Datasets:

chemNLP

/

ncbi-text

like 0

Follow

chemNLP 5

The basic building block of eukaryotic chromatin is the nucleosome, which consists of 147 bp of DNA wrapped around an octamer of the four core histone proteins H3, H4, H2A and H2B. The packaging of DNA into nucleosomes creates a restrictive environment that reduces the accessibility of DNA to factors that mediate chromatin-templated processes such as transcription. Dynamic structural rearrangements that render the chromatin permissive for transcription are therefore intimately associated with the regulation of gene expression. These structural rearrangements are believed to be facilitated by various posttranslational modifications of nucleosomal histones that affect chromatin structure either directly or by creating docking sites for the recruitment of effector proteins [for recent reviews see ()]. Among the most well-documented modifications associated with actively transcribed genes are the acetylation of various lysine residues in the N-terminal tails of H3 and H4 (H3Ac and H4Ac) (), the di- and tri-methylation of lysine 4 in the N-terminal tail of H3 (H3K4Me2 and H3K4Me3) () and the methylation of arginine 17 in the N-terminal tail of H3 (H3R17Me2) (). More recently, it has been shown that the methylation of lysine 9 in the N-terminal tail of H3 (H3K9Me3)—a modification that was until then thought to be a hallmark of silent heterochromatin ()—is actually also found within the 5′ ends of several actively transcribed mammalian genes (). It is now well established from studies in multiple model systems and species that there is a tight correlation between the presence of these histone modifications and active gene expression (). A widespread view is that histone-modifying activities are recruited by transcription factors and/or associated co-activators, and that they impact primarily on assembly of the general transcription machinery and transcription initiation (,). However, there is also growing evidence suggesting that factors associated with elongating RNA polymerase can play a key role in establishing histone modifications associated with actively expressed genes (,). A full understanding of the mechanisms that mediate the deposition and functional consequences of specific histone modification patterns requires the analysis and integration of at least three related parameters: where in the gene the various different histone modifications are made, in what temporal order they are introduced and what steps in the transcription process they are implicated in (). Yet in the majority of higher eukaryotic systems these three parameters have not been integrated into a single comprehensive and dynamic description of the spatial distribution and temporal order of chromatin modification events that occur during the activation of transcription. We have performed such an analysis in one of the most well-defined human model systems, the activation of major histocompatibility complex class II (MHC-II) genes (). Two modes of MHC-II expression, constitutive and inducible, are recognized (). Constitutive expression is restricted mainly to specialized cells of the immune system (thymic epithelial cells, dendritic cells, macrophages and B cells). Most other cell types do not express MHC-II genes unless they are exposed to interferon-γ (IFN-γ). The molecular machinery that regulates MHC-II expression has been exceptionally well defined thanks to the elucidation of the genetic defects that are responsible for the Bare Lymphocyte Syndrome, a rare hereditary immunodeficiency disease resulting from mutations in genes encoding transcription factors that are essential for MHC-II expression (,). One of these factors, the class II transactivator (CIITA) (), is a transcriptional co-activator that is exquisitely specific for the activation of MHC-II genes (A). CIITA serves as the master regulator of MHC-II genes and is expressed in a cell-type-specific and IFN-γ-inducible manner that dictates both the constitutive and inducible patterns of MHC-II expression (). It is recruited to the regulatory regions of MHC-II genes by protein–protein interactions with a multi-protein ‘enhanceosome’ complex that assembles on a characteristic enhancer known as the MHC-II S-Y module (A) (,,). A heterotrimeric DNA-binding factor called regulatory factor X (RFX) is a central component of the enhanceosome complex (A) (,). The activation of MHC-II genes by CIITA is associated with an increase in many of the major histone modifications known to correlate with active transcription, including H3Ac, H4Ac, H3K4Me2, H3K4Me3 and H3R17Me2 (). Most studies examining these modifications have restricted themselves to the promoters of MHC-II genes (,), despite the fact that several reports have shown that they may affect distal regions as well (,,). Moreover, only limited information is available on when and in what order these modifications are introduced during the activation process (,,). We have therefore performed a comprehensive analysis of the spatial distribution, order and timing of the histone modifications that are introduced during IFN-γ-induced activation of the prototypical MHC-II gene . We demonstrate that there are two spatially and temporally distinct chromatin-modification phases. The first phase precedes the initiation of transcription and is characterized by a global increase in H4 acetylation throughout a large upstream domain. All other modifications are restricted to short regions situated within or near the 5′ end of the gene and are introduced in a second phase that coincides with, and is a consequence of, ongoing transcription. With the exception of H4Ac, most modifications observed at the gene are thus established by transcription-coupled mechanisms. Me67.8 cells were grown in RPMI+Glutamax medium (Invitrogen) supplemented with 10% fetal calf serum and antibiotics, and were induced with 200 U/ml of human IFN-γ (Invitrogen) as described (). Cell surface HLA-DR expression was analyzed by FACS as described (). DRB (Sigma) was added to culture medium at a concentration of 100 μM. Chromatin preparation and ChIP experiments were performed as described using antibodies specific for RFX, CIITA and modified histones (,,). Results for modified histones were normalized using the TATA-binding protein (TBP) promoter as internal control and were compensated for fluctuations in nucleosome density by performing ChIP experiments in parallel with an antibody against unmodified H3. Histone antibodies were obtained from Abcam, Cambridge, UK (H3, Ab1791; H4K5Ac, Ab1758; H3K4Me3, Ab8580; H3K4Me2, Ab7766; H3K9Me3, Ab8898; H3K36me3, Ab9050) or Upstate, Lake Placid, NY (H4Ac, 06-866; H4K8Ac, 07-328; H4K12Ac, 07-595; H4K16Ac, 07-329; H3Ac, 06-599; H3K9Ac, 07-352; H3K14Ac, 07-353; H3K23Ac, 07-355; H3K27Ac, 07-360; H3R17Me3, 07-214). The RNA Pol II antibody was from Abcam (Ab817). Results were generated by real-time PCR using the primers listed in Supplementary Table 1. PCR was performed using the iCycler iQ Real-Time PCR Detection System (BioRad) and a SYBR-Green-based kit for quantitative PCR (iQ Supermix—BioRad). Amplification specificity was controlled by gel electrophoresis and dissociation curve analysis. Results were quantified using a standard curve generated with serial dilutions of input DNA. PCR amplifications were performed in triplicate. Experiments were repeated between two and five times with different inductions. CIITA and DRA mRNAs were quantified by real-time RT-PCR and normalized with respect to TBP mRNA as described (,). Chromatin-bound nascent and intergenic transcripts were isolated and quantified by real-time RT-PCR as described (). Primers used for nascent and intergenic transcripts are listed in Supplementary Table 1. PCR measurements were made in triplicate. Experiments were repeated between two and four times with different inductions. Cells were lysed for 1 h on ice in 50 mM Tris-HCl, pH 7.4, 150 mM NaCl, 10 mM EDTA, 1% NP40, 1 mM DTT containing 1× complete™ protease inhibitors (Roche) and a phosphatase inhibitor cocktail (Sigma). Lysates were clarified by centrifugation and protein concentrations were determined. Proteins were dissolved in 2× sample buffer and equal amounts were fractionated by SDS-PAGE. Proteins were transferred to Immobilon-P membranes (Millipore), which were incubated with rabbit polyclonal CARM1 antibodies (Upstate 07-080, used at 1/1000 dilution), rabbit polyclonal CBP antibodies (SantaCruz sc-369, used at 1/500 dilution), rabbit polyclonal pCAF antibodies (Abcam ab12188, gift from W. Herr, used at 1/600 dilution), rabbit polyclonal Set1c antibodies (gift from W. Herr, used at 1/500 dilution), rabbit polyclonal WDR5 antibodies (gift from W. Herr, used at 1/500 dilution) and mouse monoclonal S2-phosphorylated-Pol II antibodies (Covance, MMS-129R). Detection was performed using peroxidase conjugated anti-mouse or anti-rabbit IgG antibodies (Promega) and chemiluminescence visualization (ECL, Amersham) according to the manufacturer's instructions. Blots were quantified using Quantity One software. Histone modifications introduced during IFN-γ-induced activation of the gene were studied in the melanoma cell line Me67.8, which was chosen because it exhibits a robust and reproducible induction of CIITA and MHC-II expression in response to INF-γ (A) (). Me67.8 cells respond to IFN-γ in a synchronized manner, such that the entire population exhibits a homogeneous shift over time in the level of cell surface MHC-II expression (B). Time course experiments were performed to document the activation of CIITA and mRNA expression (C), as well as occupation of the S-Y modules by CIITA and RFX (D). The time courses obtained for these parameters are highly reproducible from one experiment to another (see error bars in ). CIITA mRNA accumulation starts 2 h after induction, whereas the increase in mRNA only becomes evident after 6 h (C). During this 4-h lag period, occupation of the S-Y modules by CIITA and RFX increases to reach a plateau by 6 h (D). These results define a precise 4-h time window during which key events implicated in CIITA-mediated transcriptional initiation of the gene are most likely to occur. We performed chromatin immunoprecipitation (ChIP) experiments to examine changes in the levels of H3Ac and H4Ac that are induced over time within a 7-kb domain that spans 5 kb of the upstream regulatory region and 2 kb of the 5′ end of the gene. Two spatially and temporally distinct histone acetylation phases were evident (). During the first phase, there was a strong increase in H4Ac over the entire region examined. This global increase in H4Ac was first evident after 3 h of induction and reached nearly maximal levels by 6 h. During the second phase, there was an increase in H3Ac that was restricted to a short 1-kb region situated within the 5′ end of the gene. This local increase in H3Ac occurred mainly between 6 and 12 h after the start of induction. To define the precise lysine (K) residues that are implicated in the two phases, we performed ChIP experiments with antibodies directed against specific acetylated K residues of histones H4 and H3. During the first phase, H4 was acetylated mainly at K5 and K8 (A). No obvious increases in the acetylation of H4K12 and H4K16 were evident (A). As observed for H4Ac, the increases in H4K5Ac and H4K8Ac were early events that were essentially complete by 6 h of induction (A) and concerned the entire 7-kb region (B). During the second phase, the strongest increase in acetylation was observed at K9 of H3 (A). There was a more modest increase for acetylation at H3K14, whereas no obvious changes were evident for acetylation at H3K23 and H3K27 (A). As for H3Ac, the increase in H3K9Ac was a late event occurring mainly after 6 h of induction (A) and was restricted to the 5′ end of the gene (B). The acetylation specificity that we observed is consistent with previous reports showing that H4K8 and H3K9 are among the major residues that are acetylated at the promoter in B cells and IFN-γ-induced cells (,). To situate the two phases of histone acetylation with respect to the initiation of transcription, we performed ChIP experiments to determine the timing of Pol II recruitment at the promoter and the appearance of nascent chromatin-bound transcripts (A). Pol II recruitment coincided with the increase in H4 acetylation and reached nearly maximal levels by 6 h. In contrast, nascent transcripts appeared mainly after 6 h, in parallel with the increase in H3 acetylation. The first phase of H4 acetylation is thus associated with Pol II recruitment but precedes the initiation of transcription. In contrast, the subsequent H3 acetylation phase appeared to coincide with active transcription. To confirm the latter, we performed a finer time course between 6 and 12 h (B). The results indicate that the increase in H3 acetylation and the appearance of nascent transcripts are simultaneous events. Constitutive expression of the gene in B cells is associated with intergenic transcription of its upstream regulatory region (). We therefore determined whether these intergenic transcripts are also observed in IFN-γ-induced cells (). Intergenic transcripts were induced during the early phase of H4 acetylation, prior to transcription of the gene itself. The abundance of these intergenic transcripts is highest just upstream of the promoter and in regions that flank the distal S′-Y′ module (). This IFN-γ-induced pattern of intergenic transcription is similar to the one observed previously in B cells (). To examine where and when changes in H3 methylation occur during IFN-γ-induced expression of the gene, we performed ChIP experiments with antibodies directed against H3K4Me3, H3K4Me2, H3R17Me2 and H3K9Me3 (). Increases restricted to short regions situated near the 5′ end of the gene were observed for all four modifications. The increases in H3K4Me3, H3K4Me2 and H3K9Me3 peaked at a position situated after the transcription initiation site. The strongest increase in H3R17Me2 was situated at the transcription initiation site. These methylation events occurred mainly after 6 h, although the increase in H3K4Me2 appeared to precede the others somewhat. Taken together, these results show that introduction of the H3 methylation marks overlaps temporally and spatially with increased H3 acetylation. With respect to both their timing and position, the increases in H3 acetylation and methylation coincide with active transcription of the gene, raising the possibility that they might actually be a consequence of transcription. To address this possibility, we induced cells with IFN-γ for 6 h to permit the induction of CIITA expression and completion of the early H4 acetylation phase, and then continued the induction in the presence of the drug 5,6-dichloro-1-β--ribofuranosylbenzimidazole (DRB) (A), which blocks transcription elongation by inhibiting phosphorylation of the C-terminal domain (CTD) of Pol II (B) (,). This protocol completely blocked the induction of mRNA accumulation (C) and reduced Pol II density within the body of the gene (Supplementary Figure 1). As expected, DRB did not significantly affect HLA-DRA promoter occupation by RFX and CIITA, the acetylation of H4 or recruitment of Pol II to the promoter (B and C). Events that precede the initiation of transcription had thus been completed and were not reversed by the addition of DRB. In contrast, the increases in H3Ac, H3K4Me3, H3K4Me2, H3R17Me2 and H3K9Me3, which occur mainly after 6 h, were all completely eliminated by DRB. The introduction of these five modifications thus behaves in a manner similar to that of the H3K36Me3 modification (C), which is known to be strictly dependent on active Pol II elongation (). To ascertain that the DRB sensitivity of the H3Ac, H3K4Me3, H3K4Me2, H3R17Me2 and H3K9Me3 modifications is not a non-specific consequence of a general block in the transcription of genes encoding histone-modifying factors, we examined the levels of several key proteins by western blotting experiments. These included the histone acetyltransferase (HAT) factors [cyclic AMP responsive-element binding-protein (CBP) and p300/CBP-associated factor (pCAF)], the histone methyltransferase (HMT) factors (Set1 and WDR5), and the H3R17-specific HMT CARM1 (D). We also examined the HAT GCN5 (data not shown). No drop in the abundance of any of these proteins was induced by the 6-h DRB treatment. Our results demonstrate that histone modifications associated with IFN-γ-induced gene activation are introduced during two sequential phases that differ with respect to their timing, the regions of the gene that are concerned, the precise modifications that are made and their dependence on ongoing transcription. The first phase precedes the initiation of transcription and is characterized by a rapid increase in H4K5 and H4K8 acetylation over a large upstream domain. The second phase is concomitant with active transcription and is characterized by increases in H3K9Ac, H3K4Me2, H3K4Me3, H3K9Me3 and H3R17Me2 in short regions situated at or within the 5′ end of the gene. This second phase is completely blocked by the transcription elongation inhibitor DRB, indicating that it is a consequence of active elongation by Pol II. The timing, localization and transcription-dependence of the increases in H3Ac, H3K4Me2, H3K4Me3, H3K9Me3 and H3R17Me2 suggest that these marks are introduced by HATs and HMTs that are recruited by actively transcribing Pol II, associated elongation factors and/or prior histone modifications that were introduced in a transcription-dependent manner. Moreover, they imply that most histone modifications introduced during gene activation do not play a role in facilitating assembly of the general transcription machinery and transcription initiation, but are instead likely to be implicated in subsequent processes such as promoter clearance, transcription elongation and/or the establishment of transcriptional memory. These conclusions are consistent with previous reports suggesting that phosphorylation of the CTD of Pol II by the CDK7 subunit of the general transcription factor TFIIH and the CDK9 subunit of the transcription elongation factor pTEFb are key events in CIITA-induced MHC-II gene expression (,). In contrast, our results challenge current models proposing that CIITA serves as a scaffolding protein that recruits and coordinates all key chromatin-modifying activities at MHC-II promoters, and that the modifications introduced by these activities are primarily required for transcription initiation at MHC-II genes (). Among the histone modifications we have examined, only the rapid and long-range increase in H4 acetylation coincides temporally with the recruitment of CIITA and occurs prior to and independently of active transcription of the gene. This suggests that only the early H4 acetylation phase is likely to be mediated by HATs that are recruited directly by CIITA. HATs that are believed to be able to cooperate with CIITA include CBP, pCAF, GCN5 and steroid receptor co-activator (SRC)-1. These HATs have been reported to be recruited to MHC-II promoters in B cells and IFN-γ-induced cells, can interact with CIITA and upon over expression in transfected cells, and can activate MHC-II reporter genes in synergy with CIITA in transient transfection experiments (). However, we have so far been unable to document a robust and reproducible recruitment of any of these HATs in a pattern that coincides temporally and spatially with the binding of CIITA and the early increase in H4 acetylation. One explanation that could account for this discrepancy is that these HATs associate only transiently with the HLA-DRA upstream region. Alternatively, other HATs could be implicated. Since CIITA has been reported to have intrinsic HAT activity (), one possibility is that CIITA itself might be responsible for the increase in H4 acetylation. A second intriguing possibility is suggested by the finding that intergenic transcription is induced in the upstream region according to a pattern that is similar to that observed for H4 acetylation with respect to both timing and spatial distribution. Intergenic transcription has been attributed a regulatory function in the establishment of open chromatin domains in several systems (). It has notably been suggested to contribute to the function of locus control regions (LCRs) (,,). Among other models, it has been postulated that the regulatory role of intergenic transcription could be due to chromatin remodeling activities—such as HATs—that track along the chromatin with Pol II (). Since the HLA-DRA upstream regions exhibit properties typical of LCRs (,), it is tempting to speculate that such a tracking mechanism is operating during the early phase of H4 acetylation observed during gene activation. We can however not exclude the possibility that the intergenic transcription is a consequence rather than a cause of increased H4 acetylation. Our results highlight a dominant role of transcription elongation in the recruitment of histone-modifying activities during the induction of MHC-II gene expression. This is at odds with widespread models of gene regulation postulating that HATs and HMTs are primarily recruited by DNA-bound transcription factors and their associated co-activators (,). However, there is growing evidence that a decisive role of transcription elongation in the establishment of histone modifications is not just a peculiarity of the MHC-II system. There is strong evidence for a link between deposition of the H3K4Me3 mark and transcription elongation in . Set1—the HMT responsible for methylating H3K4 in yeast—is recruited to the elongating CTD-phosphorylated form of Pol II by its interaction with the elongation factor Paf1 (,). This mode of recruitment is consistent with the results of genome-wide mapping studies, which have shown that the density of H3K4 methylation generally peaks within the 5′ transcribed portion of active yeast genes (,). The link between H3K4 methylation and transcription elongation is less well established in higher eukaryotes. Set1 homolog in higher eukaryotes—of which there are several—have in fact been reported to be recruited by interactions with specific transcription factors (). However, large-scale mapping studies in and humans, as well as a more limited study in the chicken, have demonstrated that H3K4 methylation is, like in yeast, frequently concentrated within the 5′ transcribed regions of active genes (). Moreover, a protein fraction enriched in H3K4-specific HMT activity was recently reported to introduce the H3K4Me3 modification in a transcription-coupled manner in a reconstituted human transcription system (). These findings suggest that a transcription-dependent mode of recruitment of H3K4-specific HMTs is conserved in higher eukaryotes. There are also indications that H3 acetylation may be coupled to transcription elongation at numerous genes in diverse species. Large-scale mapping studies performed in , , and humans have established that the density of H3 acetylation tends to peak, as observed here at the HLA-DRA gene, within the 5′ transcribed region of expressed genes (,,,,,,). This intragenic localization would be consistent with a widespread role of active transcription in increasing H3 acetylation. Transcription-dependent H3 acetylation could be mediated by HATs, such as the elongator complex, that travel along chromatin in association with elongating Pol II (,). Alternatively, several findings suggest that H3 acetylation could be coupled to H3K4 methylation. Large-scale mapping studies have shown that the patterns of H3Ac and H3K4Me3 often coincide (,,,,,,). Moreover, methylation at K4 makes H3 a preferential target for dynamic changes in acetylation (). Finally, histone-modifying complexes containing both HAT and H3K4-specific HMT subunits have been identified in yeast, and humans (,,). The recruitment of CARM1, the HMT responsible for making the H3R17Me2 modification, is believed to be mediated by interactions with specific transcription factors or co-activators (,,,,). However, in two systems CARM1 recruitment has been shown to be dependent on the prior acetylation of H3 (,). Tethering of CARM1 to chromatin modifications established during active transcription, such as H3 acetylation, might thus constitute an alternative pathway for CARM1 recruitment. Finally, our results extend the recent discovery that the H3K9Me3 mark is introduced into the 5′ transcribed region of mouse and human genes by a transcription-elongation-dependent process (). As observed here at the gene, this previous report also demonstrated co-localization between the H3K9Me3 and H3K4Me3 marks, suggesting that their introduction might be coupled (). Taken together, the findings outlined above suggest that transcription elongation may play a critical role in establishing histone-modification patterns associated with many actively expressed genes in species ranging from yeast to humans. The results reported here for activation of the gene are thus likely to be of widespread relevance to numerous gene regulatory systems in diverse species. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Expression arrays have progressed to a point where low technical variance, low background noise and a high degree of accuracy have encouraged the development of array-based medical devices that predict drug response, relapse potential or general prognosis (). Normalization is a critical pre-processing step for most array technologies, due to the known biases. As normalization methods get more sophisticated and perhaps more specialized, the list of pros and cons for each grows. The array user should be aware of the bottom-line consequences of the normalization methods available today. Affymetrix (Affymetrix Inc, Santa Clara, CA, USA) and Agilent (Agilent Technologies, Santa Clara, CA, USA) are leaders in expression array manufacturing. They use quite different approaches to the construction, layout, optimization, hybridization, image acquisition and data extraction methods. Much of the difference that we see is attributable to the difference between probe synthesis—photolithography (light-directed) versus liquid-based (ink-jet) oligonucleotide synthesis. Reports have found both poor () and good (,) cross-platform correlation, but the MAQC consortium have generally found that proper sample preparation is sufficient to dramatically enhance multi-lab and multi-platform correlations (,,). Quality control rules () tell us that one could fix a high-quality RNA source and identify all other variables that could cause discordant data. With that logic, we propose a system that fixes the RNA source and changes data normalization methods in order to estimate their effect on data precision, classifier error and biological interpretation. The system we developed is a simple analysis that both graphically and quantitatively shows how adjustable parameters (in this case normalization) affect discordance. Although many publications have proposed somewhat esoteric methods for measuring cross-platform reproducibility, we believe that a simple, easy-to-understand analysis will not only highlight most sources of variance, but will also enable the user to visualize how process-control techniques improve reproducibility. Data was structured as follows: data sets were log (intensity) and log (ratio) transformed as needed. summarizes the reproducibility and dispersion for each platform and tissue combination across most of the twenty-four conditions. Agilent CY3 was left out for brevity, but plots were very similar to the CY5 data. The first three columns are the intensity replicates (e.g. liver sample 1 versus liver sample 2) and graphically illustrate technical variability as a function of fluorescence intensity. Background-subtracted methods in general tended to show the highest apparent dispersion (MAS5 and dChip PM–MM) while GC-RMA, dChip PM and most of the Agilent data showed much less scatter. The third, fourth and fifth columns show the MvA (Bland–Altman) plots, indicating the of correlation between variance and intensity. Only the Affymetrix MAS5 and GC-RMA data have substantial scatter, indicating a disjunction between intensity and variance. The ratio replicate plots in columns seven, eight and nine indicate how precisely each pair of tissue samples can be used in ratios for each of the three pairwise cases. MAS5 and dChip PM–MM show comparatively high scatter, indicating higher variability across replicate ratio calculations, especially at ratios near one. The dChip PM and RAW plots, and to a lesser extent MAS5, highlight the problem of using either under-normalized or imperfectly estimated mismatch data as a reliable estimate of background. The Agilent data shows a slight trend to higher dispersion with the BSUB and PROCESSED signals showing the impact of subtracting background. The boxplots shown in (top) indicate the relative data spread, another graphical estimate of precision. Agilent MEAN and Affymetrix GC-RMA and RAW show the lowest quartile ranges, suggesting high precision. The bottom plots show the relative compression of un-normalized signals, explaining the illusion of precision due to the low dynamic range of near-RAW data. shows the effect of normalization on hierarchical clustering (Euclidean distance, average linkage, 1000 ANOVA-selected genes, GeneSpring 7.2, Agilent Technologies, Palo Alto, CA). Affymetrix data tends to form clusters based on the (relatively greater) effect of normalization while Agilent data tends to cluster by tissue regardless of the channel or normalization. The Venn diagram shows the overlap of genes for each cluster experiment; there were 699 common genes out of 1000 based on RefSeq. Precision estimates such as these are always imperfect in some way, but when taken together they provide a good estimate of relative precision. Sensitivity was calculated in several ways. We first estimated the power using normal.sample.size() in S+ or power.t.test() in R. We computed Δ (the minimum detectable fold change) at an arbitrary threshold of one potential false positive per array, or = 1/. The -value threshold used throughout this article often use 1/, or = 5.3 × 10 for Agilent and = 4.5 × 10 for Affymetrix. Calculations of delta used = 3 replicates, = 0.80 for every pairwise gene expression value across each unique tissue case, per platform and per normalization. shows the sorted Δ (black curve) calculated for each probe case with the actual ratios between the two tissues plotted as blue bars. If abs(log gene/gene) > Δ, then gene is significant by definition, as indicated by the red circles. Some circles lie below the curve Δ because the significance was calculated by a -test using log intensities rather than the log ratios in the power calculation. This is formalized below in Equation (): shows the results from three methods for calculating sensitivity. Column 1 shows the mean delta +/− the standard deviation computed by calculating power from every possible pairwise case, column 2 shows the average minimum-detectable fold-change (MDFC) across replicate measures at the 95th percentile. Equation () is the method for averaging delta for each case. Column 3 shows the median MDFC across replicate measures at the 95th percentile. Equation (1.3) clarifies the calculation for delta across the th gene and the th sample where = 22 215 for Affymetrix and 18703 for Agilent. Mean and median fold-change values across ratio replicates were averaged across all case for all ratio calculations used in sensitivity calculations. Sensitivity estimates correlate well with the replicate scatterplots in . Agilent methods BSUB and PROCESSED have the highest sensitivity followed by Agilent MEAN, Affymetrix GC-RMA and dChip PM, with the worst precision and sensitivity seen with MAS5 and dChip PM–MM normalizations. The fact that dChip PM produced better sensitivity results than dChip PM–MM is likely due to the scatter that the mismatch subtraction causes, similar to the problem that MAS5 has. Algorithms that use background subtraction methods cause low-intensity imprecision when MM > PM. This effect is manifested in MAS5 and dChip PM–MM data by a minimum detectable fold change near 2-fold, while GC-RMA and Agilent data show 1.3-fold or less MDFC. We tested Gene Ontology functions by computing lists of genes differentially expressed across each pair of tissues (). Each gene list was tested for unusual abundance using GO categories, as calculated in GeneSpring 7.2 with corroborative results obtained from OntoExpress (). Nearly identical results were obtained across the Agilent normalizations (columns 3, 5 and 7), less so among the Affymetrix normalizations, with dChip PM identifying functions that are quite unique. MAS5 and GC-RMA showed the greatest similarity to the Agilent results, suggesting that differentially expressed genes identified using GC-RMA and the Agilent samples led to a common biological interpretation. Subsequently, we wanted to see the extent of overlap given a common set of genes across the two platforms. We converted probe name to RefSeq, then to Hugo Gene Symbol, then to HUGO gene name and selected the intersection between the two platforms. We also used GeneSpring's Translate Genome function, and obtained a similar overlap. Using this common genome of probes, we selected the 1000 most significant genes from a Model I ANOVA (). The highest overlap across the two platforms exists between Affymetrix dChip PM–MM and Agilent PROCESSED (243 genes out of 1000, G) which, given the precision results, was a little surprising. Overall the overlap among MAS5, PM–MM and RAW (127 genes, K) is higher than across dChip PM and GC-RMA (39 genes, I). The Agilent normalizations were very similar to each other, with MEAN having the highest unique set of genes (288, B) among the three normalizations. An interesting finding is the relatively high overlap between the Affymetrix background subtraction methods (dChip PM–MM and MAS5) versus the Agilent data (C). In contrast, the more precise measures of dChip PM and GC-RMA versus the Agilent data (I) showed very little overlap, again suggesting that the most aggressive and platform-specific normalizations improved precision at the cost of accuracy. The highest overlap between GO functions was found between MAS5 or dChip PM–MM and Agilent PROCESSED, again suggesting that high Type I error may not affect a GO analysis as dramatically as Type II errors. Using more detailed GO nodes did not clarify the differences between our normalizations, nor did it change the rank of best–to–worst. We feel this functional analysis is suitable as a 10 000 foot view of biological consistency. However, we wished to examine another biological analysis, and GenMapp, Biocarta, Kegg and Cytoscape all yield sufficient discrimination to quantify biological differences based on gene lists. We performed pathway analysis of 100 significant genes from each list () using . Interestingly, once again we see that MAS5 and to a lesser extent dChip PM–MM match the Agilent data well, with Affymetrix RAW consistently identifying pathways outside consensus. By comparing the pathways from , we find that the pathways tend to validate the GO analysis from a different biological and mathematical perspective. We demonstrate how feature selection and classification can be compromised by comparing classifier error rates across platforms and normalizations (). We used a two-feature sequential forward floating search (,) with bolstering error estimation to score the feature sets, and linear discriminant analysis (LDA) as the classification rule (). Overall error was estimated using cross validation with 500 replicates to reduce internal variability. Initially, we applied the selection routine to whole data sets containing the full complement of genes, obtaining in all cases zero misclassification error. In order to introduce some variability, we iteratively removed 500 of the most significant (by -test) probes until less than 500 probes remained for both platforms; removal was done within the cross-validation step to reduce error. In , we show the error rates per normalization and per case for lung:spleen, liver:spleen and liver:lung, and in we compute the area under each curve as a relative rank of error. The Y-axis is the classifier error; the X-axis is the percentage of probes removed per iteration. In all cases the trends are generally consistent; Agilent data (dashed lines) are generally below the dChip PM–MM and RAW Affymetrix normalization methods, and are similar to GC-RMA. It is likely that a rapid increase in error indicates that the best predictive genes were removed fairly quickly, implying that good predictive features are not necessarily those with high statistical significance. Another characteristic of this group is the instability in error after ∼40% of the most significant probes were removed. The error rate for MAS5 shows a linear increase in error suggesting that this gene list contains features that contribute evenly to classification, whereas other groups rise and fall quite suddenly. This variability in error is likely not due to cross-validation since we performed 500 replicates, sufficient to converge to a stable error estimate. This instability likely results from the disconnect between a classifier error and the distributional tests we used in the removal step. A random removal method with more replication might have yielded a better estimate of error, but the computation time would be excessive. The areas under the curve (, columns 2, 4, 6) show Agilent MEAN data to be marginally better than PROCESSED and BSUB, but the confidence intervals overlap indicating that these three normalizations are equivalent. MAS5 and RAW tended to show the highest Affymetrix error while GC-RMA showed the lowest, again reflecting improvements caused by technical precision, but also on bias, since the RAW data was much more precise than the MAS5 data. The percent of total genes that are significant at < 5.3 × 10 for Agilent and < 4.5 × 10 for Affymetrix reflects the pool of genes tested in the classifier. The Affymetrix RAW data which is known to be biased also contains many significant genes, showing that our classifier is not compromised by inaccurate and biased signals. The RAW classification resulted in high error, seen in . GC-RMA had lower misclassification than any group or platform, but we were less convinced that this was the best normalization scheme for these tissues since the GO and pathway GC-RMA results differed from consensus. We wanted to determine the probe position for the best and worst correlated probes for the best normalizations for classifier error: GC-RMA and PROCESSED (,). We sorted the probes for the best and worst correlation across Agilent's CY5 PROCESSED data and Affymetrix's GC-RMA data for liver and spleen. We determined the probe location by identifying the probe sequence (or exemplar) on Human Build 36 using BLAT. In nearly all of the best and worst correlated cases, discrepancy occurred when the probes were physically separated (), but the degree to which this was the case varied. Within an Affymetrix probeset, physical distance often resulted in poor intra-probe correlation as well. xref sup #text Intra-lab and intra-platform correlation and calibration can optimize data quality and reduce lab- and platform-dependent biases. In industrial Six Sigma Quality Control, the most influential parameters affecting process quality are identified to reduce faults in order of importance. In the case of expression data, poorly correlated data is often caused by RNA quality. This is prevalent even given the differences in probe location () and platform idiosyncrasies. Array users may be unable to obtain the advertised performance figures for a commercial microarray due to difficult-to-extract tissues, such as plant cells. We propose that precision, power and pathway analysis can pinpoint samples that lie outside a consensus, especially in large experiments or with public data. Clustering has seen a backlash against graphical interpretation of data, but taken in context and with an understanding of the limitations, it presents array data in a richly informative way. Degraded RNA causes signal compression and high background which show clearly in clustering analysis. Power and sample size calculations also pinpoint degraded RNA or poor labeling by showing greatly reduced sensitivity and delta values. Classification has become a much-used method in disease prognosis and diagnosis (); it is therefore important to understand the causes of misclassification. Microarray normalization methods, especially loess () and model-based (), often cause large non-linear changes that attempt to improve the reliability of measuring relative differences across samples (). High precision methods like GC-RMA can affect the classifier, resulting in very low error, but classifiers are less affected by highly biased data than significance tests. As seen in , highly aggressive normalizations combined with very differential tissues, can cause mis-clustering. However, genes identified as either up or down between tissues across normalization methods can be quite comparable if one quantizes to the level of ‘up’, ‘down’ and ‘unchanged’ by using the appropriate confidence interval. Agilent data is almost unaffected by channel and normalization effects, but the normalizations were much more subtle than Affymetrix methods. Normalized expression data often exaggerates the magnitude of ratios and inflates false positives over comparable qRT-PCR data (,). That effect alone will change the rank of genes, and will change the biological pathways identified (). It is increasingly difficult to identify biomarkers that work independently of the platform (,,,), but appropriate normalization choice may ameliorate this effect somewhat. Affymetrix MAS5 and Agilent MEAN share 256 genes, MAS5 and BSUB share 261 genes and PM–MM and PROCESSED share 243 genes, the highest overlap between platforms. These low-precision but high-accuracy methods, while often underpowered, can also provide genes that are more platform-neutral. Although the background subtraction methods generally provide the highest false positives, their conservative nature tends to avoid strong and potentially inaccurate biases ( and ). Based on these outcomes, we recommend MAS5 or dChip PM–MM and Agilent PROCESSED normalizations for feature selection and classification, and for biological pathway analysis, especially when identifying platform-neutral biosignatures. If comparisons across laboratories or expression platforms will be done, the most conservative estimate of Affymetrix data is best. We caution the user that the power of detection drops considerably with MAS5 and dChip PM–MM, and more technical replicates should be used to obtain the same detection limit as GC-RMA or dChip. Most public expression databases provide the MAS5-normalized data (e.g. the SOFT file format from GEO), but increasingly the .CEL files are being made available. We recommend GC-RMA normalization when large data sets are used, high sensitivity is needed, and samples are not terribly different from one another. GC-RMA provides a good signal that has been shown to have good sensitivity and accuracy in the context of distinguishing disease subtypes or other subtle phenotypes. When a moderate-to-small number of samples are used, dChip PM is an excellent choice since it strikes the best compromise between variance reduction methods and background subtraction methods. If single-color analysis is needed, extracting one of the two Agilent channels works well, but Agilent recognized the need for a single-color product and now offers one-channel protocols. In , we show the relationship between probe distance and the correlation between liver:spleen ratios between Agilent CY5 (PROCESSED) and Affymetrix (GC-RMA). In general, the best correlation occurred when the probes were relatively close to one another, the worst correlations occurred when the probes were distant, an effect previously reported (,,). This effect actually occurs within a probeset on the Affymetrix platform, but the effect is not as pronounced. This effect is easy to measure since the probe sequences for these arrays are available from the manufacturer. When contrasting qRT-PCR and array data, one should carefully design RT primers that are uniformly spaced across the gene, rather than a single probe in the same location as the microarray. This principle reveals array limitations, but also gives the best RT results. In summary, we provide three simple, qualitative methods of analysis to identify discrepancy in expression data sets. Precision and sensitivity measurements are useful in finding the minimal detectable fold-change and raw performance values for an array platform (or qRT-PCR). Biological comparisons such as the Gene Ontology and pathway analyses are a valuable way of examining and comparing the actual biological interpretation. Differences in pathways indicate consistency problems. This inconsistency can be quantified by counting the differentially expressed genes between platforms that move in different directions. Finally, classifier error provides a way of identifying misleading transcriptional signals. When sufficiently large numbers of informative genes exist, one can identify a platform-neutral set of genes that provide both low error across multiple platforms and low classifier error by utilizing the selection criteria mentioned above. Taken together, precision, biological interpretation and multiple platform data sets will allow better selection of genes that yield clinically useful biosignatures

