role as they were co-developed with the ICF. Moreover, the high level of internal consistency (Cronbach's α = .94-.96 within domains, .98 across domains) virtually guarantees that the domains will cross-correlate with each other, thus defeating the intent of the study ([@bibr31-0898264315609907]). In addition, the instruments used focused on important aspects of an older person's life, such as mobility (GUGT), overall physical activity and energy expenditure (IPAQ), and social networking (LSNS-6). There is very little overlap between these instruments, thus enabling the constructs to be clearly distinguished ([@bibr3-0898264315609907]). The instruments are also not disease specific, thus allowing generalization across a variety of conditions. The choice of instruments was made to reflect the conceptual framework behind the ICF model. For example, mobility (GUGT) was chosen to represent body functions and structures (ICF "b" and "s" codes, respectively) as the majority of impairments (mental, sensory, cardio-respiratory, metabolic, and neuro-musculoskeletal, corresponding to ICF chapters 1, 2, 4, 5, and 7, respectively) will result in reduced mobility. Similarly, physical activity (IPAQ) and social networking (LSNS-6) were chosen to represent activities and participation (ICF "d" codes) as these are consequent manifestations of an individual's mobility in relation to the ICF model ([@bibr30-0898264315609907]). Although the WHODAS-II has been superseded by the more recent WHODAS 2.0, the 12-item instrument used is virtually identical in wording with the updated version ([@bibr34-0898264315609907]). This means that the findings and conclusions made in this article equally apply to the newer instrument and the ICF in general. Conservative values were chosen for sample size calculation as a significant proportion of the survey was conducted in sparsely populated rural areas. A design effect correction of 3 was used rather than the recommended value of 1.5 as a study demonstrated that in such circumstances this value can range from 1 to 5 ([@bibr25-0898264315609907]). The expected response rate was also reduced as certain allocated EBs were impossible to access due to geographical isolation. Considering this, the overall non-response rate for the study was within expectation. The individual item response rates were very good aside from GUGT, which was likely because some older people can be fairly averse to physical testing. Whereas the complete ICF model depicts the interplay between disease states, human functioning, and environmental factors, this study is primarily scoped toward disease states and human functioning. A related study by the authors examined the contribution of environmental factors (ICF "e" codes) in greater detail ([@bibr12-0898264315609907]). Some limitations of the study are worth highlighting. There is a systematic bias against severe disability due to the sampling method, which excluded people who were unable to complete the survey because of this. However, there is no good way of working around this issue as the alternative would be to rely on secondary information from caregivers, which would be vulnerable to confirmation bias. Although there may have been under-sampling of the Chinese population due to language barriers, the proportion of ethnic Chinese in the study is similar to that from the census data, suggesting that the magnitude of under-sampling was small ([Table 1](#table1-0898264315609907){ref-type="table"}). Moreover, this should only give rise to minimal bias as WHODAS-II scores in this ethnic group are fairly similar to the other major races (Malays and Indians). Although there was possibly some minor element of bias from the GUGT non-response, this should not have affected the conclusions drawn as the results were all highly significant with large effect sizes. Most of the instruments used in this study were validated in English, and there were no validation studies either in Malay or the many languages and dialects used locally. Once again there is no realistic way around this issue, and this study should be interpreted with this in mind. Although the validation article for MMSE-M recommended that serial 7's or serial 3's be tried as a first choice, followed by the spell backwards method, we found that most of the elderly in rural communities could not cope with this due to low levels of education. The MMSE-M was validated in city hospitals with subjects who had much better educational exposure, so the project team made a decision to use the best result of the three methods as a compromise ([@bibr14-0898264315609907]). Although this may potentially alter the diagnosis of dementia to a small degree, this risk was judged to be acceptable as cognition was not the main focus for the study. Lastly, certain aspects of the study were either abbreviated or simplified to keep the questionnaire time to less than an hour. Although extending the time or using a multi-stage interview might have allowed collection of additional data, there was a strong likelihood that those with high levels of functional impairment would drop out or refuse in a developing country setting ([@bibr24-0898264315609907]). This would introduce significant bias, particularly in a study focusing on disability. The cognitive aspect of disability was also not examined in depth, as other parallel studies by the authors will address this issue separately in greater detail ([@bibr12-0898264315609907]; [@bibr18-0898264315609907]). It would have been desirable to examine the impact of each disease group individually, but the CDBS was used in this study as a compromise between survey time and greater detail. In any case, the relationship between CDBS and WHODAS-II was still found to be highly significant on logistic regression testing (*p* \< .001), suggesting that this instrument was adequate for the limited purposes of this study. Conclusion {#section7-0898264315609907} ========== Our study builds upon the body of evidence from earlier disease-specific studies by demonstrating the validity of the earlier observations based on the ICF model using empirical data from a large sample of community-dwelling older adults with a broad range of medical diagnoses. Consistent with earlier studies, our results confirm that disability is strongly related to chronic disease burden, and that functional impairment correlates with activity performance and consequently with social participation. The novel finding from our study is that at low levels of functional impairment, there is considerable variability in the relationship between functional impairment and activity performance, but this diminishes at high impairment levels especially when complicated by cognitive deficits. This suggests that modification of the environment benefits those with moderate levels of functional impairment, but not those with higher grades of functional loss. The authors would like to extend their gratitude to the director general of Health Malaysia for his kind support, funding of the research, and allowing the article to be published. **Author Contributions:** Seng Cheong Loke conducted the data analysis, interpreted the results, and prepared the article. Wee Shiong Lim and Yoshiko Someya assisted with data interpretation and critically appraised the article. Tengku A. Hamid and Siti S. H. Nudin were involved in study design. Tengku A. Hamid directed the study implementation. **Authors' Note:** Siti S. H. Nudin was responsible for quality assurance and control. All authors reviewed the article and approved the final version. **Declaration of Conflicting Interests:** The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article. **Funding:** The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was paid for using internal funding from the Ministry of Health Malaysia. The Ministry of Health through the Institute for Health Behavioural Research was fully involved in all aspects of the study as co-investigators.

1. Introduction {#sec1-biology-06-00022} =============== 1.1. Intraguild Predation {#sec1dot1-biology-06-00022} ------------------------- In many aquatic ecosystems, invertebrate predators such as predatory crustaceans (IG-prey) compete for food with the planktivorous fish (IG-predator), which, at the same time, may also prey on the predatory crustaceans, thereby creating an intraguild predation (IGP) relationship in the system \[[@B1-biology-06-00022],[@B2-biology-06-00022],[@B3-biology-06-00022]\] ([Figure 1](#biology-06-00022-f001){ref-type="fig"}). IGP is probably more common in lakes than so far documented \[[@B4-biology-06-00022]\]. The theory predicts that productivity and the relative efficiency of resource (shared prey) utilization determine the outcome of interactions between IG-predators and IG-prey \[[@B2-biology-06-00022],[@B5-biology-06-00022]\]. 