With completion of the human and mouse genome projects, a large volume of data for genes with unknown function will be available. In the present and forthcoming era of functional genomics, sequenced bacterial artificial chromosomes (BACs) (), P1 vectors () and P1 artificial chromosomes (PACs) () will be very useful in the analysis of gene function, since they can be used for construction of complex vectors and subsequent creation of transgenic, knockout and knock-in mice. The difficult manipulation of such large episomes with conventional cloning techniques has until now limited their complete usefulness. However, the recent development of techniques based on homologous recombination in () has improved, simplified and facilitated the manipulation of both large and small vectors, thus becoming a very important tool in the design of targeting vectors and the subsequent functional analysis of newly discovered genes. Manipulation of plasmid DNA by homologous recombination was first reported in yeast () and since then techniques have been developed to include phage-based vectors for convenient and efficient use in . The use of these plasmids to carry out genetic engineering has been called or (). Recombineering exploits the phage derived protein pairs, either RecE/RecT from the phage or Redα/Redβ from the λ phage, to assist in the cloning or subcloning of fragments of DNA into vectors without the need of restriction enzyme sites or ligases (,). A limitation of the original homologous recombination technique was due to the fact that bacterial RecBCD nuclease degrades linear DNA and initially the event had to be studied in RecBCD-deficient strains (). This was overcome by the discovery that Redα and Redβ were assisted by Redγ, which inhibits RecBCD nuclease activity making it possible to use the technique in and other commonly used bacterial strains (). In addition, the recombination efficiency was increased 10–100 times (). The combination of these three enzymes (α, β and γ, or E, T and γ) in one vector was named Red/ET recombination and the basic principles of the method are that it requires two homology regions of >42 bp in a linear fragment, double strand breaks (DSBs) in both ends, and another linear or circular plasmid in order for recombination to take place. DSBs are essential so that RecE or Redα can bind and degrade one chain of the DNA (5′ to 3′) and at the same time load RecT or Redβ to the single strand chain that is exposed (). The single DNA strand loaded with the RecT or Recβ recombinase finds a perfect match sequence and joins the two sequences by either chain invasion or annealing. However, this system requires insertion of homology arms (HAs), which are included in the oligonucleotides that are used for amplification of the PCR products used as linear substrates for the recombination event. The limitations of the PCR reaction with long primers make it difficult to generate large quantities of fragments that are longer than ∼4–5 kb, and even in the event of being able to PCR longer templates the rate and risk of mutations/mismatches increases with the size of the template. If longer fragments of DNA are needed for the cloning procedures then the HAs may be inserted with conventional restriction/ligation techniques into primary constructs that are excised and used for recombination. Moreover, the insertion of DNA by Red/ET cloning is highly efficient only if a selection marker gene is included, which may be undesirable in the final construct. This has been addressed by the use of elegant selection/counter-selection systems (,) but the limitations of the PCR-size of the insert has not yet been solved. Insertion of large fragments with recombineering may thus be a time consuming multi-step process. BACs and large plasmids (e.g. subcloned genomic fragments from BACs into minimal vectors) are important tools for the creation of transgenic organisms since they contain most of the regulatory elements needed for optimal expression. There are many large mammalian genes which are of the same order of magnitude or larger than the average insert of the BAC libraries and it is often difficult to find a single BAC spanning the entire gene including its controlling elements. It may thus be desirable to fuse or exchange fragments from different BACs in order to obtain larger functional regions of DNA or to create mouse genes, where the gene of interest in the mouse BAC is exchanged for the human genomic counterpart. Fusion of the regulatory elements of one BAC with another BAC that contains the coding regions of a gene of interest thus becomes a powerful strategy to study tissue or cell specific gene expression. This can be achieved by using one end of a BAC that overlaps an area of another as homology regions or by adding the HAs into the insert-containing vector by either recombination or conventional cloning (). However, finding overlapping BACs limits the uses of this strategy. Moreover, the method cannot introduce other non-overlapping fragments such as the same gene from different species or a different gene and it seems to be efficient only if a selection marker (Sm) is present in the inserting fragment. We report here a method, Assisted Large Fragment Insertion by Red/ET-recombination (ALFIRE), where we have overcome both the size limitations and the restrictions of the cloning/subcloning procedures used in the Red/ET recombineering system. We describe the transfer of a very large DNA fragment (∼55 kb) from one BAC into another unrelated BAC by the means of a modified -counter-selection (CS) system () combined with or restriction enzyme digestion (,). ALFIRE thus represents an improvement of tools that can be used in order to facilitate construction of large and complex vectors utilized in molecular biology and functional genomics. The Red/ET recombination system (pSC101BADγβαA[tet]), and the -neo counter-selection cassette (CS) were purchased from Genebridges, Dresden, Germany. BACs containing the entire mouse or human luteinizing hormone/chorionic gonadotropin receptor genes () were purchased from Research Genetics (clones RP23-18D7 and RP11-73L19, respectively). Individual BAC clones were propagated in in Luria–Bertani (LB) media containing chloramphenicol (Cm, 20 μg/ml)/streptomycin (Stp, 75 μg/ml) and purified using the Large construct kit (Qiagen, Valencia, CA, USA). PCR reactions using oligonucleotides carrying HAs were performed using a proofreading polymerase and buffers included in the kit Triple Master Mix (Eppendorf, Hamburg, Germany). The standard reaction conditions were 96°C for 3 min, followed by 30 cycles with 96°C, 45 s, 57°C 45 s, 72°C, 1–3 min, depending on the size of the product, and a final elongation of 5 min at 72°C. An aliquot of each PCR product was analysed by gel electrophoresis. The rest of the PCR product was ethanol-precipitated, resuspended in Tris–HCl (10 mM pH 8.5) and used for transformation into electrocompetent bacteria. PCR was also used to screen for positive recombinants and to verify correct integration. The short primers (20–30 nt) were purchased from TAG, Copenhagen, Denmark, whereas the longer primers (60–140 nt) were purchased from Thermo Electron, Hamburg, Germany. Oligonucleotide sequences used for design of the I-SceI-CS--BAC: (CCAGCATACTGGCCTAGCCACCGGAGCTCACACTCAGGCTGGCGGGCCATGAA GCAGCGGTTCTCGGCGCTGCAGCTGCTGAAGCTGCTGCTGCTGCT agggataacagggtaatGGCCTGGTGATGATGGCGGGATCG); (GTGGACTTTTTTGGGGGGAACATATTTAGATACAATTCAGTAATGCAGTTAACA CTCTGTGTAGCGAGTCTTGTCTAGGAGAGCTGTACCTTGACAGT attaccctgttatccctaTCAGAAGAACTCGTCAAGAAGGCG). Recombination events were performed in electrocompetent DH10B or HS996 cells harbouring the pSC101BADγβαA[hygro] or the pSC101BADγβαA-I-SceI[amp] plasmids and using standard Red/ET protocols (). Electrocompetent cells were prepared as described in the general manuals from Genebridges. Electroporation was performed at 1.35 kV, 25 μF, 200 Ω using an Eppendorf electroporator. Bacteria were incubated in 1 ml LB media at 37°C, 1100 rpm for 1–2 h before plating on agar plates conditioned with the appropriate antibiotic(s). The pSC101BADγβαA[hygro] plasmid was based on the pSC101BADγβαA[tet] vector (GeneBridges) and was constructed by standard Red/ET recombination procedures () using the primers (5′ AATGCGGTAGTTTATCACAGTTAAATTGCTAACGCAGTCAGGCACCGTGTATGGATA GATCCGGAAAGCCTGAA 3′) and (5′ TCCAATTCTTGGAGTGGTGAATCCGTTAGCGAGGTGCCGCCGGCTTCCAT CTATTCCTTTGCCCTCGGACGAGT 3′) in order to amplify the from the pIRESHYG3 plasmid (BD Biosciences, Palo Alto, CA, USA). Positive recombinants were selected on LB-agar plates supplemented with Hyg 30 μg/ml. The pSC101BADγβαA-I-SceI[amp] vector was assembled by PCR amplification of the I-SceI gene together with the Tet-Repressor, Tet-Promoter and gene sequences from the PsiI-digested pST98AS plasmid (AF170483) () [oligo sequences (AATGCGGTAGTTTATCACAGTTAAATTGCTAACGCAGTCAGGCACCGTGTGACC AATTCGGGTCGACTTAT)/ (TCCAATTCTTGGAGTGGTGAATCCGTTAGCGAGGTGCCGCCGGCTTCCATT GGTCATGAGATTATCAAAAAGGA)]. The PCR product which contained HAs for pSC101BADγβαA[hygro] was co-transformed with the pSC101BADγβαA[hygro] vector into HS996 electrocompetent cells harbouring the pSC101BADγβαA[tet] vector and recombination was achieved using standard Red/ET protocols () and selection with Ap 50 μg/ml. Single BAC clones were grown in LB medium conditioned with the Cm/Stp and screened by PCR over the inserted areas (the promoter to the intron1 area; and the exon 11 to the 3′ area) using primers (CCAGCATACTGGCCTAGCCAC)/ (AGTACACAGCGCTCCCGTC) and (CGAAACCCAGAATTAATGGCTA)/ (CAATTCACCTGAAGTGCTTAAAGA), respectively. Correctly integrated clones were further screened by PCR of areas covering exons 5 and 6 of the inserted using primers (GCATGAGGGACTTCTAAATTGC)/ (TGCTCTTTTTAAGCCAGGAAAG), and by lack of PCR amplification of either the -BAC [primers / (GTTTTTAGTGTGGCAGTGGTCA)] or the -BAC [primers (TACCCTTACAGTCATCACTCTGGA)/]. xref fig #text The pSC101BADγβαA-I-SceI[amp] vector (Supplementary Figure 1) was constructed to promote enzyme digestion followed by recombination and thereby facilitating modification of BAC clones using ALFIRE. It was assembled by introducing the tetracycline (cTc)-inducible I-SceI homing endonuclease fragment from the pST98AS plasmid () to the pSC101BADγβαA[hygro] plasmid with standard Red/ET protocols (). The entire gene was thus exchanged with the I-SceI-Tet-Rep/Tet-Prom-amp unit generating pSC101BADγβαA-I-SceI[amp]. Positive (Ap selected) clones were verified by SspI restriction enzyme digestion (b) and three clones were sequenced over the modified areas. The vector data were submitted to EMBL (AM403094). The vector pSC101BADγβαA-I-SceI[amp] encodes the Redγ, Redβ, Redα and RecA proteins as well as the homing endonuclease I-SceI. Upon activation with -arabinose and chloro-tetracycline (cTc), respectively, these enzymes will be expressed and cells harbouring this plasmid will become capable of performing digestion and recombination of the resulting ends containing the homologous sequences. Plasmids or BACs containing the 18 bp recognition site of the homing endonuclease I-SceI can be linearized and the double-strand breaks thus formed promote recombination to identical homology sequences. Efficient recombination is especially important when large DNA fragments are being recombined such as in BAC fusion experiments. To test the functionality of the I-SceI meganuclease, 10 bacterial colonies harbouring the I-SceI-CS--BAC and the pSC101BADγβαA-I-SceI[amp] were grown overnight at 30°C in LB media with Km/Cm/Ap. Next day, three parallel tubes with 1 ml LB media were set up and inoculated with 30 μl of each culture and incubated at 1100 rpm with the following conditions () cTc/Cm/Ap, 37°C for 1 h; () cTc/Cm/Ap, 30°C for 5 h; () Cm/Ap, 30°C for 6 h. None of the ten clones propagated in the cTc-induced tubes incubated at 37°C, showing that I-SceI was expressed and that the I-SceI-CS--BAC was linearized at the I-SceI sites. All of the ten clones grew in the non-induced tubes incubated at 30°C indicating that the I-SceI endonuclease was not expressed and that the I-SceI-CS--BAC clone and the pSC101BADγβαA-I-SceI[amp] vector could continue to propagate. Another control experiment was also set up where a clone harbouring only the I-SceI-CS--BAC was incubated at 37°C in LB media (cTc/Cm) for 6 h, 1100 rpm. This clone also propagated indicating that cTc is not toxic to at the appointed concentration. ALFIRE is based on the rationale that HAs can be easily inserted directly into the accepting vector at the same time when the integration of an insert/selection cassette is performed. We used long oligonucleotides containing ∼50 bases homologous to the acceptor plasmid insertion site, ∼55 bases homologous to the insert, and in order to generate the DSBs needed for recombination a unique restriction site (RS) sequence was included prior to the primer binding sequence. The counter-selection cassette -neo (CS) was used for primary selection in order to reduce the background of the undigested accepting vector. Accepting and donor vectors were linearized and transformed into Red/ET competent bacteria for recombination to take place (). With this strategy ALFIRE can be used efficiently irrespective of the vector or insert size and we show here the exchange of a very large fragment (∼55 kb) from one BAC () to another unrelated BAC (). In the initial step, a CS cassette flanked by recognition sites of I-SceI and HAs to the insert and the acceptor vector was designed (see and Supplementary Figure 2). The primers were ∼140 nt long and this was the limit set by the supplier to guarantee good quality primers. The I-SceI sequence was chosen because the 18 bp long recognition site is rare even in genomic DNA. In fact, the chromosome does not contain a single I-SceI site and the meganuclease has been used to generate DSBs into foreign DNA amplified in (, ). First, the CS flanked by two I-SceI restriction sites (I-SceI-CS) was inserted to the mouse (-BAC by Red/ET recombination with the following procedure: the I-SceI-CS cassette was amplified by PCR with primers [CS (I-SceI) to hmLhcgr-F/CS (I-SceI) to hmLhcgr-R] containing (a) the HAs for insertion (∼50 nt) to the promoter and 3′ regions; (b) HAs (∼55 nt) to subclone the human (); (c) the 18 nt recognition site of the homing endonuclease I-SceI (TAGGGATAACAGGGTAAT); and (d) the primer sites (24 nt) for the -neo ( and Supplementary Figure 2). Red/ET competent bacteria carrying the -BAC were transformed with the I-SceI-CS PCR product and plated on LB agar plates supplemented with Km/Cm. Positive clones were screened by PCR (primers mLhcgrP-F/mLhcgr3-R) and further selected by their inability to grow (negative selection) on LB agar plates supplemented with Stp/Cm. In addition, the resulting PCR product was digested with I-SceI (NEB) in order to check for functionality of the inserted I-SceI restriction sites. One of three clones contained a mismatch in the I-SceI sequence leading to reduced or abolished cleavage by this meganuclease. Five positive clones were further selected for sequencing of the entire HA area, which needed to be faultless to allow efficient Red/ET recombination. Four out of five clones were without mismatches and one was selected for further experiments. The I-SceI-CS--BAC construct can thus be used as a general PCR template (I-SceI-CS) and the I-SceI recognition sites would not need to be added each time in the oligonucleotide sequences in future experiments, making the oligos shorter and eliminating the need of checking the I-SceI sites every time. The use of homologous recombination to generate and manipulate large fragments of DNA has opened up new avenues in the study of functional genomics. A limitation of traditional recombinogenic engineering is the difficulty of inserting large fragments into large episomes in a single step. It is necessary to overcome these limitations in order to facilitate the manipulation of large genomic sequences needed for the production of transgenic, knockout and knock-in organisms. We have overcome the major difficulty of inserting HAs into large fragments by incorporating the homologous sequences in the acceptor vector together with unique restriction sites thereby promoting generation of DSBs and facilitating recombination. It is important to note, that often such restriction sites are palindromes and therefore the choice of the endonuclease should be carefully addressed to make sure that the resulting ends have little homology in order to reduce vector re-circularization. It is known that as little as ≥6 nt homology near each end might result in ‘end joining’ (). Homing endonucleases have long recognition sites that are not palindromes and the presence of these sequences is rare even in genomic size DNA. We chose the homing meganuclease I-SceI recognition sequences as unique sites to generate DSBs in the acceptor BAC carrying the HAs for the insert. However, when the entire (∼55 kb) was subcloned into the region between the promoter and the polyA of a BAC containing the mouse homologue (), the efficiency after digestion (I-SceI) and isolation was relatively low (6.3%). The poorer recombination efficiency is most probably due to the limitations of transforming the large BACs, in particular if they are linear, into bacteria and the risk of shredding of the BAC during the DNA purification steps. To overcome these problems we constructed an all-in-one universal vector (pSC101BADγβαA-I-SceI [amp]) to be used in This system expresses the recombinases Redγ, Redβ and Redα as well as the I-SceI homing endonuclease under the tight control of -arabinose and chloro-tetracycline, respectively, and makes it possible to linearize the acceptor vector intracellularly followed by recombination between homologous sequences. In this way only the BAC was linear resulting in improved transformation efficiency and recombination (69% positive clones). Thus, a 10-fold higher efficiency was achieved by compared to DSBs generation by I-SceI () and the background was also reduced significantly with enzyme digestion. We show that by using appropriate unique restriction enzymes sites, and subcloning-HAs, large fragments can be copied in 2 or 3 steps (see and Supplementary for general overview of the ALFIRE method). By adding the HAs for subcloning into the oligos used for amplification of the I-SceI-CS cassette we eliminated the time-consuming use of transfer constructs. The presence of these short exact sequences (∼55 nt) is thus sufficient to result in a highly efficient and accurate recombination. We also show that long oligos, up to ∼140 nt, can generate satisfactory HAs for insertion and for subcloning. Alternatively, the double HAs could be generated by PCR using four shorter oligos (>80 nt) instead of two in the same PCR reaction. The methodology described is highly efficient and an easy protocol to follow in order to insert, fuse, exchange and manipulate virtually any size of DNA or plasmid without long template PCR and undesired selection markers. It is thereby an extra tool in addition to existing methods (,) for the generation and modification of large constructs, and it has unlimited uses from insertion of point mutations to incorporation of large fragments of DNA into BAC, PAC or cosmid vectors. The fact that a 55 kb fragment from genomic DNA can be subcloned into a BAC suggests that high throughput manipulation of BAC libraries and cloning of large fragments of genomic DNA is possible with this straightforward methodology, as well as generation of complex transgenic, knockout and knock-in constructs. A . R . - M . a n d S . L . c r e a t e d t h e i d e a , p e r f o r m e d t h e e x p e r i m e n t s a n d d r a f t e d t h e m a n u s c r i p t . I . H . s u p e r v i s e d t h e r e s e a r c h . A l l a u t h o r s r e a d a n d a p p r o v e d t h e f i n a l m a n u s c r i p t . p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Human mitochondrial DNA (mtDNA) is a closed circular molecule of ∼16.5 kb and was sequenced 25 years ago (,). The two strands of mtDNA are denoted as the Heavy-(H)-strand and the Light-(L)-strand on the basis of their mobility in a denaturing caesium chloride gradient. The strand-asynchronous or strand-displacement model for mammalian mitochondrial DNA replication was first proposed in the early 70s (). In this model, synthesis of the nascent H-strand starts at a fixed point in the major non-coding region (NCR) of mtDNA denoted O. O was originally defined by mapping the 5′ ends of the so called D-loop and is located on the L-strand upstream of three conserved sequence blocks. Leading-strand (nascent H-strand) synthesis proceeds two-thirds of the way around the molecule, displacing the parental H-strand in the process with mitochondrial single-stranded DNA-binding protein (mtSSB) suggested to provide protection against the action of nucleases and other insults such as reactive oxygen species. Following exposure of the lagging-strand initiation site (O) synthesis of the nascent L-strand begins (,). More recently, Holt and co-workers proposed two models of mtDNA replication, one a more conventional strand-synchronous theta mode () where mtDNA replication initiates bidirectionally at various sites across an initiation zone (OriZ). In this case termination occurs at or near O. The other mode of replication is similar to the strand-asynchronous mode of replication so that the nascent L-strand DNA was suggested also to be synthesized with a considerable delay. Initiation is essentially unidirectional and occurs in the NCR, importantly however RNA is deposited on the displaced H-strand rather than mtSSB, thus forming ribonucleotide incorporation throughout the lagging strand (RITOLS) intermediates, which is a crucial difference from the strand-asynchronous model (). Although the high levels of mtSSB () could be seen as supporting the strand displacement model, also for example is estimated to have several thousands of molecules of SSB () even though it contains a single copy genome and replicates via conventional theta replication. SSB is nevertheless essential, as it would be in mammalian mitochondria, not only at the replication fork but also in repair, recombination and other DNA maintenance processes. Given the various essential functions of SSB, the high levels might simply reflect a cell's precaution to ensure it is readily available. The RITOLS model requires that the RNA is replaced by DNA to produce a dsDNA lagging-strand. It was shown that the RITOLS replication intermediates (RIs) are prone to RNaseH degradation during mtDNA purification leaving a single-stranded parental H-strand (), thus producing RIs originally predicted by the strand-asynchronous model. Strand-asynchronous RIs are therefore considered purification/degradation artefacts. In rodent and chick liver and cultured human cells under normal culture conditions RITOLS intermediates are the predominant class (,,). However, in cultured human cells recovering from mtDNA depletion, the majority of the replication intermediates are essentially double-stranded DNA suggesting a switch from the RITOLS replication mode to more conventional theta replication (). Alternatively, initiation of lagging-strand DNA synthesis occurs more frequently resulting in an increased rate of conversion of RITOLS RIs to dsDNA RIs. All proteins responsible for mammalian mtDNA maintenance are encoded in the nucleus, translated by cytosolic ribosomes and imported into the mitochondrial compartment. So far, a limited number of proteins has been identified. These include the mitochondrial DNA polymerase gamma (POLG1) and its accessory subunit (POLG2) [see, e.g. ()], the mitochondrial DNA helicase Twinkle (,), mitochondrial single-stranded DNA-binding protein (mtSSB) () and various proteins with a more general role in mtDNA maintenance. The POLG holoenzyme, Twinkle and mtSSB can form a minimal mitochondrial replisome capable of genome length DNA synthesis on an artificial template (). Some of the components of the mitochondrial replisome and transcription machinery show similarity to their counterparts in T-odd bacteriophages suggesting that a T-odd phage ancestor contributed to the early ‘mitochondrial’ endosymbiosis event (). For example, Twinkle shows striking similarity to the T7 phage primase/helicase protein gp4 (T7 gp4) (). The Metazoan primase domain of Twinkle has diverged from the ones of more primitive Eukaryotes and T-odd phages suggesting it has lost its primase function (). The genes for Twinkle, POLG1 and more recently POLG2 () have been implicated in human diseases. Autosomal dominant (ad) mutations in Twinkle are associated with Progressive External Ophthalmoplegia (adPEO)(), while a single recessive mutation is associated with infantile onset spinocerebellar ataxia or IOSCA (). Mutations in POLG1 are associated with a variety of disorders, including dominant and recessive PEO, various types of ataxia, Parkinsonism and the severe mtDNA depletion Alpers syndrome (see, e.g. () and references therein, and ). The catalytic subunit of polymerase gamma, POLG1, is well-characterized biochemically (see (), bearing similarity with prokaryotic A-type DNA polymerases such as DNA polymerase I and T7 DNA polymerase. Conserved regions include a C-terminal domain responsible for polymerase activity and an N-terminal 3′-5′ exonuclease domain involved in proofreading. Several disease associated POLG1 mutations have been studied using purified recombinant enzyme. These include the common autosomal dominant Y955C and other mutations, which result in a moderate to severe decrease in polymerase activity, reduced nucleotide selectivity or reduced processivity (). , the properties of POLG1 have also been partly characterized in yeast and in cultured human cells (). In both cases, expression of a mutant form of the protein deficient in 3′-5′ exonuclease activity results in the accumulation of mtDNA mutations. An exonuclease deficient variant in mouse also results in a mutator phenotype and shows a whole-organism phenotype of reduced lifespan with a variety of tissue specific ageing associated defects (,). Expression of an adPEO associated Twinkle mutation in transgenic mice has shown a late onset phenotype with striking similarities to late onset PEO (). Although there is a need for further biochemical characterization of POLG1 and Twinkle mutants, and animal models can provide a wealth of information on disease aetiology and pathogenesis, both approaches have their limitations. We therefore chose an alternative approach of inducible expression of wild-type and mutant POLG1 and Twinkle in cultured human cells, allowing us to study protein function and mtDNA replication dynamics . Using this inducible system in combination with two-dimensional neutral/neutral agarose gel-electrophoresis (2DNAGE), we show here that the induced expression of either Twinkle or POLG1 mutants results in distinct replication stalling phenotypes suggesting defined roles for both proteins in mtDNA replication and in particular the frequency of initiation of lagging-strand maturation/synthesis. The full-length cDNA of POLG1 and Twinkle variants were originally cloned in the pcDNA3.1(−)/Myc-His A (Invitrogen, Carlsbad, CA, USA), as previously described (,). All constructs were re-cloned in the pcDNA5/FrT/TO vector (Invitrogen) taking advantage of two PmeI restriction sites flanking the multiple cloning sites of the original pcDNA3 vectors and target vector. The resulting fusion proteins contained the sequence of the respective proteins followed by the Myc-His. All resulting plasmid-constructs were confirmed by DNA sequencing. The Flp-In™ T-REx™ 293 host cell-line (Invitrogen), a HEK293 variant containing a Flip recombination site at a transcriptionally active locus, was grown in DMEM medium (Cambrex Bioscience, Walkersville, MD, USA) with 2 mM L-glutamine (Cambrex Bioscience), 10% FCS (Euroclone, Milan, Italy) and 50 μg/ml uridine (Sigma, St. Louis, MO, USA) supplemented with 100 μg/ml Zeocin (Invivogen) and 15 μg/ml Blasticidin (Invivogen) in a 37°C incubator at 8.5% CO. Two-day prior to transfection cells were split to 10 cm plates and grown to ∼80% confluence in medium lacking antibiotics. Cells were co-transfected with TransFectin (Bio-Rad, Hercules, CA, USA) according the manufacturer's protocol with the appropriate pcDNA5/FrT/TO construct (0.4 μg) and pOG44 (Invitrogen; 3.6 μg), a plasmid encoding the Flp-recombinase necessary for targeted stable integration. Six hour later, transfection medium was replaced with regular fresh medium lacking antibiotics. Twenty-four hour after transfection the selective antibiotics Hygromycin (150 μg/ml) (Invivogen) and Blasticidin (15 μg/ml) were added. Selective medium was replaced every 2 days for cell maintenance. All inducible cell lines were created according this method. To induce expression the indicated amount of doxycycline (Sigma) was added to the growth medium, and cells were processed for further analyses. With longer than 2 days induction medium was refreshed every 2 days. Cell lysates were prepared and analyzed for protein expression by immunoblotting after SDS-PAGE (). A primary monoclonal c-myc (Roche Molecular Biochemicals, Nutley, NJ, USA) antibody was used for detection of recombinant proteins. Peroxidase-coupled secondary antibody horse-anti-mouse was obtained from Vector Laboratories, Burlingame, KS, USA. Enhanced Chemiluminescence detection was done essentially as described (). The copy number of mitochondrial DNA per cell was determined by real time PCR of using the gene for amyloid precursor protein as a nuclear standard as described (). Briefly, crude nucleic acid extracts were obtained from cells by lysis, proteinase K digest and subsequent isopropanol precipitation, and copy numbers of and were determined in a duplex Taqman PCR on an Abiprism 7000 (Applied Biosciences, Foster City, CA, USA) using pCR 2.1-TOPO (Invitrogen) vectors containing the and amplicon as standards. Immunofluorescent detection was done essentially as described previously (). For the detection of mtDNA we used a monoclonal anti-DNA antibody AC-30-10 (PROGEN, Shingle Springs, CA, USA) as described previously (). Secondary antibodies were anti-mouse IgG-Alexa Fluor®488 (Invitrogen; Myc) and anti-mouse IgM-Alexa Fluor®568 (DNA). Image acquisition using confocal microscopy was done as described (). In vitro assays for determination of helicase activities were performed with highly enriched Twinkle preparations derived from 293 Flp-In™ T-REx™ cells. The cells were induced with 50 ng/ml doxycycline (Sigma) for 2 days, harvested and mitochondria isolated by hypotonic lysis and differential centrifugation (). The mitochondrial pellet obtained was lysed in high salt buffer (50 mM KHPO pH 7.0, 1 M NaCl, 1% Triton X-100, 1 × complete Protease inhibitors EDTA-free, Roche) and sonicated on ice (Sonics Vibra-cell, 1 min 40% amplitude, 1 s pulses with 2 s break). The insoluble DNA fraction was pelleted for 10 min at 12 000 g and 4°C. Supernatant was incubated with Talon metal-affinity resin (Clontech, Palo Alto, CA, USA) for 1–2 h at 4°C to allow binding of His-tagged proteins. Resin was washed twice with high salt buffer and twice with low salt buffer (25 mM Tris–HCl pH 7.6, 40 mM NaCl, 4.5 mM MgCl, 10% glycerol, 100 mM L-Arginine) containing 20 mM Imidazole. Elution was carried out with low salt buffer containing 500 mM Imidazole. The supernatant of this step was shock-frozen in liquid nitrogen and stored at −80°C. The assay was performed by incubating 1 ng Twinkle protein in 40 μl helicase buffer (25 mM Tris pH 7.6, 40 mM NaCl, 4.5 mM MgCl, 100 mM -Arginine-HCl pH 7.6, 10% glycerol, 3 mM UTP, 1 mM DTT, 5 μM unspecific oligonucleotide) with 2 amol substrate for 30 min at 37°C. The reaction was stopped by adding 10 μl loading buffer (90 mM EDTA, 6% SDS, 30% glycerol, 0.25% bromophenol blue). Twenty microliter reaction mixes were separated on a 15% acrylamide gel in 1 × TBE, dried on a vacuum gel drier and exposed to X-ray film or quantified by phosphoimager. Point mutation levels in the NCR and region of mtDNA of POLG1 cell lines were measured as previously described (). Mitochondrial nucleic acids were extracted using cytochalasine (Sigma-Aldrich) as described (). Purified mtDNA was digested with HincII and where mentioned further treated with RNase H or S1 nuclease (Fermentas, Hanover, MD, USA) with the indicated amounts and time. The fragments were separated by 2DNAGE as described (,) and the gels were blotted and hybridized with a P-labeled DNA probe for human mtDNA nts 14 846–15 357 (). In order to study the mtDNA maintenance machinery in cultured human cells, inducible cell lines expressing wild-type and mutant variants of Twinkle, the Twinkle splice variant Twinky () or POLG1 were established using HEK293 Flp-In™ T-REx™ cells (see for a list of all variants). Twinkle mutants included a lysine mutation (K421A) in the highly conserved Walker A motif implicated in ATP binding and hydrolysis; a mutation (G575D) in helicase motif H4 and implicated in DNA binding (); a deletion of 31 amino acids (▵346–376) of the region that shows similarity with the T7 gp4 linker region that has been implicated in multimer formation [see e.g. ()]; a large deletion (▵70–343) in the region of the protein that shows similarity with the T7 gp4 primase domain (). POLG1 mutants included two polymerase deficient mutants (D890N and D1135A), one exonuclease deficient mutant D198A and a non-deleterious deletion mutant (ΔCAG) of 10 consecutive glutamines in the N-terminus, all as previously described (). All cell lines and >99% of all cells expressed the recombinant proteins upon doxycycline (DC) induction and expressed proteins were all targeted to the mitochondrial compartment ( and not shown). Since DC is a mitochondrial protein synthesis inhibitor at μg/ml concentrations, we first determined the lowest possible levels of DC to achieve full induction. A shows an increase of protein expression of wild-type (wt) Twinkle in cells with increasing DC concentrations (0–1000 ng/ml.) Both after one or three days, maximum induction levels were reached at low ng/ml concentrations, but at slightly lower concentrations after three days induction. B shows the expression of all proteins used in this study confirmed by immunoblotting, following 72 h of treatment with 0, 3 and 10 or 20 ng/ml DC. All proteins were detected using their respective epitope tag and gave bands of the expected size () upon induction. In the absence of DC we could detect leaky expression of most Twinkle variants, but only when films were overexposed considerably (not shown). However, immunofluorescence in the absence of DC induction did not result in any mitochondrial signal above background fluorescence suggesting the expression levels were very low (see also below and A). Some of the analyzed Twinkle variants, such as Twinky, showed reproducibly lower protein levels with full induction of expression, indicating differences in protein (or mRNA) stability of these variants. Similarly, the POLG1 D1135A mutant showed lower expression levels than the other POLG1 variants, suggesting that the mutant protein is less stable. As a final test for the inducible expression system, we created a cell-line expressing a POLG1 variant (D198A) in which exonuclease activity is abolished. We have shown previously that constitutive expression of D198A in cultured human cells results in the accumulation of point mutations in mtDNA (). To validate the obtained inducible cell-line we determined the point mutation levels in two regions of mtDNA (C). After 60 days of induction, both the and control region showed elevated mutation levels in the D198A cell-line while non-induced D198A cells showed low mutation levels similar to cells expressing POLG1 wt. The relative mtDNA copy number in the various inducible cell lines was compared by quantitative real-time PCR (QPCR) using the nuclear amyloid precursor protein (APP) gene as a standard () (). The absolute copy number determined by us for the HEK293 Flp-In™ T-REx™ and the majority of non-induced transgenic cells was ∼3000 copies/cell (2798 ± 450 ( = 4) for the non-induced parental cell line). Induced overexpression of POLG1 wt or Twinkle wt did not significantly change mtDNA copy number per cell, indicating that abundance of these proteins is not rate-determining for mtDNA replication at least in cell culture. More importantly it