1.2. Modeling Intraguild Predation {#sec1dot2-biology-06-00022} ---------------------------------- To fully understand an ecosystem that includes IGP dynamics, a separate analysis of the IGP components and their impact on the ecosystem is necessary. This kind of analysis can be carried out using methods in which the abundance, distribution, diet, consumption rates, and predation pressure of the related IGP components are studied \[[@B6-biology-06-00022],[@B7-biology-06-00022]\]. However, rigorous long-term empirical tests of equilibrium IGP interactions are rare \[[@B5-biology-06-00022],[@B8-biology-06-00022],[@B9-biology-06-00022],[@B10-biology-06-00022]\], and long-term studies of IGP in natural systems are even scarcer \[[@B11-biology-06-00022],[@B12-biology-06-00022]\], where conditions are far from equilibrium and influenced by external drivers. An alternative approach to explore food web dynamics that include IGP dynamics is by using numerical ecosystem models. Food webs in which invertebrate predators are important cannot be modeled as a simple chain, as is regularly done in aquatic ecosystem models. Adding a compartment for invertebrate predators into lake food web models introduces an element of IGP \[[@B1-biology-06-00022],[@B2-biology-06-00022],[@B8-biology-06-00022]\]. It is also necessary to add an element of cannibalism/self-limitation, if observed in reality \[[@B13-biology-06-00022]\]. Such models are solved analytically and are used to explore ecological issues such as coexistence in IGP systems \[[@B14-biology-06-00022],[@B15-biology-06-00022]\], alternative stable states \[[@B16-biology-06-00022]\], and the possibility of unstable dynamics, as well as problems in applied ecology such as the biological control of pest species, the transfer of pollutants through the pelagic food web \[[@B17-biology-06-00022]\], and the conservation of threatened species \[[@B18-biology-06-00022]\]. Although there is much variability among ecosystems, a known result of these models is that coexistence is feasible if the IG-prey is better than the IG-predator at competing for the shared prey, and if the IG-predator accrues a sufficient gain from attacking the IG-prey \[[@B1-biology-06-00022],[@B19-biology-06-00022]\]. An IG-predator that is superior at suppressing the target herbivore population would stress an IG-prey population to extinction through a combination of competition and predation \[[@B1-biology-06-00022],[@B19-biology-06-00022],[@B20-biology-06-00022]\]. Additionally, Holt and Polis \[[@B1-biology-06-00022]\] showed that ecosystems with IGP have the potential to generate alternative stable states, with unstable dynamics that lead to transient phases. However, these models are based on several arguable assumptions, such as: (1) the system is assumed to be at equilibrium and the models are solved to equilibrium with constant meteorological forcing data \[[@B8-biology-06-00022],[@B21-biology-06-00022]\]; or, (2) the two predators compete for only a single species of shared prey. Yet, according to Rosenheim and Harmon \[[@B22-biology-06-00022]\], the inclusion of multiple herbivore prey into IGP models excludes the prediction that the IGP is uniformly disruptive to biological control (i.e., the ability of the IGP-predator to suppress populations of herbivores); or, (3) the direct intraguild interactions are sufficiently common to be important for the IGP dynamics \[[@B23-biology-06-00022]\]. While simple models can often reproduce the elementary processes operating in complex systems and reveal fundamental ecological patterns \[[@B24-biology-06-00022]\], more complex mechanistic food web models are able to respond to the underlying variability in the physical-chemical conditions over a range of time-scales, and may better describe the natural feeding interactions \[[@B25-biology-06-00022]\]. A sufficient level of detail in IGP models is therefore essential for developing a deeper understanding of the predation rates and dynamics of zooplankton and fish, and their spatial and temporal variability \[[@B26-biology-06-00022]\]. However, to date, the explicit implementation and analysis of IGP in freshwater ecosystem models has been surprisingly rare. 1.3. Intraguild Predation in Lake Kinneret {#sec1dot3-biology-06-00022} ------------------------------------------ The IGP component in the Lake Kinneret food web consists of the dominant fish in the lake, the zooplanktivorous *Mirogrex terraesanctae* \[[@B27-biology-06-00022]\], commonly known as Lavnun, and the predatory invertebrates (cyclopoid copepods); both feed on herbivorous zooplankton, while the fish also feed on the cyclopoid copepods. For more than two decades (1970--1993), the Lavnun constituted an important share of the Lake Kinneret commercial fishery, with a fairly constant catch of \~1000 t·y^−1^ \[[@B28-biology-06-00022]\]. In the winter of 1991/92, and again in 2002/03, the Lavnun fishery collapsed as the population became devoid of individuals of commercially harvestable sizes (\>12 cm, \[[@B29-biology-06-00022]\]). This was in tandem with exceptionally high precipitation in the winter of 1991/92, and again in 2002/03, leading to \>4 m rise in the lake's water level, with major changes to littoral habitats where the Lavnun spawns. Hydroacoustic surveys indicated that the abundance of mainly sub-commercial sized Lavnun in Lake Kinneret in 1993, and again in 2004, increased by 8--10 fold in comparison to fish abundance levels prior to the flood years \[[@B30-biology-06-00022]\]. These exceptional increases in fish density were the consequence of the unusually successful recruitment of Lavnun as a result of the extreme water level increases. Hereafter, the term "atypical" will refer to an exceptionally high fish abundance (an abundance which is eight fold or higher than the multiannual average), as was observed approximately one to two years after an atypical high inflow winter. During the 1980s and early 1990s, crustacean (predatory) zooplankton biomass dropped precipitously, reaching its lowest ever annual mean value (0.57 g~ww~·m^−3^) by 1993, in tandem with the collapse of the Lavnun fishery \[[@B31-biology-06-00022]\]. Furthermore, the predatory zooplankton underwent a regime shift due to the marked changes in abundance, evident as a gradual replacement of the larger-bodied *Mesocyclops* sp. with the smaller-bodied *Thermocyclops* sp. \[[@B32-biology-06-00022]\]. Gophen et al. \[[@B33-biology-06-00022],[@B34-biology-06-00022]\] argued that Lake Kinneret demonstrates a top-down controlled system in which the Lavnun serves as the major predator of herbivorous zooplankton. Based on this assumption, a subsidized dilution program of the Lavnun was initiated as a management policy in 1995 and continued until 2006, whereby an annual amount of 400--500 t of commercial and sub-commercial size Lavnun fish were harvested. The objective of this subsidized harvest was to reduce the predation pressure on zooplankton, leading to an increase in their abundance and a reduction of phytoplankton, through grazing pressure, and ultimately, an improvement of the water quality. However, no correlation was found between the subsidized harvest and the Kinneret Water Quality Index \[[@B35-biology-06-00022]\] or other water quality indicators, suggesting that top-down control was not occurring as hypothesized. Blumenshine and Hambright \[[@B6-biology-06-00022]\] compared the potential predation pressure on Lake Kinneret herbivorous zooplankton by Lavnun with that of the cyclopoid copepods. Despite having a much lower biomass than Lavnun, cyclopoid copepods accounted for a greater portion of the predation mortality of herbivorous zooplankton. This suggested that the intentional removal of Lavnun would not result in a subsequent increase in herbivorous zooplankton biomass as expected according to top-down theory. Instead, a reduction in Lavnun predation pressure may allow for an increase in the abundance of cyclopoid copepod and thereby result in a net increase in the predation pressure on herbivorous zooplankton. A food web model developed for Lake Kinneret by Hart et al. \[[@B36-biology-06-00022]\] also predicted that planktivorous fish may serve as minor predators of herbivorous zooplankton, with the majority of its top-down regulation associated with cyclopoid copepods. Hart \[[@B13-biology-06-00022]\] explored the effects of IGP and self-limiting factors on the top-down and bottom-up properties of Lake Kinneret using the different models constructed based on the Lake Kinneret carbon flux model. The results indicated that IGP can reduce or even reverse the top-down effects predicted by the food chain theory, as a decrease in planktivorous fish is accompanied by an increase in the predation of zooplankton on invertebrates, and that the degree of self-limitation among the IG-prey is a key factor in determining the direction and strength of the top-down response. Although models such as Hart et al. \[[@B36-biology-06-00022]\] and Hart \[[@B13-biology-06-00022]\] extended our understanding of ecosystems that have an IGP component, they are limited in their ability to reveal insights into the details of IGP variability. In particular, we are interested in understanding how these IGP-dynamics respond to a highly variable physico-chemical environment and the associated complex patterns of planktonic succession \[[@B36-biology-06-00022],[@B37-biology-06-00022],[@B38-biology-06-00022],[@B39-biology-06-00022],[@B40-biology-06-00022],[@B41-biology-06-00022]\]. Moreover, Roelke et al. \[[@B42-biology-06-00022]\] analyzed a long-term data-set of Lake Kinneret within the framework of an alternative states model and revealed a possible complex triggering mechanism and system hysteresis (e.g., a change in a variable threshold where alternative states are possible, to a threshold where alternative states will no longer be possible). Gal and Anderson \[[@B32-biology-06-00022]\] identified the occurrence of a regime shift in the zooplankton population. Therefore, to explore the IGP relationship under the non-equilibrium conditions of Lake Kinneret, we applied a complex aquatic ecosystem model (DYCD-FISH). Unlike the models of Hart \[[@B13-biology-06-00022]\], that capture the most important first-order interactions by means of simple physics, DYCD-FISH is able to simulate the seasonal dynamics of vertical stratification and the most important chemical and biological ecosystem components. DYCD-FISH has previously been configured to include the IGP trophic triangle \[[@B43-biology-06-00022]\], accounting for both the bottom-up and top-down pathways shaping the seasonal dynamics of biogeochemical and ecological processes. However the intricacies of the IGP relationship, as described above, were not considered. The aims of this study were therefore: (1) to improve our understanding of the predator-prey interactions between the dominant fish species in Lake Kinneret, *Mirogrex terraesanctae*, and zooplankton; (2) to study the sensitivity of the key factors controlling the IGP triangle (e.g., predation rate, feeding preference, and vulnerability to fish predation); (3) to compare IGP dynamics during "atypical" conditions, whereby an extreme increase in the fish population occurs, relative to "typical" conditions; and (4) to explore the likely impacts of fishery biomanipulation on fish and zooplankton biomass. 2. Materials and Methods {#sec2-biology-06-00022} ======================== DYCD-FISH was validated for Lake Kinneret as previously described in Gal et al. \[[@B39-biology-06-00022]\] and Makler-Pick et al. \[[@B43-biology-06-00022]\]. The model is briefly described below. 2.1. Ecological Configuration {#sec2dot1-biology-06-00022} ----------------------------- The mechanistic model DYCD-FISH was used to simulate the interactions between the hydrodynamics, biogeochemistry, bacteria, phytoplankton, zooplankton, and fish within Lake Kinneret. DYRESM-CAEDYM (DYnamic REservoir Simulation Model-Computational Aquatic Ecosystem DYnamics Model, DYCD) is a one-dimensional coupled hydrodynamic-ecological model. DYRESM predicts the vertical distribution of temperature, salinity, and density, and models surface heat, mass and momentum transfers, mixed layer dynamics, hypolimnetic mixing, benthic boundary layer mixing, inflows, and outflows \[[@B44-biology-06-00022]\]. The model represents the lake as a series of homogeneous horizontal Lagrangian layers of variable thickness that expand or contract according to the degree of stratification and mixing, and the inflows or outflows entering or leaving the lake. CAEDYM is a generic aquatic ecological model in which a series of ordinary differential equations are solved to describe changes in the concentrations of nutrients (C, N, P, Si), detritus, dissolved oxygen, phytoplankton, and zooplankton, as a function of environmental forcing and ecological interactions for each layer represented by DYRESM. The variables of irradiance, temperature, salinity, and density are passed to CAEDYM, typically at a 1-h time step, and are used to determine the rates of change of biomass and chemical constituents for each of the ecological state variables. DYCD-FISH couples a fish population model with DYCD to create a combined ecosystem-fish model capable of integrating the relatively slow metabolism of fish with other ecosystem components that often respond faster than fish. The model also accounts for the direct and indirect feedback between fish dynamics and the biogeochemical and planktonic modules. Fish directly affect the concentrations of the zooplankton, the particulate and dissolved organic matter (POM and DOM, respectively), oxygen, and the dissolved inorganic carbon in the different layers where the fish are located. Specifically, the Lake Kinneret version of DYCD-FISH is configured to simulate the carbon, nitrogen, phosphorus, and oxygen cycles, along with the biomass and metabolic processes of five phytoplankton groups (the dinoflagellate *Peridinium gatunense*; the filamentous diatom *Aulacoseira granulata*; the toxin-producing, N-fixing filamentous cyanobacterium *Aphanizomenon* sp.; the toxic colonial cyanobacterium that does not fix gaseous N, *Microcystis* sp.; and a general group of small-cell species, termed nanoplankton), three zooplankton groups (predatory zooplankton: adult stages of the predatory copepods and predatory rotifers; herbivorous zooplankton: cladocerans and copepodites; and micro-zooplankton including flagellates, ciliates, and copepod nauplii), heterotrophic bacteria (all simulated in units of mgC·L^−1^), and the population dynamics of the dominant fish in Lake Kinneret, *Mirogrex terraesanctae* \[[@B27-biology-06-00022]\]. The fish sub-model is based on a generic bioenergetics formulation, whereby an energy balance equation equates the energy gained as the difference between that consumed and metabolic expenses. Specifically, energy gain occurs through the difference between consumption and respiration, excretion, egestion, specific activity, and reproduction. The energy gained affects the growth rate and fish wet weight, and therefore, the growth rate of an individual fish is represented as the daily changes in wet weight per unit of fish weight per day (g~ww~·g~fish~^−1^·day^−1^). In DYCD-FISH, the bioenergetics model is applied to many "super-individuals", where each "super individual" represents numerous similar individuals \[[@B45-biology-06-00022]\], which each experience processes such as recruitment, and natural and fishing mortality that dynamically affect the number and biomass of fish. The population model takes advantage of the 1D hydrodynamic model, and at each time step, based on ambient conditions, the fish are redistributed in the water column with some stochastic variability, imitating the natural spatial movement and spread of the fish. Since the model is spatially (vertically) explicit, zooplankton, phytoplankton, and fish are located in varying water layers, such that the feeding intensity depends on the habitat overlap between predator and prey. Additionally, an explicit recruitment model was developed to simulate the reported linkage between the change in water level and Lavnun spawning success. The specific parameters and customization used in the current fish population model are described in Makler-Pick et al. \[[@B43-biology-06-00022]\]. The fish-related variables examined in this study included fish biomass and the fish predation rate for zooplankton. Since the focus of the manuscript is on fish-related food web dynamics associated with the IGP, we will now describe relevant components and interactions included in the Lake Kinneret implementation of DYCD-FISH. The Lavnun (the IG-predator) is an endemic, visually orienting zooplanktivore. As evident by the positive electivity values (i.e., preferences to food source) for Cladocera and Copepoda \[[@B46-biology-06-00022]\], it mainly feeds on Cladocera and Copepoda. The diet composition of the Lavnun also includes micro-zooplankton and particulate detrital material. When there are multiple prey types, the actual consumption depends on prey densities, the vulnerability of each prey item to the predator, the half-saturation constants governing the rate of feeding, and the maximum daily ration at a particular mass and temperature. Prey vulnerability, defined in the literature as the product of the encounter rate between predator and prey and the capture success of the predator \[[@B47-biology-06-00022]\], is dependent on several factors, such as abiotic conditions, the relative size of predator and prey, and predator activity and movements. Here, however, because of the lack of vulnerability experimental data, we used stomach content analysis \[[@B48-biology-06-00022],[@B49-biology-06-00022]\] as an indicator of prey vulnerability and set the initial food vulnerability of the predatory zooplankton, herbivorous zooplankton, and micro-zooplankton to fish predation rates of 0.4, 0.5, and 0.02, respectively. The advantage of the current model is that it enables an exploration of the theoretical impacts of imprecise parameters, such as food vulnerability, the predation rate, and feeding preference, by substantially modifying their values. The diet of the predatory zooplankton, the IG-prey, consists of micro-zooplankton, herbivorous zooplankton, and the early stages of predatory zooplankton (via self-limiting predation) with preferences of 0.5, 0.35, 0.15, respectively. The main dietary source of herbivorous zooplankton (the IGP basal resource) is a multi-species group of small-celled phytoplankton termed the nanoplankton. 2.2. Model Base-Case Simulations {#sec2dot2-biology-06-00022} -------------------------------- The simulations were configured to run from Jan. 1997 to Sep. 2003, at a 1 h time step for DYCD and a daily time step for the fish model. The initial conditions, forcing data, and input data are as described in Gal et al. \[[@B39-biology-06-00022]\] and Makler-Pick et al. \[[@B43-biology-06-00022]\]. The initial number of fish was set to 100 million fish in the lake (hereafter, ×1 or base level), as validated in Makler-Pick et al. \[[@B43-biology-06-00022]\]. The model time-series output files were used to study the temporal dynamics of zooplankton and fish biomass, as well as the predation rates of both fish and predatory zooplankton (the model is available upon request from the software developers). Multiple simulation runs verified that the stochastic component within the individual-based fish model resulted in negligible differences in fish population behavior for repeated simulation runs. 2.3. IGP Scenarios and Sensitivity Analysis {#sec2dot3-biology-06-00022} ------------------------------------------- To simulate IGP dynamics when the number of fish is higher than the multiannual average, a series of scenarios were conducted, in which the number of fish represented by each "representative" super-individual was changed. The lake-wide fish population was initialized with 1000 "representative" fish, each set to represent 100,000 fish in the lake, equating to 100,000,000 fish in total. Then, this number was changed from an initial number of 100,000 (×1) to 200,000 (×2), and then up to 800,000 (×8) fish per "representative" fish, which was higher than the estimated increase in fish density recorded in 2004 \[[@B30-biology-06-00022]\], but enabled us to examine an extreme situation. Based on the outcome of the scenarios, we studied the impact of increasing fish numbers on the zooplankton biomass and predation rates. The results of complex models often suffer from limitations due to various sources of error and uncertainty, such as the initial conditions, input data, model structure, model parameters, and validation data \[[@B50-biology-06-00022]\]. However, a large source of model uncertainty is associated with the parameter values \[[@B50-biology-06-00022],[@B51-biology-06-00022]\]. When a high uncertainty in the value of a parameter coincides with a high sensitivity of the model to that parameter, the reliability of the model predictions may be very low \[[@B52-biology-06-00022]\]. Makler et al. \[[@B53-biology-06-00022]\] and Gal et al. \[[@B54-biology-06-00022]\] evaluated the sensitivity of DYCD model parameters. Here, the sensitivity of the herbivorous zooplankton to different key factors playing a part in the IGP triangle are explicitly studied. The initial values of the following parameters: the maximum predation rate of the predatory zooplankton (g~max~, gC·m^−3^ (gZ·m^−3^)^−1^·day^−1^), the feeding preference of predatory zooplankton to the early stages of predatory zooplankton (cannibalism or self-limiting predation) (P~zk1~), and the vulnerability of predatory (V~11~) and herbivorous (V~12~) zooplankton to fish predation, were modified. The changes were made over the course of a series of scenarios by changing the initial value of each parameter, one at a time, by ±50%. The average concentration of the herbivorous zooplankton over the seven-year simulation was calculated and served as the basis for comparison. 