also showed that overexpression per se did not otherwise interfere with mtDNA replication. Similarly, overexpression of Twinky, Twinkle Δ346–376, POLG1-D198A and POLG1-ΔCAG did not influence steady-state mtDNA levels. In contrast even low-level expression of Twinkle mutants K421A or G575D and POLG1 mutants D890N or D1135A lead to a dramatic decrease of mtDNA levels within a few days. The Twinkle K421A and G575D cell lines showed a significant steady-state reduction in mtDNA copy number of ∼60% even prior to induction, presumably caused by the slightly leaky expression of the Twinkle variants. This suggests these mutants are strongly dominant in nature. In contrast, the POLG1 D890N and D1135A mutants did not show a significant copy-number reduction without induction (not shown). Notwithstanding the minor leakiness, depletion upon induction was dose-dependent, as higher expression levels lead to a faster depletion than low-level expression (data not shown). For the Twinkle K421A and G575D as well as the POLG1 D890N and D1135A mutants, the mtDNA levels after three days of full DC induction were ∼20–30% compared to non-induced cells, indicating a complete abolishment of successful replication and dilution of mtDNA by cell division. The localization of Twinkle variants was analyzed by immunofluorescence using the myc-tag of the recombinant proteins (A). Overexpressed Twinkle wt showed the typical punctuated pattern within mitochondria, indicating the normal localization in mtDNA nucleoids. Twinky and ▵346–376 both showed diffuse mitochondrial staining. The N-terminal deletion variant ▵70–343 showed almost normal punctuate nucleoid-like localization. The variant G575D could be detected in punctate foci, but in addition showed enhanced diffuse staining in mitochondria. The K421A variant having a mutation in the WalkerA motif was detected in few enlarged spots, indicating either abnormal nucleoid segregation or protein aggregation. To differentiate between these possibilities we used an anti-DNA antibody to see to what extent this and other Twinkle mutants co-localized with mtDNA. As previously shown (), Twinkle wt showed excellent co-localization with DNA as did ▵70–343 (not shown). The K421A variant, despite its abnormal appearance of enlarged foci, did co-localize with mtDNA (B). The number of DNA foci, however, was severely reduced compared to non-expressing or Twinkle wt expressing cells. For Twinkle G575D, the number of nucleoids was again severely reduced (B) but most of the protein-foci co-localized with mtDNA. Both Twinky and the ▵346–376 variant showed normal nucleoid numbers judging from detection with the anti-DNA antibody. The helicase activity of Twinkle variants was compared using His-affinity purified protein in an helicase assay. Twinkle wt showed the expected helicase activity and was able to unwind a DNA substrate with a 5′ overhang of >20 bases, as long as the double-stranded part was less than 25 base pairs ( and data not shown). No such helicase activity could be detected with ▵346–376 or Twinky. Similarly, proteins bearing the mutation K421A in the Walker A motif or the G575D mutation in the helicase motif H4 also had less than 5% residual unwinding activity. ▵70–343 had ∼40% helicase activity of the wild-type protein. The effect of overexpression of the various mutant proteins on replication was studied using two-dimensional neutral/neutral agarose gel electrophoresis (2DNAGE) and Southern blotting, as first established by Brewer and Fangman (,). This method allows visualization of RIs, as DNA fragments are separated both by size and shape. When applied to the analysis of mtDNA isolated from cultured cells, Holt and co-workers showed that various types of RIs can be detected [see e.g. ()]. shows a schematic figure of human mtDNA, indicating the appropriate restriction fragments and probes used for detection of these fragments on 2DNAGE gel blots. The first 2 panels of A show the example of a HincII digest of purified HEK293 mtDNA run on a 2DNAGE gel and probed for the 3.9 kb mtDNA fragment (nucleotide number [nt] 13 636–1006) that includes the whole NCR (for a detailed explanation of the various RIs see Supplementary Data). All the analyses shown in the subsequent figures consider this HincII fragment. Analysis of a second region of the genome was also performed (Supplementary Figure 1). We applied the 2DNAGE methodology to analyze the effects of overexpression of POLG1 and Twinkle variants on mtDNA RIs and in particular to test the hypothesis that the variants depleting mtDNA do so by causing non-site specific, general replication stalling or pausing. First we examined the effect of overexpression of wild-type Twinkle by comparing mtDNA RIs with and without induction (A, right two panels). Very little change in quality or quantity of RIs was observed, with the exception of a presumed resolution intermediate which markedly decreased after induction with >3 ng/ml DC. Similar results were observed with induced expression of untagged wild-type Twinkle (not shown). Increased expression of ▵346–376 or Twinky showed no effect on overall replication (B) or RIs. When overexpressing the Twinkle mutant K421A or G575D, a considerable increase in y- and bubble RIs was observed (C). Since at the same time the total amount of mtDNA decreased strongly, these results indicate a severe, non-site specific, slowing down of replication fork movement. In parallel there was a concomitant decrease in RNA containing RIs, RITOLS or partially single-stranded RIs. Thus, the bubble arc was not only sharper and longer than normal (C) but was substantially resistant to RNase H treatment (Supplementary ). In K421A and especially G575D expressing cells the majority of mtDNA molecules were found on the bubble arc, indicating stalling occurred in the early phase of replication as replicating molecules forming bubble arcs on 2DNAGE harbor initiation site(s) in the fragment. The fraction of molecules in a replicative state was ∼30% in the 3.9 kb HincII fragment, compared to <5% in control cells. In contrast, the ▵70–343 mutant showed a somewhat milder stalling phenotype (C), with the majority of RIs in the upper part of the bubble arc and on the descending area of the y-arc suggesting that replication is not aborted so frequently in the very early stages as was the case for the K421A and G575D mutants. Nevertheless, in this case also RITOLS and single-stranded RIs were considerably reduced. Finally, S1 nuclease treatment for various lengths of time with a fixed amount of enzyme resulted only in very minor changes in the abundance of the stalled RIs in Twinkle stalling mutants (D shows the treatment of ▵70–343 isolated mtDNA). Only with the longest treatment was there some reduction in the bubble arc intensity with a slight concomitant increase in the ascending part of the y-arc, suggesting some single-strandedness in the bubbles resulting in broken bubbles upon S1 nuclease treatment. This is not unexpected as some single-strandedness is always expected close to the junction point of the bubble structure of replication intermediates. The ▵70–343 stalling in this case still showed some RITOLS intermediates and these were efficiently removed by the S1 treatment (compare regions marked by arrows in the right two panels of D). Nevertheless, most RIs were insensitive to the S1 nuclease showing that they are essentially dsDNA (this was further supported by the analyses shown in Supplementary Figures 5 and 6). In contrast, simultaneous S1 treatment in samples with abundant RITOLS showed a strong reduction in RITOLS RIs, illustrating the effectiveness of the S1 treatment (compare regions marked by arrows in the left two panels in D). Overexpression of POLG1 wt (A) and POLG1 ▵CAG (not shown) did not result in any obvious effect on the RIs whereas even modest expression of the mutant variants D890N and D1135A lead to a clear increase in y- and bubble arcs, suggesting these POLG1 mutants also caused replication stalling (B). However, unlike the stalling Twinkle variants, RITOLS and single-stranded RIs persisted in the case of D890N and D1135A POLG1 mutants (B), even after full induction of the mutant proteins for three days (not shown). The proofreading deficient D198A POLG1 showed little change in the appearance of RIs. Three ng/ml DC slightly enhanced both the Y- and bubble arc (C). Only after full induction did we observe a phenotype suggestive of stalling, similar to but clearly less severe than that observed with D890N and D1135A overexpression at low induction levels. Comparison of Twinkle induced replication stalling with POLG induced replication stalling showed a considerable quantitative difference in RITOLS and single-stranded RIs (see also Supplementary Figures 3 and 5). We hypothesized (see also Discussion) that this difference could be explained by involvement of POLG1 in initiation of lagging-strand DNA synthesis or maturation. In this scenario, expression of POLG1 mutants would not only result in stalling of the leading-strand but also in delayed lagging-strand DNA synthesis. To test whether inhibition of POLG1 function could delay lagging-strand DNA synthesis in cells overexpressing either Twinkle wt or the stalling mutations ▵70–343, G575D and K421A, cells were treated with the cytidine analogue dideoxycytidine (ddC), a competitive inhibitor of POLG. Incorporation instead of deoxycytidine can result in chain termination. Surprisingly, already a short 3–4 h treatment with a high dose of ddC (200 μM) in cells showing modest stalling due to Twinkle mutants like G575D, resulted in the reappearance of considerable levels of RITOLS, while accumulated y- and bubble-arc RIs were as prominent as in untreated cells (). The appearance of RIs under the applied conditions was highly similar to the appearance observed with POLG1 stalling (compare lower panels of with those in B). Due to the slightly leaky nature of the Flp-In™ T-REx™ system, the strongest Twinkle stalling mutants G575D and K421A already showed signs of stalling without induction. In particular, the G575D mutant had already lost most RITOLS in the absence of DC. Treatment of these cells with ddC showed a similar reappearance of RITOLS (). The same regime of ddC treatment did not result in obvious changes in RIs in cells expressing Twinkle wt or in non-induced Twinkle wt cells (not shown). Finally, ddC treatment of cells showing a strong stalling phenotype caused by higher level expression of Twinkle mutants did not result in a clear increase in RITOLS (not shown), suggesting that replication was completely abolished under these conditions (see Discussion section). Because mtDNA encodes some of the central components required for cellular energy metabolism, its maintenance is essential for development and overall cell function. While human and mouse mtDNA were sequenced 25 years ago, for a long time the knowledge of the replication and repair machinery was lacking. This made it very difficult to test predictions of replication models for example by reconstituting the various components needed for replication or by manipulating the individual enzymes. Using inducible overexpression of wild type or mutant variants of Twinkle and POLG1 in cell culture, we show here that replication stalling results in changed patterns of RIs that can best be understood by considering the proposed mechanisms of lagging-strand synthesis. Most notably our results show that stalling induced by deficient Twinkle results in RIs that mimic conventional strand-coupled RIs and suggest that initiation of lagging-strand DNA synthesis or maturation occurs at multiple sites across the genome. We furthermore propose that this maturation involves POLG1. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

One of the most intriguing aspects of non-coding RNAs (ncRNAs) is their post-transcriptional modification. Post-transcriptional modifications in RNAs are of three main types (): (i) conversion of uridine to pseudouridine (5-ribosyluracil, Ψ); (ii) methylation of 2′ hydroxyls (Nm) and (iii) alterations to bases, generally methylations of different positions (mN). High-pressure liquid chromatography (HPLC) coupled to mass spectrometry (MS) is the first method of choice for the analysis of post-transcriptional modifications because it is both very sensitive and specific. Furthermore, nearly all modifications cause a change in mass, making detection with LC/MS fairly straightforward. Up till now, post-transcriptional modifications have been studied mostly in tRNA and oligonucleotide models (). On the other hand, 16S rRNA is much less investigated and consequently, the number of organisms for which the post-transcriptional modifications in the 16S rRNA are characterized is much smaller (). is the only mesophilic bacterium for which the rRNA (16S and 23S) appears to be fully characterized () and recently, the 16S rRNA modification map of was published (). The collective importance of post-transcriptional modifications for efficient protein synthesis has been demonstrated by the superior performance of authentic rRNAs compared to unmodified 16S rRNA () and 23S rRNA () counterparts. Further analysis of the function of post-transcriptional modifications in 16S rRNA would most definitely benefit from a larger number of fully characterized organisms to allow comparison of the data (). As a first step in the production of these data, we analyzed the 16S rRNA. was selected in order to add a Gram-positive, anaerobic and mesophilic bacterium to the short list of bacteria for which the 16S rRNA modification map has been completed (). Furthermore, the rather low-G content of the 16S rRNA (30%) leads to larger oligonucleotides when the rRNA is hydrolyzed with the G-specific endonuclease T1 and this facilitates the localization of modifications. At first, modified nucleosides were identified solely based on their chromatographic mobility (e.g. P and/or C-labeling and 2D electrophoresis combined with thin layer chromatography (TLC), anion exchange chromatography and HPLC). However, these methods suffer from poor specificity and reproducibility and identification becomes problematic as the number of modifications or RNA chain length increases. In contrast, MS is a better technique for the analysis of post-transcriptional modifications, because nearly all modifications produce a change in mass of the canonical nucleosides (). The application of MS to nucleic acids has long been restricted due to the experimental difficulties associated with the ionization of polar compounds such as nucleosides, nucleotides and oligonucleotides. Since the development of electrospray ionization (ESI) and matrix-assisted laser desorption ionization (MALDI), these molecules can be ionized and analyzed by MS. ESI currently holds the advantage because of its greater accuracy and ease with which it can be coupled to chromatographic separation systems. Combining the high-resolving power of the HPLC system and the high specificity of the mass measurement offers a method which is superior to either method alone (). More than a hundred different modifications are known () and the majority have been characterized by some form of MS. The general setup for identification and sequence placement of post-transcriptional modifications in large RNAs is based on the LC/ESI-MS analysis of two enzymatic digests. First the rRNA strand is digested to the nucleoside-level and analyzed by LC/ESI-MS. The combination of chromatographic retention times and mass measurements allows identification of nearly all the modifications that are present in the original intact RNA strand. Distinction between isomers (e.g. mC and mC) is made based on comparison to tandem MS (pseudo MS, see the Methods section) analysis of reference samples. In a second step, sequence-specific endonucleases are used to cleave the intact RNA at specific positions (e.g. RNase T1 cleaves the 3′-side of all unmodified guanosines and RNase A cleaves at the 3′ end of all pyrimidine nucleotides). This way, oligonucleotides restricted to one or two 3′-nucleotide types are produced, which are subsequently analyzed by LC/ESI-MS. Comparison of the observed mass data with data predicted from the gene sequence identifies oligonucleotides that contain modifications. These anomalous oligonucleotides are analyzed by MS/MS for exact sequence placement of the modification in the oligonucleotide and thus in the RNA sequence. Two types of MS/MS experiments were used: In a first condition without precursor selection in the quadrupole, a higher collision energy is used to release monomer ions from the oligonucleotide. These monomers are often base ions, but also nucleoside phosphates and cyclic phosphates can be used. The Time-of-Flight (TOF) analyzer screens both the low-mass region for the modified monomer and the high-mass region for the intact oligonucleotide. Corresponding signals in the reconstructed ion chromatograms (RICs) identify the nature of the modification in the oligonucleotide. In a second condition, MS/MS analysis, with precursor selection, is used for sequencing of the oligonucleotide and exact sequence placement of the modification that was identified in the base-release experiments. Using LC/ESI-MS nearly all the post-transcriptional modifications can be mapped in the RNA sequence. Pseudouridine, however, does not allow straightforward analysis by MS because it is isobaric to the omnipresent uridine. Patteson . () and Mengel-Jörgenson . () reported addition of a mass tag to all Ψ by specific derivatization. We developed a method to increase Ψ detection sensitivity by derivatization with methyl vinyl sulfone (). Pomerantz and McCloskey identified Ψ solely based on identification of signature masses for Ψ (). Unfortunately, the presence of Ψ in the sequence GΨG (as in the case of 16S rRNA) makes these methods not useful because they are all based on preliminary RNase T1 digestion, which will produce numerous UG isomers. Pomerantz and McCloskey propose in this case to replace RNase T1 by nuclease U2, but this enzyme is no longer commercially available. Therefore, a reverse transcriptase assay was used for placement of pseudouridines in the 16S rRNA sequence. Pseudouridine bases were selectively derivatized with -cyclohexyl-′-β-(4-methylmorpholinium) ethylcarbodiimide -tosylate (CMCT). Upon subsequent reverse transcription, CMC derivatized Ψ blocks elongation of radioactively labeled primers, highlighting the position of the Ψ on a PAGE gel (). Even without CMC derivatization, the reverse transcriptase approach can be used for sequence confirmation of modifications found by LC-MS, because most modifications block the reverse transcriptase elongation. Therefore, reverse transcriptase assays can be used to determine the sequence location of modifications when this is not possible by LC-MS analysis. Examples are the identification of a modified oligonucleotide in a group of isomers or when problems arise during MS/MS sequencing (for examples, see the result section below). In any case, the reverse transcriptase stops should be checked by simultaneous analysis of transcribed RNA, because stops can be caused by secondary structure of the RNA, even at elevated temperatures (,). (ATCC824) was grown in 2 × YT medium containing 16 g bacto-tryptone, 10 g yeast extract, 4 g NaCl and 10 g glucose per liter. Cultures were grown anaerobically in a modular atmosphere-controlled system (MACS, Don Whitley Scientific, Shipley, UK) at 37°C in 50 ml medium. Growth phase was monitored by turbidity measurement at 600 nm and cells were harvested at an OD of ±1.5 by centrifugation at 10 000 r.p.m. for 10 min. Multiple isolates of these 50 ml batches were combined and pellets were stored at −80°C. 16S rRNA was isolated in two steps. A first step consisted of the isolation of total RNA from the bacterial pellet using lysozyme digestion and phenol/chloroform extraction. In a second step, the 16S rRNA was purified from this total RNA mixture by rate zonal centrifugation through a 10–30% sucrose gradient. The samples were centrifuged at 94 000 on a Beckman L7-55 ultracentrifuge (SW 28 rotor) for 24 h, and 750 μl fractions were collected. To produce the nucleoside mixture, 100 μg (3 μg/μl) 16S rRNA was hydrolyzed with 10 U nuclease P1 (Amersham, Chalfont, UK), 0.01 U snake venom phosphodiesterase I (E.C.3.1.15.1, Sigma, St. Louis, MO, USA), and 2.5 U shrimp alkaline phosphatase (E.C. 3.1.3.1., Amersham) as described by Crain (). Digests were stored at −80°C until analysis by LC/ESI-MS. To produce the oligonucleotide mixture, 100 μg 16S rRNA (10 μg/μl in 10 mM Tris–HCl, 1 mM EDTA acid, pH 7.4) was hydrolyzed with RNase T1 (E.C. 31.27.3, Amersham-GE Healthcare) at 37°C for 30 min. The amount of enzyme was 1 U RNase T/3 μg rRNA and digests were also stored at –80°C until analysis by LC/ESI-MS. With these conditions, mG and mG modifications were resistant to RNase T1 hydrolysis. For the nucleosides mixture, 2–5 pmol of rRNA hydrolysate were injected directly onto an Atlantis dC18 column (0.32 × 150 mm, 3 μm diameter particles, Waters, Milford, USA). The column was eluted at a flow rate of 5 μl/min using an ammonium acetate (50 mM in Millipore Milli-Q water, pH 6.0)/acetonitrile gradient (). Diode array UV absorbance data were acquired from 240–300 nm. The chromatographic effluent was conducted without splitting into the mass spectrometer and two different scan functions were used. For the first scan function, the cone voltage and collision energy were set to 30 V and 10 eV, respectively, allowing mass measurement of the intact nucleoside. Qualitative interpretation of data was carried out as previously described (). Data were acquired over a mass range of 100–700. For the second scan function, pseudo MS was accomplished by fragmentation in the source, selection of the base anion in the quadrupole and fragmentation in the collision cell. The cone voltage used for MS was 30–40 V and the collision energy 30 eV. Data were acquired in positive mode over a mass range of 50–250. With this second scan function, distinction between nucleoside isomers (e.g. mC and mC) is possible. The same instruments were used as for the analysis of the nucleoside mixture. 2–5 pmol of the oligonucleotide mixture were injected directly onto a PepMap C18 column (0.3 × 150 mm, 5 μm-diameter particles, LC-Packings, Amsterdam, The Netherlands). The column was eluted at a flow rate of 5 μl/min, using a 1,1,1,3,3,3-hexafluoro-2-propanol (HFiP, Acros, Geel, Belgium) based solvent system (60 mM HFiP in MQ water, adjusted to pH 7.5 with triethylamine) (). The organic modifier was acetonitrile and a 2-step linear gradient (0.5%/min during 50 min and 5%/min during 10 min.) was used. Diode array UV absorbance data were acquired from 230 to 320 nm. Different scan functions were used for data acquisition. During the first scan function, cone voltage and collision energy were kept at 30 and 10 eV, respectively. Thus, minimizing fragmentation, intact oligonucleotide molecular masses can be detected. Data were recorded in continuum mode over a 500–1500 mass range. For the second scan function, the collision energy was increased to 40 eV to release monomers from oligonucleotides. Data were acquired over a 100–1500 mass range and modified bases and intact precursor ions (the modification containing oligonucleotides) are recorded simultaneously. Oligonucleotide sequencing was performed in a third scan function at 30 V cone voltage and collision energies varying from 20 to 40 eV. During this scan function, precursor ions were selected in the quadrupole. Finally, to confirm the detection of sugar methylated nucleosides, a fourth scan function was used: The cone voltage was increased to 60 V in order to release modified nucleoside cyclic phosphate ions. After selection of these ions in the quadrupole, a collision energy of 20 eV was used to release the unmodified base and the methylated ribose ion (pseudo MS). Mongo Oligo Mass Calculator () generated a mass-ordered list of all oligonucleotides produced by RNase T1 digestion of 16S rRNA of . This list is calculated based on the gene sequence (GenBank accession number NC_003030) and the software allows calculation of the electrospray series and the selection of modified nucleosides. In-house software was used to search the 16S rRNA sequence for oligonucleotides corresponding to masses that were found in the mass data from the samples analyzed. It allows the user to restrict the search on the basis of the 5′ or 3′ end, number and residue masses of modifications, undercut residues and phosphorylation state. Simple oligonucleotide sequencer (SOS) was used for the sequencing of modified oligonucleotides based on the fragmentation spectra produced by CID (). The gene sequence corresponding to 16S rRNA was isolated from genomic DNA by a PCR with primer 27f: 5′-AGAGTTTGATCCTGGCTCAG-3′ and primer 1492r: 5′-GGCTACCTTGTTACGACT-3′ (Sigma Genosys). For this PCR, a mixture of 10:1 Taq: Pfu Turbo polymerases was used and primer concentrations were 0.5 μM. Buffer conditions were as provided by the TOPO cloning protocol. After 3 min at 94°C, 20 cycles of 1 min at 94°C, 1 min at 55°C and 3 min incubation at 72°C were used. Incubation of the samples for 7 min at 72°C guaranteed full-length PCR products with 3′-A overhangs, necessary for TOPO cloning. After purification with a PCR purification kit (Qiagen, Crawley, UK), the amplified 16S rRNA gene was cloned, downstream of the T7 promotor sequence, into a pCR4-TOPO vector (Invitrogen, Carlsbad, CA, USA) following the protocol provide by the supplier. After heat shock transformation of TOP10 cells, distinct colonies were incubated overnight in 0.01% ampicillin containing LB. For the selection of colonies with inserts in the correct direction, PCR was performed with two combinations of primers. For a first PCR (the positive control), these primers were 27f (sequence as above) and M13r: 5′-GGAAACAGCTATGACCATG-3′. If the insert is in the correct position, a band should be visible at 1.6 kB after agarose gel electrophoresis. A second PCR (negative control) with primers 27f and M13f: 5′-GTAAAACGACGGCCAGT-3′ was used for confirmation. If the insert is in the correct position, no band should be visible after agarose gel electrophoresis. Conditions were as above, except that primer concentrations were 0.25 μM and no final 7-min incubation step at 72°C was necessary. Correct clones were digested with PmeI, which produces blunt ends downstream of the rRNA sequence. After cleanup with a PCR purification kit (Qiagen), RNA was produced using the Ribomax T7 kit (Promega, Madison, WI, USA). After RNeasy Mini Kit (Qiagen) purification, 1–2 μg/μl 16S rRNA was obtained. Fifteen to twenty-five microgram of total RNA was derivatized as in ref. (). To remove the CMC groups of U and G, the mixture was placed for 4.5 h at pH 10.4 on 37°C. After precipitation with sodium acetate and ethanol the RNA was reverse transcribed with Avian Myeloblastosis Virus (AMV) reverse transcriptase (GE Healthcare, Brick, NJ, USA) using 5′-P-labeled primers. The primer complementary to residues 548–569 ( numbering) was used for identification of Ψ-516. Other primers used for Ψ-screening were complementary to residues 1098–1219, 1221–1243, 1435–1475 and 1492–1510. For hybridization, the mixture containing 6 pmol RNA, 0.3 pmol primer and hybridization buffer (55.5 mM HEPES Potassium salt pH 7.0 and KCl 111 mM) in a total volume of 13.5 μl was heated at 80°C for 2 min and cooled gradually to 40°C in 30 min. For the extension reaction, 1 μl hybridization mixture and 1 μl 100 μM ddNTP were added to 3 μl extension mix containing: 0.166 mM Tris–HCl pH 8.4, 13.3 mM MgCl, 13.3 mM DTT, 0.066 mM each dGTP, dATP, dCTP and dTTP (GE Healthcare) and 0.33 U AMV reverse transcriptase. Samples were incubated 5 min at room temperature and 35 min at 42°C. After reaction and precipitation with sodium acetate and ethanol, the samples were dissolved in 6 μl urea loading buffer containing 8 M urea, 20 mM Tris–HCl pH 7.8, 1 mM EDTA and 0.02% xylene cyanol and bromophenol blue dyes. The samples were loaded on 7 M urea − 8% polyacrylamide gels and run in 1× TBE buffer at 45 W until the bromophenol blue dye reached the bottom of the gel. Radioactive products were visualized using a PhosphorImager. Except for the CMC derivatization, sequence location of mC at position 1409 ( numbering) was performed by reverse transcription of radiolabeled primers as described above. The radiolabeled primer was: 5′-CCAAAAGGTTACCTCACGGG-3′ complementary to nucleotides 1435–1475 ( numbering). Modified nucleosides were identified by LC/ESI-MS analysis of the 16S rRNA. The chromatographic separation of a total nucleoside digest is shown in . The presence of nine different modified nucleosides in the 16S rRNA of is indicated, Ψ, D, mC, Cm, mG, mU, mCm, mG and mA. Most of the assignments could be derived from relative retention times and correspondence of RICs for the protonated base and nucleoside masses. For mC, mCm and mU, specific placement of the modifications on the base or sugar residue was done based on LC/ESI-pseudo MS comparison to standard samples (data not shown). The absence of specific tRNA nucleosides (in particular -threonylcarbamoyl-adenosine) indicates that no tRNA contamination is present (). The mass silent modification pseudouridine (Ψ) was not detected by LC/ESI-MS of the nucleosides digest and it was added in based on literature data (). Even with the method described in (), no reliable identification was possible, probably because of the very low concentration, short retention time on the reversed phase column and higher number of fragment ions (). The structures of the modified nucleosides that were characterized in this study can be viewed at (). The modified nucleosides depicted in were mapped onto the 16S rRNA sequence by LC/ESI-MS analysis of an RNase T1 digest of the intact purified 16S rRNA. Comparison of the measured data to data predicted from the gene sequence identified eight modified oligonucleotides. These oligonucleotides and the values for modified monomers inside these oligonucleotides are shown in . The modified monomers are released during LC/ESI-MS/MS analysis with fragmentation in the collision cell, without precursor selection in the quadrupole. This experimental setup allows detection of the released modified bases and nucleotide monomers at the same time as the molecular ion signals of the oligonucleotides from which they originate. illustrates these experiments for the oligonucleotide AAmGmCAACGp. Exact sequence placement of these modifications and of the sugar modifications was done in a third step where fragmentation with precursor selection was used to sequence the modified oligonucleotides. Using sequencing software () unambiguous mapping of the modification in the modification containing oligonucleotides and thus in the rRNA could be established (). Sequencing analysis by MS/MS of the oligonucleotide UCACACCAUGp (mass: 3196.4), for sequence placement of the extra methyl group, was problematic. Both precursor ions with sufficient intensity for MS/MS fragmentation ( values 1064.5 and 798.1) co-eluted with other oligonucleotides with nearby values (e.g. the 3′-terminal AUCACCUCCUUUCU, without 3′ phosphate, with mass 4262.5 and 1064.6 and the oligonucleotide AUUAAUACCGp at pos. 155 with mass 3204.4 and 800.9). The masses of the precursor ions could not be selected separately without sacrificing sensitivity. Fortunately, mC blocks reverse transcription and sequence location of this modification could be confirmed by a reverse transcriptase assay (result not shown). The phosphorimage of reversed transcriptase products identifying Ψ516 in 16S rRNA is depicted in . This figure also shows a band caused by mG, which is also important because of the redundancy of the CCGCGp oligonucleotide produced by RNase T1 (three occurrences). With the reversed transcriptase technique, ∼30% of the 16S rRNA, containing the Ψ modification sites in other organisms (), were scanned. No other Ψs were detected in . Of the 11 post-transcriptionally modified nucleosides that were reported in 16S rRNA, seven were also detected in 16S rRNA, while position 966 contains a mG in and a mG in . Interestingly, the same seven modifications, and the mG at position 966, were recently reported in the 16S rRNA of (), indicating that these modifications are perhaps indispensable for the ribosomal biosynthesis or maturation. Indeed, they are situated in or near helices 18, 31, 34, 44, 45 (), which are functionally important sites in the small subunit (SSU), where contacts to tRNA and the large subunit (LSU) are made (,). This location would allow them to directly influence these interactions or play a role in the 3D folding of these important domains. As shown in , three modifications in are not present in , while three modifications in are absent in . Some of these differences are situated in helix 34 in the head domain of the SSU. The Cm at position 1195 ( numbering) in is not present in , although the base sequence is nearly the same: … AGACGUAAGUCA … in and … UGACGUAAAUCA … in . The mG modification at position 1207 in is not present in , but this is not a surprise as the primary sequence of the 16S rRNA is different at this position (C in , G in ). Helix 34 interacts with A site tRNA, but the influence of these differences in modification maps on ribosomal function is unclear. Our data also indicate the presence of a dihydrouridine nucleoside at position 1211 or 1212, in between helices 32 and 34. However, both the LC/MS/MS data and RT assays do not allow identification of the exact location site. This is the first report on the presence of this modification in 16S rRNA, but further research is necessary to confirm the exact location of this modification and to determine its function. Other differences are located in the 3′ minor domain. In helix 44 of 16S rRNA, a mC is present at position 1407, while in this modification is present at position 1409 ( numbering). Alignment of incomplete modification maps of all organisms in the SSU database indicates that both are rather conserved modification sites (). Why these two adjacent cytosines are so often modified is not known, both are situated in the functionally important helix 44, at either site of the bulged out A1406. G1516 in helix 45 of 16S rRNA, is methylated at position 2 of the base, while contains an unmodified A at this position. Probably the purine at this position requires a rather hydrophobic site at position 2. Again, no further insight in the role of this modification is known. In general, the level of modification appears to be equal in and 16S rRNA (each 11 modification sites), while 16S rRNA is more extensively modified (14 modification sites). However, it appears that can only represent mesophilic bacteria to a certain extent as some modifications are absent from 16S rRNA, while some modifications are absent from 16S rRNA. The bacterial ribosomal SSU is a target for many antibiotics. Both aminoglycoside () and cyclic peptide antibiotics () target the top of helix 44 and tetracyclines bind multiple sites in the SSU, e.g. helix 34. It has been reported that some post-transcriptional modifications alter the resistance to these ribosome-targeting antibiotics. Therefore, knowledge of the natural modification maps might be interesting for the analysis of this type of resistance mechanism. As is an important industrial bacterium, novel antibiotic resistance genes might be used for plasmid maintenance during genetic engineering. Furthermore, more information on this resistance mechanism is also interesting from a therapeutic point of view, because the genus contains several pathogenic species. The influence of post-transcriptional modifications on the sensitivity to aminoglycosides has been the best studied, however, anaerobic species are resistant to these antibiotics because they lack active membrane transport. Furthermore, resistance to the cyclic peptide antibiotics (capreomycin and viomycin) has been reported to be caused by the loss of a 2′--methyl group from C1409 ( numbering) in (), other species without this modification (e.g. ) are less sensitive. The influence of the 5-methylation of C1409 in or the loss of this methylation on the sensitivity to capreomycin or viomycin remains to be studied. is sensitive to tetracyclines, but no links between resistance to these antibiotics and rRNA methylation has been reported yet (). The results for were compared to the T1 catalog data, detected by Tanner . () and compiled in the SSU database () These short sequences were used in early studies to establish patterns of phylogenetic relatedness. Although itself was not determined, good correspondence was found with the other gram-positive anaerobic species (). Not surprisingly, the best correspondence was found with , phylogenetically closest to . During the last decade, several methods have been published that use MS for the analysis of post-transcriptional modifications (). Quadrupole instruments are very popular because of the flexibility of their scan functions. Unfortunately, some of these scan functions are not available on a oa-qTof instrument. On the other hand, to determine the nature of modified bases in an oligonucleotide, the oa-qTof allows the use of a different scan function from the quadrupole scan functions. In quadrupole instruments, an increased cone voltage is applied to release modified bases or nucleotides from the oligonucleotides. Careful time alignment of chromatic profiles of runs with and without increased cone voltages is needed to identify modified oligonucleotides from which the modified bases or fragments originated. In contrast, in oa-qTof instruments, the release of these fragments can be initiated by increasing the collision voltage in the collision cell, thus producing monomer ions, which are measured in the TOF analyzer. The broad mass range of the TOF analyzer allows the simultaneous detection of the small base ions ( 100–150) and the large oligonucleotide molecular ions ( 500–1500) from which the base ions originate. With this different method of collision cell fragmentation without precursor selection, time alignment of chromatographic profiles is no longer necessary and nearly unambiguous connection of modified base ions and modification containing oligonucleotides is possible. It is another extension of the already wide variety of MS based methods for the analysis of post-transcriptional modifications in large RNAs. In source fragmentation instead of collision cell fragmentation is not an option on the oa-qTof, because the quadrupole filter, even when used in rf-only mode, restricts the transmission of a wide range. This makes it impossible to simultaneously detect base and oligonucleotide anions. Although LC/ESI-MS is extremely powerful in the analysis of post-transcriptional modifications, it seems that in some cases the combination with reverse transcriptase assays is necessary. For example, if both RNase T1 and RNase A digests produce redundant oligonucleotides (e.g. CCGCGp) or when MS/MS sequencing is problematic because of co-eluting oligonucleotides with the same values (e.g. 1064.5 for UCACACCAUGp, = −3, and the 3′ terminal oligonucleotide AUCACCUCCUUUCU, without 3′ phosphate with 1064.6 and = −4). As was illustrated in this study for the oligonucleotide CCmGCGp, reverse transcriptase is stopped at base modified nucleosides and, although exact identification of the modification is not possible, it can be used to discriminate redundant modified oligonucleotides, when comparison to data predicted from the gene sequence can not reveal the exact location of the modification. Furthermore, analysis of pseudouridine by LC/MS is not straightforward because it is isobaric to uridine. Given the fact that the only known pseudouridylation site in and 16S rRNA is present at position 516 ( numbering) () and given the great correspondence between modifications in and modifications found in , Ψ was expected to be present at position 516 in . 16S rRNA as well. Therefore, all methods based on preliminary RNase T1 digestion can not be used as they will produce 35 UG isomers from the 16S rRNA. Pomerantz and McCloskey propose in this case to replace RNase T1 by nuclease U2, but this enzyme is no longer commercially available. We have chosen the approach described by Bakin and Ofengand (): All U-like and G-like residues were derivatized with CMC followed by an alkaline removal of all CMC groups except those linked to the N3 of Ψ. CMC derivatized Ψ blocks the reverse transcription reaction of radiolabeled primers, highlighting the position of the Ψ on a PAGE gel. For both the analysis of pseudouridine and other modifications by reverse transcription assays, comparison of wild-type 16S rRNA to produced 16S rRNA is necessary to differentiate between bands on the gel that are caused by secondary structure of the 16S rRNA and bands that are caused by post-transcriptional modifications. Thus unambiguous identification of modification sites is possible.