2.4. Biomanipulation {#sec2dot4-biology-06-00022} -------------------- Fish biomanipulation (namely, the positive size-selective harvest of mainly non-commercial size fish) is a possible management action aiming to control high fish recruitment, such as occurs in Lake Kinneret in atypical years, to reduce phytoplankton and ultimately improve the water quality. The size-selective harvest of the Lavnun was modeled by increasing the value of the fishing mortality (exploitation rate) parameter of non-commercial size fish from 0 to 50 (%·year^−1^), and the value of the fishing mortality parameter of commercial size fish from 28 to 50 (%·year^−1^), based on the scenario of an atypical year (×8 scenario). 3. Results {#sec3-biology-06-00022} ========== 3.1. IGP Dynamics under Typical Conditions {#sec3dot1-biology-06-00022} ------------------------------------------ The herbivorous zooplankton in the base simulation level (×1) showed seasonal biomass peaks from fall to spring, with low inter-annual variation throughout the simulation period ([Figure 2](#biology-06-00022-f002){ref-type="fig"}A). The predatory zooplankton (cyclopoid copepods) showed annual biomass peaks in late winter to early spring, with low interim values ([Figure 2](#biology-06-00022-f002){ref-type="fig"}B). The average simulated lake-wide total biomass of predatory cyclopoid over the years 1997--2003, was 799 × 10^6^ g~ww~, with a mean daily predation rate of 109 mg~prey~·g~pred~^−1^·day^−1^. During this period, the Lavnun had an average total lake biomass of 1281 × 10^6^ g~ww~ and the mean daily predation pressure on herbivorous zooplankton was 28 mg~prey~·g~pred~^−1^·day^−1^. The simulated predation pressure by both the Lavnun and the predatory zooplankton varied seasonally. During winter and early spring (November-February), predation by predatory zooplankton accounted for the majority of the predation mortality for herbivorous zooplankton ([Figure 3](#biology-06-00022-f003){ref-type="fig"}A), imposing a predation pressure 10--20 times higher than the Lavnun predation pressure ([Figure 3](#biology-06-00022-f003){ref-type="fig"}B). At the time of its annual peak, in December, the monthly average predation rate of predatory zooplankton was 7.0 μgC·L^−1^·day^−1^ (10% of herbivorous zooplankton biomass per day), and remained high in January, before declining to very low levels (0.004 μgC·L^−1^·day^−1^) between March and October ([Figure 4](#biology-06-00022-f004){ref-type="fig"}A). During this time period, the fish predation rate was higher than the zooplankton predation rate ([Figure 4](#biology-06-00022-f004){ref-type="fig"}B). The annual cycle of the fish predation rate was more moderate than that of the predatory zooplankton, ranging between a minimum of 0.2 μgC·L^−1^·day^−1^ in September, to a maximum of 0.7 μgC·L^−1^·day^−1^ in November (1% of herbivorous zooplankton biomass concentration per day). From spring to autumn, fish predation was the dominant loss of herbivorous zooplankton, in tandem with a 40% lower herbivorous zooplankton biomass in comparison to late winter (November--February, [Figure 2](#biology-06-00022-f002){ref-type="fig"}A and [Figure 5](#biology-06-00022-f005){ref-type="fig"}). However, in late winter, during the short period of the annual biomass peak of the predatory zooplankton, there was an extremely high predation mortality (\>80% of total herbivorous zooplankton) associated with zooplankton predation. 3.2. IGP Dynamics during High Fish Abundance {#sec3dot2-biology-06-00022} -------------------------------------------- Throughout most of the high fish abundance simulation (×8), the concentration of predatory zooplankton and micro-zooplankton was lower than in the base level scenario (ratio \< 1 in [Figure 6](#biology-06-00022-f006){ref-type="fig"}A,C). The herbivorous zooplankton biomass was also affected by the changes in fish densities. This impact, however, mainly occurred in the periods between the seasonal biomass peaks of the herbivorous zooplankton. In the presence of higher fish numbers, the concentration of herbivorous zooplankton during its simulated seasonal biomass peaks (January--February) was similar or even higher than the concentration simulated at the base level of fish. This phenomenon is indicated by values higher than one in [Figure 6](#biology-06-00022-f006){ref-type="fig"}B. In the presence of ×8 more fish, fish predation of herbivorous zooplankton dominates the predation in this group for most of the simulation period (values \< 1 in [Figure 3](#biology-06-00022-f003){ref-type="fig"}B), with some exceptions during the fourth, sixth, and seventh year of the simulation, when the ratio of zooplankton predation to fish predation was greater than one. Substantial differences between the zooplankton predation pressure on herbivorous zooplankton in the ×8 scenario and the ×1 simulation are mainly observed during the annual peak of the herbivorous zooplankton biomass, demonstrating a considerable seasonal decrease in the net predation by predatory-zooplankton at the ×8 scenario ([Figure 7](#biology-06-00022-f007){ref-type="fig"}). Intriguingly, the lower predation pressure may allow for an increase in herbivorous zooplankton biomass during these periods ([Figure 6](#biology-06-00022-f006){ref-type="fig"}B). At all other times, model simulations suggest that the herbivorous zooplankton biomass is mainly controlled by fish predation. The mean annual predation by fish and zooplankton on the different zooplankton groups at varying levels of fish abundance, are given in [Table 1](#biology-06-00022-t001){ref-type="table"}. At the base level (×1), the mean annual predation of predatory zooplankton is 2.5, 5.3, and 7.7 times higher than fish predation on herbivorous, predatory, and micro-zooplankton, respectively. These results demonstrate the predation superiority of the predatory zooplankton (over the fish) at the base level of fish. An increase in the number of fish resulted in a lower mean annual predation rate of predatory zooplankton, both on itself (i.e., self-limitation) and on the other zooplankton groups. The main impact was on the predatory zooplankton itself; the self-limitation was lowered to 9% of the base level in the presence of ×8 more fish. On average, the outcome of the ×8 scenario was a 70% lower total zooplankton predation (on all zooplankton groups), and on the other hand, a higher fish predation, on all its prey types (\>200%). Interestingly, the total herbivorous zooplankton eaten annually, whether by fish or by zooplankton, was close to 44 × 10^3^ ton year^−1^ for all simulated scenarios, regardless of fish biomass. The total predation in the system (the sum of zooplankton predation and fish predation) ranged from 52 × 10^3^ to 60 × 10^3^ ton. At the base level of fish, and even when the number of fish was doubled, the ratio between the annual average of zooplankton predation to the annual average of fish predation on all zooplankton groups was greater than one, indicating that, in total, the main predator in the ecosystem is the predatory zooplankton ([Table 1](#biology-06-00022-t001){ref-type="table"}). A higher number of fish (×8) resulted in a ratio lower than one, indicating the dominance of fish in consuming zooplankton under these conditions. 3.3. The Effect of Biomanipulation {#sec3dot3-biology-06-00022} ---------------------------------- Manipulating the fishery by imposing particularly heavy fishing mortality on non-commercial (\<12 cm) and commercial size fish, at the ×8 scenario, demonstrated a decrease in the total fish biomass ([Figure 8](#biology-06-00022-f008){ref-type="fig"}A) and an increase in the zooplankton biomass. The increase was particularly considerable for the predatory zooplankton ([Figure 8](#biology-06-00022-f008){ref-type="fig"}B) and more moderate for the herbivorous zooplankton, where no substantial impact was shown in the timing of the annual peaks ([Figure 8](#biology-06-00022-f008){ref-type="fig"}C). 3.4. Sensitivity Analysis of the IGP Component {#sec3dot4-biology-06-00022} ---------------------------------------------- The relative change (%) in comparison to the base model results is presented in [Table 2](#biology-06-00022-t002){ref-type="table"}. The results indicate that the greatest impact on the average herbivorous zooplankton biomass was by the maximum predation rate of the predatory zooplankton parameter (g~max~). Decreasing its value by 50% resulted in an increase of more than 27% in the herbivorous zooplankton biomass concentration, while increasing its value by 50% resulted in a 10% decrease in the herbivorous zooplankton concentration. Decreasing the vulnerability of herbivorous zooplankton to fish predation (V~12~) by 50% had a positive impact, resulting in a 10% increase in population size, and a comparable but negative impact was observed when the vulnerability was increased by 50% (−7%). 4. Discussion {#sec4-biology-06-00022} ============= The simulation results indicated that predatory zooplankton (and not the Lavnun) control herbivorous zooplankton predation when the abundance of fish is similar to the multi-annual average ([Figure 3](#biology-06-00022-f003){ref-type="fig"}, [Figure 4](#biology-06-00022-f004){ref-type="fig"} and [Figure 5](#biology-06-00022-f005){ref-type="fig"}); this accounts for a maximum daily mortality rate of 10% of the population, in comparison to 1% imposed by the fish. The predatory zooplankton consumption varied seasonally, and at the time of the annual biomass peaks it exerted a predation pressure that was 10--20 times higher than fish predation. Predation rates are typically measured by feeding experiments with different food types and/or radioactively or fluorescently labeled food, serial dilution/concentration experiments, stable isotope experiments, gut content analyses, and bioenergetics modeling \[[@B55-biology-06-00022]\]. However, none of these methods are continuous, long-term, or in-situ, and therefore limit the validation of the model. Thus, to assess the simulation results, we compare several indirect metrics of the model simulation with available data from Lake Kinneret and other sites. For example, Blumenshine and Hambright \[[@B6-biology-06-00022]\], reported an order of magnitude higher mass specific consumption rate on Lake Kinneret herbivorous zooplankton by cyclopoid copepods than the Lavnun (mean value of 921 mg~prey~·g~pred~^−1^·day^−1^ for cyclopoid copepods in comparison to 74 mg~prey~·g~pred~^−1^·day^−1^ for Lavnun), and an exertion of 66% of the predation pressure on