The La protein is a highly abundant phosphoprotein first described in human as an autoantigen in patients suffering from the rheumatic diseases, systemic lupus erythematosus and Sjögren's syndrome (,). It is an RNA-binding protein involved in many aspects of RNA metabolism () and is present in a wide range of eukaryotes including budding and fission yeasts, vertebrates, insects, worm () and trypanosome (). The La protein is one of the first proteins to bind to primary polymerase III (pol III) transcripts due to the specific recognition of the 3′-UUU-OH motif present in these precursors (). The La protein (named Lhp1p) also binds polymerase II (pol II) transcribed small RNAs that terminate in 3′-UUU-OH such as precursors to the U3 snoRNA (mall uclelar ) or U snRNAs (mall uclear ) (). From yeast to human, genetic and biochemical studies have shown that La protects these small RNAs from 3′–5′ exonucleases (,). For example, the binding of La to pre-tRNA precursors prevents exonucleolytic nibbling of their 3′-trailer and promotes its endonucleolytic removal () and Lhp1p stabilizes U snRNAs and U3 precursors from exonucleolytic degradation (,,). In addition, the La protein fulfils an RNA chaperone activity () involved in the assembly of several RNPs (,) and in the structural stabilization of pre-tRNAs (,). Finally, the La protein most probably takes part in the quality-control mechanism of newly synthesized non-coding RNAs such as pre-tRNAs (,,). To accomplish its various functions in the biosynthesis of small stable RNAs, most of the La protein logically accumulates in the nucleoplasm as assessed by steady-state subcellular localization experiments (). Nevertheless, the La protein subcellular distribution is highly dynamic as this protein was shown to shuttle through the nucleolus in association with several precursor RNAs () and strongly accumulates in the nucleolus during late G1 and early S phases for yet unknown reasons (). But not all La protein is present in the nuclear compartment. It has been demonstrated that 2–4% of the La homologue accumulates in the cytoplasm () and that the human La (hLa) protein shuttles between nucleus and cytoplasm (). Moreover, a major pool of La protein is redistributed to the cytoplasm under various stress conditions such as apoptosis () or viral infections (,). These observations are in agreement with several reports suggesting that, beside their primary role in pol III and non-coding pol II stable RNAs biosynthesis, La homologues could be implicated in mRNA translation enhancement [for review see ()]. For example, by binding to their 5′-UTR, the cytoplasmic La protein stimulates the internal ribosome entry site-mediated translation of viral mRNAs (,) as well as certain cellular mRNAs (,). Also, La is involved in the cap-dependent translation of 5′-terminal oligopyrimidine stretch (TOP) containing mRNAs (). La proteins are modular polypeptides whose molecular weight ranges from ∼50 kDa in vertebrates to 32 kDa in yeasts. Their NH2-terminal domain (NTD) is extremely well conserved and always contains a 60–80-amino-acid-domain called the La-motif, also found in a number of otherwise unrelated (La-like) proteins (), closely followed by a canonical NA ecognition otif (RRM1) (). Diverse structural () and functional studies (,) emphasized the importance for a La protein to display this particular NTD organization at least for efficient and specific binding of 3′-UUU-OH-containing substrates. The COOH-terminal domain (CTD) of La proteins is more variable. The hLa CTD contains an atypical RRM (RRM2) ending with a long helix comprising a nuclear retention element (,), followed by a hort asic otif (SBM), several phosphorylation sites (), a uclelar ocalization ignal (NoLS) () and a uclear ocalization ignal (NLS). The RRM2 motif is found in La proteins from all vertebrates, but is absent from the very short CTD of the yeast proteins and was not detected for La homologues from some metazoans such as fly or worm (). In higher plants, a functional homologue of the La protein is yet to be identified. We report here that higher plants are exceptional compared to other eukaryotes by having two distantly related proteins that display every structural feature of genuine La proteins. protein ) is a true functional homologue as it is able to fulfil the nuclear La functions related to RNA pol III transcripts maturation and stability. We also demonstrate that T-DNA insertion in the gene leads to embryonic lethality showing that AtLa1 function is required for plant viability. The amino acid sequences of the different La motifs were aligned using ClustalW multiple-alignment program (). Evolutionary distances were calculated using the probability matrix from blocks (PMB) model () of the Protdist program (PHYLIP package version 3.6, available at the following web site: ). The coefficient of variation of the gamma distribution (to incorporate rate heterogeneity) was obtained by pre-analysing the data with the Tree-Puzzle program (), and the significance of the various phylogenetic lineages was assessed by bootstrap analyses (1000 trials) (). The phylogenetic tree and the consensus tree were inferred using the Neighbour-Joining (NJ) and Consense programs (both from the PHYLIP package), respectively. cDNAs corresponding to At32 and At79 isoforms were PCR amplified from a cDNA library derived from Col0 suspension cell line (). PCR products were inserted in pBluescript vector and two clones for each cDNA were fully sequenced. Plasmids were named p116 (At32 in pBSK) and p114 (At79 in pBSK), respectively. The centromeric yeast shuttle vector pFL38 (-) () was used as a platform to assemble the different genes under the control of the promoter region () with or without the Protein A tag (). First, a 1169-bp SacI-BamHI and a 264-bp BamHI-HindIII restriction fragments, both from the pGALPATG1L vector (obtained from K. Hellmut and E. Hurt, University of Heidelberg), containing the GAL1-10 promoter region fused with the Protein A tag, and the GAL4 transcription terminator region, respectively, were cloned into pFL38 to produce the pFL38GALProtA vector (p131). The open reading frames (ORFs) were PCR amplified from plasmids p116 and p114, respectively, while the gene was amplified from the pATL vector [generous gift from Sandra Wolin (HHMI, New Haven)]. All PCR amplifications were performed using a 5′-primer with a terminal BamHI site (to produce the tagged version) or with a terminal SphI site (to produce the untagged version) in combination with a 3′-primer ending with an XhoI site. The tagged and untagged final constructs were produced by cloning the different PCR products in pFL38GALProtA after digestions with BamHI-XhoI (tagged version) or with an SphI-XhoI (untagged version). By this way, we obtained the pGAL [numbers p140 (), p139 (At32) and p138 (At79)] and pGALPa [numbers p133 (), p132 (At32) and p135 (At79)] collections. To construct the high-copy vectors, the regions of plasmids p131, p133 and p132 were excised by ClaI digestion and replaced with the 2 μL region of pFL44L () to produce the p2μGALPa plasmids [p144 (Protein A alone), p150 (), p145 (At32) and p146 (At79)]. Restriction sites (NheI-SmaI at the 5′-end and BamHI-XbaI at the 3′-end) were added by PCR amplification to the eGFP ORF. The PCR amplification product was inserted into pBluescript vector at sites HincII and XbaI to obtain pBSK-eGFP (p315) and verified by sequencing. The At32-coding region was subsequently inserted into the BamHI and XbaI sites of the pBSK-eGFP resulting in NH2 GFP fusion (vector p316), and into the XhoI and NheI sites to give COOH fusion (vector p317). After sequencing, each fusion as well as the eGFP ORF were transferred at the KpnI and XbaI restriction sites of the plant transformation vector pBIN-HYG-TX () under the control of the cauliflower mosaic virus 35S promoter (CaMV35S) giving plasmids p320 (GFP), p321 and p322 (GFP-At32 and At32-GFP). Growth and handling of were by standard techniques (). The red/white sectoring medium contained 1% bactopeptone, 0.5% yeast extract and 4% galactose. Fluoroorotate tests were performed on Yeast Nitrogen Base (YNB) medium containing 2% galactose, supplemented as required in amino acids and containing 1mg/ml of proline and 1 mg/ml of fluoroorotic acid (Melford Lab. Ltd, ref F5001). The BP1 (a /pATL), CY2 (α ) and CY3 (α ) strains (,) used in the present work were kindly provided by Sandra Wolin (HHMI, New Haven). Yeast transformations were performed as described () except that 6% DMSO was added prior to heat shock and the final pellet was resuspended in 0.15 M NaCl. The BP1 strain was transformed with the different sets of yeast plasmids and plated on minimal medium lacking uracil and containing 2% glucose. The sectoring assays were always performed as follow, with two transformants for each plasmid previously selected at least twice on YNB plates lacking uracil and containing 2% galactose. The transformants to be tested were grown overnight at 30°C in liquid minimal medium lacking uracil and containing 2% galactose until cultures reached 1 OD . They were then diluted in the same medium and monitored for growth until they reached 0.5 OD . Cells were then plated on sectoring medium at 200–500 cells per plate. After 4 days growth at 30°C, the percentage of red/white sectoring was determined. For further tests, red sectors were streaked on rich galactose medium as many times as necessary to give only solid red colonies. An independent red clone was then tested on appropriate plates for auxotrophy to tryptophan and uracil and for its ability to grow on fluoroorotic-acid-containing plates. The CY2 strain was transformed with the p2μGALPa plasmids and plated on minimal medium lacking uracil and containing 2% glucose. Two transformants for each plasmid were streaked at least twice on YNB plates lacking uracil and containing 2% galactose. Cultures were then conducted as for the sectoring assays and total RNAs were extracted as described (). To analyse tRNAs and U3 snoRNA, 5 μg of total RNAs were fractionated on polyacrylamide 8.3 M urea gels. RNAs were then blotted on nitrocellulose Hybond N+ membranes (Amersham Biosciences). Hybridizations were performed as described using previously published oligonucleotide probes (,). Total protein extracts were prepared as described () from 5 OD  cell cultures prepared for sectoring assays or northern blot analysis. Proteins were fractionated on 10% SDS-PAGE gels and blotted on nitrocellulose membrane with Trans-Blot semi-dry system (BioRad). The blots were reacted with rabbit anti-Nhp2p [generous gift from M. Caizergues-Ferrer (LBME, Toulouse)] () at 1/5000 dilution as primary antibody and donkey anti-rabbit IgGs horse radish peroxydase linked (Amersham Biosciences) at 1/10 000 dilution as secondary antibody. Antibodies were produced using the Eurogentec double X immunization programme followed by affinity-column purifications. Briefly for each protein, rabbits were inoculated with a mix of two synthetic peptides (pep1 and pep2) corresponding to specific regions of the protein. For both proteins, pep1 is located in the central region between RRM1 and RRM2 (At32pep1: HN-CQPQKGSANQKNGSDH-CONH, At79pep1: HN-CLGKSESHNEFRRGQI-CONH,) and pep2 corresponds to the very last 16 or 15 amino acids of the proteins (At32pep2: HN-CDSPGGRWNKSQKVEA-COOH, At79pep2: HN-CFENVQPTKKARKEP-COOH). Sera from rabbit's final bleeding were divided into two and each sample was affinity purified against pep1 or pep2, respectively. To assess the immunogenicity of each purified serum, we performed western blot analysis on total extracts prepared from yeasts expressing At32 or At79, wild-type cell suspension or 2-weeks-old seedlings. For both At32 and At79, only the sera fraction purified against pep2 gave a satisfying immunogenic response and was used in subsequent western blot analyses. The T-DNA insertion lines SAIL 548H11 () and GABI 870F12 () were obtained from the ABRC Stock Centre and the Max Planck Institute for Plant Breeding Research, respectively. To prepare transgenic cell suspension lines expressing the different GFP constructs, T87 cells were transformed with the different plant transformation plasmids using , as previously described (). For each construct, cell lines were screened by western blot analysis with anti-GFP monoclonal antibodies (Clontech) for expression of the transgenic protein. Cell lines were maintained under constant hygromycine selection (25 μg/ml) as liquid suspensions and as calluses growing on solid media, as previously described (). Here, 2 g of cell suspension expressing GFP fusions, obtained by filtration of 3–4-day-old liquid culture is resuspended in 6 ml of ice-cold lysis buffer (150 mM NaCl, 50 mM Tris-HCl pH 8, 2.5 mM MgCl, 0.1% Triton 100×) with 1 mM DTT, 10 mM vanadyl ribonucleoside complex (Biolabs), 1% protease inhibitor cocktail (P9599 Sigma), 2 mM benzamidine, 1 mM phenyl methyl sulphonyl fluoride and 10 μM decarboxylase inhibitor. The suspension is loaded in the pre-cooled chamber of a ‘one-shot’ cell disrupter system (Constant Systems Ltd), and cells are lysed under a pressure of 552 bar. The extract is subsequently cleared by centrifugation (30 min, 4200 , 4°C). For immunoprecipitation, crude extracts were incubated with a 1/300 dilution of anti-GFP full-length polyclonal antibody (Clontech) on a rotary shaker for 3 h at 4°C and then mixed with magnetic nanoparticles conjugated with Protein A (Bio-Adembeads Protein A, Ademtech S.A) and incubated for an additional hour. After three washes with lysis buffer, beads were divided as follows: one-ninth of the beads was resuspended in Laemmli buffer and heated 5 min at 95°C for protein analysis and the remaining beads were eluted with urea 8 M 5 mM EDTA buffer and extracted once with phenol:chloroform:isoamyl alcohol. RNAs were then precipitated by adding 40 μg of glycogen and 2.5 volumes of absolute ethanol and resuspended in RNase-free water. To compare the different transgenic lines, equivalent fractions of proteins or RNAs were used for western blot or RT-PCR analysis. Proteins were analysed by western blot with an anti-GFP monoclonal antibody (Clontech). Chemiluminescent signal was quantified using the VersadocImaging system (BioRad). RNAs were treated with the DNA-free kit from Ambion following the manufacturer's ‘Rigorous DNase treatment’ protocol. A fraction of the RNA (usually one-fifth) was reverse transcribed with expand reverse transcriptase and hexanucleotides as primer (Roche). cDNAs were then PCR amplified with primer pairs specific to pre-tRNA, pre-tsnoR43.1 or 5.8S RNA. Protoplasts were prepared from cell suspensions expressing the GFP fusions as described by Sheen,J. (2002, A transient expression assay using mesophyll protoplasts. ) using an overnight digestion in the dark. Images were obtained with a Zeiss Laser Scanning Microscope LSM 510. Wild-type and mutant seeds were collected from hemizygous siliques at different stages of maturity. The seeds were fixed in ethanol:acetic acid (3:1) for 20 min, followed by a slow rehydration in a series of ethanol–water solutions. After rehydration, the seeds were cleared in Hoyer's solution (2.5 g gum arabic, 100 g chloralhydrate, 5 ml glycerol in 30 ml of HO) and observed using Nomarski optics with a Zeiss Axioskop2 microscope (Carl Zeiss, Germany). We searched (by a protein BLAST at ) the - and rice ()-expressed genomes for proteins presenting similarities with the eukaryote consensus La-motif (SMART accession number SM00715, ). Our search revealed the presence of eight and nine rice putative proteins containing a La motif (). We used a phylogenetic approach to test the relationship of and rice La motifs with those present in several eukaryote La and La-like proteins (A). Our results show that two (At4g32720 and At1g79880) and two rice (bad19607 and cae03115) proteins form a well-supported cluster (bootstrap of 1000) and that the association of this cluster with the and La proteins is fairly well supported (bootstrap of 728, A). The NTD of La proteins contains a typical RRM (SMART accession number SM00715) (referred to as RRM1) that closely follows the La motif. In addition to At4g32720, At1g79880, bad19607 and cae03115, three and four rice proteins also present a RRM closely following the La motif (B), but these proteins group in a cluster intermediate between genuine La and La-like proteins (A). The use of the first RRM, instead of the La motif in the phylogenetic analysis also clearly distinguishes these proteins from La homologues (not shown). The CTD of the hLa protein possesses a novel type of RRM (called RRM2), the structure of which is atypical compared to canonical RRMs (). This motif is very difficult to detect based on primary sequence analysis alone. We hence searched for the presence of such atypical RRMs in the plant putative La homologues using the hidden-Markov-model (HMM)-based protein structure prediction program SAM-T02 (). This program was successful in predicting the correct topology for the human atypical RRM2 (see Supplementary online). We found that amongst the plant proteins displaying both the La motif and the canonical RRM, At4g32720 and At1g79880 from and bad19607 and cae03115 from rice are the only ones likely to possess an atypical RRM2 domain in their CTDs (B and Supplementary online). We also found that with the exception of the and homologues, each of the genuine La proteins we tested is likely to display such motif (B, Supplementary Data online). Therefore, the presence of an atypical RRM in the La protein CTDs is apparently not a characteristic restricted to vertebrate proteins, but is likely to be a conserved feature of genuine La proteins with the exception of the yeast homologues, which possess a much shorter CTD. Altogether these data suggest that At4g32720 and At1g79880 and rice bad19607 and cae03115 are the most likely La homologues of the two species. The two proteins, At4g32720 and At1g79880, have an overall amino acid identity of only 44%, with the highest identity found in the La motif (54%) and in the first RRM (59%) (see Supplementary Data online), suggesting that they are not produced from recently duplicated genes. In fact our phylogenetic studies show that At4g32720 is more closely related to the two rice proteins than it is to At1g79880. The presence of two distantly related proteins in with characteristics of La proteins is exceptional as the eukaryote La function has always been associated to a single La protein [with the exception of two closely related (>90% amino acids identity) La proteins ()]. Searches of the information resources (TAIR at ) and NCBI databases revealed that full-length cDNAs corresponding to both loci have been sequenced. For the At4g32720 locus, all characterized ESTs and full-length cDNAs code for a single protein reported as At4g32720.1. For the At1g79880 locus, three different isoforms are suggested by EST data but only two of those (called At1g79880.1 and At1g79880.2) are supported by full-length cDNAs. The At1g79880.2 protein differs from At1g79880.1 by the deletion of the first 44 amino acids, including more than half of the La motif. Since At1g79880.2 lacks a complete La motif and since the third isoform (At1g79880.3) is only supported by a single EST sequence (corresponding to a small portion of the protein CTD), we decided to conduct further structural and functional studies on the At1g79880.1 protein. From now on, the At4g32720.1 and At1g79880.1 proteins will be referred to as At32 and At79, respectively. To monitor the developmental expression profiles of At32 and At79, we searched the expression atlas of development microarray data (AtGenExpress) using the Genevestigator web site () (). We observed that mRNAs corresponding to both genes are present at all developmental stages in every plant tissue tested (see Supplementary online). At32 mRNA levels are highest in tissues composed of actively dividing cells (such as root tips, radicles, seedling, callus or cell suspensions), while At79 mRNAs is in general more abundant in tissues composed of mainly differentiated (non-dividing) cells. In average, At32 mRNAs are four times (4×) more abundant than that of At79 but this ratio is highly variable and is highest (10×) in ‘young’ tissues and lowest (2×) in 35-day-old senescent and cauline leaves (see Supplementary online). Western blot analysis using antibodies specific for At32 and At79 confirmed that both proteins are present in all developmental stages tested (see Supplementary online). To further characterize the roles of At32 and At79, we first conducted functional studies in yeast and tested the ability of these plant proteins to complement phenotypes linked to the full depletion of Lhp1p, the La homologue (). While Lhp1p is not required for growth in wild-type cells, it becomes essential in specific genetic backgrounds (). The Sm-like protein 8 (Lsm8p) is an essential member of the Lsm2p–Lsm8p ring-shaped complex (,). In the nucleus, Lsm8p associates with U6 snRNA and is important for its stability, for the formation of U6-containing snRNPs and for pre-mRNA splicing (). The allele is not lethal in normal growth conditions but becomes essential when combined with a deletion of the gene. In the background, Lhp1p is essential to stabilize newly synthesized U6 RNAs and to facilitate the U6 snRNP assembly (). The viability of the BP1 (-Δ, ) colethal strain is maintained by an extra-chromosomal copy of the gene carried on an plasmid (pATL). As a red pigment accumulates in mutant strains (), cells that retain pATL will form white colonies, while cells that grow without the plasmid will form solid red colonies (,). The ability of plant proteins to restore growth in the -Δ background can be monitored by the capacity of BP1 to form colonies with red sectors when transformed with a plasmid expressing the coding sequence of interest. DNA fragments corresponding to the , At32 and At79 ORFs were cloned in fusion at their NH2 terminus with the Protein A tag () and placed under the control of a galactose-inducible promoter on a low-copy vector () to create the pGALPa plasmids collection. BP1 was transformed with the different pGALPa plasmids and grown under galactose-inducing conditions. To compare protein production levels in the different transformants, proteins were prepared from an aliquot of each culture and western blot analysis performed using a rabbit anti-Nhp2p (a snoRNP H/ACA protein) antibody () that, as for most IgGs, also binds to the Protein A tag (). As expected, the anti-Nhp2p antibody recognizes in all extracts a single protein migrating above 20 kDa, corresponding to Nhp2p (A). Additional bands corresponding to the expression of the Protein A alone (A, lane 2) and to the different fusion proteins (A, lanes 3–5) were detected in extracts from the BP1 strain transformed with the different pGALPa plasmids. Using the Nhp2p signal as a loading control, we can observe that all fusion proteins accumulate to similar levels (A). The galactose-induced cultures of transformed and untransformed BP1 strains were plated on sectoring medium and the number of colonies with red sectors was counted and expressed as a percentage of the total number of colonies (B). As expected, the ProtA-Lhp1p positive control restores red/white sectoring in the BP1 strain giving 22.2% of colonies with red sectors while background, as observed for the untransformed and Protein A-expressing colethal strain, is between <0.2% and 0.6%. Among the proteins, only At32 gives red/white sectoring above background (2.9%). Using fluoroorotic-acid-containing medium (), we confirmed that the red colonies growth is strictly dependent on the presence of the pGALPa vectors expressing ProtA-Lhp1p or ProtA-At32. The inability of the At79-tagged protein to restore growth in the BP1 strain might be due to the presence of the tag or to the fact that a higher level of this protein is required. To test these possibilities, we placed the untagged plant and ORFs under the control of the promoter on the plasmid (pGAL vectors) and the galactose controlled protA-tagged ORFs on a high-copy μ plasmid () (p2μGALPa vectors). Each set of plasmids was transformed into the BP1 strain and tested by western for equivalent protein expression levels using anti-Nhp2p antibodies for p2μGALPa vectors and At32- and At79-specific antibodies for the pGAL vectors (not shown). As with the pGALPa plasmids collection, we tested the ability of the new BP1 transformants to produce red sectors. The untagged proteins gave results similar to the previous ones showing that the presence of the NH2 tag has no clear impact on the expressed proteins (B). As expected, the use of high-copy vectors leads to higher levels of red sectors for ProtA-Lhp1p and ProtA-At32 giving 45 and 8% red/white sectoting colonies respectively, whereas the At79 sectoring percentages remain at the background level (B). Altogether these data show that the At32 putative La homologue is able to complement the colethal -Δ phenotype and suggest that At79 is not able to do so. Inactivation of the gene in an otherwise wild-type background leads to several molecular phenotypes including a differential accumulation of U3A snoRNA (referred to as U3 from now on) and tRNA precursor species (,). To test the ability of the plant proteins to complement molecular phenotypes observed upon Lhp1p depletion, we expressed At32 and At79 from the p2μGALPa set of vectors in the CY2 () yeast strain which bears a fully inactivated allele of the gene (). The transformed as well as untransformed CY2 and the isogenic CY3 wild-type strains () were grown on galactose-inducing medium. We confirmed by western blot analysis the proper and equivalent expressions of the different fusion proteins in the CY2 background under these growth conditions (data not shown). Total RNAs were extracted from the different transformants and the accumulation of U3 and tRNAs precursors assessed. Three independent experiments were performed and gave similar results. Lhp1p has been shown to protect two U3-3′ extended forms from exonucleolytic trimming by binding to poly(U) stretches present in the 3′-end region of pre-U3 molecules (A). These extended forms, named U3-3′I (U3 + 12) and U3-3′II (U3 + 18) are readily detected by northern blot in wild-type conditions (A lane 2 and drawing). Following inactivation, the two extended forms are replaced by a heterogeneous population of molecules whose sizes range from +12 to +8 (). Consequently, the impact of inactivation on pre-U3 3′-maturation can be assessed by monitoring the level of the U3-3′II extended form. Our analysis shows a strong impact on the accumulation of the U3-3′II form upon inactivation of (A lane 1) but by contrast to previous report (), there is no complete depletion of the band. This might result from strain discrepancies since we are using a different genetic background () than the one used by Kufel (). In extracts from the CY2 strain transformed with the p2μGALPa-LHP1 plasmid (A lane 4), we observe a strong over-accumulation of U3-3′II (and likely of U3-3′I as well) as compared to the CY3 wild-type strain (A compare lanes 2 and 4 and see also Supplementary A online for quantification of the signals). Since the expression of the Protein A tag alone under the same strong promoter and from the same high-copy vector, has no influence on U3-3′II accumulation (A lane 3), we conclude that the over-expression of Lhp1p is likely to be responsible of this effect (see footnotes on Supplementary online). In the presence of the ProtA-At32 protein, the U3-3′II precursor accumulates ∼45% of the wild-type levels, which is around three times the residual levels observed in the untransformed CY2 strain (A lane 5 and Supplementary online). Expression of At79 gives only a 5% increase in U3′-II levels as compared to the untransformed or Protein A expressing CY2 strain (A lanes 1, 3 and 6 and see Supplementary online). tRNAs are transcribed with 5′- and 3′-extensions and many contain intervening sequences. Removal of the 3′-extension is catalysed by an endonucleolytic event occurring most of the time after 5′-end excision (,,). The La proteins protect the 3′-trailer from exonucleolytic trimming and, at least in yeasts, stimulate its endonucleolytic maturation (,). Although there is no clear defect in mature tRNA levels in the absence of Lhp1p, the mechanism of 3′-end tRNA maturation and the order by which it occurs are both altered. In -Δ background, the mature 3′-end is produced by exonucleolytic trimming, a process leading to characteristic modifications in the accumulation patterns of most tRNA precursors and intermediates [(,) and compare also lanes 1 and 2 B]. We monitored these patterns for five intron-containing tRNA families upon expression of the plant proteins in the CY2 (-Δ) strain (B and Supplementary online). The identities of tRNA precursors were determined based on their relative electrophoretic mobility, hybridization patterns and comparison to previous reports (,,). The maturation defects we observed in the -Δ strain for tRNAs serine and proline (B lane 1) are as previously reported (). In these cases, upon Lhp1p depletion, the unspliced 5′- and 3′-extensions containing precursors migrate faster and under accumulate, and the processing intermediates corresponding to unspliced 5′-processed, 3′-unprocessed pre-tRNAs are undetectable (B compare lanes 1 and 2). As for pre-U3 3′-end processing, expression of the Protein A tag alone in CY2 has no impact on the tested pre-tRNA patterns (B lane 3 and Supplementary Data online) showing that the effects described below most probably arise from the factor fused to the tag. As expected, expression of the ProtA-Lhp1p in the -Δ background qualitatively restores a wild-type processing pattern for all tested tRNA families (B lane 4 and Supplementary Data online). Nevertheless, we can observe a striking over-accumulation of the 3′-extended intermediate of tRNA (and of other tested tRNAs, see Supplementary online), a more limited one for tRNA, and a slight increase in all unspliced 5′-, 3′-extended transcripts. As for U3, we speculate that all these