herbivorous zooplankton. In that research, fish consumption was estimated by using the observed Lavnun growth rate as the input for a bioenergetics model. The model output was an estimate of the amount of energy needed to manifest the observed growth rate. The monthly temperature-dependent specific ingestion rates for predatory zooplankton were based on data for adult female *M. ogunnus* from Gophen \[[@B56-biology-06-00022]\]. The difference between the simulation results in the current study and those reported by Blumenshine and Hambright \[[@B6-biology-06-00022]\], can be partly attributed to the different methods employed. For example, the DYCD model calculates zooplankton predation as the product of the predation rate coefficient, temperature dependence function of predation, and the overall potential rate of carbon consumption, which is regulated by a Michaelis-Menten function. Alternatively, Blumenshine and Hambright \[[@B6-biology-06-00022]\] calculated zooplankton predation rates based on the temperature-dependent empirical quadratic equation. Placing our model simulated data (i.e., simulated predation rate) and measured biomass of predatory zooplankton and herbivorous zooplankton into this equation, gives an average predatory zooplankton predation rate of 1029 mg~prey~·g~pred~^−1^·day^−1^, which is comparable to the 921 mg~prey~·g~pred~^−1^·day^−1^ reported by Blumenshine and Hambright \[[@B6-biology-06-00022]\], providing an additional indirect validation to DYCD-FISH. In research conducted at the temperate Lake Ontario, Gal et al. \[[@B7-biology-06-00022]\] found that the consumption rates of *Mysis* (IG-prey) and fish (IG-predator) varied widely with the season, and that *Mysis* predation (2.6 × 10^−3^--1.3 g~prey~·m^−2^·day^−1^) is superior over fish predation (1.4 × 10^−3^--0.5 g~prey~·m^−2^·day^−1^) in the summer, but that fish predation is dominant in October. Morin \[[@B5-biology-06-00022]\] examined the interactions between two freshwater protists, *Colpidium striatum* (the IG-prey) and *Blepharisma americanum* (the IG-predator), in laboratory microcosm experiments. Their results supported the predictions of the IGP theory and the dominance of the IG-prey. The outcome of exceptionally rapid and large increases in water levels in Lake Kinneret, and the subsequent increase in the number of Lavnun fish, can be described in terms of top-down control. The simulation results demonstrated that an eightfold increase of fish causes: (1) a shift of fish body size towards smaller sizes, resulting in most of the fish being of a sub-commercial size \[[@B57-biology-06-00022]\]; (2) lower zooplankton biomass ([Figure 6](#biology-06-00022-f006){ref-type="fig"}); and (3) lower predation of zooplankton ([Table 1](#biology-06-00022-t001){ref-type="table"}). These results can be compared to reported data in the years 1993 and 2004 \[[@B58-biology-06-00022],[@B59-biology-06-00022],[@B60-biology-06-00022]\], in terms of the linkage between large changes in the water level, the following increase in the number of fish, and zooplankton abundance. Furthermore, these results support the hypothesis that large increases in fish abundance following extremely wet winters cause increased fish predation on zooplankton, thus drastically reducing the total zooplankton densities. Lake-based data from the years 1990--2004 confirm an exceptional decrease in the mean monthly densities, mainly of predatory zooplankton and herbivorous zooplankton one to two years after atypical wet years \[[@B32-biology-06-00022],[@B58-biology-06-00022]\]. However, the decline is not consistent throughout the whole year; during certain months, zooplankton densities were high, relative to the levels observed in "typical" years. For example, in 2004, the observed annual average zooplankton density was one of the lowest ever recorded. Yet the density of herbivorous zooplankton in June was fairly high (70 organisms L^−1^) and the density of all zooplankton groups in that month was similar to the multi-annual mean \[[@B58-biology-06-00022],[@B61-biology-06-00022]\]. The comparable results simulated in the ×8 more fish scenarios, serve not only as indirect validation of the model processes, but clearly illustrate the possible effects of an IGP component in the food web. For example, from December to February, in the presence of a higher number of fish (×8), the lower predation pressure by predatory zooplankton allows for a temporary higher biomass of herbivorous zooplankton in comparison to the ×1 fish scenario ([Figure 6](#biology-06-00022-f006){ref-type="fig"}B). This finding illustrates how, in the presence of an extremely high number of fish, the system seasonally shifts the food web between being directly dominated by fish predation (when the food web somewhat resembles a linear food chain), to periods when the direct dominant effect, on herbivorous zooplankton, is the substantial release from the predatory zooplankton predation pressure caused by the increased fish predation. The outcomes of the long-term (1994--2006) biomanipulation (removal of 400--500 ton fish year^−1^) employed in Lake Kinneret can be, at least partly, explained within the framework of this model and the complexities of IGP dynamics. In the first five years of the biomanipulation program, zooplankton abundance in Lake Kinneret increased, reaching a peak in the year 2000, similar to those measured in the 1970's \[[@B31-biology-06-00022],[@B58-biology-06-00022]\]. Since then, and disagreeing with the expectations, zooplankton abundance has steadily decreased, reaching one of its lowest ever mean annual biomass values in 2004 \[[@B58-biology-06-00022]\]. Simulating biomanipulation with DYCD-FISH showed: (1) a decrease in fish biomass; (2) an increase in predatory zooplankton biomass; and (3) a moderate increase in herbivorous zooplankton biomass in periods when predatory zooplankton biomass is relatively low and an insignificant change when predatory zooplankton is at its seasonal peak ([Figure 8](#biology-06-00022-f008){ref-type="fig"}). Blumenshine and Hambright \[[@B6-biology-06-00022]\] claim that the lower-than-expected impact of Lavnun removal on herbivorous zooplankton biomass in Lake Kinneret can be attributed to the possible boosting of cyclopoid copepod abundance, caused by the reduced Lavnun biomass, and the following increase in the net predation pressure on herbivorous zooplankton. The simulation results confirm this claim, interestingly also indicating that the total zooplankton loss to predation, and particularly that of herbivorous zooplankton, is largely indifferent to the amount of Lavnun. The specific predation on herbivorous zooplankton is relatively constant (44 × 10^3^ ton·year^−1^), but channeled differently into either predatory zooplankton or fish, depending on the fish density in the lake ([Table 2](#biology-06-00022-t002){ref-type="table"}). This was true not only for a small increase (×2) in the number of fish, but also for a substantial increase (×8). The simulated biomanipulation results suggest a possible reason why the culling program (the subsidized harvest) implemented in Lake Kinneret did not have the desired impact on the water quality. By removing fish from the lake (and decreasing fish predation pressure), zooplankton predation on herbivorous zooplankton is increased (mainly at the time of the seasonal peak), weakening the cascading effects of top-down control. This serves as an example for compensatory effects within the food web, where trophic interactions mediate resilience to the overall system in the sense that perturbations at the top of the food web are dampened and only marginally affect the lower parts of the food web \[[@B6-biology-06-00022]\]. The end result regarding the predation of herbivorous zooplankton is approximately the same, regardless of whether there are high or low numbers of fish. The sensitivity test of the simulated herbivorous zooplankton biomass was conducted by changing the values of the grazing-related parameters, such as the "maximum predation rate of predatory zooplankton" (g~max~). Based on laboratory experiments \[[@B6-biology-06-00022],[@B56-biology-06-00022]\] and model calibrations, the initial value of the parameter was set to 3.03 gC·m^−3^ (gZ·m^−3^)^−1^·day^−1^. The sensitivity analysis indicated that g~max~ had the highest impact on herbivore biomass, when compared with parameters such as "predatory zooplankton diet preference" and "zooplankton vulnerability to fish predation". Our findings also indicate that increasing (or decreasing) g~max~ by 50% does not affect the seasonal predation dynamics or the conclusion, mentioned earlier, that a higher number of fish can seasonally reduce the predation pressure on the herbivorous zooplankton. According to the sensitivity test, the parameter "predatory zooplankton self-limitation" had a relatively low impact. It should be noted, however, that the self-limitation refers to the self-cannibalism of the predatory zooplankton, while in reality, it would be mainly due to adults consuming herbivorous copepodites or nauplii. According to Hart \[[@B13-biology-06-00022]\], the self-limitation among the IG-prey is a key factor in determining the direction and strength of the top-down response and the IG-predator will only induce an increase in the basal resource if IG-prey self-limitation is adequately strong. Exploring IGP is challenging, both empirically and in terms of modelling. Many IGP systems are embedded in communities with several alternative prey species; therefore, there are often more species involved in intraguild predation than just the three used in the basic conceptual model. This can explain why simple theories and models fail to capture real-world data-sets \[[@B11-biology-06-00022],[@B26-biology-06-00022]\]. As was demonstrated by a set of structured models \[[@B62-biology-06-00022]\], depending on the trophic position where self-limitation occurs, it can entirely modify the dynamics and structure of three-species IGP systems. Furthermore, empirical tests of ecological theory are limited by the mismatch between the short-term population dynamics data that are readily obtained from field experiments (i.e., changes in population size over one to at most several generations), and the long-term dynamics described by models (i.e., whether two species will coexist stably for many generations). The short duration of most field studies relative to the generation times of the organisms disrupt our ability to make strong conclusions about population stability or prolonged coexistence \[[@B5-biology-06-00022],[@B11-biology-06-00022],[@B63-biology-06-00022]\]. The difference between the DYCD-FISH and other models can be attributed to the more conceptual and theoretical nature of these models. Most IGP models described in the relevant literature are relatively simple \[[@B1-biology-06-00022],[@B13-biology-06-00022],[@B15-biology-06-00022],[@B20-biology-06-00022]\]. The models employed by Hart \[[@B13-biology-06-00022]\], for example, were deliberately kept as simple as possible, so that they could be easily compared to simple food chain models. However, this approach is not able to account for major factors such as inedible algae, nutrient cycling, the nutrient limitation of consumers, and the seasonal dynamic interactions between all of these components and other food web components, making it unclear if they are able to induce dynamics which are different from those predicted by the classic food chain theory. In the current model, the IGP interactions are affected by variability in lake-hydrodynamics, and chemical and biological interactions. The model employed here does not assume that the IGP-predator and the IG-prey compete over a single prey species or that the populations are in equilibrium. In particular, the IGP dynamics in this study are examined under non-equilibrium conditions, where the IG-prey population shows a strong boom-bust response relative to the fish whose role is more constant, thereby capturing the natural environmental dynamics that influence the different IGP components. The explicit integration of complex networks, and non-linear and non-equilibrium modeling allows for the exploration and examination of long-term various relationships and dynamics in a system resembling the real-world food web. In the current approach, a complex model is confronted with theoretical concepts and paradigms \[[@B64-biology-06-00022]\]. The outcome complies with the conclusion of the simpler models that the IG-prey must be competitively superior to the IGP-predator, but uniquely shows the seasonal dynamics of an IGP-predator that is more effective at suppressing the target herbivore population (i.e., the ×8 scenario in this study), thus reducing the IG-prey and the inter-and intra-seasonal impact on the resource (herbivore population) and other related ecosystem components, such as the micro-zooplankton. In many aspects, the model is validated both directly and indirectly. Nevertheless, there are limitations and inaccuracies in the model, producing differences between observed and simulated outcomes. For example, zooplankton dynamics do not always match empirical observations, particularly for predatory zooplankton. The model underestimates predatory zooplankton abundance; the simulated biomass is low for most of the year and the amplitude is biased towards a high value, while the observed seasonal pattern is not as clear and obvious as that seen in the model output. The herbivorous zooplankton biomass is also underestimated. However, it seems consistent that increasing its biomass will increase its predation by the predatory zooplankton and will support our conclusion that the dominant predator of the herbivorous zooplankton is the predatory zooplankton. The differences between the simulated and observed zooplankton biomass can be attributed to: (1) the variability of the field data resulting from long-term changes that occurred in the lake ecosystem \[[@B32-biology-06-00022],[@B55-biology-06-00022]\]; (2) changes in the sampling protocols, with a shift from a vertical mix sample to depth specific sampling that indicates a large degree of variation across the various months for all three zooplankton groups, thus hindering the detection of any statistically significant seasonal patterns \[[@B58-biology-06-00022]\]; (3) inadequate accuracy in the model at the base of the food web \[[@B39-biology-06-00022]\] and the absence of a self-contained model for zooplankton that accounts for the maturation of juvenile copepods in the predatory compartment. This obstacle could be eliminated by adding a stage-structured model, but at the cost of introducing additional complexity, error, and uncertainty into the model \[[@B65-biology-06-00022],[@B66-biology-06-00022]\]; and (4) the averaging of spatial processes taking place in the lake in the one-dimensional modeling approach. However, the aim of the research, which was to gain a better understanding of the processes and interactions within the food web, is achieved using the current complexity of the model. The simulation results can be further validated by future research, both in Lake Kinneret and in ecosystems where the abundance of fish or zooplankton has changed as a result of natural or anthropogenic actions (e.g., overfishing, stocking), or in ecosystems where the species composition was altered as a result of species invasion. Additionally, employing the approach presented here within a 3D hydrodynamic model would allow an exploration of the role of spatial variation in IGP interactions and is suggested as an area of future research. 5. Conclusions {#sec5-biology-06-00022} ============== IGP dynamics vary seasonally and influence the structure and function of aquatic food webs. When the biomass of the IGP-predator is significantly higher, like in atypical meteorological years, the effect is a shift from a bottom-up controlled ecosystem to top-down control, with seasonal exceptions leading to an increase of resource (shared prey) biomass, to extents similar to the base level (typical meteorological years). The analysis suggests that IGP dynamics should be considered in food web models, in order to more accurately capture mass transfer and trophic interactions. We wish to thank James Shapiro for collecting and making available Lavnun fishery data. We thank Kinneret Limnological Laboratory (KLL) staff for their assistance in data extraction, organization and for running DYCD. Support for Vardit Makler-Pick was provided by the Yohay Ben-Nun Scholarship fund and The Australia-Israel Scientific Exchange Foundation. Additional support was provided by the Israeli Water Authority. Vardit Makler-Pick conceived the study, designed the study, performed the research, analyzed the data, interpreted the results and wrote the paper. Matthew Hipsey conceived the study, contributed to the study design and analysis approach, supported the model application and contributed to the preparation of the article. Tamar Zohary has contributed to the understanding of the functioning of the Kinneret ecosystem and its food web and to the write-up of the manuscript. Yohay Carmel has conducted parts of the statistical analyses, advised on other parts of the analyses, and improved sections of the text. Gideon Gal conceived the study, assisted in the analysis and interpretation of the results and improved the text. The authors declare no conflict of interest. {#biology-06-00022-f001} {#biology-06-00022-f002} {#biology-06-00022-f003} {#biology-06-00022-f004} {#biology-06-00022-f005} {#biology-06-00022-f006} {#biology-06-00022-f007} {#biology-06-00022-f008} biology-06-00022-t001_Table 1 ###### Comparison of the fish (F) and zooplankton (Z) average annual predation rate (thousand ton year^−1^) and ratio of zooplankton predation to fish predation for each zooplankton group, at base level of fish (×1), at ×2 fish, and at ×8 fish. Values in parentheses indicate the relative fraction (in %) of predation pressure compared to the base level. Z1~z1~, Z1~z2~, and Z1~z3~ are predatory zooplankton predation on predatory (Z1), herbivorous (Z2), and micro (Z3) zooplankton, respectively. F~z1~, F~z2~, and F~z3~ are fish predation on predatory, herbivore, and micro zooplankton, respectively. Variable ×1 ×2 ×8 ---------------------------------------- ------- -------------- -------------- Z1~z1~ 12.62 6.89 (54%) 1.17 (9%) F~z1~ 2.37 3.48 (146%) 5.01 (211%) Z1~z2~ 31.81 25.32 (79%) 12.11 (38%) F~z2~ 12.83 18.57 (144%) 33.10 (258%) Z1~z3~ 0.59 0.45 (75%) 0.16 (27%) F~z3~ 0.08 0.13 (162%) 0.35 (438%) Total Z~zi~ predation 45.02 32.66 (76%) 13.43 (30%) Total F~zi~ predation 15.28 22.19 (145%) 38.45 (252%) Total predation 60.3 54.84 51.9 Total predation on Z2 (Z1~z2~ + F~z2~) 44.64 43.89 45.21 Z1~z1~/F~z1~ 5.32 1.98 0.23 Z1~z2~/F~z2~ 2.48 1.36 0.37 Z1~z3~/F~z3~ 7.7 3.4 0.45 biology-06-00022-t002_Table 2 ###### The change in the average concentration of herbivorous zooplankton (% of ×1) following changes of +50% and −50% in the parameter values. Initial parameter values are provided. Parameter Initial Parameter Value −50% Change to Parameter Value +50% Change to Parameter Value --------------------------------------------------------------------------------------------- ------------------------- -------------------------------- -------------------------------- Maximum predation rate of predatory zooplankton (g~max~) 3.03 +27% −10% Preference of predatory zooplankton for predatory zooplankton---self limitation (P~zk1~) \* 0.15 −7% +1% Vulnerability of the predatory zooplankton (V~11~) 0.4 −8% +1% Vulnerability of the herbivorous zooplankton (V~12~) 0.5 +10% −7% \* The preferences of predatory zooplankton to all of its prey types must sum to 1. Therefore increasing or decreasing the preference of predatory zooplankton to the predatory zooplankton (self-limitation) by ±50% requires a change in the preference to the other prey type. The preference to micro-zooplankton was modified to comply with this requirement.