molecular phenotypes are the consequence of Lhp1p over-expression (see footnotes of Supplementary online). In the presence of the ProtA-At32 fusion, we clearly observe the restoration of a normal pattern of precursors and intermediates for all tested tRNA families (B lane 5 and Supplementary Figure 3B). We also observe, as in the ProtA-Lhp1p-expressing strain but to a more limited extent, the over-accumulation of the 5′-, 3′-extended intron-containing transcripts (B compare lanes 2 and 5). Pre-tRNA patterns from the ProtA-At79 expressing CY2 strain are similar to that of the untransformed and Protein A expressing CY2 strains (B lane 6 and Supplementary Figure 3B) even after longer exposures (not shown). La homologues can associate with highly diverse coding or non-coding RNAs from viral or cellular origins [for review see ()]. This multi-functionality is in part explained by the capacity of La proteins to specifically bind with high-affinity RNAs presenting a terminal 3′-UUU-OH motif (,,). All primary RNA pol III transcripts end with such a motif but mature forms do not, as it is subsequently removed by a 3′-end processing step. To determine whether At32 is able to bind to 3′-UUU-OH ending RNAs in plant, we asked whether RNA pol III precursors could be co-immunoprecipitated with green fluorescent protein (GFP) tagged versions of the protein. However, since most plant pol III precursors have a very short 3′-extension and are difficult to distinguish (by hybridization or PCR amplification) from mature forms, we had to select rare situations where this distinction would be possible. One such situation is provided by the presence of an intron in a subgroup of pre-tRNA methionine (pre-tRNA) (tRNA database: ) and another by the presence in plant of a dicistronic gene organization, where a tRNA is co-transcribed with a snoRNA in a single precursor molecule (pre-tsnoR43.1) by the RNA pol III complex (). Stable transgenic cell suspension lines expressing At32 fused at its NH2 or COOH terminus to the GFP, as well as a cell line expressing the GFP alone as negative control were produced. We prepared native extracts of each of these cell suspension lines and performed immunoprecipitations with anti-GFP antibodies. Two independent experiments that yielded similar results were conducted for each fusion and no significant difference was observed between the NH2 and COOH translational fusions. Western blot analysis performed on identical fractions of each eluate showed that both GFP and At32-GFP are efficiently immunoprecipitated (A). We checked for the presence of pre-tRNA and pre-tsnoR43.1 in each eluate fraction by RT-PCR analysis (B). As negative control, to test the specificity of our immunoprecipitation experiments, we also performed RT-PCR amplifications targeting 5.8S ribosomal RNA since La proteins are not known to associate to this highly abundant transcript. PCR reactions performed on total genomic DNA demonstrate that primers are efficient and give specific signals (B lane 2) and since we were not able to amplify from non-reverse transcribed RNAs (B lanes 3, 5, 7 and 9), we conclude that the signals obtained with reverse-transcribed samples arise from cDNAs and not from contaminating genomic DNA. Also, in each case, we were able to amplify a specific product using input cDNAs (B lanes 4 and 8) showing that each of the three tested RNAs were present in corresponding crude lysates. PCR reactions using cDNAs produced from the At32-GFP immunoprecipitated fraction generated the expected product for pre-tRNA and pre-tsnoR43.1, but not for 5.8S rRNA (B, lane 10), while PCR reactions using cDNAs produced from the GFP-alone fraction did not produce any PCR product (B lane 6). We conclude that pol III transcribed precursors to tRNA and tRNA-snoR43.1 specifically co-immunoprecipitate with At32-GFP. Despite the fact that the GFP protein is efficiently immunoprecipitated and is more abundant than At32 in the eluate (A), no such RNA is co-immunoprecipitated with GFP alone strongly supporting the conclusion that At32 is able to specifically associate at least to these RNA pol III transcripts and probably to 3′-UUU-OH ending RNAs in general. To test the capacity of At79 to bind pol III precursor RNAs , we also produced stable transgenic cell suspension lines expressing At79 fused at its NH2 terminus to the GFP protein. Using the same immunoprecipitation procedure as above, we were unable to co-immunoprecipitate pre-tRNA or pre-tsnoR43.1 with the At79 fusion protein whereas the protein is efficiently recovered in the eluate fraction (not shown). These negative results were reproduced when performing the PCR reactions on twice the amounts of cDNAs from GFP-At79 eluates. protein . Although displaying characteristic evolutionary and structural features of genuine La proteins, At79 failed to complement yeast phenotypes linked to inactivation and at least in our experimental conditions, does not seem to bind to the tested RNA pol III transcripts in plants. To get a better understanding of the La function in plant, we further studied AtLa1 in , starting with its subcellular localization. Cell lines expressing GFP (as a control) or AtLa1 fused to the GFP, were prepared as protoplasts and fluorescence was observed by confocal laser scanning microscopy (). As expected, the GFP alone is distributed throughout cytoplasm and nucleoplasm with exclusion of the nucleolus (A–C). In the vast majority of observed cells, GFP-AtLa1 (NH2 fusion) or AtLa1-GFP (COOH fusion) (D–F and not shown) display a diffuse pattern throughout the nucleoplasm and no obvious general labelling of the nucleolus. The plant nucleoli have the particularity as compared to other eukaryotes, to display a central entity named nucleolar cavity that appears as a clear space when observed by electron microscopy (,). The precise function of the nucleolar cavity is not well defined but amongst other factors small nuclear and nucleolar RNAs were shown to localize in this space (). We observed the accumulation of the AtLa1 NH2 and COOH GFP fusion proteins in the nucleolar cavity for a small fraction of cells (G–I and not shown), whereas we never observed the GFP in this nucleolar subcompartment. This observed subcellular localization is unlikely to be an artefact linked to the production of protoplasts from cells, since the direct observation of a transgenic BY2 tobacco cell line stably expressing GFP-tagged AtLa1 gave the same result (not shown). To further study AtLa1 functions, we searched several collections of T-DNA insertion mutants for disruption of the At4g32720 locus. We identified two candidates, one from the SAIL (n 548H11) () and one from the GABI-KAT (n 870F12) () collections. PCR analysis and sequencing allowed us to map each T-DNA insertion sites. Both Sail (allele ) and Gabi (allele ) T-DNAs are inserted at the beginning of the At4g32720 exon VIII, three bases apart (A). Sail and Gabi T-DNAs confer plant resistance to glufosinate (Glufo) and sulfadiazine (Sul) respectively. Plants bearing and alleles can hence be selected for growth on medium containing one of these drugs. After selection for several generations, no homozygous plants were recovered for neither allele, suggesting that the homozygous gene At4g32720 inactivation is lethal. The self-progenies of / and / always yielded a segregation for drug-sensitive to drug-resistant plants of 1:2 (1135 Glufo:601 Glufo, χ = 1.29, > 0.05 and 885 Sul : 410 Sul, χ = 1.63, > 0.05, respectively). These results suggest that the T-DNAs are inserted at a single locus in each case and that the two mutations are nuclear, recessive and sporophytic. Observation of siliques from hemizygous mutant plants under the microscope showed that they contained one-quarter of aborted seeds confirming the lethality of homozygous mutant embryos (B and ). Initially aborted seeds appeared smaller and pale brown at a stage where wild-type and hemizygous siblings were green (B). Later they were dark brown and completely dried out. Allelism tests were performed using the two independent T-DNA insertion lines. Immature siliques produced by crosses were opened under the binocular and analysed. The presence of 26.5% of seeds (156/589, χ = 0.69, > 0.05) confirmed that the two mutations are allelic and that embryonic lethality directly results from the alteration of AtLa1 function. We examined the embryonic phenotype of seeds for the two mutant lines ( and ). Compared to wild-type embryos at the torpedo/cotyledonary stage, the embryos from both lines were much smaller and composed of <50 cells (A panels b, c, e and f). Based on their morphology, we conclude that the terminal stage of mutant embryo development is early globular (A panels c and f). Comparison of hemizygous siliques from the two mutant lines, taken at the torpedo/cotyledonary stage for wild-type seeds, revealed that the percentage of brown desiccated seeds is almost twice that of seeds (). Moreover embryos proper of were often marginally larger and composed of more cells than embryos (A compare panels a–c and d–f). These data suggest that although the development of the embryos in both alleles is arrested at the same stage (), mutation is more severe than . We also observed that most if not all embryo cells display abnormally large nucleoli (B). We measured the diameter of nucleoli from and embryos arrested at early globular stage (∼20 embryos were analysed for each allele with 4–6 nucleoli measured per embryo) and observed that they are 1.8–2 times larger than nucleoli from wild-type embryos at globular stage. We report here the identification of two distantly related proteins from At32(/AtLa1) and At79 that display every structural feature of La proteins. Microarray expression data from the AtGenexpress project and our western blot analysis clearly indicate that both corresponding loci (At4g32720 and At1g79880) are expressed . protein ) is a genuine La protein. The fact that AtLa1 can restore correct U3 and tRNA precursor patterns in -Δ background and growth of the -Δ strain is a strong indication that it can bind to all RNAs that terminate with an oligouridylate motif (not only to pol III transcripts) and participate in their biogenesis. In yeast, U3 snoRNA is produced by the RNA pol II complex and binding of Lhp1p to precursor molecules is a key step of its maturation process (). AtLa1 is able to restore correct U3 precursors pattern in an -Δ background showing that it can bind these pol II encoded RNAs and participate in their processing in yeast. This suggests that AtLa1 has the potentiality to bind to pol II-encoded 3′-UUU-OH RNAs such as snoRNAs () or snRNAs () and to participate in their processing, stability and/or RNP formation. AtLa1 is also able to restore normal pre-tRNA patterns in an -Δ background and even to stabilize intermediates above wild-type levels. This high level of complementation by a plant protein of the yeast La function in 3′-end pre-tRNAs processing is consistent with previous reports showing that this maturation is evolutionary well conserved in eukaryotes (,). This together with the fact that AtLa1 is able to bind to plant tRNA precursors allow us to propose that the mechanism of 3′-processing and protection of pre-tRNAs involves the La function and is conserved in . Finally, the capacity of AtLa1 to restore growth in the -Δ strain is a strong indication that it can participate in U6 snRNA biogenesis. In yeast, both the ring-shaped complex Lsm2p-Lsm8p and the Lhp1p proteins are involved in the biogenesis of several non-coding RNAs such as U3 snoRNA, tRNAs and U6 snRNA (,,). However, while or inactivation has no impact on the steady-state accumulation of mature U3 or tRNAs (,), U6 snRNA levels are reduced by 50% in the background (,). A 2–3-fold over-expression of Lhp1p in this background restores U6 snRNA levels to 75% of wild-type and expression of U6 snRNA in extra copies allows viability of the -Δ strain (,). These data suggest that Lhp1p is needed in the background to help in U6 snRNP biosynthesis and/or accumulation and the ability of AtLa1 to restore growth of the -Δ strain is a strong indication that it can at least partially replace Lhp1p in this role. In summary, our data strongly suggest that in higher plants as in other eukaryotes, oligouridylate ending transcripts, whether encoded by RNA pol II or RNA pol III, require the La function for normal biogenesis and that this function is fulfilled by AtLa1. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Comparative phylogenetics is a powerful tool, not just for establishing evolutionary relationships among taxa, but also for discerning biochemically significant aspects of a locus, versus those always present but variable. The 3′ nuclear ribosomal transcribed spacer region [second internal transcribed spacer (ITS2)] sequence is much used for phylogenetic studies at the species and genus level. More recently, additional phylogenetically useful information from these sequences has emerged as a consequence of the solution of their putative transcript secondary structure (). Initial analysis of potential folding homologies was presented by Wolf . (). From folding of transcripts of many additional phyla, it now appears that conservation of secondary structure of ITS2 itself among essentially all eukaryotes is far more remarkable than had ever been anticipated. Although the highly conserved helix regions revealed are presumably essential for rRNA processing, the details of this complex activity are not yet fully understood. Here we point out homologies as they apply across animals, plants and protistan phyla to call attention to the breadth of information already available to both phylogenetic studies and to detailed analyses of ITS processing. Phylogenetic studies have relied on a variety of DNA loci. Protistans present probably the most difficult choice of locus to sequence, for the broad and ancient variety of cell types encompassed is greater than in fungi, animals or plants. Among protistans, plastid L has been sequenced for many algae and mitochondrial cox I for many non-photosynthetic protistans. However, some protistans are photosynthetic and some not, even in the same class; and there are numerous examples of host cells with plastids transferred from a different eukaryote. As for mitochondria, some protistans are even lacking standard mitochondrial genomes. These facts make organelle genes less appealing for broader comparisons. A single common nuclear locus would seem most useful, as long as it is appropriate for the task. What should be its attributes? Such a locus should be present in all of the chosen taxa, and there should be no known case of horizontal transfer; and it should identify the organism to a unique species. One candidate is the single most frequently sequenced DNA region, with variability suitable to the species level, the ITS2 in the nuclear ribosomal gene cistron (). Here we review the ITS2 region across eukaryotes and find it is not just specific to species but also laden with surprisingly useful information concerning higher relationships, and clearly constrained in its evolution to maintain certain regions of transcript secondary structure universal among eukaryotes. Aspects of these conserved regions should be useful and very important to future studies of rRNA processing. xref #text From ITS sequencing and analysis of RNA transcript secondary structure, a set of near-universal eukaryote ITS2 structures has emerged, as highlighted in for a diatom, a green alga, a red alga and yeast. The features common to eukaryotes were first noted during analyses of the many genera and species of the green algal group Volvocaceae, and their comparison with terrestrial plants (). ITS2 typically has four helices (), and helices I and IV are the most variable (species and subspecies level specificity). At least within a genus, the basal pairings of helices I, II and IV are conserved while the most conserved portion of helix III is distal. The two helices that we term II and III, and their adjacent single-stranded regions, contain the most conserved regions of primary sequence (, cartoon). However, not all eukaryotes have the same number of helices, and only helices II and III are recognizable and common to essentially all. Hence, helix II (recognized by its hallmarks) may be the first helix in, for example, the typical ciliate ITS2 structure (). In some groups, insects particularly, there is commonly a helix between the recognizable Helix II and Helix III, which we have termed IIA. Like all the other helices except II, it may be branched. For global comparisons, it is less confusing to ignore the total number of helices and concentrate rather on the two helices that have individually recognizable motifs. Helix II is recognizable because it is short (rarely more than 12 pairings), it lacks any branching, and it posesses a pyrimidine–pyrimidine mismatch (arrows in ), which makes a bulge near the base of the helix. of () illustrates the many variant forms of helix II among angiosperms, all with the pyrimidine mismatch. The sole exception to the rule of shortness and no branching is spectacular. In the genera of ticks studied by Hlinka . () there is an insertion of hundreds of nucleotides of long repeats at the tip of helix II, leading to complicated branching. These are the longest known ITS2 sequences. Helix III, by contrast, is typically longer than helix II and is frequently branched. The region of greatest absolute sequence conservation in the whole ITS2 is on the 5′ side of helix III near the tip. Even when helix III is branched, the highly conserved stretch of nucleotides is found on the 5′ side of one of the branches. This sequence, marked in , is typically conserved absolutely at the family or even higher level. Coelomates and some other taxa (e.g. 8) have a branched, sometimes much branched, helix III. Historically this characteristic has affected the research field, because mammalian and particularly human, helix III of ITS2 is long and multiply branched. Hence not only is the common secondary structure difficult to perceive, but the great number of relatively poorly conserved nucleotides in ITS2 initially gave the locus a reputation for high variability and perhaps uselessness for phylogenetic purposes. Our secondary structure analyses and comparisons show that, no matter what the total length of ITS2, ∼100–115 nt positions, consistent in their position in the secondary structure, are relatively conserved. These include the basal 10 pairings of helix II and, in helix III, those 18 pairings that include and surround the single most absolutely conserved sequence, that on the 5′ side of helix III (see cartoon). The known or proposed cut sites for transcript processing in yeast and mammals () are all in the relatively highly conserved regions in the 5′ half of ITS2 (approximately up to the tip of helix III). The significance of actual sequence nucleotides has as yet been explored experimentally only for the pyrimidine–pyrimidine bulge region of helix II in yeast (). The hallmarks of ITS2 secondary structure can now be recognized in the major groupings of algae and various other protistans as well as animals, fungi and plants (see ). We have derived the major transcript secondary characteristics of many clades, including in column 7 the most highly conserved sequence on the 5′ side of helix III, using sequence comparisons. The sequence given in column 7 is common to all of the taxa cited in column 7. Significant omissions from , either for lack of sufficient sequences in GenBank or from lack of attention to transcript secondary structure, include various lesser known animal phyla, plus reptiles and birds. Among protistans, the trypanosomes and euglenoids, some Heterokont groups (Dictyochophytes, Pinguiophytes) and various poorly represented (and some perhaps yet unknown) protistan groups of flagellates and amoeboid types (including the photosynthetic Chlorarachniophytes) are also missing. There is no reason to expect any exceptional structure of their ITS2; for trypanosomes, the locus has already proven useful in delineating species (). In all phyla, the relatively highly conserved hallmark sequences on the 5′ side of helix III in display a certain sameness, a high purine content, particularly guanine, presumably of functional importance to processing. In the majority of examples in , a YGGY can be found here. Comparison of transcript foldings at higher taxonomic levels fails to support any clade-constant positioning of pairing versus bulges in this most highly conserved sequence, implying some other aspect as the clue to processing. Finally, the Table serves not only as a guide to expectation when working on a group of taxa, but also as a source of citations concerning detailed RNA folding structure of the phylum, useful for resolving secondary structure in further taxa. These citations also serve as an entree into the phylogenetic analysis literature. The full significance of the transcript secondary structure parallels will not become apparent until the biochemical details of processing are deciphered. There are two general types of exception to the rule that ITS2 has a recognizable helix II and III. The first type of exception is that helix II is present, but a recognizable helix III seems not to be. Only three disparate groups exemplify this condition, the coral genus and its immediate relatives (as opposed to the remaining stony corals), at least some of the marine siphonaceous green algae, and four genera ( and ) of the Valkampfiidae of the Heterolobosea. Whether their total ITS2 region is long or quite short, nothing recognizably comparable to a standard helix III is found. These organisms all presumably undergo normal processing of their initial RNA transcript to produce SSU, 5.8S and LSU RNAs. The second type of exception is found among a few genera of parasites, the Diplomonadida and the genera of Microsporidia, and . These organisms lack a free 5.8S RNA molecule, instead incorporating its homolog within the 5′ end of the LSU gene, as in prokaryotes. The resulting transcript needs no processing of the type requiring guidance by ITS2 structure. In all of these exceptions, the region of the 5.8S and what lies between it and the standard LSU, though short, is still adequate for identifying to species (e.g. 15,16). The term ITS traditionally refers to the entire region between the nuclear genes for the ribosomal small subunit (SSU) and the ribosomal large subunit (LSU) RNAs; ITS1 is the first spacer, followed by the 5.8S RNA gene, and then the ITS2. Both the ITS1 and ITS2 regions of the long RNA transcript are removed, by ‘processing’ enzymes in the nucleolus, which produce the final SSU, 5.8S and LSU RNAs for ribosomes. The ‘processing’ is not yet fully understood in biochemical terms, but it is clear that the folding pattern, the secondary structure, of the initial RNA transcript plays a role in guiding processing (). Hence, not only the sequence but also the transcript folding patterns of ITS1 and of ITS2 have been objects of study. The second spacer, ITS2, has received vastly more attention, for plants, animals and protistans, than the ITS1. More than 100 000 ITS2 sequences can be found in GenBank, and numerous studies have used this region for phylogenetic analyses. When both ITS2 and another locus, plastid or mitochondrial, have been studied in parallel, the ITS2 has been found to contain at least as much information, and usually more (e.g. 17). Its sequence is unique to a species, and usually to the subspecies level. Basically, the reason for this is that ITS2 combines both remarkably conserved with relatively labile stretches of nucleotides, as shown in the alignment of Hershkovitz and Lewis (). The explanation for this apparent alternation of conserved with variable regions was presented by Mai and Coleman (). The regions of greatest sequence conservation contribute heavily to the pairings in the helices of the secondary structure that the initial RNA transcript takes on, as shown in the cartoon in . This secondary structure, in turn, guides processing. An initial concern with ITS was the fact that there are typically several hundred copies of the ribosomal RNA locus in tandem in the nuclear genome ()—hence there was the potential for intragenomic variation. It is now clear that, except as described below, these repeats are subject to ‘concerted evolution’ (), the result of a poorly understood process of homogenization that renders the ITS repeats of an organism identical over very short evolutionary time. Intragenomic variation, if present at all, is typically only in very few extremely variable positions that are never paired in secondary structure. Thus, the ITS2 can be treated as a single gene (). For an exhaustive analysis of ITS2 intragenomic variation, see Pröschold . (). The other potential problems, discussed at length in Alvarez and Wendel () and Bailey (), concern hybridization and polyploidy. Clearly if two organisms differing in their ITS2 sequences hybridize and crossover occurs, the outcome can confuse the phylogenetic analysis. There may ultimately be more than one locus of rDNA genes in the nucleus, there may be mixed ITS2 sequences within and between arrays, resulting from differing parents and from crossingover, and there may even be pseudogene sequences of ITS resulting from degeneration of one set. Non-functional pseudogenes, in fact, are readily recognizable by their imperfect 5.8S and for absence of some or all of the relatively conserved regions of ITS2, aspects obvious from transcript secondary structure knowledge. Hybridization, polyploidy and their consequences, more common in plants than in animals, can engender confusion, but such situations are generally recognizable and already suspected in particular groups under study. The resulting ITS sequences, containing two or more repeat types, may then actually prove highly informative in resolving the phylogenetic problems. For the four-helix model of ITS2, using the hallmarks of the helix II pyrimidine mismatch and the longer helix III with TGGT on the 5′ side, Schultz . () have automated GenBank searching for ITS2 sequences and their folding. Subsequently, Wolf . () and Schultz . () have extended automation and set up a website (http://its2.bioapps.biozentrum.uni-wuerzburg.de) with eukaryote-wide representation of exemplar ITS2 folds. In addition, there is a growing literature on programs to handle simultaneously sequence alignments and their secondary structure characteristics, as in Siebert and Backofen (), Seibel . () and Wolf . (). Biffen . () provide a recent example of the application of such programs, plus an interesting analysis of the types and rates of compensatory base change in SSU versus ITS. These methods appear to work best for relatively short ITS2 sequences, averaging ∼200 nt, and thus have proven particularly applicable to plants and green algae, fungi, dinoflagellates and some metazoan groups. A unique advantage of the ITS2 as a choice for sequencing is that the resulting alignment contains information related to the level of the biological species (). This is an empirical observation that has borne up for all eukaryote groups investigated so far. The fundamental correlation arises from the highly conserved regions of sequence. In taxonomic groups with fairly short ITS2 sequences, where these are all identical, the organisms are observed to be able to intercross experimentally, and if no compensatory base change is present, at least to some degree. For groups with longer sequences, the additional regions appear to show lesser evolutionary constraint, so that one must limit the comparison to only the most conserved paired positions (the 10 basal pairings in II and the 18 pairings including and immediately surrounding the highly conserved 5′ sequence of helix III); these should be identical, or lack any compensatory base change, for any interbreeding to be possible. Such an analysis requires a plethora of data, not only ITS2 sequences but experimental interbreeding data; yet examples have been found among protists (,,), plants () and animals (,) to test the hypothesis. An additional data set arises from the clades of angiosperms endemic to either the Hawaiian Islands or to Macaronesia, where only a limited evolutionary time has been available to produce the endemic genera and species groups now present. Breeding studies have suggested that none of these groups has managed to evolve to the point of sexual isolation (). Comparisons of the ITS2 sequences, in the nine genus and species swarms where ITS2 is available, now agree that all have the expected ITS2 nucleotide identity (Coleman, in preparation). In sum, ITS2 is present in essentially all eukaryotes, and even when truncated, the region is still sufficient to allow identification to species and lower. There is no evidence of any horizontal gene transfer (). ITS2 PCR and sequencing are straight forward, and there is rarely any excessive length. A recognizable short pyrimidine bulge-containing helix (‘helix II’) and downstream, a longer helix with highly conserved nucleotide motif on the 5′ side (‘helix III’) are essentially universally present. For the full biochemical understanding of how ribosomal RNA processing occurs, the many ITS alignments available should prove invaluable to test models, and those few eukaryotes groups (e.g. corals) lacking helix III and hence its detailed guidance role in processing pose a further challenge since they obviously produce ribosomes. Furthermore, thanks to its conserved secondary structure aspects, one has a guide, from sequence and structure alone, to the group of taxa probably capable of interbreeding. This correlation allows interesting comparisons of breeding potential with the idiosyncracies of taxonomic practice at the species level across eukaryote groups. The ITS2, once considered a highly variable and largely uninteresting locus, has proven in fact to be one containing eukaryote-wide homology, undoubtedly a consequence of its guidance role in what must be a eukaryote-wide biochemistry of ribosome formation.