Role of monocytes/macrophages in atherosclerosis, coronary artery disease, acute coronary syndrome and vascular inflammation ============================================================================================================================ Myocardial infarction (MI) is one of the leading causes of morbidity and mortality in western societies (Hausenloy, [@B27]). Many risk factors like arterial hypertension, diabetes mellitus or lipid metabolism disorders can result alone or in combination in stenosis of the coronary blood vessels. Within the last years multiple genetic factors and the influence of the immune system were described in the pathogenesis of atherosclerosis. Innate immunity pathways have long been suspected to contribute to the initiation and progression of atherosclerosis (Shibata and Glass, [@B51]). Hence, atherosclerosis and its sequela can be considered as both a chronic inflammatory disease and as a metabolic disorder. The sequence of neutrophils and monocytes ----------------------------------------- Polymorphonuclear leukocytes (PMNs, reflecting mainly neutrophil granulocytes) dominate the initial influx of leukocytes to sites of acute infection and inflammation followed by emigration of monocytes. In 1968 Ward et al. described first that there exists a causal link between initial extravasation of PMNs followed by later emigration of monocytes (Ward, [@B63]). Gallin et al. identified that in patients who suffer from specific granule deficiency had a reduced effect on chemotactic effect on monocytes (Gallin et al., [@B21]). In different mouse studies it seemed that recruited PMNs trigger monocyte recruitment by releasing soluble factors (Doherty et al., [@B12]; Fillion et al., [@B15]; Janardhan et al., [@B29]). Different granule proteins of PMNs, for example cathepsin and human neutrophil peptide 1--3, with monocyte chemotactic activity could be identified (Territo et al., [@B58]; Chertov et al., [@B9]). Peripheral blood monocytes are a heterogeneous group of circulating leukocytes, which can be distinguished in different subsets with diverse functions. These subsets have different phenotypes in mice and humans (Tacke et al., [@B56]; Ley et al., [@B34]). Geissmann et al. identified two subsets in murine blood that can be discriminated by the receptor for the chemokine fractalkine, the CX~3~CR~1~: The short lived CX^lo^~3~CR~1~CCR2^+^GR1^+^ (more specifically, Ly6C^hi^) cells, which are pro-inflammatory, are chemoattracted by MCP-1 and migrate to sites of inflammation and a CX~3~CR^hi^~1~CCR2^−^GR1^−^ subset (Ly6C^lo^) not attracted by inflamed tissues, that was later termed reparative. The CX~3~CR^lo^~1~CCR2^+^GR1^+^ cells correspond to CD14^hi^CD16^−^ monocytes, whereas the latter were equivalent to CD14^lo^CD16^+^ monocytes in humans (Geissmann et al., [@B22]). They express different levels of chemokine receptors and cytokines depending on their function (Weber et al., [@B64]). CD14^hi^CD16^−^ monocytes produce high levels of pro-inflammatory cytokines such as tumor necrosis factor alpha (TNF-α) and low levels of anti-inflammatory cytokines such as IL-10. In contrast, CD14^lo^CD16^+^ monocytes have been shown to produce high levels of IL-10 and low of TNF-α (Patino et al., [@B45]; Belge et al., [@B5]; Skrzeczynska-Moncznik et al., [@B52]). Shortly after acute MI in patients CD14^hi^CD16^+^ monocytes were increased and decreased at day 7 (Tsujioka et al., [@B60]). All subsets of monocytes can differentiate into macrophages after being recruited to lesions and then can be found in all tissues, where they have functional diversity (Wynn et al., [@B68]). The macrophages can be divided in M1 and M2 macrophages, M1 are activated by interferon gamma (IFN-γ) and have a more inflammatory phenotype by producing nitrogen intermediaries and pro-inflammatory cytokines (IL-1β and TNF-α) and participating in TH1 polarization. M2 macrophages act more in an anti-inflammatory way (IL-12^low^ and IL-10^high^) and participate in TH2 responses (Murray and Wynn, [@B41]). In mouse models of atherosclerosis, an infiltration of M2 macrophages in the early lesions could been identified, whereas M1 macrophages were predominant in advanced stages of the disease (Khallou-Laschet et al., [@B31]). Interestingly, it was recently shown, that experimental MI accelerates atherosclerosis in high fat diet fed ApoE^−/−^ mice driven by inflammatory Ly6C^hi^ monocytes (Dutta et al., [@B14]). This study provided a mechanistic explanation for the fact that history of MI is a risk factor for cardiovascular disease. It also highlightens the essential role of monocyte driven vascular inflammation for coronary atherosclerosis and MI. The inflammatory response in cardiac ischemic injury ---------------------------------------------------- Immediately after cardiac ischemia, different inflammatory signals recruit neutrophils to the infarct zone within 24 h, and monocytes/macrophages shortly thereafter (Figure [1](#F1){ref-type="fig"}). These homed leukocytes degrade extracellular matrix constituents and macromolecules released by injured cells. Neutrophils play an important role in the clearance of dead cardiac myocytes and their debris (Nahrendorf et al., [@B42]), release oxidants (mainly via their phagocyte type NADPH oxidase) and proteases and secrete mediators for inflammatory cell recruitment. Romson and coworkers could show in 1983, that depletion of these neutrophils in animals undergoing reperfused MI led to a marked decrease in infarct size, suggesting that a significant amount of myocardial injury induced by coronary artery occlusion followed by reperfusion may be neutrophil-dependent (Romson et al., [@B46]; Litt et al., [@B36]). The detrimental effects seem to be mediated by ICAM-1-dependent neutrophil-cardiomyocyte adhesion, a primary ligand of Mac-1 and CD11b/CD18 integrin. In an ICAM-1 ko mouse model it could be shown that there was less myocardial injury after MI (Metzler et al., [@B38]). Infarcted hearts modulate their chemokine expression profile over time, and sequentially and actively recruit Ly6C^hi^ and Ly6C^low^ monocytes. Ly6C^hi^ monocytes dominate early and exhibit phagocytic, proteolytic and inflammatory functions. In the later process Ly6C^low^ monocytes predominate. Consequently, Ly6C^hi^ monocytes digest damaged tissue, whereas Ly6C^low^ promote healing via myofibroblast accumulation, angiogenesis and deposition of collagen (Nahrendorf et al., [@B42]). In 2010 Panizzi et al. found that Ly-6Chi monocytosis disturbs resolution of inflammation in murine infarcts and consequently enhances left ventricular remodeling (Panizzi et al., [@B44]). ![**Phases of the inflammatory response in ischemic injury**. After ischemic injury, e.g., myocardial infarction, a sequence of inflammatory events has been proposed. As first line of defense, neutrophil granulocytes rush in clearing debris and paving the way for monocytosis by specific chemotactic patterns (Soehnlein et al., [@B54]; Wantha et al., [@B62]). Neutrophils are followed by inflammatory Ly6C^hi^CCR2^+^CX~3~CR^−^~1~ monocytes (day 2--4) which can either transdifferentiate into or are replaced by reparative Ly6C^low/-^CCR2^−^CX~3~CR^+^~1~ monocytes (day 5--10). Both subsets can give rise to macrophages and/or dendritic cells (starting from d7 to 10) or modulate the activity of those cells already residing in the heart. Alternatively, those macrophages or dendritic cells which are essential for post ischemic remodeling and scar formation can also emerge from tissue resident precursors or bone-marrow-lineage independent macrophages. How natural killer cells crosstalk to those innate cells in the setting of MI is poorly understood.](fphys-05-00295-g0001){#F1} After recruitment in the infarcted territory, monocytes may differentiate into macrophages. Local upregulation of macrophage colony-stimulating factor (M-CSF) may play an important role in this process (Frangogiannis et al., [@B18]). Macrophage accumulation in the healing heart is regulated by the renin-angiotensin-aldosterone system. Angiotensin II (ATII) allows deployment of myelomonocytic cells via the ATII receptor type 1 (AT~1~R) from the spleen and directs them to the inflamed tissue (Swirski et al., [@B55]). ATII leads to release of the mineralocorticoid aldosterone from the adrenal glands. Aldosterone has pro-inflammatory effects via its mineralocorticoid receptor (MR) on cardiomyocytes. Selective MR blockade immediately after MI improved healing (Frantz et al., [@B19]), an effect that was blunted by macrophage depletion (Fraccarollo et al., [@B17]). Cardiomyocyte specific genetic ablation of the MR improved healing, partially via increased RANTES/CCR5 dependent chemotaxis of myelomonocytic cells supporting healing (Fraccarollo et al., [@B16]). This indicates, that recruitment of inflammatory cells not necessarily *impairs* healing, but is possibly also *required* to kick off the outbalanced sequence of cellular events involved in healing. Indeed, mice that were either chemically or genetically depleted of inflammatory myelomonocytic cells, died from cardiac rupture and atherothrombosis propably due to impaired wound healing (Frantz et al., [@B20]). The beneficial impact of macrophages on remodeling can partially be assigned to the release of soluble lipid mediators like resolvins, lipoxins and maresins. These factors have been shown to contribute to resolution of inflammation in general (Serhan, [@B49]). In ApoE^−/−^ mice fed a high fat diet, failure to form the inflammation-resolving lipids lipoxin A~4~, resolving D1 and protectin D1 accelerated atherosclerosis progression, while macrophage-specific overexpression of 12/15 lipoxygenase reduced atherosclerotic burden via the formation of these mediators (Merched et al., [@B37]). If resolution of MI related cardiac injury is improved by these mediators remains to be tested. Mutual activation of NK cells and monocytes in vascular dysfunction ------------------------------------------------------------------- Besides CD8^+^ T cells and T-bet^+^CD4^+^ T cells, Natural Killer (NK) cells are a major source of IFN-γ. In type 1 inflammatory immune responses including atherosclerosis and acute phase of MI, IFN-γ is known to play a major role in activating monocytes and macrophages (Boehm et al., [@B8]; Schroder et al., [@B47]) and driving them toward inflammatory phenotype (e.g., M1 macrophages or Ly6C^hi^ monocytes) (Abbas et al., [@B1]). NK cells are key players in the innate immune response and produce a variety of cytokines such as IL-1beta, tumor necrosis factor alpha (TNFα), and IFN-γ and thereby participate in regulating innate and adaptive immunity (Yokoyama et al., [@B69]; Shereck et al., [@B50]). They represent a specialized lymphoid population, and act in an activating and inhibiting way by expressing different receptors (Moretta and Moretta, [@B40]). NK cells and monocytes/macrophages undergo reciprocal patterns of activation, that include receptor based cell-cell-contact triggered pathways mainly via CD154-CD40, 2B4, and NKG2D interactions and NKp46 conserved in a variety of species including mouse and human (for review, please see Michel et al., [@B39]). Of note, natural killer cell activation is in many aspects unique in humans as compared to mice, and even mice strains differ in the expression of certain surface markers and clusters of differentiation, like NK1.1, which is only found in C57BL/6 mice. In contrast, macrophage/NK cell interactions through soluble factors does not necessarily require cell-cell-contact for efficient signaling, which has gained growing attention in recent years. In a seminal work, Interleukin 12 (IL-12) was identified as *the* natural killer cell stimulatory factor (D\'andrea et al., [@B10]) released by activated monocytes. Shortly thereafter, IL-12 and TNF-α released by activated macrophages were identified to be powerful inducers of IFN-γ by NK cells (Tripp et al., [@B59]). It has been identified that macrophages can activate NK cells by producing interleukins such as IL-12 or IL-18 or through direct cell-to-cell contact (Aranha et al., [@B2]; Atochina and Harn, [@B3]). On the other side NK-cell-derived IFN-γ plays an important role in initializing the differentiation of monocytes into macrophages or dendritic cells, which are producers of IL-15, IL-12, and IL-18 (D\'andrea et al., [@B10]; Welte et al., [@B65]; Goldszmid et al., [@B23]). By synergizing of IL-12 with IL-18, production of IFN-γ in NK cells increases, which is due to a positive feedback loop that represent an important mechanism in the early innate inflammatory response. Mice deficient in the helix-loop-helix transcription factor inhibitor of differentiation (Id2(--/--)) which lack Langerhans cells, are defective in CD8^+^ T cells and dendritic cells as well as NK cells, were partially protected from ATII induced hypertensive damage. However, a clearcut effect of NK cells in this setting could not be delineated from these findings (Gratze et al., [@B24]). Interestingly, monocytes depend on T-bet for production of IL-12 and therefore for sufficient activation of NK cells to release IFN-γ (Soderquest et al., [@B53]). Kossmann et al. could recently show that angiotensin II-induced vascular dysfunction depends on vascular entry and IFN-γ production by NK cells. They identified IFN-γ was produced by NK cells in the aortic wall which initiate vascular oxidative stress, inflammatory cell recruitment and reciprocal innate immune cell activation in the vessel wall. In addition, T-bet deficient monocytes were not capable to induce ATII-dependent immune cell influx and NK-cell activation and depletion of NK-cells lead to protection from AT-II induced vascular dysfunction (Kossmann et al., [@B33]). This effect was equivalent to the effect of myelomonocytic cell depletion in LysM^iDTR^ mice in the same model of arterial hypertension that had been reported before (Wenzel et al., [@B66]), highlighting the importance of NK cell/monocyte interaction and mutual activation for induction and maintenance of vascular inflammation. More evidence supporting the importance of NK cells for vascular dysfunction was demonstrated by introducing the NK cell gene complex derived form C57BL/6 mice into the genome of BALB/C mice, rendering the strain as succeptible to vascular injury as the C57BL/6 mice in a hypertension model (Taherzadeh et al., [@B57]) and in a balloon injury model (De Vries et al., [@B11]). Neither it is known, how these findings extrapolate to the situation in acute MI, nor if NK cells preferentially cooperate with Ly6C^hi^ (proinflammatory) or Ly6C^lo^ monocytes in vascular inflammation. Most pathways triggering an activating pattern of communication between NK cells and monocytes are hallmarked by TLR-ligands and proinflammatory cytokines like TNF-α, IFN-γ, IL-12, and IL-18 (Kloss et al., [@B32]; Vivier et al., [@B61]; Michel et al., [@B39]). This suggests, that NK cells preferentially cooperate with proinflammatory myelomonocytes in vascular inflammation and MI, or at least skews these cells toward this phenotype. This hypothesis, however, remains to be tested. Interaction of NK cells and monocytes in coronary artery disease and myocardial infarction ========================================================================================== Little is known about the interaction of NK cells and monocytes/macrophages in the setting of atherosclerosis or acute MI. It was shown that NK cells are present in atherosclerotic plaques in humans as well as in mice (Whitman et al., [@B67]; Bobryshev and Lord, [@B7]), but the *in vivo* role of these immune cells in (coronary) atherosclerosis and MI is incompletely understood. It was shown *ex vivo* that NK cells produce more IFN-γ fostered by the CD48-2B4 axis (Figure [2](#F2){ref-type="fig"}) when co-cultured with monocyte derived dendritic cells, that had been exposed to oxidized LDL (Dong et al., [@B13]). This finding can at least in part explain the proatherogenic effects of NK cells. ![**Vascular inflammation, NK cell/monocyte crosstalk and angiotensin II in cardiac ischemic injury**. In cardiac ischemia reperfusion injury and remodeling after MI, vascular inflammation plays a central role. Monocytes transmigrate into the infarcted zone and transdifferentiate into macrophages, that can exert various functions in remodeling depending on the microenvironment provided by cytokines (e.g., IL-12, IL-18, IFN-γ), chemokines (e.g., RANTES/CCL5, MCP-1/CCR2), integrins (e.g., VCAM-1, ICAM-1) and soluble lipid factors like resolvins, lipoxins, and maresins. Signaling through angiotensin II and its downstream hormone aldosterone plays a central mechanistic role in this process. It leads to deployment of AT~1~R^+^ monocytes from the spleen to the infarcted area, increases (in red) or attenuates (in blue) chemokine and cytokine expression from the endothelium and from leukocytes, fosters monocyte transmigration and -differentiation and supports a reciprocal program of activation between NK cells and monocytes. While parts of this crosstalk have been partially investigated in this setting (e.g., the T-bet/IFN-γ/IL-12 pathway), most aspects of mutual NK cell/monocyte activation are not known in the setting of vascular inflammation and cardiac ischemic injury. This includes interaction between CD40 and CD154 (Bellora et al., [@B6]), CD48 and 2B4 (Nedvetzki et al., [@B43]) and NKG2D and retinoic acid inducible-1 (RAI-1) or members of the MHC class-I related chain (MIC) family of proteins, like MICA (Hamerman et al., [@B26]; Kloss et al., [@B32]). All of these are known to increase IFN-γ production by NK cells and CD69 expression on NK cells in infection or lipopoplysaccharide-driven models. Abbreviations: CD, cluster of differentiation; IL, interleukin; ATII, angiotensin II; AT1R, ATII receptor type 1; MR, mineralocorticoid receptor. Red letters: induced/activated by ATII/Aldosterone. Blue letters: attenuated by ATII/Aldosterone.](fphys-05-00295-g0002){#F2} Different research groups described a reduced NK cell activity in patients with CAD in comparison to healthy subjects (Jonasson et al., [@B30]; Hak et al., [@B25]). This effect was much more pronounced in patients in unstable conditions or acute coronary syndrome (Backteman et al., [@B4]; Hou et al., [@B28]). NK cells show a biological variation over time and in a recently published 12-month follow-up study of patients with a coronary event, no evidence was found that reduction of NK cells was associated with aberrations in NK cells cell phenotype at any clinical stage of the disease. However, failure to reconstitute NK cell levels was associated with a persistent low-grade inflammation, suggesting a protective role of NK cells in CAD (Backteman et al., [@B4]). The mechanism why there is a deficit of NK cells in CAD is not identified yet, but an increased apoptosis of these cells in CAD patients has been reported (Li et al., [@B35]). On the other side data from experimental studies demonstrated that NK cells infiltrate the vessel wall and promote atherosclerotic lesion development (Whitman et al., [@B67]). A recently published work of Selathurai and coworkers demonstrated evidence that NK cells are atherogenic and their production of perforin and granzyme B contributes to atherosclerosis and the expansion of necrotic cores (Selathurai et al., [@B48]) representing vulnerable lesions prone for atherothrombosis as in MI. RAAS signaling might play an important role in promoting mutual activation of NK cells and myelomonocytic cells, given its key role in hypertension, atherogenesis and MI (summarized in Figure [2](#F2){ref-type="fig"}). Expression and activity of many chemokines, integrins and cytokines and its receptors (e.g., MCP-1/CCR2, IFN-γ/IFN-γR, VCAM-1/VLA-4) as well as pro-atherogenic enzymes (e.g., NADPH oxidase) were shown to be driven by ATII/aldosterone on monocytes and the vessel, but not yet on NK cells, although their battery of signaling structures (e.g., CCR2, CCR5, LFA-1, VLA-4) share many similarities with monocytes (Figure [2](#F2){ref-type="fig"}). Conclusion {#s1} ========== Natural killer cells and monocytes/macrophages with their mutual programs of interactions are an important asset to the innate immune response. Its role in vascular inflammation, atherosclerosis, coronary artery disease and MI is now starting to catch interest from immunologist and vascular biologist (for scheme, see Figure [2](#F2){ref-type="fig"}). The molecular and physiologic interactions are nevertheless poorly explored and incompletely understood, although some potential targets have arisen from experimental research, that might be attractive for therapeutic targeting. Sources of funding ================== Maike Knorr and Philip Wenzel are supported by the German Federal Ministry of Education and Research (BMBF 01EO1003). Philip Wenzel is supported by grants from the German Research Foundation (DFG 4361/3-1 and 4361/4-1) related to this work. Conflict of interest statement ------------------------------ The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The expert graphical assistance of Margot Neuser is gratefully acknowledged. [^1]: Edited by: Christian Albert Gleissner, University of Heidelberg, Germany [^2]: Reviewed by: Dan Predescu, Rush University, USA; Giuseppe Pignataro, Federico II University Of Naples, Italy [^3]: This article was submitted to Vascular Physiology, a section of the journal Frontiers in Physiology.