The meta-server technique represents one of the major progresses in the field of protein tertiary structure prediction during recent years (). It generates 3D structure predictions by taking the consensus models from a variety of individual (mainly threading/fold-recognition) servers. Various benchmarking and blind test experiments demonstrate that the consensus meta-server predictions outperform the best individual threading server (,). There are, however, several drawbacks in the current meta-servers. First, all the meta-servers, including 3D-Jury () and GeneSilico (), take the initial threading inputs from remote computer servers installed in other laboratories. Because of the differences in the available computer resources among different laboratories, it is difficult to collect the threading results from the individual servers, which influences its usefulness in the large-scale protein structure prediction (,). Especially, some remote individual servers can be occasionally shut down or become not available. In the 3D-Jury meta-server, for example, there was only one server from FFAS03 () that was available during the CASP7 season. The absence of sufficient initial threading inputs will influence the performance of the final meta-server results. The second drawback of the current meta-servers is the instability of the algorithms of the remote servers. To achieve the best performance, the meta-servers need to balance various cutoff parameters for the selection and combination of the final models. This requires careful tuning and training of the meta-server algorithms based on all the individual servers. However, the inconsistent updating and modifications of the remote individual servers make the development of a steady and robust meta-server algorithm difficult. In this work, we developed a new meta-threading-server, LOMETS, where all nine individual threading servers are installed locally. This will allow us to control and tune our meta-server algorithms in a consistent manner, and make the users able to obtain the comprehensive predictions of all servers quickly. In addition to the construction of the best possible 3D models, the LOMETS server also provides the C and side-chain contact and distance map predictions, combined from all threading alignments. These constraints can be used to guide the structure construction procedures such as MODELLER (), ROSETTA () and TASSER () for generating protein tertiary models. LOMETS server takes predictions from nine different servers that represent a diverse set of state-of-the-art threading algorithms, i.e. FUGUE (), HHSEARCH (), PROSPECT2 (), SAM-T02 (), SPARKS2 (), SP3 (), PAINT, PPA-I and PPA-II. The first six programs were copied from other laboratories and the last three developed in our own lab. All the nine servers are installed and run in our local computer cluster with template libraries updated every week. The algorithms were selected to cover different threading methods. Here, we give a brief introduction of the methods. where (, ) is the frequency of the th amino acid at the th position of the query sequence when a PSI-BLAST search of the query sequence runs against a non-redundant sequence database () with an -value cutoff of 0.001; (, ) is the log-odds profile of template sequence in the PSI-BLAST search; () is the secondary structure prediction from PSIPRED () for the th residue of the query sequence and () the secondary structure assignment by DSSP () for the th residue of the template; δ((),()) equals to 1 if () = () and 0 otherwise. The weight factor is an adjustable parameter for balancing the profile term and the secondary structure matches; the shift constant is introduced to avoid the alignment of unrelated regions in the local alignment (). The Needleman–Wunsch () dynamic programming algorithm is used to find the best match between query and template sequences. A position-dependent gap penalty in the dynamic programming is employed: no gap is allowed inside the secondary structure regions; gap opening () and gap extension () penalties apply to other regions; ending gap-penalty is neglected. The four parameters [i.e. , , in Equation (), and , of gap penalties in dynamic programming] are decided by trial and error on the ProSup benchmark (). Models in LOMETS are selected from individual servers purely based on consensus, i.e. the structure similarity of the considered model with other threading alignments. For the best performance, 30 models are taken from the top predictions of the nine servers sequentially from PPA-I, SP3, PPA-II, SPARKS, PROSPECT, FUGUE, HHSEARCH, PAINT and SAM-T02, where the order of the servers are based on their performance on independent test runs. The 30 models are taken as following: First, select the first model of PPA-I and then the first model from SP3. This procedure proceeds until all the first models from nine servers are collected. Then, all the second models from nine servers are collected in the same order. The collection process proceeds and stops until 30 models have been reached. During the collection, the templates of very short alignments, i.e. the number of aligned residues is less than a quarter of the query sequence length, are neglected. The consensus score of each (th) of the 30 models is calculated by the average TM-score (): We note that, when running the TM-score program with model and model, the TM-score is by default normalized by the length () of the second model (i.e. model). But in Equation () TM-score should be uniformly normalized by the query sequence length (). To do this, one can first run the TM-score program with an option of ‘−’ with to obtain TM-score (). The normalized TM-score can be then obtained by TM - score() / . Here, purpose of the option ‘−’ in the TM-score program is to assign the new-defined length scale of to the Levitt–Gerstein score (). Finally, the models are ranked based on 〈 TM - score 〉 , i.e. the models with higher average TM-score to other models are ranked higher. For each protein, threading models are categorized as ‘good’ or ‘bad’ depending on whether the inherent Z-score (the energy in standard deviation units relative to mean) of the alignment is above or below a threshold Z-score. The threshold cutoff is determined by the minimization of the false positive (high Z-score but with low TM-score) and false negative rate (low Z-score but with high TM-score) of each threading program based on an independent benchmark set of 1489 non-redundant proteins (). For PPA-I, SP3, PPA-II, SPARKS2, PROSPECT2, FUGUE, HHSEARCH, PAINT and SAM-T02, the Z-score are 8.2, 8.0, 7.0, 8.8, 4.0, 6.0, 11.0, 0.5 and 9.5, respectively. If the total number of ‘good’ models is more than nine (i.e. on average at least one ‘good’ model from each server), the target is defined as an ‘Easy’ target; if there is no ‘good’ model at all in all the servers, the target is a ‘Hard’ target; otherwise, it is a ‘Medium’ target. For Easy/Medium/Hard targets, (=20/30/50) highest confident models are selected from the servers for the next constraint construction. The ‘good’ models and then the ‘bad’ models are taken in a sequential server order as mentioned above until models are selected. The logic for the decision of is the following: for ‘Easy’ targets where we have good templates, about top two (good) templates on average are taken from each program while including more templates with bad quality will bring more noises for the good templates. For the ‘Medium’ and ‘Hard’ targets where we do not have good templates and constraints overall, we will take more templates to enhance the consensus information because there are usually some partially correct substructures even in the low rank templates which may be identified by the consensus selections. There are four types of spatial constraints that are collected from the selected threading alignments: Here (, ) was obtained by calculating the average distance of side-chain centers of mass of the contacted residues and with at least one pair of heavy atoms in and < 4.5 Å in 6379 non-homologous PDB structures. Δ(, ) is the SD of (, ). The data of (, ) and Δ(, ) can be seen at our website . In the side-chain contact file of LOMETS server, we list the identities of all the contacts with contact order ⩾5, as well as the confidence score that is defined as the number of occurrences of the contacts divided by the total number of templates that have both residues aligned. ext-link #text We have developed a quick and automated meta-server, LOMETS, for protein structure predictions. Different from other on-line meta-servers, all nine component-threading servers are installed and run in our local computer cluster. The local installation of the servers greatly speeds up the coherent generation of initial threading alignments, as well as facilitates the development of a robust and well-tuned meta-server algorithm. The consensus prediction taken from LOMETS servers is at least 7% more accurate than all the individual servers. The difference is also statistically meaningful with a -test at 0.1% of significance level. The average CPU time for a medium size protein (∼200 residues) is less than 20 min when the programs are run in parallel on nine nodes of our cluster. In addition to the threading alignments, LOMETS also provides highly accurate contact and distance predictions for the query sequences. In our benchmark testing of 620 proteins, the average accuracy of side-chain center contacts is 0.42 with coverage of 91%; the average accuracy of C contacts is 0.61 with coverage of 41%. The average errors of the best long- and short-range distance map prediction are 3.5 and 1.2 Å, respectively. These data can be easily used as constraints to guide the tertiary structure modeling procedures such as MODELLER (), ROBETTA (), TASSER (,). Last but not the least, the template libraries of all nine servers are kept updated every week. We have managed to generate template files in our local computers for SAM-T02, PROSPECT2, SPARKS2, SP3, PPA-I, PPA-II and PAINT. The template library for FUGUE and HHSEARCH are automatically downloaded from the authors’ websites (i.e. and ), which are also kept updated each week. LOMETS will be open to add new and efficient threading programs when they become available. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

DNA is a flexible and polymorphic structure, whose conformation changes as a response to the presence of ligands, variations in the temperature or modifications in the solvent (). Such flexibility is implicitly coded in the structure of the DNA () and is crucial to its biological functionality, since, for example, binding of DNA-regulatory proteins can require dramatic changes in the structure of nucleic acid. Understanding DNA's flexibility and conformational transition is thus a necessary step in transforming structural data into biologically important information. In general, the determination of DNA structure is no longer the challenge for experimental techniques had it used to be, both NMR and X-ray techniques are now able to provide accurate structures for most sequences (more than 2500 (experimentally-solved) DNA structures are deposited in the Protein Data Bank in June 2006). Unfortunately, these experimental techniques are less effective at discovering flexibility or tracing conformational transitions, and so for this we currently make major use of simulation techniques like molecular dynamics (MD) to study these transitions at the atomic level. However, many conformational transitions in DNA occur on the micro or millisecond time scale, while timescales currently accessible by atomistic MD are in the 5–50 ns range. To overcome this problem, it is common to use biasing techniques to move simulations along a reaction coordinate at an ‘artificially’ high rate, but many of the transitions we wish to study involve large conformational changes, where many internal degrees of freedom move in a coordinated way. In this situation, it is difficult to apply standard biasing techniques based on the regular change of internal degrees of freedom along the reaction coordinate. In summary, the analysis of large conformational changes is still a major challenge for both experimental and theoretical techniques. A clear example is the B⇔A transition of DNA duplexes (). It has been known since 1953 that physiological DNA is mostly B-form, while physiological RNA is always in the A-conformation (,,). However, the binding of some proteins to DNA can induce local B→A changes, which are needed to form some protein-DNA complexes crucial for the control of gene functionality (). The B→A transition can also be brought about by other stress conditions, like crystal lattice restrictions (16 and references therein), or the addition of a large proportion of ethanol, which is believed to displace water molecules hydrating the helix, leading to a pseudo-anhydrous environment where the most compact A-form is more stable (,,,). The B⇔A conformational transition has been also the subject of numerous theoretical studies, aimed at understanding the mechanism of the transition and the atomic reasons for the A/B preference in water and other solvents. The latest generations of both CHARMM and AMBER force-fields (the ones most used in simulations of nucleic acids) recognize the A-form as the only stable conformation of duplex RNA, and the B-form as the most important conformation of DNA duplex in aqueous solution (). In fact, MD trajectories of DNA duplexes started in the A-form convert very rapidly to the B-conformation (), showing that MD simulations are able to drive the DNA from an incorrect conformation to a correct one. Unfortunately, MD simulations are unable to detect reversible transitions due to the limited length of current trajectories, which makes very unlikely to sample unstable regions of the conformational space. This lack of reversibility in the transition precludes the determination of the free energy associated with the conformational change and the determination of the atomic mechanism of the transition. For this reason, these simulations need to be biased to force them to sample reversibly the B⇔A transition. In this article, we re-visit the B⇔A transition for a duplex DNA dodecamer using the techniques of essential dynamics, unbiased molecular dynamics simulations and a combination of targeted MD (tMD ()) and the weighted histogram analysis method (WHAM ()) in both water and a mixture of 85:15 ethanol/water. Using these techniques, we were able to trace the B⇔A transition in a smooth reversible way providing estimates of the associated free energy and of the molecular mechanism of this conformational change. We felt that in order to obtain reliable conclusions on B/A preference, a DNA duplex containing at least one helix turn should be used as the model system (i.e. at least 10–12 base pairs); smaller duplexes might not provide a good model of interactions along the major and minor grooves and would maximize the effect of artifactual end-effects. Unfortunately, the increase in the size of the duplex leads to a parallel increase in the computational difficulty of the simulation. Thus, as a compromise, we decided to use Dickerson's dodecamer (dGCGCAATTGCGC) a B-type 12-mer oligo () for which several experimental structures are available (see http://ndbserver.rutgers.edu) and for which dozens of simulations have been published (,). Input coordinates for trajectories started in the B-form were generated by taking Dickerson's crystal geometries (). The experimental structure was neutralized by adding Na in the best regions according to classical molecular interaction potential (CMIP) calculations () and immersed in a rectangular box of water (aroubnd 4454 TIP3P molecules). As described elsewhere (), MIP neutralization protocol allows the definition of a reasonable ionic atmosphere around DNA based on Poisson–Boltzman potentials, reducing then the time needed for counterion equilibration. The solvated systems were then optimized, thermalized and pre-equilibrated using our standard protocol (). The final conformations were then re-equilibrated (constant pressure and temperature: 1 atm, 298 K) for 1 ns more to ensure the stability of the trajectories. Simulations starting in the A-conformation were generated by first building a standard A-type DNA duplex of the desired sequence, which was then neutralized and hydrated as below. Optimization, thermalization and pre-equilibration was carried out following the same procedure than for B-DNA but adding an harmonic restraint ( = 160 kcal/mol.rad) that restrained all the δ angles to 80°. The same restraints were used in equilibration (4 ns) and in production runs; when they are removed the structure changes quickly to the B-form. Ten different structures obtained during the equilibration of the A-form (with angles restrained to 80°) were used to perform 10 parallel un-restrained 3 ns MD simulations. Within this short simulation time the A→B transition was completed in all replicas, which allowed us to obtain reasonable statistics on the microscopic mechanism of the ‘unbiased’ transition. Equilibrated boxes containing 85% ethanol: 15% water were generated from Monte Carlo simulations (300 million configurations) at constant pressure (1 atm) and temperature (300 K). These boxes were then used to solvate neutralized B and A-DNAs (starting conformations of B and A-DNAs were those obtained at the end the respective MD equilibrations in water (see above)). The solvated systems were thermalized ( = 298 K) and pre-equilibrated using 2 ns of MD where only the solvent was free to move. Finally, all the systems were equilibrated for 4 ns in both A and B-conformations. The δ angles were restrained to 80° in A-simulations and 120° in B-simulations. Fifteen structures obtained during the last 1 ns of the restrained B-type simulations were used to start 3 ns unbiased MD simulations intended to capture spontaneous B→A transitions in (85:15) ethanol/water. Mirror calculations with unbiased trajectories started in the A-form were used to confirm that the A-form is a stable minimum in the ethanol/water mixture considered here. These simulations drive a transition by introducing a harmonic penalty that forces the sampled structure to be at a given RMSd from a reference(s) structure(s). By changing smoothly the target RMSd, we can then approach or separate the molecule from reference geometries. Several formalisms can be introduced to introduce the harmonic restraints (), but in our hands smoother and more reliable A⇔B transitions were obtained by using eq. () (see results). As described below, a selected group, rather than all the atoms () were used to compute the RMSd. After several preliminary tests, simulations were performing using windows of 0.25 Å with = 0.1 kcal/mol Å × atom for all windows except the terminal ones, where we used spacing 0.1 Å and = 8 (water) or 2 (ethanol/water) kcal/mol Å × atom. Unless otherwise noted, each window was simulated for 1.2 ns, the first 0.2 being considered as equilibration. To make transitions as smooth as possible, the end point of each window was used in general as starting point for the next. Using these conditions smooth transitions and good overlap between windows was obtained. The biased samplings obtained were used to derive potentials of mean force (PMF) for the transition using the WHAM method (). Free energies were recovered by integrating the PMF using suitable boundaries (see below). Unbiased B-DNA and A-RNA 10 ns trajectories were used to derive the essential dynamics of relaxed B and A-forms (,,,,). The first eigenvector (that describing the most important deformation mode) of the relaxed B-DNA (or A-RNA) trajectory was compared with the B→A transition vector (see eq. ()) to obtain an estimate of the overlap between normal B-DNA fluctuations and B→A transitions. As described in Results, the first eigenvector of the essential dynamics of B-DNA correlates well with the B→A transition vector. We then explored the harmonic energy needed to animate this eigenvector (from the B-form) to reach conformations close to the A-form. For this purpose, we add vibrational energy to the mode obtaining the associated displacement through eq. () from which perturbed geometries were determined. Pursuing this approach, we performed calculations of the dimension-less Mahalanobis distance (; see eq. ()) between B and A forms. The Mahalanobis distance is a unit-less metric, directly related to the deformation energy (see eq. ()) which defines the minimum pathway in essential space between two conformations. Thus, using calculations, we can trace the B→A transition by activating many different essential modes of the B-equilibrium trajectories computing then the energy needed to reach structures closer (to a certain threshold) to the A-form. where is the displacement along individual eigenvectors and stands for the corresponding eigenvalue (in distance units). The sum extends of -important essential movements, in this article we consider the first 10th ones, which accounts for more than 85% of the DNA variance. All simulations were carried in the isothermal–isobaric ensemble ( = 1 atm, = 300 K). Monte Carlo simulations (for preparing the hybrid solvent box) were performed using the BOSS3.4 computer program () allowing internal rotations in the solvent molecules, but no other internal changes. A non-bonded cutoff of 12 Å was used in conjunction with periodic boundary conditions to reduce the number of non-bonded interactions in these Monte Carlo simulations. Molecular dynamics (MD) calculations were carried out using the AMBER8 computer programs () and long-range electrostatic effects were taken into account by means of the Particle Mesh Ewald method (). PARM99 was used to represent DNA interactions (,), while TIP3P () and all-atoms OPLS () parameters were used for the solvents. The use of SHAKE () to maintain all the bonds at equilibrium distance allowed us to use 2 fs as the integration step size. Analysis of trajectories was carried out using PTRAJ module included with the AMBER8 release as well as software developed in house. All simulations were performed using the supercomputer at the Barcelona Supercomputer Center. As reported by others (), MD simulations of DNA using the PARM99 force-field produces a spontaneous A→B conformational change in water in short simulation times, which makes possible the study of the atomic mechanism of such an unforced transition. Ten different trajectories starting from well-equilibrated A-form conformations move to the B-conformation within the 3 ns simulation times. Extension of five of these trajectories to 10 ns did not shown any back-transition to the A-form (data not shown, but available upon request). The analysis of the RMSd plots from canonical A and B-forms show that the transition starts very quickly, and after around 500 ps, the lines of RMSd(versus A) and RMSd(versus B) cross for the first time (see ). After this period, the structures fluctuate around a conformation with B-like properties (see ; individual plots for trajectories are shown at http://mmb.pcb.ub.es/A-B), but still not far from the A-form in terms of RMSd. For all 10 trajectories, the transition in RMSd only happens after major changes in sugar puckering, supporting the idea that sugar re-puckering is the driving force for A→B transition (). Once delta change, helical parameters like roll and twist are adjusted to the standard B values. Thus, in just 100 ps half of the sugars which were originally in the N-conformation ( below 95° have moved into the E and S regions ( greater than 95°) and after 500 ps only four sugars are (in average) in the N-region. Slow re-puckering of these residual North sugars occur along the 1–3 ns simulation period. Analysis of individual sugars shows that in general, sugars in purine nucleotides change faster than those in pyrimidines. Within purines G shows faster transitions than A and within pyrimidines C is that with the slower rate of N→S transitions. Very interestingly, the transition vector correlates very well (see ) with the first deformation mode of both relaxed B and A-forms (taken from ref. ()). This indicates that the deformation pattern needed to perform the biologically relevant B⇔A transition is implicitly coded in the polymeric structure of these nucleic acids and that in a rough approximation, the A-form can be interpreted as a vibrational-activated state of B-DNA. This hypothesis is supported by analyzing the structures obtained by moving the B-DNA along the first essential mode (see ). Thus, a simple deformation along the first mode yields structures with RMSd(A) = RMSd(B) for vibrational energies around 4 kcal/mol, and for deformation energy around 12 kcal/mol to structures that are at only 2 Å from the A-form (see ). The activation of additional deformation modes (up to 10) using the Mahalanobis metric allowed us to slightly reduce the RMSd from target A-form (to 1.7 Å), but at the expense of a very large vibrational energy (19 kcal/mol). Clearly, B→A transitions follows closely the first mode, and local rearrangement from a distorted to the real A-conformation requires some local, probably non-harmonic deformations. Our previous work () and the unbiased simulations discussed above suggested that sugar puckering could be a good variable to discriminate between B and A-form. Accordingly, we undertook a study of the B→A transition by means of restrained MD simulations, slowly changing the angles of all sugars from 140 to 70°. As shown in , and in data from equilibrium simulations in , the B→A transition happens for values around 90°. Note that such a transition is clear not only in the RMSd from A and B-conformations, but also in the width of the minor groove () and in Olson's () parameter (see ). Thus, a structure can be assigned to the A-family if RMSd(A) < RMSd(B), «» > 1.0; «δ» < 95° and ‘’ > 14 Å. When only the RMSd criteria is fulfilled, the structure should be labeled as distorted B-form. These threshold need to be consider to validate when a tMD simulation is really driving the structure to the target conformation and not to a distorted geometry, which might have small RMSd to target conformation but very poor internal geometry. While the use of restraints on delta permitted the required transition, and was extremely useful to define clear threshold values to classify structures as A or B, it does not allow us to easily recover the free energy associated to the conformational change. Thus, we decided to follow the A⇔B transition by defining the RMSd function considering only heavy atoms in the sugar ring (Reference set 1, see ). Additional calculations were performed considering an extended set of restraint atoms which include all ring atoms plus O5’ (Reference set 2) and an alternative set of restraint atoms formed by the phosphate group and O4’ (Reference set 3). Using any of these three RMSd definitions and the restraint function shown in eq. (), we were able to reproduce correct B→A transitions (see and Figure S1 in Supplementary material), where both starting and end structures fits the canonical B and A-forms. It is worth noting that such smooth transitions cannot be so easily obtained with other choices of the RMSd function. Caution is then necessary against a un-supervised pure-force use of the targeted MD technique to follow complex conformational transitions. It is worth to note that the transition pathway closely follows the direction of the first essential mode of relaxed DNAs and RNAs (see ), confirming the results obtained for unbiased A→B transitions. This finding demonstrates that the ability to change between the B and A-conformers is implicitly coded in the structure of nucleic acid duplexes. Interestingly, irrespective of the set of atoms used to define the RMSd with respect to target structures the transition is similar: for RMSd() ≈ 2.0 Å, the RMSd(A) and RMSd(B) lines cross, but until very late in the reaction coordinate (RMSd() ≈ 1.0 Å) full transition is not achieved (see and Supplementary Figure S1). The PMF associated to the B→A transitions in water are well converged and seems robust to the selection of the set of restrained atoms and to changes in the ‘forward’ or ‘reverse’ direction of the transition (see ). As predicted, the A→B PMF takes the form of a downhill landscape, where the A-form is not defined as a minimum. Integration of the PMF profiles provides a direct estimate of the free energy associated with conformational transitions. Inspection of (see text) allowed us to classify the PMF into three regions: (i) pure A-form (RMSd() ≤ 1.0 Å), pure B-form (RMSd() ≥ 2.0 Å), and distorted B-forms (2.0 Å > RMSd() > 1.0 Å) Using these values, the deformation of B-DNA to achieve structures with RMSd(A) ∼ RMSd(B) requires around 3 kcal/mol (see ), while a full B→A transition is associated with a free energy penalty of 11.4 kcal/mol. After sending the first version of this manuscript for publication, a colleague addressed us to the work by Zhurkin's group () who reported empirical scales for the free energy of B→A transition in water. Applying Zhurkin's data in our duplex, we obtain an ‘empirical’ estimate of 11 kcal/mol for the B→A transition, a value that matches perfectly our estimate. It is also worth noting the very good agreement between rigorous tMD/WHAM estimates and rough values obtained by essential dynamics in , something that confirms that the B→A transition is a pure ‘up-hill’ process which can be simplified as a vibrational activation of the first deformation mode of B-DNA. A small amount of energy is enough to distort the DNA to conformations that are not far in terms of RMSd from the A-form, however, a full transition is very unlikely and requires major changes in the environment, such as the introduction of proteins or co-solvents. In summary, all our simulations suggest that in water the A→B transition is a downhill process, and that the A-form is not a stable conformation of Dickerson's dodecamer in water. This result is in full agreement with all previous MD simulations irrespective of the force-field, sequence or simulation conditions used (,,), which argues against force-field or simulation protocols. Special agreement is found with data by Banavali and Roux (), who using other force-field a shorter oligo, and a different functional for the restrain energy obtained quite similar results, demonstrating that the method, when applied with common sense is robust. However, these theoretical results have been severely criticized by Jose and Porschke (,), who found that experimentally the transition time for the A⇔B change (around 70:30 ethanol/water) is in the range 10–10 μs, which should indicate the existence of a sizeable transition barrier and the presence of two well-characterized canonical forms. We will show below that the criticism is not justified since the shape of the free-energy curve in pure water or ethanol-rich solutions is completely different, and conclusions obtained in rich ethanol mixtures cannot be simply extrapolated to water solution (see next sections). It is experimentally known that in the presence of large concentration of ethanol, the DNA changes from a B to an A-type conformation (,,,,,). Up to 15 MD simulations of Dickerson's dodecamer starting from the A-conformation remain in this region, as previously found by McConnell and Beveridge (). Extension of one of this simulation to 10 ns does not show any dramatic change from the shorter ones (see Supplementary Figure S2), suggesting that the force-field recognize the A-form as a stable conformation of DNA in ethanol/water solution. However, 10 unbiased 3 ns trajectories starting in the B-conformation remains within this conformational family (see ), confirming again previous results by other authors (,,). Overall, unbiased MD simulations suggest that in the presence of high concentration of ethanol, the free-energy landscape is different to that in pure water, since at least two minima (close to the B and A-forms) exist, separated by a sizeable energy barrier. Targeted MD simulations (using restraint set 1) are able to drive a smooth and complete A→B transition (see ) in the presence of a high concentration of ethanol. Analysis of RMSd and internal geometry changes along the transition pathway demonstrate that the A→B transition in 85:15 ethanol/water starts with a fast change (RMSd() ≈ 3.3 Å) of sugar puckering from N to S which leads to an immediate change in the parameter, and later (RMSd() ≈ 2.0 Å) to the crossing of the RMSd(A/B) lines and to the reduction of the minor groove width to canonical B-values. In summary, we see in ethanol/water a mirror copy of the transition found in pure water. Combining results in ethanol/water and pure water, we can obtain a quite complete picture of the nature of the A⇔B conformational change. Accordingly, when the DNA moves from the A to the B-form, the driving force is a change in the sugar puckering that is soon followed by a global arrangement of the structure. On the contrary, for the B→A transition a global change of shape and reduction of the width of the minor groove happens first, and only when the global structure is close to the A-form do the sugars adopt a North puckering, defining at that point a true A-type structure. As suggested by unbiased simulations, PMF profile for the A→B transition in ethanol/water shows the existence of two minima regions separated by wide region of higher free energy (see ). Current tMD simulations cannot be used to determine precisely the free energy barrier, but a rough estimate of ∼2 kcal/mol (A→B) and ∼1 kcal/mol (B→A) can be derived from the PMF curve in . Clearly, the shape of the free-energy profile associated with the A⇔B transition is strongly dependent on the solvent, the barrier-less A→B transition found in water is changed to a barrier-limited B→A transition in 85:15 ethanol/water. The apparent discrepancy between MD simulations (in water) and experimental studies (in high ethanol concentration) of the B⇔A transition can then be easily understood, and illustrates that caution is always necessary when MD simulation results are interpreted in the light of experimental measures performed under different conditions. Analysis of geometrical variation along the tMD simulation pathway allows us to define boundaries for integration of the PMF curve. Thus, the pure A-form is obtained just in a narrow valley (4 Å ≥ RMSd() ≥ 3.3 Å), an intermediate distorted A-conformation is obtained in the region: 3.3 Å ≥ RMSd() ≥ 2.0 Å. Finally, a canonical B-form is defined in the region: 2.0 Å ≥ RMSd(). Note that these partitions agree well with the positioning of maxima and minima in the PMF curve (see ). Using these definitions, we conclude that in 85:15 ethanol/water the A-form (considering both canonical and distorted species) is 0.8 kcal/mol more stable than the B-one, while the distortion of the canonical A-form is disfavored by only 0.4 kcal/mol. In summary, tMD simulations show that the presence of large amounts of ethanol produces a dramatic change in the A/B conformational equilibrium, in good agreement with experimental data (see Introduction). The A-form, which is negligibly populated (ratio / = 10) in water, is the favored conformer in 85% ethanol (ratio / = 0.2). Clearly, as discussed elsewhere, the differential screening of water and ethanol/water explains this dramatic solvent effect (,). However, the important point to emphasize here is that despite the different and numerous sources of errors, force-field simulations coupled to tMD and WHAM are able to reproduce a dramatic solvent-induced conformational transition.

Homologous recombination is a universal process in living organisms. The central enzyme of this reaction appears to have been conserved in all kingdoms of life () and in viruses: RecA in bacteria, Rad51 in eukaryotes, RadA in archaea and UvsX in phageT4 (). In each case, the enzyme functions by catalysing the exchange of single-stranded DNA into intact DNA duplex, generating homologous pairing and promoting recombination (). Furthermore, each orthologous protein binds single-stranded DNA in the form of a nucleoprotein filament, which has a highly similar structure in (), () and (). The purpose of homologous recombination is to repair DNA double-strand breaks () and to restart stalled or collapsed replication forks (). In eukaryotes, it is also used in meiotic recombination () and, in some circumstances, telomere maintenance (). Beyond these generalized functions, homologous recombination has been co-opted into specific functions in a diverse set of organisms. One example is mating type switching in yeast, where homologous recombination is induced by site-directed DNA lesions (). In many pathogenic organisms, including bacteria, fungi and protists, homologous recombination can play a similarly specialized role in host immune evasion (). One way pathogens evade immunity is by antigenic variation, the periodic switching of surface antigens. In , a protistan parasite of mammals in Africa, antigenic variation involves switches in the variant surface glycoprotein (VSG) coat. The success of this strategy relies upon a cell expressing a single VSG coat type at any one time, and the ability to switch to an antigenically distinct version, selected from an enormous archive of >1000 silent genes (); for recent reviews see (). Singular VSG expression has involved the evolution of telomeric transcription units, termed expression sites (ES), and transcriptional control mechanisms that act upon them. switching is dependent on recombination of the silent s into the ES, and a number of such reactions have been described. The most commonly observed, at least early in infections, are gene conversions that generate a copy of a silent and transfer it to the ES. Such gene conversions encompass the ORF and extend to regions of homology upstream and downstream. -associated 70-bp repeats frequently demarcate the upstream conversion boundary (,), whilst the reaction can extend downstream to short blocks of homology in the 3′ ends () or to the telomeric repeats (). Crossover exchanges between chromosome ends, termed reciprocal switches, are also seen (). Finally, segmental gene conversions are found where multiple pseudogenes are recombined together to form novel, mosaic (,). These have been considered rare events, found late in infections. However, sequencing the genome has revealed that pseudogenes represent the substantial majority of the archive (), arguing that this is likely to be a significant mechanism of VSG switching (,). Growing evidence suggests that switching is closely linked to homologous recombination, since mutations in several key factors of homology-directed strand exchange, including RAD51 (), a RAD51-related protein called RAD51-3 () and BRCA2 (C.Hartley and R.McCulloch, unpublished data), impair the immune evasion process. However, switching presents several unusual characteristics. First, the reaction can occur at very high rates (up to 1 × 10 switches/cell/generation)(,), significantly more frequent than the rates of general homologous recombination, which are more typical of random mutation (). Second, recombination of frequently relies on flanking sequences, such as the 70 bp repeats, that are rather short and share limited homology (,), despite the fact that the mismatch repair (MMR) machinery regulates homologous recombination to favour well-matched sequences, and estimates suggest that around 100 bp of homology are needed for efficient RAD51-mediated recombination (). Finally, there appears to be a hierarchy in the substrates that are used by general homologous recombination in other organisms: sister chromatids appear to be the favoured substrate in both yeast () and mammals (), while allelic sequences on chromosome homologues and homologous sequences at ectopic locations are progressively disfavoured (,). Recombinational activation of allelic sequence on the sister chromatid would not result in a coat switch, compelling the reaction to search for silent s throughout the genome. It has been argued that this is why the archive is predominantly sub-telomeric, as these locations appear to escape such substrate constraints (). Beyond that, it is unclear how the other characteristics of switching are accommodated by homologous recombination. One study has suggested that the homologous recombination is distinct from that of and its relatively close cousin , in that the reaction can act on short stretches of homology and may have an elevated rate of strand exchange (), perhaps indicating modifications of the recombination machinery. Another study has suggested that it may be necessary for to suppress MMR to allow homologous recombination to act during switching (). Here, we have sought to examine the parameters of homologous recombination in in order to address these questions further. To do this we have used a transformation assay that allows us to measure the efficiency of recombination, and to assess the pathways that operate. In previous work, we characterized the MMR machinery (), which plays a critical role in maintaining genome stability and is conserved throughout evolution. The function of MMR is to recognize and excise base mismatches, which arise through replication errors, by chemical damage or during recombination between incompletely sequence-matched DNA molecules (,). Eukaryotic MMR is catalysed by homologues of bacterial MutS and MutL proteins, though the machinery has been elaborated, since most eukaryotes encode 3-7 MutS-related proteins (termed MSH 1-7) and 2-4 MutL-related proteins (MLH1-3 and PMS1-2). Mutation of either MSH2 or MLH1 in demonstrates that MMR functions in correcting errors in the nuclear genome (). Furthermore, MSH2, and by implication MMR, plays a role in constraining homologous recombination to occur between well-matched sequences (). A similar anti-recombination role for MMR has long been appreciated in other organisms, where it contributes to the suppression of excessive genome rearrangements and to speciation (), though how it occurs remains to be determined fully. In this work, we extend the above analysis to show that homologous recombination is strictly dependent on substrate length, in keeping with findings in other organisms. In addition, by comparing the recombination of diverged substrates of different lengths in wild-type cells and in MSH2 mutants, we have identified an MMR-independent homologous recombination pathway and find that a short-patch MMR pathway can correct base mismatches during recombination. bloodstream form cells were used and grown at 37°C in HMI-9 medium (). The cells were of strain HTUB (), which was derived from the MITat1.2 cell line by insertion of the hygromycin phosphotransferase () ORF into the tubulin array; heterozygous (+/−) and homozygous (−/−) mutants in this strain have been described before (). Transformations to assay recombination efficiency were carried out by electroporation with 3 μg of DNA that had been PCR-amplified using Herculase (Stratagene) high-fidelity DNA polymerase (see below). Electroporation conditions were a single pulse at 1.4 kV, 25 μF using a Bio-Rad Gene Pulser II, and at least three transformations were performed for most constructs. For the transformations, cells were grown maximally to 3 × 10 cells·ml, and minimally to 1.5 × 10 cells·ml, harvested by centrifugation at 600 for 10 min at room temperature, and then re-suspended in Zimmerman post-fusion medium supplemented with 1% -Glucose () to a concentration of 5 × 10 cells·ml. A total of 2.5 × 10 cells were used per transformation and were allowed to recover following electroporation by growth in 10 ml of non-selective medium for 18 h before antibiotic selection. For this, the cells were harvested as before and then resuspended in HMI-9 containing 2.5 μg.ml phleomycin (Cayla) and spread in 1.0 ml aliquots over a 24-well dish. Between 2 and 10 × 10 cells were plated out in this way, depending on the construct being used (see Supplementary Figure 1). Transformation rates were measured by the number of wells containing phleomycin-resistant cells after 8–14 days growth. We have assumed that the cell population in each well is clonal, arising from a single transformant, though this underestimates the true number of transformants. However, calculating the likely correct number of transformants in each plate using the number of negative wells, and assuming a Poisson distribution of clones (), does not alter the nature of the relationship between transformation rate and substrate length or homology (data not shown). In addition, such analysis does not alter significantly the calculated minimum efficient processing segment, which was determined as described by Shen and Huang. For each transformation construct, 24 PCRs were performed, the products pooled and purified using a Qiagen PCR purification kit according to manufacturer's instructions; 1 QIAquick column was used per six PCRs, and the pooled PCR products resuspended in distilled HO. Each of the different sized PCR products were PCR-amplified either from pwt::, p05:: or p11:: (), generating constructs with 0, 5 or 11% base mismatches relative to the HTUB (wt) sequence. The oligonucleotide primers used to generate each construct were size-purified prior to the PCR and are named in Supplementary (sequences can be provided on request); the precise length of each targeting flank in the different PCR products, as well as the number of mismatched bases relative to wt, is also detailed. For reasons that are unclear, a 125 bp construct with 11% sequence divergence could not be PCR-amplified and was omitted from the study. All the purified PCR products were examined by agarose gel electrophoresis to ensure the lack of visible contaminating DNA molecules, and their concentration was determined spectrophotometrically. The hygromycin sensitivity or resistance of the i cells was determined by replica passaging 100 μl of the phleomycin-resistant transformants into 1.5 ml of either non-selective HMI-9, or media containing 5 μg.ml hygromycin (Roche). Growth was assessed microscopically 48 h later. For Southern analysis of the transformants, a 15 ml culture was grown to a density of ∼4 × 10 cells·ml, harvested by centrifugation as before, resuspended in 500 μl of 1 mM EDTA, 100 mM NaCl, 50 mM Tris-HCl pH 8.0 and lysed overnight at 37°C following the addition of SDS to 1% and proteinase K to 100 μg.ml. DNA was recovered by phenol/chloroform extraction and ethanol precipitation, and then resuspended in distilled HO. The genomic DNA samples were digested by restriction enzymes as described by the manufacturers and separated by electrophoresis, typically at ∼30 V overnight, on 0.8% agarose gels (Seakem LE agarose, BioWhittaker Molecular Applications) made with 1× TAE buffer (40 mM Tris, 19 mM acetic acid, 1 mM EDTA) containing 0.2 μg.ml ethidium bromide (Sigma). DNA was blotted by capillary transfer onto hybond-XL membrane (Amersham), probed with α-P labelled DNA generated by random priming and washed to 0.2× SSC, 0.1% SDS at 65°C. Separation of intact chromosomes was carried out on a Bio-Rad CHEF-DRIII apparatus. For this, each agarose plug contained ∼4 × 10  cells, which had been grown in HMI-9 to a density of ∼2 × 10 cells·ml, centrifuged as before, washed by resuspending the pelleted cells in 10 ml PSG (1 × PBS, 1% w/v glucose), re-centrifuged and then resuspended in PSG at a concentration of 1.6 × 10 cells·ml. The cells were then warmed at 37°C for 1 min and an equal volume of 1.4% Microsieve low-melt agarose (Flowgen) in HO added and mixed. Disposable plug moulds (BioRad) were filled with ∼50 μl agarose and placed at 4°C for ∼4 h to set. The agarose plugs were then removed from the moulds, incubated in NDS buffer (0.5 M EDTA, 1 mM Tris base and 34.1 mM lauroyl sarcosine) pH 9.0 containing 1 mg.ml proteinase K at 50°C for ∼24 h, transferred into NDS buffer pH 8.0 containing 1 mg.ml proteinase K at 50°C for ∼24 h, and finally transferred into NDS buffer pH 8.0 for storage at 4°C. For electrophoresis, the plugs were washed four times at room temperature in 1 ml of 1× TB(0.1)E (0.089 M Tris-borate pH 8.0,).2 mM EDTA) for 1 h each, then placed in the wells of a 1.2% agarose (Seakem LE, BioWhittaker Molecular Applications) gel. The gel was electrophoresed at 15°C at 2.5 V.cm for 144 h with a 1400–700 s switch time and visualized by staining with 0.5 μg.ml ethidium bromide and UV illumination. Sequences of the integrated DNA constructs were determined by performing PCR amplifications using oligonucleotide primers corresponding to the first and last 20 nt of the ORF and DNA polymerase (ABgene). The resulting PCR products were purified and sequenced using oligonucleotides primers that read upstream and downstream from the bleomycin resistance cassette common to each construct. Previously, we described an assay to examine homologous recombination efficiency in (); this is summarized in A. The assay relies upon a hygromycin phosphotransferase ORF () integrated into the tubulin array () on chromosome 1, providing a unique site for recombination. Recombination efficiency is determined by measuring the transformation rate of linear constructs containing a bleomycin resistance cassette () flanked by sequences derived from . The advantage of this approach is that a single, defined site for recombination is analysed from which a number of parameters can be varied (see below). In addition, using a foreign sequence as a target reduces the potential for recombination into related sequences elsewhere in the genome, an issue that has influenced other studies where endogenous sequences are used both as a genomic target and recombination substrate; e.g. (,,). Although transformation of any organism is likely to be affected by a number of factors, including DNA concentration, transformation conditions and growth of the cells, this appears to be a reliable measure of recombination in . The same assay has demonstrated the importance of MMR in controlling homologous recombination between sequences containing base mismatches ()(see below). Moreover, related transformation assays have quantified the role in recombination of a number of genes, including (,), (), () and two -related genes ( and ) (), and have looked at the importance of target copy number (). Most likely, this approach succeeds because virtually all stable transformants in integrate linear DNA by homologous recombination, rather than by end-joining processes, and the formation of extra-chromosomal episomes following such transformation has not been described (). Using this assay we have shown previously that homologous recombination, acting on substrates with targeting flanks of 450 bp, becomes increasingly less efficient as the level of sequence homology between the flanks and the genomic target decreases (). One percent of sequence divergence resulted in a 2.8-fold reduction in transformation efficiency, and 11% divergence caused a near 100-fold reduction. By mutating the gene encoding MSH2, we showed that transformation efficiencies of substrates with 2–11% sequence divergence increased by around 9-fold, indicating that MMR is an important regulator of homologous recombination. Nevertheless, it is not the sole factor that determines the success or failure of recombination on such substrates, as the same decline in transformation efficiency with increasing divergence was seen in both -/- and wild-type cells. In this study, we have examined two further features of homologous recombination. First, we tested the relationship between substrate length and recombination efficiency. Second, we asked if MMR has the same influence on homologous recombination between diverged substrates when the length of homology becomes short. To do this, a series of constructs with targeting flanks varying in size from 25 to 200 bp were generated (B). These were derived by high-fidelity PCR from the previously described constructs with 450 bp flanks () that are either perfectly homologous to the genomic sequence (0% sequence divergence), or that contain base mismatches resulting in either 5 or 11% divergence. Linear DNA was prepared by PCR amplification, rather than by restriction digestion, because non-homologous overhangs at the DNA ends, while insignificant for an integration flank of 450 bp, could have larger effects on recombination efficiencies mediated by the shorter substrates. The transformation efficiency of each construct was determined in −/−, +/− and wild-type bloodstream stage strains (,). The results of this analysis are graphed in and , and summarized in and . Comparing the transformation efficiencies of constructs with 100% sequence identity with the genomic target indicates that homologous recombination efficiency is dependent on substrate length. Over the range 200–50 bp, a linear relationship was found between transformation efficiency and substrate length (). The average transformation rate for the 50 bp construct in the MMR-proficient cells (MMR+; either wild type or +/−) was 0.80 ± 0.45 transformants × 10 cells, corresponding to a near 13-fold reduction in efficiency compared with the 200 bp construct (10.16 ± 0.38 × 10). Beyond this range the linear relationship broke down. It appears that around 50 bp of sequence homology represents a lower threshold for recombination, at which point the reaction becomes extremely inefficient: the 25-bp substrate displayed an average transformation rate in MMR+ cells (0.02 ± 0.06 × 10) that was around 500-fold lower than the 200 bp substrate and 40-fold lower than the 50 bp substrate. It is notable, however, that transformants can be generated with 25 bp flanks, and that these can integrate by homologous recombination ()(). By comparing the transformation rate of the 200 bp construct with the restriction-digested plasmid containing 450 bp of homologous flank used previously (), it appears that the rate of transformation reaches a plateau around 200 bp. This may indicate that recombination becomes no more efficient with substrates longer than 200 bp, though it is also possible that this is not a reflection of recombination but represents the maximum transformation efficiency achievable in bloodstream stage cells under these conditions. Mutation of MSH2, resulting in cells with impaired MMR (MMR−), had no discernible effect on recombination for any of 100% homologous constructs in the range 25–200 bp. This contrasts with the 450 bp parental construct, which showed a small (1.6-fold; , ) but significant increase in recombination in −/− cells relative to MMR+ cells (). Elevation of recombination between similar-sized (350 bp), sequence-matched substrates is seen also in MMR mutants (,). The basis for this is unknown, but its absence on shorter substrates may indicate that MMR acts to suppress recombination on sequence-matched substrates only when they are of a significant length, perhaps because they are more prone to secondary structure during strand exchange. On 450 bp substrates containing base mismatches relative to the genomic sequence, increasing levels of sequence divergence had an exponentially deleterious effect on homologous recombination (). Furthermore, mutation of MSH2 resulted in a mean 9-fold elevation in the rate of transformation when sequence divergence was >2% (, ). The transformation measurements in this study show that base mismatches and MMR regulation do not have a uniform effect on recombination over the range of substrate length analysed. This can be seen by examining the data in two ways. First, we examined the effect that mutating MSH2 had on recombination of the 5 and 11% diverged substrates ( and , ). On the 5% diverged substrates, a statistically significant elevation in transformation efficiency in the MMR− cells relative to the MMR+ cells was observed only on the 450 and 200 bp substrates. In contrast, though the 5% diverged substrates smaller than 200 bp appeared to show a trend towards a slight increase in transformation in MMR− cells, this was not significant. Comparing the average transformation rates of these constructs confirms this (): the 450 and 200 bp substrates displayed 9.3- and 5.3-fold increases, respectively, in transformation rate in the MMR− cells relative to MMR+ cells, whereas averaging the data from all substrates below 200 bp revealed a mean 3.4-fold increase (range 1.3–6.7). On the 11% diverged substrates, impairment of MMR caused no significant elevation in transformation efficiency relative to the MMR+ cells on any substrate other than the longest (450 bp). The second way we examined these data was to compare the transformation rates of the 5 and 11% diverged constructs, in both the MMR+ and MMR− cells, relative to the perfectly matched substrates. This is quantified in by determining the extent of the reduction in average transformation rate of each 5 or 11% diverged substrate relative to the sequence-matched substrate of cognate length. In both the MMR+ and MMR− cells, at virtually all substrate lengths, increasing sequence divergence caused a progressively more severe impairment in transformation rate. However, in MMR+ cells this effect was more pronounced on 450 and 200 bp substrates than on any substrate shorter than 200 bp. In contrast, in MMR− cells the extent to which sequence divergence impaired transformation was more uniform across all lengths. The average level of reduction in transformation of 5% diverged substrates smaller than 200 bp in MMR+ cells was 8.1-fold, compared with a 3-fold reduction of all substrates in MMR− cells. For 11% diverged substrates, the reduction in transformation rate for substrates smaller than 200 bp was the same as all substrates in MMR− cells (26.6- and 26.7-fold, respectively). Taken together, these data argue that MMR has an important role in regulating recombination between diverged sequences when the substrates are long (around 200 bp or longer), but MMR has a less significant role on shorter substrates. Higher levels of sequence divergence presumably exacerbate this because the lengths of uninterrupted homology stretches in such substrates are reduced. An important consideration in the above analysis, and in previous work using the same assay (), is whether or not the experimental approach measures homologous recombination on all substrates. For instance, it may be that MMR has less influence on recombination of short substrates because primarily non-homologous pathways of recombination act upon such sequences. Indeed, we have shown that some DNA integration in can occur by a pathway mediated by sequence microhomology (). Homologous recombination of the constructs into the genomic locus should lead to loss of hygromycin resistance (Hyg) due to disruption of the gene by . We therefore tested the antibiotic resistances of a large number of transformants (), revealing that a significant proportion of transformants in wild type, +/− (data not shown) and −− cells retained hygromycin resistance (Hyg). There was, however, no clear relationship between the retention of functional and either substrate size or level of sequence homology. Although the numbers of Hyg transformants appeared to increase in wild-type cells as substrate length decreased from 200 to 50 bp at 0% divergence, the same trend was not apparent in −− cells () or in +/− cells (data not shown). Determining the average level at which Hyg cells arose, using the accumulated data for all substrate lengths, showed that this is not influenced by the extent of divergence: 15.1, 9.2 and 16.7 of transformants were Hyg for the 0, 5 or 11% diverged substrates in wild-type cells, respectively, compared with 28, 20.5 and 26.9% of transformants in −/− cells. These latter data do, however, indicate that Hyg transformants appeared to be slightly more prevalent (around 2-fold) when MSH2 was mutated, perhaps indicating that MMR has an influence on the process that leads to their formation. To examine the recombination pathway(s) that gives rise to Hyg transformants, a number of clones were examined by Southern analysis. The large majority of Hyg transformants had integrated the constructs by homologous recombination. In −− cells, 20 of the 24 clones examined had maps indicative of disruption by homologous integration of (). Re-probing the blot with a portion of the ORF revealed that the antibiotic marker had been duplicated (based on the equivalent signal intensity of the two bands) in these clones, explaining why they retained resistance to hygromycin. −/− transformants had not integrated into the tubulin target, but instead into unmapped genomic locations, and the marker was unaltered. A broadly similar pattern of integration was seen for Hyg transformants in wild-type cells (data not shown). Here, 24 clones were examined, and 23 had disrupted by homologous recombination but retained a functional copy, whilst 1 had integrated into an aberrant genomic location. The greater number of aberrant integrations in the −/− cells may indicate that this minor recombination pathway gains prominence in the absence of MMR. Furthermore, most aberrant integrations were seen using relatively short or diverged substrates: Three of the −/− clones had arisen from 11% diverged substrates of 100 bp (, lanes 12, 13 and 23) and one had arisen from a 50 bp, 100% homologous substrate (, lane 3); the wild type aberrant integrant arose from a 100 bp, 100% homologous substrate. The above data suggest that two putatively distinct pathways operate in the context of this recombination assay to generate hygromycin-resistant transformants. A minor pathway is integration events that do not target the tubulin locus, as directed by the terminal targeting flanks, but into other genomic loci. Although we have not mapped these integrations, the background rate at which they occur is very reminiscent of the microhomology-mediated reactions we have described before (). The prevalent pathway, in contrast, is a homologous recombination reaction associated with retention of a functional gene. A number of processes could account for this. First, it is possible that duplication (or amplification) of occurs at a relatively high rate in the growth of these clonal lines, perhaps due to recombination within the multigenic tubulin locus. Second, larger scale changes in chromosome copy number (for instance, trisomy arising from replication errors) could be rather common in . Finally, it is possible that a form of recombination, termed break-induced replication (), could be commonly induced by such construct integrations, leading to the duplication of large stretches of chromosome 1, or even the complete chromosome. To examine this, we characterized a selection of Hyg and Hyg transformants from wild-type cells by pulsed field gel electrophoresis (). One Hyg transformant (Hyg5), which arose as a result of an aberrant integration, contained a novel chromosome, 370 kb in size, that contained and sequence (data not shown). Such alterations in chromosome structure have been seen previously in microhomology-mediated reactions (), providing further indirect evidence as to the recombination pathway involved in these integrations. In contrast, no differences in karyotype were seen in the other Hyg transformants (that had utilized homologous recombination; Hyg1-4) when compared with the Hyg cells or the parental strain, suggesting that large changes in chromosome structure are not associated with these events. Furthermore, probing of a Southern blot of these transformants (Supplementary Figure 3) indicated that there was no difference in the relative amount of chromosomes 1 and 2. Since the tubulin array in which is inserted is located on chromosome 1, this suggests that integration of did not result in the generation of a new copy of the chromosome. Taken together, it seems likely that Hyg in some transformants is not a consequence of targeted integration by , but results from events that amplify in the tubulin locus and are a constant background process in the HTUB cell lines. To examine the mechanisms that contribute to the recombination and processing of the constructs during integration, we sequenced the DNA surrounding the homologously integrated marker in a number of transformants clones generated in both wt and −/− cells using constructs with 50, 100, 150 or 200 bp flanks of 5 or 11% sequence divergence. We did not examine any clones that had integrated aberrantly. The results in depict the pattern of residues in the transformant DNA that are mismatched between the constructs and wt target. Most of the transformants had a pattern of sequences, in which the majority of mismatched residues corresponded to the sequence of the construct DNA on one side of and the genomic target on the other; no difference in this pattern was observed if the transformants were Hyg or Hyg (data not shown). Only clone 9 (from wt cells using a 100 bp, 5% diverged construct) was substantially different, with sequence in both directions. Clone 23 (−/− cells with 200 bp, 5% construct) was somewhat different also, with predominantly -derived residues. Such a pattern is consistent with a model for targeted gene replacement in yeast () and mammals () involving independent strand invasions by both arms of the construct (). Such a model predicts that heteroduplex DNA can form by strand invasion of each construct end, which can result in ‘sectored’ or mixed sequence at the mismatched residue positions following replication of the heteroduplex (see ). However, no such mixed sequence was observed in this study: at each position where a base mismatch might from during strand invasion, the sequences were clearly either construct-derived or - derived. This infers either that heteroduplex does not form or that repair of such mismatches is normally rapid and occurs before replication. The distribution of construct and -derived sequences was very similar in the −/− and wt transformants, indicating that if mismatch repair occurs it can proceed independently of MSH2. In each transformant in which construct-derived residues were present in the transformed DNA, this was not continuous along the length of the flank, but was combined with -derived residues. Most likely, -derived residues that are distal to in the flanks of transformants with an otherwise continuous tract of construct-derived sequence (clones 6, 22, 35, 30, 24, 37, 38) result from nucleolytic degradation of the ends of the linear molecules following transformation and before integration. This was most extensive in clone 30 (maximally 76 bp), whereas in a number of other clones mismatched residues close to the ends of the construct molecules had been integrated (e.g. within 15 bp in clones 41 and 15, and 32 bp in clones 37 and 38), arguing that such degradation need not be substantial. In a number of other transformants, the flanks contained a discontinuous pattern of construct and -derived sequence tracts (clones 1, 15, 41, 23), and some of the residues that were patterned in this way were positioned in close proximity (e.g. separated by only 2 and 8 bpin clones 15 and 23, respectively). This patterning is most simply explained by the formation of heteroduplex DNA during homologous integration and the repair of base mismatches prior to replication by short-patch mismatch repair. The primary conclusion from this work is that homologous recombination has substrate characteristics that are typical of those described in other organisms. The significance of this lies in our understanding of the relationship between homologous recombination and the critical immune evasion process of antigenic variation, which involves switches in expression. Genetic evidence points to antigenic variation being closely linked to homologous recombination, despite the fact that gene switching occurs at high rates (,), often acts upon rather short, diverged sequences (,,) and frequently recombines s on different chromosomes to the ES (,), characteristics that are unusual for stochastic homologous recombination. We have shown previously that the efficiency of homologous recombination is dependent on substrate homology, which is at least partially controlled by the MMR machinery (). We now show here that recombination efficiency is also dependent on substrate length, at least over the range 25–200 bp. This indicates that specific features of switching have not arisen through modifications of the recombination machinery in these two key features. This conclusion is perhaps not surprising, as the primary function of homologous recombination is to repair DNA damage () and ensure the completion of replication (), and therefore fundamental reaction changes could have far-reaching effects on genome integrity. Any specificity of homologous recombination during switching must therefore involve elements or factors that have not yet been uncovered and may be unique to . Similar conclusions have been reached regarding antigenic variation in pathogenic sp., where convergent evolution has produced a related immune evasion reaction that is also closely linked to homologous recombination (). The relationship we describe in between substrate length and recombination rate (measured indirectly as transformation rate) appears to be conserved throughout evolution. In , three studies have recorded reductions in recombination efficiency over the range 74–20 bp (), 405–27 bp () and 200–25 bp (). In eukaryotes, recombination rate appears to decrease over 960-80 bp () or 2 kb–26 bp (), and in mammalian cells a similar relationship has been recorded between an upper length of around 6.8 () to 10 kb () and a lower length between ∼160 () and 330 bp (). Variations in whether the relationship between recombination rate and substrate length is linear or exponential and in the range of substrate sizes involved, both within and between organisms, presumably reflect differences in the assays used, meaning that it is difficult to compare the absolute sequence requirements of the recombination machineries between each organism. Nevertheless, a sharp drop in recombination efficiency below a lower threshold, referred to as the minimal efficient processing segment (MEPS)(), has been described in (), () and mammalian cells (). We observed the same phenomenon in , reinforcing our view that the recombination machinery in the parasite operates in the same manner. Despite this broad evolutionary conservation, previous reports have suggested that recombination efficiency is not affected by substrate length over the range 50–400 bp (,), which is difficult to reconcile with the findings detailed here. One of these reports compared constructs that target distinct genomic locations (), and it may be that differences in target accessibility (for instance, through chromatin structure or transcription level) affect recombination efficiency and confounded the analysis. Another explanation may be revealed by linear regression analysis (), which suggests that the MEPS for on sequence-matched substrates is around 31 bp, shorter than the estimates in either or mammals (∼250 bp) (), and closer to (around 25 bp). Intriguingly, transformation data from , a related kinetoplastid parasite, failed to recover any integrants when DNA constructs with less than around 220 bp of targeting flank were used (). Though this dichotomy with may reflect the limitations of transformation in the two parasites rather than recombination, linear regression analysis of the (albeit limited) data set from Papadopoulou and Dumas is consistent with a MEPS of 190 bp in (Supplementary Figure 2), closer to the size predicted in the two other eukaryotes. Perhaps, therefore, has evolved to allow homologous recombination to operate, with reduced efficiency, on shorter substrates. More analysis will be needed to examine this, but it could be the result of differences in the activity of RAD51, differences in the factors that mediate RAD51 function, or a more active short-sequence recombination pathway. The second conclusion from this work is that contains an MMR-independent, or at least an MSH2-independent, pathway for homologous recombination that has not previously been described. On 5% diverged substrates in the size range 50–175 bp, and on 11% diverged substrates in the range 50–200 bp, this pathway appears to assume considerable significance, but is subordinate to MMR-dependent recombination on longer substrates (450 bp) containing these levels of sequence divergence. This argues that the reaction assumes a greater role during homologous recombination as substrate length decreases. Characterization of transformants by antibiotic resistance and Southern analysis suggests that the considerable majority of recombination, at all sequence lengths, occurs by homology, suggesting that the MMR-independent reaction is capable of precise integration. We have described a DNA repair process in based on the joining of DNA molecules using very short stretches of homology, typically 5–15 bp in length (), and similar reactions have been described in other eukaryotes (,) and potentially in bacteria (). We suspect that this is not the pathway responsible for most MMR-independent recombination, for two reasons. First, examination of the transformation rates of the 5% diverged substrates 175 bp or smaller, and of the 11% diverged substrates 200 bp or smaller, suggests that transformation by the MMR-independent pathway can be as efficient as 1 × 10 transformants ·cells, since these are conditions in which it is likely that most integrations occur by this route. This is considerably higher than the maximum transformation rate (0.1 x 10) described for the microhomology reaction in wild-type (). Second, in most reactions where we have demonstrated that microhomology guides construct integration, this is associated with visible karyotype changes (), suggesting that the reaction may be an end-joining process that exploits random DNA breaks, leading to genomic rearrangements. In contrast, such rearrangements are very rare in the extensive numbers of transformation events we have characterized here, including the conditions in which MMR-independent recombination would be expected to be prominent. The presence of a putative MMR-independent recombination reaction in is reminiscent of pathways described in , at least superficially. Although Rad51 is a central factor in DNA strand exchange during homologous recombination, some reactions in can occur in its absence; reviewed in (). Rad52 appears to be essential for nearly all recombination in yeast, suggesting it contributes to both Rad51-dependent and -independent recombination. A relative of Rad52, termed Rad59, was identified in a search for factors required for recombination in mutants (), and the suggestion that both proteins contribute to Rad51-independent reactions is supported by findings that each can catalyse strand annealing (). Rad59-dependent recombination requires less sequence homology than the Rad51-dependent reaction (), and displays a lower degree of regulation by Msh2, and hence MMR, during recombination of diverged DNA sequences (). A prediction, therefore, is that that the MMR-independent recombination we see in is RAD51-independent. We have not tested this directly, but some data appear consistent with this hypothesis. Linear regression analysis of the data presented here on substrate length and transformation rate is best accounted for by two lines of best fit, consistent with two pathways operating (). Above around ∼150 bp a pathway predominates that has a MEPS of ∼100 bp, whereas below ∼150 bp a distinct pathway with a MEPS of ∼31 bp. Interestingly, the longer MEPS is consistent with previous estimates of RAD51-dependent recombination acting on long (450 bp) substrates (). Despite the considerable inefficiency of homologous recombination on substrates shorter than the MEPS, the assay we have used here shows that such recombination can occur, as has been found in yeast (), mammalian cells () and (where recombination of 25 bp substrates is predominantly RecA-independent)(). Indeed, 80% of the transformants that arose from the constructs with only 25 bp of homology were Hyg (), demonstrating that they had integrated by homologous recombination. In , recombination of short substrates around 29–40 bp is not only Rad51-independent, but mutation of Rad51 increases the reaction efficiency (). In , transformation of constructs with very similar-sized regions of homology (24 bp) occurs at essentially the same frequency in −/− and wild-type cells, suggesting the action of a RAD51-independent pathway (). Despite these similarities, it is unclear what factors would catalyse RAD51-independent homologous recombination in . Rad52 and Rad59 belong to a superfamily that is not conserved universally (), and homologues of both proteins are either absent from the genome (), or are sufficiently diverged in sequence to have escaped detection. It is therefore possible that contains recombination factors, thus far unidentified and distinct from RAD51, that can perform the functions equivalent to the pathway in which Rad59 and Rad52 act in . Identification of an MMR-independent pathway of homologous recombination could be important in understanding antigenic variation and genetic variability in . Although antigenic variation is impaired in RAD51 mutants, gene conversion reactions can still be catalysed (). It is plausible that the MMR-independent pathway, if it is RAD51-independent, could explain these residual switching events. There is also no evidence that switching is influenced by MMR (), which appears to be at odds with the requirement for RAD51, since this pathway should be suppressed by MMR. However, although assays for recombination in yeast mutants reveal functions for Rad59 in defined recombination pathways, it is very likely that in wild-type cells Rad51, Rad59 and Rad52 actually act together (,). VSG switching may therefore occur by a specific recombination pathway that requires RAD51, but is directed towards an MMR-independent route by unidentified factors that act like Rad59. Indeed, this could explain the ability of VSG switching to use rather short and dissimilar DNA sequences as substrates, such as the 70 bp repeats upstream of genes. Furthermore, although nothing is known about the genetic requirements of segmental gene conversion involving the pseudogenes, it is clear that genes share very little primary sequence homology (<25% identity between encoded amino acids in >95% of the repertoire; L.Marcello and J.D.Barry, personal communication). It is therefore possible that an MMR-independent reaction, active on short stretches of homology, would be involved. It is important to note that to date we have only assayed homologous recombination at one interstitial site. Although some work has suggested that recombination occurs at equivalent efficiencies in different genomic locations (), it is important to examine this systematically and to assess the pathways used. For instance, it is clear in other organisms that other factors, such as transcription (), can influence recombination, and some work has suggested that different pathways of recombination are active in interstitial relative to subtelomeric environments (). Moreover, recombination can contribute to the maintenance of telomeres (), and the pathway(s) that acts in this regard in is unknown (). In any model in which MMR triggers the rejection of recombination between mismatched DNA substrates, heteroduplex DNA must form during the strand exchange step. In MMR+ cells in which recombinants escape rejection, such heteroduplex is likely to be repaired, whereas it should be visible in MMR− cells. An MMR-independent pathway may or may not generate heteroduplex, since it could avoid MMR surveillance by using short, perfectly matched sequences or because the strand exchange mechanism between mismatched molecules avoids triggering MMR. We cannot readily distinguish these possibilities through sequencing the integrated DNA in these experiments. In no transformant, even in MMR− cells, did we find evidence for the DNA having a mixture of construct and genomic sequence at the mismatched residues, which would arise if heteroduplex formed and was not repaired prior to replication, as illustrated in . Instead, we saw a predominant pattern of construct sequence to one side of the integrated marker and sequence on the other. This is most readily explained by independent strand invasion and annealing between a 3′ single strand on each side of the marker and the genomic target, suggesting that the mechanism of homologous recombination during targeted gene replacement in is equivalent to that in yeast and mammals (,). The one exception we found, where all the sequence is -derived, occurred in wild-type cells (clone 9, ). This could result from MMR of the heteroduplex in favour of the recipient, DNA. Alternatively, it could indicate that construct integration occasionally occurs by annealing of a single strand encompassing both arms of the construct, followed by mismatch repair; such a mechanism has been seen in yeast (), though appears to be rare (). One distinction between targeted gene replacement in and is revealed here: mutation of in either increased the frequency of transformation or had no effect, depending on substrate length, whereas the same mutation in results in reduced integration rates (). This may indicate that the parasite MSH2 protein does not promote recombination in the ways it has been found to do elsewhere (). The lack of clear evidence for heteroduplex DNA is perplexing. In wild-type cells, the lack of mixed sequence at mismatched residues could be explained by MMR that acts following construct integration and is directed towards one or other DNA strand by, for instance, the direction of replication (). However, the same pattern was seen in the −/− transformants. In addition, we did not see any striking difference in this pattern as construct length changes, which could arise through a shift in the substrate requirements of the recombination pathways being used. It is conceivable that in all these conditions most strand exchange is limited to perfectly matched sequences between the construct and target, and crossovers then occur on these intermediates to incorporate both strands of the construct DNA. However, a number of findings argue against this. First, we did not observe construct sequence on both sides of , and it is not clear how crossover integration could be directional in this way. Second, it is difficult to explain by this model the rather common appearance of transformants in which tracts of construct sequence were interrupted by patches of sequence in the ‘left’ arms (clones 1, 15, 41, 23; ). Finally, the length of homology that mediates strand pairing would have to have been very short in some cases (e.g. 15 bp in clone 41, and 12 bp in clones 6, 35 and 30; ). For these reasons, we suggest that heteroduplex does form during homologous integration and that it is repaired by short-patch MMR. This would explain the predominant pattern of sequence we see in the integrated DNA (). Furthermore, it most readily explains transformants in which construct-derived and -derived residues were found in close proximity. Such a situation is consistent with repair of two putative mismatches in different directions (in favour of the invading and recipient strands), which is distinct from long-patch MMR, where a mismatch triggers excision of extensive regions of DNA to allow repair (). However, given that the pattern of sequence in the transformants was predominantly with construct sequence to the ‘left’ of and sequence to the ‘right’, and that many tracts were continuous stretches of either construct or residues, it seems likely that some feature(s) biases the direction of short-patch repair. The nicks that are present during strand invasion cannot explain this, as repair would then be in the same direction for each end of the construct. Replication or transcription seems like a reasonable alternative, though this cannot be absolute, since we see discontinuous tracts in a number of transformants, and one clone has reversed the predominant pattern (clone 23; ). Short-patch MMR has not been described in , but has been found in other organisms (including during recombination), though the molecular machinery is still being described (,). A consequence of the lack of identifiable heteroduplex DNA in this study is that we cannot determine how extensive the lengths of strand exchange intermediates are in these experiments, nor whether this differs in wild-type and −/− cells, as has been reported in yeast (,). In addition, this approach does not allow us to test our hypothesis that the putative MMR-independent pathway we propose utilizes shorter substrates. However, given the existence of mixed tracts of construct and sequence at all substrate lengths, it is likely that heteroduplex is formed in all conditions. A footnote in this study, which was not appreciated in our previous analyses (), is that we find significant levels of duplication of the locus in this assay. The available evidence suggests that this is due to the generation of additional copies of in the tubulin array during growth of the clonal HTUB cell lines, rather than being a consequence of the targeted gene replacement. This ‘spread’ of a resistance marker in tubulin has been described previously (), and shown to be due to unequal sister chromatid exchange. At least in our experiments this appears to be rather frequent. Though this may be a result of the antibiotic exerting a selective pressure for increased gene product, it is notable that high frequencies of allelic gene conversion have been described in () and that considerable variation in repetitive sequences has been described between strains (). Our data also hint that impairment in MMR may enhance amplification, perhaps because MMR-mediated suppression of recombination between divergent sequences is alleviated, leading to greater rearrangements in the genome. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Eukaryotic gene transcription is controlled by a network of transcription factor (TF) proteins (,). TFs bind to specific DNA elements near transcription start sites, and through cooperative interaction, guide Polymerase-II complex to the transcription start site. Identification of -regulatory elements for the TFs is an important first step towards deciphering regulatory networks. This, however, remains a practical challenge because TFs often bind to highly diverse sequences resulting in degenerate binding models or motifs, and searching for putative binding sites using these degenerate motifs results in too many false positives. It is now well established that regions in the genome that have been conserved over long evolutionary periods are more likely to be functional (). Fortunately, such highly conserved regions make up only a small fraction of the genome (). Thus by restricting the search for putative binding sites in evolutionarily conserved sequences, one can drastically reduce false positives. This is exactly the premise underlying the, now well established, approach of (). However, for a genomic region to be functional, evolutionary conservation is neither necessary (,) nor sufficient (). Besides conservation, what are other important characteristics of functional elements? The regulation of gene transcription depends on interactions among transcription factors and the polymerase. This imposes location constraints on the corresponding DNA elements. For example, several elements occur at a specific distance relative to the transcription start site (TSS) (). Additionally, several elements occur in the same promoter with restricted spacing between them. For example, in the adenovirus 2 promoter, increased spacing between the GC-box and the TATA-box diminishes transcription significantly (). There are other examples of such positional and spacing restrictions (). Previous works have exploited the co-occurrence of promoter motifs to predict interacting TFs (,), to model expression regulation (,), and to detect regulatory modules (,), and some of these works impose specific distance constraints between co-occurring motifs. Positional constraints provide distinguishing characteristics of -regulatory elements in addition to evolutionary conservation, but have not been systematically exploited for motif discovery. Here, we show that in human promoters a large fraction of known motifs exhibit significant positional constraint and a large number of motif-pairs exhibit significant inter-motif distance constraint. The target genes that have position-specific motifs or the distance-specific motif-pairs tend to be co-expressed and have similar functions. A large majority of these positionally constrained motifs are not conserved between human and mouse; this underscores the importance of positional constraints in discovering -regulatory motifs. Finally, to discover novel motifs, we assess the position and distance specificity of all words (7 bases long) and word-pairs in human promoters that do not overlap a known motif. After clustering of similar motifs, this resulted in 168 position-specific novel motifs and 3708 distance-specific pairs involving a novel motif. Several of these are highly correlated with specific expression and function of the target genes. We extracted 600 bp human promoter sequences (+500, −100) corresponding to 30 927 transcription start sites from DBTSS version 5.2 (). We also extracted the human–mouse conservation for these regions from UCSC's axtNet database (UCSC hg17 release). TRANSFAC v8.4 () describes 546 vertebrate TF positional weight matrices (PWM). Often PWMs corresponding to evolutionarily related TFs are highly similar. To minimize the bias caused by this redundancy we clustered the PWMs based on their similarity and then retained 175 representative PWMs (methods). For these 175 PWMs, we scanned the 600 bp promoter sequences using our PWM_SCAN tool () with a stringent -value threshold of (chance expectation of one hit every 10 kb of human genome). Ten of the 175 PWMs did not have any match in our promoter set; our analysis is based on the remaining 165 TRANSFAC motifs. We used the Novartis tissue survey data () for gene expression profiles and GO () for functional annotation of genes. We only use the GO ‘biological process’, and to avoid non-specific biological processes we only include processes that are associated with at most 500 genes. This includes 99% of all the GO terms and eliminates non-specific GO terms. To quantify motif conservation, motif positional specificity and motif-pair distance specificity, we use a generic procedure. Let be the total occurrences of a motif (or motif-pair). Among these let be the number of ‘successful’ occurrences. An occurrence could be called ‘successful’ if for instance, it is conserved. Given the expected success rate , we assume a binomial distribution for the number of occurrences and estimate the Z-score as [ − ( × )]/[√ ( ×  × (1 − )]. A similar procedure was used in (). Precisely what we mean by ‘successful’ and how do we estimate depends on the context and will be described later. We say that a motif match is conserved if the mouse sequence aligned with the human site also matched the PWM with a -value ≤ . To estimate the expected conservation rate , we permuted each column of PWM to shuffle the nucleotide preferences in each position and generate a set of control matrices (five for each of the 165 TRANSFAC PWMs with a total of 819). We use these control matrices to obtain an overall expected conservation rate . shows the conservation Z-score distribution of 165 motifs. The core factors have high conservation Z-scores: CAAT-box (100.3), Sp1 (90.4), Oct-1 (10.1), TATA (8.4), etc. also shows the Z-score plot for the 819 randomized PWMs. Based on this plot, we arbitrarily categorize the motifs into three classes: (i) 27 highly conserved motifs (Z-score ≥ 8), (ii) 27 medium-conserved motifs (3 ≤ Z-score < 8) and (iii) remaining 111 non-conserved motifs. Here, we assess whether a motif preferentially occurs at a specific position relative to the transcription start site. Given the total occurrences of a motif and the subset of occurrences in a window (defined by the start position and the length), and the expected fraction of occurrences in the window, , we compute the Z-score. We compute the Z-score for windows of length 20 bp starting at each position in the 600 bp promoter region and retain the maximum value for each motif among all windows; we call this the Z-max. An important concern in estimating is the GC-composition of the motif and GC-composition of various parts of the promoter. We have experimented with three different controls (see Supplementary Data). Here, we report the results based on our most stringent control. To preserve the base composition of the motif, we randomly permute the columns of the PWM. To ensure a good representation we generate five permuted PWMs for a given TRANSFAC PWM. We estimate based on pooled occurrences of five permuted PWMs on the real promoter sequence. Note that is specific to each PWM and each window. It is easy to see that for instance, a G-rich motif is not going to differ from its permuted copies and thus will not have a high Z-score. At the risk of missing such cases, we decided to pursue this highly stringent control to minimize the risk of false discoveries. a shows the distribution of Z-max for the 165 motifs. As a negative control for the Z-max distribution, we repeat the above process on randomized promoter sequences. The randomized promoter is generated so as to preserve the base composition at each position along the 600 bp region. As above, we estimate based on pooled occurrences of five permuted PWMs on the randomized promoter sequence. As shown in a, a large fraction of TRANSFAC motifs occur in a position-specific fashion. As a reference, we show the positional Z-score distribution of three core motifs that exhibit high Z-max (b). Positional preferences of several core promoter motifs have been previously investigated. In our analysis, the GC-box binding TF, Sp-1 has a maximum Z-score at 66 bp upstream of the TSS, also observed in (). We found CAAT-box binding TF NF-Y to have the maximum Z-score at position 86 bp upstream of TSS. Xie . have reported a preferred position of 89 bp upstream (). TATA-box binding TF TBP is most frequent at ∼35–31 bp upstream of TSS (). However, the maximum Z-score in our analysis is achieved at 45 bp upstream of the TSS. Note that in contrast to frequency, we compute Z-score, which controls for the base composition. There is a peak of A + T frequency around 35–30 bp upstream of TSS (), which would lower the Z-score exactly at those positions, and thus our Z-score peak is slightly shifted. Based on these distributions, we define a set of 39 (23%) motifs to be position-specific (Z-max ≥ 5) (Supplementary Table T1) and another set of 38 motifs to be position-nonspecific (Z-max ≤ 3). Our numbers are consistent with the previous report where 25% of the known motifs were found to be position specific (). Furthermore, we found that the position-specific motifs also tend to be conserved. Among the 39 position-specific motifs, 42% are highly conserved (conservation Z-score ≥ 8) whereas among the 38 position-nonspecific motifs, only 3% are highly conserved (chi-square -value = 3 × 10). As expected, several known core factors like CAAT box (bound by NF-Y), Muscle TATA box, TBP, Sp1, etc. show a very high position specificity (Supplementary Table T1). We investigated whether the presence of a position-specific motif in the gene promoter is correlated with the gene's function or expression pattern. We assessed the functional coherence of a set of target genes based on GO annotation using the Ontologizer tool (Robinson ., 2004). For each target gene-set, we also randomly select the same number of genes and subject them to same analysis. To assess how significantly the position- specificity of a motif correlates with the target gene function, we used three criteria to select the target gene-sets: (i) genes that contain position-specific motifs at specific position range where Z-max is achieved, (ii) genes that contain position-specific motifs at locations other than Z-max positions and (iii) genes that contain position-nonspecific motifs at any location. Each of the gene-sets also has a corresponding random gene-set of the same size. The GO Ontologizer tool compares a given gene-set with several functional categories for significant overlaps. To control for multiple testing, we pool the GO -values for all gene-sets and their corresponding randomized gene-sets for the three criteria and based on the pooled set of -values we estimate a -value cutoff corresponding to a false discovery rate, FDR ≤ 5% (). shows the fraction of motifs under each of the three criteria whose target genes show significant GO association and also the average number of associated GO processes. These numbers are also shown for the matched random gene-set. Additionally, among the motifs that do show a significant GO association, the table also shows the fraction that is conserved. As shown in , GO association is the greatest for position-specific occurrence of motifs and some of these could not be detected using conservation criterion alone. We have listed the significant GO associations under criterion A in Supplementary file ‘Motif2GOAssociation’. These include mRNA processing and metabolism, protein localization, transcription, etc. Next we investigated whether the gene targets of position-specific motifs are differentially expressed in specific tissues. For each of the 79 tissues from the Novartis dataset, we assessed using Wilcoxon rank sum test, whether a target gene-set had differential (up or down) expression relative to all other genes. Each pair of gene-set and tissue results in a -value and based on pooled -values as before we estimate a cutoff corresponding to FDR ≤ 1%. We consider a motif to have differential expression if in at least one of the 79 tissues the target genes are differentially expressed with a significant -value. Approximately 1% of the random gene-sets show significant differential expression at a FDR cutoff of 1%. Thus if 1% of the -values are significant by chance, we expected ∼55% (1 − 0.99) of the motifs will show differential expression in at least one tissue by chance alone. The relative-enrichment number (column 6 of ) is the ratio of the ‘actual% of motifs that show differential expression’ to 55. The average number of tissues in which the gene-set is differentially expressed is also of interest. shows the fraction of motifs under each of the three criteria whose target genes are differentially expressed, as well as the average number of tissues. These numbers are also shown for the matched random gene-set. Additionally, among the motifs that are differentially expressed, the table also shows the fraction that is conserved. Similar to our conclusions based on GO analysis, the tissue-specific differential expression is most prevalent among the position-specific occurrence of motifs and many of these motifs could not be detected using conservation criterion alone. Only about half of tissue-associated motifs are conserved. The top five most significant tissues under criterion A are SuperiorCervicalGanglion, Skin, PrefrontalCortex, PB-CD8 + TCells, Ovary and Atrioventricular node. Thus our results highlight the importance of position specificity of cis elements and that it should be used in conjunction with conservation to identify -regulatory motifs. Supplementary Table T1 lists 39 representative position-specific TRASNFAC motifs. Most of these are known to be involved in condition-specific regulation (as opposed to basal transcription). Next we assessed how often a motif-pair preferentially occurs at a specific distance from each other. Given the total number of occurrences of a motif-pair and the subset that falls within a distance range, defined by the minimum distance and the range, and the expected fraction, , we compute the Z-score. Here, the locations of individual motifs are irrelevant and only the distance between them is of concern. Much like position specificity analysis, we compute the Z-score for 50 bp ranges starting at each position and retain for each motif-pair the maximum value, the Z-max. Our control is analogous to that for the position specificity analysis, i.e. we randomly permute the columns of each PWM to generate the background matches. shows the distribution of Z-max for the 21 777 motif-pairs on real promoters and permuted PWMs on the random promoter sequences with positionally conserved GC composition. Two position-specific motifs will obviously manifest as distance specific. To unambiguously reveal distance specificity, we require that at least one of the motifs in the pair must be position nonspecific. also shows the distribution of Z-max for this reduced set of motif-pairs, and for comparison the motifs pairs where both motifs are position specific. From the reduced set (pairs with at least one non-position-specific motif), based on Z-max distributions, we define a set of 915 motif-pairs to be distance specific (Z-max ≥ 4) (listed in Supplementary Table T2) and another set of 865 motif-pairs to be distance nonspecific (Z-max ≤ 2). The distance-specific pairs involve 41 position-specific motifs and 38 position-nonspecific motifs. For comparison and subsequent analyses, we also included pairs where both motifs are position specific, using the same Z-score cutoff (Z ≥ 4) we get 410 motif-pairs. We performed the GO and tissue expression analysis using same procedure as that for the analysis of position-specific motifs above and the result summarized in is organized similarly to . The gene promoters containing distance-specific motif-pairs that are both position specific at a preferred distance have highest association with GO process and highest fold-enrichment for differential expression in tissues. By far, most of these motifs are more conserved compared with the gene-sets using other criteria. Particularly of interest, criterion B, that includes gene promoters containing distance-specific motif-pairs where at least one motif is non-position specific, show significant association with GO processes and differential expression in tissues. Moreover, the functional and expression associations for criterion B is stronger than that for criterion C, which includes all occurrences of the motif-pair, as opposed to only distance-specific occurrence in criterion B. Supplementary Table T2 lists the top 100 of the 7804 distance-specific motif-pairs where at least one of the pair is position nonspecific. In fact in 826 pairs, both motifs are position nonspecific. A cursory inspection of literature shows support for many of these. Muscle-specific motif (derived from actin promoter among other genes) has a Z-score of 11.63 with NF-Y which is known to form a complex on the alpha-actin-4 promoter (). Ets and Pax5 (BSAP) with Z-score = 6.91 are known to interact physically to regulate a B-cell specific promoter (). SREBP-1 and NF-Y show a Z-score of 5.6; expression of mouse gene ACBP is induced in hepatocytes by SREBP1 and this induction also requires a NF-Y-binding site (). We have listed the significant GO associations under criteria A and B in Supplementary file ‘Motif2GOAssociation’. These include protein localization/transport, regulation of lipid metabolism, biopolymer metabolism, RNA processing/metabolism, regulation of transcription, negative regulation of biological process, etc. The top five most significant tissues under criteria A and B include several CD cells, dendritic cell, T cell, ColorectalAdenocarcinoma, PrefrontalCortex, OccipitalLobe, Subthalamicnucleus and Hypothalamus. This is consistent with previous report by Xie ., where the two main groups of motif-associated tissues were found to be brain and immunity related (). For a majority of human transcription factors, their DNA-binding specificities are not known. motif discovery thus remains important in analyzing transcriptional networks. Our analysis of known motifs shows that position and distance constraints, in addition to conservation, are important attributes of -regulatory motifs and thus can be used to detect novel motifs. To ensure that the motifs we detect do not correspond to any of the 175 representative TRANSFAC motifs, we start by masking all positions in all human promoters that matched any of the TRANSFAC motifs based on -value threshold (see above). We then extract all 7-mers from the unmasked portion of the promoters, while allowing for at most two bases overlap with the masked portion on either side. We then cluster these 7-mers; two 7-mers were clustered if either they had at most 1 mismatch or they had 6 identical bases (after 1 base shift) including the reverse complement. All 7-mers within a cluster were aligned and a PWM was derived from each cluster. The set of 661 PWMs were then subject to same analysis as the TRANSFAC motifs to compute their conservation, position specificity and distance specificity. Seventy four of these 661 were conserved (Z ≥ 8) and 168 were position specific (Z ≥ 6) and 3708 pairs were distance specific (see below). We report a subset of 74 novel conserved motifs in Supplementary Table T3 along with their corresponding position and distance specificity properties. The Logos of the top 10 conserved motifs are shown in Supplementary Table T3a. We found that a large fraction of position-and distance-specific motifs are conserved and this fraction increases with increasing position or distance specificity. and show the relationship between conservation and respectively the position and the distance specificity. We categorized the 661 7-mer motifs into 168 (25%) position-specific motifs (Z ≥ 6) and 123 position-nonspecific motifs (Z ≤ 3). In a previous analysis, Xie . have reported 35% of novel motifs to be position specific (Supplementary Table T4 shows these 168 motifs and associated information). The Logos of the top 10 position-specific motifs are shown in Supplementary Table T4a. We applied the previously described three criteria to select the target gene-sets and shows the results of the GO and tissue expression analysis. As was the case for the TRANSFAC motifs, very few motifs show strong GO association, but a large fraction show a strong association to tissue expression. The trend across the three criteria is, however, similar to that for the TRANSFAC motifs (compare and ). We have listed the significant GO associations under criterion A in Supplementary file ‘Motif2GOAssociation’. These include RNA/mRNA metabolism/processing, reproduction and pregnancy. The top five most significant tissues under criterion A are PrefrontalCortex, OccipitalLobe, Hypothalamus, Amygdala and Thyroid. This is consistent with what was reported in (). For distance specificity between these 661 7-mer motifs, we have a total of 285 762 motif-pairs that have at least 100 non-overlapping occurrences in all the promoter regions. A further requirement that at least one of the motifs be position-nonspecific results in 79 576 pairs to analyze. We categorized these motif-pairs into 3708 distance-specific motif-pairs (Z ≥ 5) and 9204 distance-nonspecific motif-pairs (Z ≤ 2). The distance-specific motif-pairs are provided in a Supplementary file. Because the numbers are very large, we randomly selected 200 pairs from each group and did the GO and tissue expression analysis. We applied the previously described four criteria to select the target gene-sets and shows the results of the GO and tissue expression analysis. Overall the results are consistent with that for the TRANSFAC motif-pairs (compare and ). We have listed the significant GO associations under criteria A and B in Supplementary file ‘Motif2GOAssociation’. These include regulation of transcription, regulation of cellular/biological process, protein/biopolymer modification, phosphorylation/phosphate metabolism, chromatic modification, reproduction, pregnancy, etc. The top five most significant tissues under criteria A and B include several CD cells, dendritic cell, B cell and prefrontal cortex. Comprehensive identification of genomic -regulatory elements is an important long-term goal. -regulatory elements are often characterized by evolutionary conservation. In order to reduce false positives, the search for -regulatory elements is traditionally restricted to evolutionarily conserved regions of the genome. However, both, lack of conservation among -regulatory elements, as well as lack of functionality among conserved elements has been previously reported (). Here, we have assessed the importance of two additional attributes of -regulatory elements—their position specificity from the TSS, as well as the spacing between them (). Several previous works have shown -regulatory motifs to be positionally constrained. Xie . have reported a large number of motifs in the human genome based on multiple genome comparison (). Although they have shown that some of the novel motifs are position specific relative to the TSS, this position specificity of the motifs was not used in the discovery process itself. In another work, motifs in were ranked based on the enrichment of specific spacing between them and were experimentally validated for their functionality in binding sites (). We have previously reported a promoter model that captures the position and distance specificities of motifs (). Other previous works have exploited the co-occurrence of promoter motifs to predict TF interactions and TF modules (). Here, we have explicitly and extensively assessed the importance of two attributes of -regulatory elements—position and distance specificity—independent of evolutionary conservation. Our results indicate that even though evolutionary conservation is the most important attribute of -regulatory elements, these additional attributes are important, especially to detect the species-specific elements. We emphasize that our work does not represent a novel motif discovery tool, in the traditional sense of the term. Traditionally, the term ‘motif discovery’ refers to identification of motifs potentially mediating the regulation of a set of co-expressed genes (). Our work, by exploiting the positional and distance constraints, however does identify a global set of motifs in the human gene promoters that potentially mediate transcriptional regulation. Although we have controlled for base composition, we have not controlled for dinucleotide composition, especially for CG dinucleotide frequency. Clearly, using more stringent di- and tri-nucleotide controls will result in increased specificity in motif detection, however at the risk of decreased sensitivity. The compositional bias in certain genomic regions may have been preserved over evolutionary period precisely for the maintenance of -regulatory elements and thus using the extremely stringent local composition as a control will result in failed detection of these -regulatory motifs on statistical grounds. We have mentioned the example of TATA box earlier, where, because of a local (A + T) frequency peak at ∼35 bp upstream, where TATA box is most abundant, we detect the position of most specificity at a slightly shifted position of 45 bp upstream. The CG dinucleotides are of special concern because of their association with DNA methylation and transcriptional regulation (). Several of the detected position-specific motifs have one or more CG dinucleotides; indeed there is an enrichment of CG-containing motifs among the position-specific motifs, both in known and novel motifs. Similar to previous observation, these position-specific motifs mostly occur in 100 bp upstream of the TSS (). We have performed a number of cautionary analyses to ensure that CG-containing motifs are not detected simply because of lack of control for CG dinucleotides. We have summarized these analyses in the Supplementary material. For both, known and novel motifs, our analysis shows the importance of position specificity in determining the functional and tissue association of the target genes. These findings are consistent with the shifting view of core promoter as an active participant in the regulation of eukaryotic gene expression. Besides the top five tissues mentioned earlier that are enriched for targets of position-specific motifs, fetal brain and fetal liver are also among the significant tissues; this is also true for distance-specific motif-pairs. Although our result shows a greater tissue enrichment for position- and distance-specific motifs, the specific tissues revealed by our analysis may be over-interpreted; permutation-based test for tissue enrichment may be more stringent. The relative enrichment of tissue-associated motifs is much higher when the position-specific motifs occur at their preferred positions (compare column 6 for rows A and B in and ) thus underscoring the importance of position specificity for -regulatory motifs. Furthermore, although there is a strong correlation between position specificity and the conservation (), more than half of the position-specific motifs that drive tissue-specific expression are not conserved (last columns in and ), thus underscoring the importance of position specificity for motif discovery independent of conservation. Similar conclusions can be made regarding the importance of distance specificity in the discovery of -regulatory motifs. Distance-specific motif-pairs that are comprised of position-specific motifs (row A in and ) by far show the most functional association, tissue association and conservation. In fact there is a correlation between distance specificity and conservation (). Nevertheless the distance-specific motif-pairs that include non-position-specific motifs (row B) show a comparable tissue association and slightly lower functional association, and yet a large majority of these are not conserved and would be missed by a conservation-only based approach. Thus, based on an unbiased comprehensive analysis we have shown the importance of position and distance specificity in discovering -regulatory motifs beyond the use of evolutionary conservation alone, which is likely to miss species-specific -regulatory motifs. We have previously reported an information-theoretic approach to compute the similarity between a pair of PWMs (). Here, we introduce the approach briefly and provide the details in the Supplementary Data. To compute the similarity between a pair of PWMs, we use a symmetric derivative of the standard relative entropy measure (). This measure is transformed into a Z-score, and eventually into a -value, based on empirically derived distributions. We allow for shifts between the PWMs and our measure accounts for the PWM widths. Using an appropriate -value threshold for the pairwise similarity, we then compute clusters of similar PWMs; we use bi-connected components (in contrast to single-linkage clusters) as our clusters. Finally, for each cluster, we select the median PWM as the cluster representative. See Supplementary Data for further details. We use Z-score to quantify motif conservation, motif positional specificity and motif-pair distance specificity. Assuming a binomial distribution, the Z-score is defined as: For both, position specificity and distance specificity, gene-sets are generated for particular motifs or motif-pairs according to different criteria (see Results Section). For each of these gene-sets, a matched random control set of genes is generated with equal number of genes. These sets of genes are then fed into GO Ontologizer tool () and -values are generated for significance of motif to a GO-term association. A 5% FDR cutoff are applied to these -values. We only used the biological process GO terms that have at most 500 genes to avoid ubiquitous classes. Similar to GO analysis, real and random set of genes are generated for particular motifs or motif-pairs. For each gene-set, and for each of the 79 tissues in the GNF tissue survey data (), we test whether the genes in the set are differentially expressed in the tissue using Wilcoxon rank sum test. Each pair of gene-set and tissue results in a -value and based on pooled -values, we estimate a cutoff for FDR ≤ 1%. We consider a gene-set differentially expressed, if it is differentially expressed (FDR ≤ 1%) in at least one of the 79 tissues. All 7-mers motifs that are not covered by a TRANSFAC match are ranked according to their conservation Z-score with highly conserved motifs appearing nearer to the top of the list. We then walked down the list and clustered current motif with motifs we have already encountered according to the following criteria. Two 7-mers were clustered if either they had at most 1 mismatch (including the reverse complement) or they had 6 identical bases (after 1 base shift). All 7-mers within a cluster were aligned and a PWM was derived from each cluster. There are three additional files. They are as follows: File1: PositionalMotifs_AdditionalFile, this includes Supplementary results. File2: Motif2GOAssociation, this includes the significant GO associations for known and novel position/distance-specific motifs. File3: DistanceSpecificMotifs, this includes all novel distance-specific motif-pairs. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R o n l i n e .

Phage Ø29 protein p4, in synergy with viral protein p6, effects the transcriptional switch that divides bacteriophage Ø29 infection into early and late phases (,). Protein p4 binds to two regions of DNA, each containing two binding sites in tandem, and each binding site (named sites ) consists of imperfect inverted repeats. The consensus binding sequence is: 5′-AACTTTTT-15 bp-AAAATGTT-3′ (see ; ). Site 3 is the highest affinity one, followed by sites 1 and 2. Site 4, with the most imperfect inverted repeat sequence, is the lowest affinity-binding site. Protein p4 crystallizes as a dimer, and each monomer has an α/β fold and a novel N-terminal β-turn substructure, the N-hook, for DNA interaction (). In the p4–DNA co-crystal, the DNA presents a B conformation with locally narrowed and widened minor grooves. Each p4 monomer hook intrudes into the DNA major groove where Gln5 and Arg6 establish hydrogen bonds with the DNA bases T ± 15 and G ± 13, respectively (). In addition, 14 contacts between the protein dimer and the DNA phosphates are observed. Substitutions of Gln5 and Arg6 with Ala provided evidence that the Arg6-G ± 13 interaction is required for p4-binding site recognition through a direct readout mechanism. Since B-DNA sequences have dissimilar distribution of hydrogen-bond donor and acceptor groups on their bases, but similar sugar–phosphate backbones, p4–phosphate backbone contacts seemed insufficient, , to explain p4's affinity and sequence specificity. Therefore, recognition of other aspect of the sequence structure (indirect readout) could also account for p4 sequence specificity. It is known that indirect readout of DNA sequences plays a role in determining the stability and/or specificity of many protein–DNA complexes, but precise knowledge of how the sequence modulates DNA structure is limited. Although DNA–protein crystal structures identify atomic interactions, they do not always reveal the specificity of those interactions, or the induced conformational changes in the protein and DNA. Here, we studied the structural stability of the protein and DNA in the p4–DNA complex, and the relative roles played by direct and indirect readout in sequence-specific recognition by protein p4. Molecular dynamics (MD) simulations () represent a highly suitable tool to explore the dynamic behaviour of p4 and its binding site, and the mechanism of these interactions. Therefore, we analysed p4–DNA interactions by MD simulations, and by focussing on the interaction of p4 to sequence-modified binding sites. Bacteriophage Ø29 p4 protein was expressed and purified as described (). The oligodeoxyribonucleotides (Isogen) used are shown in Figure S1. To obtain each 60-bp double-stranded DNA, two complementary oligodeoxyribonucleotides were used. One of the oligodeoxyribonucleotides from each pair was 5′-end labelled using [γ-P] ATP and T4 polynucleotide kinase (). The labelled strand was purified from unincorporated [γ-P] ATP through a mini Quick Spin Column (Roche). Complementary oligodeoxyribonucleotides were annealed to yield double-stranded DNA by mixing labelled and unlabelled oligonucleotides in a 1:10 ratio in 80 µl of 25 mM Tris-HCl (pH 7.5), 200 mM NaCl, heated for 2 min at 90°C, and allowed to cool gradually (14–24 h) to 20°C. Band-shift assays were performed with a fixed amount of DNA and increasing concentrations of p4 for each of the DNA substitutions. From the data obtained, experiments were carried out using those p4 concentrations that give rise to a linear response with each DNA. Binding reactions (20 µl) contained labelled DNA (5 fmol), 25 mM Tris-HCl (pH 7.5), 10 mM MgCl, 100 mM KCl, 0.5 µg of poly[d(I-C)], 1 µg of bovine serum albumin and protein p4 at the concentrations indicated in each figure. Incubation was for 15 min at 4°C, and the reaction mixture was loaded onto a non-denaturing 6% polyacrylamide gel after addition of 4% (v/v) glycerol. Electrophoresis was performed at 4°C at 20 mA/gel. Gels were dried, and the label present on free DNA and p4–DNA complexes, quantified in a GS-710 Imaging Densitometer, gave the total amount of DNA; the amount of DNA complexed with the protein was calculated as a fraction of total DNA. MD simulations were performed using the PMEMD module of AMBER8 and the parm 99 parameter set (). The X-ray structure (PDB code 2FIO) was used for the MD simulation. The system includes the two p4 monomers (except residues 114–124 corresponding to the flexible helix α4) and the DNA molecule with the following sequence: 5′-TAACTTTTTGCAAGACTTTTTTATAAAATGTTGA-3′. Independent simulations were carried out on the p4–DNA complex (Native complex) and the unbound form of the DNA derived from the structure of the complex but devoid of the protein (free DNA). A third simulation was carried out on the p4–DNA complex where the Arg6 on each monomer of the protein was substituted to Ala (R6A complex). An adequate number of Na ions were added to neutralize the net negative charge of the systems (57 Na ions in the native system, 65 Na ions in the free DNA system and 59 Na ions in the R6A system). The counterions were placed in a shell around the system using a Coulombic potential in a grid. The neutralized complexes were then immersed in a truncated octahedron solvent box keeping a distance of 12 Å between the wall of the box and the closest atom of the solute. The counterions and the solvent were added using LEAP module of AMBER. Initial relaxation of each complex was achieved by performing 10 000 steps of energy minimization using a cut off of 10.0 Å. Subsequently, and to start the MD simulations, the temperature was raised from 0 to 298K in a 200-ps heating phase, and velocities were reassigned at each new temperature according to the Maxwell–Boltzmann distribution. During this period, the positions of the Cα atoms of the solute were restrained with a force constant of 20 kcal mol Å and the Watson–Crick bonds between all the base pairs of the DNA were constrained. This constraint with a force constant of 10 kcal mol Å was maintained during the equilibration steps to impede a spurious disorganization of the structure during the heating of the system from 0 to 298K. During the last 100 ps of the equilibration phase of the MD, the force constant was reduced stepwise until 0 except for the distance corresponding to the Watson–Crick hydrogen bonds between the first and the last base pairs of the DNA molecule which was maintained in order to mimic the cooperative stabilizing effect of base pairs present at both DNA ends that are not included in our system. This constraint was maintained during the productive phase of the simulations. The SHAKE algorithm was used throughout to constraint all hydrogen bonds to their equilibrium values so that an integration time step of 2 fs could be employed. The list of non-bonded pairs was updated every 25 steps, and coordinates were saved every 2 ps. Periodic boundary conditions were applied and electrostatic interactions were represented using the smooth particle mesh Ewald method with a grid spacing of ∼1 Å. The trajectory length was 5 ns for all the complexes. Analysis of the trajectories was performed using the CARNAL module of AMBER 8. The fact that only two bases, G13 and T15, on the target site are bonded by protein p4 suggests that sequence-specific recognition by the protein may involve factors other than direct amino acid side chain–base interactions. The protein contacts backbone phosphates neighbouring three A-tracts (). A-tracts have special properties, including bifurcated hydrogen bonds, high propeller twist and buckle of base pairs, suitable for protein recognition (). To elucidate the molecular basis of p4 selectivity for its cognate sequence, we studied the binding affinity of p4 to site 3 sequences containing base modifications at the A-tracts (). The ability of p4 to bind to modified site 3 bearing C·G base pairs substituting each A·T base pair on the external A-tracts adjacent to G ± 13 was analysed by band-shift assays (). Base-pair substitutions at positions ±8 did not affect p4 binding, while the amount of complex formed was reduced by 3- or 4-fold when the base pair was modified at positions ±9 and ±12, respectively. Interestingly, p4 binding was drastically impaired when C·G substituted the A·T base pairs at the centre of the A-tracts (positions ±10 or ±11); a faint band of p4–DNA complex was produced with the DNA modified at positions 11 and no p4–DNA complex was detected when the modification was at position 10. Thus, the affinity decreased 100-fold for substitutions at positions ±11, and >200-fold when the substitution resided at position ± 10. Therefore, the A·T ± 10 and the A·T ±11 base pairs are critical for p4 binding. Since there are no bases or phosphates contacted by the protein at these positions, the effect should be due to specific sequence recognition by indirect readout. The failure of p4 binding to site 3 bearing C·G substitutions at positions ±10 and ±11 could be explained if, when present, the guanine amino group exerts a detrimental effect on p4 binding through the formation of a third hydrogen bond with the pairing cytosine, which renders the base pair less deformable than the A·T base pair. Alternatively, the 2-amino group of guanine in the minor groove may sterically or electrostatically interfere with p4–DNA interaction. If the presence of the amino group in the minor groove and the increase in the number of hydrogen bonds between the bases interfered with p4 binding, the affinity of p4 for DNA would be drastically decreased upon substitution of adenine with the base analogue diaminopurine (DAP). DAP substitution at positions ±10 reduced p4 binding >200-fold (). Conversely, removal of the extra hydrogen bond and the amino group using C·I or T·A base pairs at positions ±10 greatly reduced the deleterious effect (). These latter substitutions decreased the affinity of p4 for the DNA by only 3-fold for T·A or 6-fold for C·I. The C·I base pair mimics the A·T base pair in the minor groove, while being identical to the C·G base pair in the major groove (,). Protein p4 has ∼8-fold lower relative affinity for site 2 than for site 3 (). There are two main differences at the A-tract level: site 2 has only three adenines in one of the external repeats, and its central A-tract is shorter with respect to the homologous site 3 A-tract (). To analyse further the role of the external A-tracts, and to elucidate the relevance of the central A-tract, we studied the binding affinity of p4 to the site 2 sequence modified at its A-tracts. The relative affinity of p4 for site 2 did not substantially increase by substituting the G·C base pair by A·T at position −9 (, Site 2B), while enlarging the central A-tract by substitutions at positions 0 and +1 (, Site 2A) yielded an amount of p4–DNA complex similar to that formed with the site 3 sequence. Therefore, and in agreement with the data from modified site 3, an A·T base pair is not critical at position −9. However, when present at positions 0 and +1, it enhances p4 binding. In order to gain insight into the solution conformation of the DNA and the induced conformational changes in the protein and DNA upon complex formation, we explored the p4–DNA complex by MD simulations. Simulations were carried out after standard structural equilibration. The three structures studied are referred to as the native complex (C), the p4R6A complex where the protein has been modified by changing Arg6 into Ala, and the free DNA dynamics of the site 3 sequence with the conformational modification of the p4–DNA crystal structure but devoid of p4. To study the functional difference of the sequences contacted by each protein monomer, we analysed p4 binding to site 3 with symmetric inverted repeats. shows that protein p4 has higher affinity (4-fold) for site 3 bearing the sequence 5′-AAAATGTT-3′ (B) than that for site 3 with the sequence 5′-AACTTTTT-3′ (A) in both external A-tracts. To corroborate this data, we substituted individually each guanine at positions ±13 by adenines. Even if p4-binding affinity was clearly diminished when either of these guanines was removed we found that the amount of complex formed with a site 3 devoid of guanine on the monomer A sequence (position −13; GA) was reduced 10-fold. In contrast, there was a greater than 30-fold decrease when the guanine was mutated on the monomer B sequence (position +13; GB). In agreement with this finding, single substitution of DAP at position +10 diminished >10-fold the relative affinity of the protein for DNA, while binding was diminished 5-fold when DAP was present only at position −10 (not shown). These data and the MD simulations trajectories support the conclusion that each protein monomer displays different binding entropies due to the slight asymmetry of the inverted repeats on site 3. Protein p4 is a transcriptional regulator that binds to four target sites with different relative affinities (). In this article, we studied the principles determining p4 binding specificity. A structural study of the p4–DNA complex by X-ray crystallography showed two direct base–amino acid interactions; Arg6 and Gln5 of each protein monomer contact G13 and T15, respectively. However (i) alanine substitution of each amino acid showed that only the Arg6–G13 interaction is required for p4 binding; (ii) substitution of either guanine at position 13 significantly reduced the relative affinity of p4 for DNA, and simultaneous mutation of both guanines abolished p4 binding ( and S4); (iii) MD simulation on the p4–DNA complex showed stable contacts along the trajectory between the Arg6 of both monomers and the guanines at position ±13, while the distances from the Nε of Gln5 of either monomer to the O4 of T15 are excessively large for hydrogen bonding in >90% of the trajectory. It is conceivable that the interaction of Gln5 with T15 in the X-ray structure resulted from the bend of the DNA toward the protein. Hence, specific sequence recognition by direct readout relies exclusively on Arg6–G13 interactions. Taking into account that two guanines on complementary DNA strands separated by 25 bp is a frequent event along the Ø29 genome while p4 binds specifically only to the four target sequences depicted in , other characteristics of the p4-binding site sequence should contribute, in addition, to the p4-binding specificity. Based on the X-ray structure, the p4–DNA complex with the sequence of site 3 includes three patches of amino acids and three A-tracts. Those A-tracts are present in the other p4-binding sites, as well as in the binding sites of the p4 homologous protein of Nf, a phage closely related to Ø29 (). Residues Thr4 and His10 contact the Ø29 site 3 DNA backbone precisely at one border of the two external A-tracts, with Tyr33 and Lys36 contacting the opposite border of the A-tracts. Substitution of A·T with C·G at each base pair of those A-tracts influences or abolishes the affinity of p4 for DNA. Our findings show that placing an amino group on the minor groove of the base pair located at the centre of the A-track (positions ±10) abolishes p4–DNA complex formation. The A·T base pair exhibits a higher intra-base pair propeller deformation than C·G or DAP·T base pairs, a consequence of two Watson–Crick hydrogen bonds between base pairs rather than three. DNA bearing the amino group in the minor groove is under-wound with respect to sites lacking this group, which may directly impact the DNA helical twist and twisting flexibility by mechanical occlusion. Either way, the main negative effect results in a less deformable base pair and, therefore, an increase in the energy required to distort the DNA. Furthermore, A·T or T·A base pairs at position 10 allow the DNA to be more easily bent in the direction of the minor groove that faces the protein at this position. Base substitutions at the A-tracts assert both sequence recognition by indirect readout and p4 over-winding of the minor groove at the centre of the A-tracks. Examination of the unbound and bound DNA conformations by MD provides clues as to the dynamics of the system. The absolute value of the RMSD for DNA in the p4–DNA complex, lower than 2 Å, can be considered small taking into account the large size of the simulated system. However, the RMSD values for DNA have a higher mean value (up to 8 Å) in the free DNA MD, demonstrating considerable DNA immobilization mediated by p4 binding. The RMSD values of the protein with respect to the starting value have a constant mean value of ∼1.5 Å along the trajectory, suggesting high stability of the protein moiety in the complex. On the other hand, comparison of the free and DNA-bound crystal structures of p4 does not reveal significant structural changes upon p4 binding with a C RMSD of 0.665 Å (). The DNA in the p4–DNA crystal structure is curved towards the protein due to local compression of the minor grooves at the areas where the protein contacts the backbone phosphates. In the MD simulation, the stable form of free DNA released from p4 does not show overall bending or a stable local narrow minor groove. In contrast, the DNA bound to p4 displays a stably narrowed minor groove between T10 and T14, T9 and G13, and T8 and A12 or T12 phosphates. Since in the absence of p4 the free DNA relaxes to a significantly different conformation, the structure of the DNA in the complex is not a stable or meta-stable substrate but a consequence of the induced conformational modification impressed by p4, which in turn does not reveal significant structural changes upon DNA binding. In fact, p4 binding does not require intrinsically bent DNA (). The MD data, the disadvantage of sequences containing disrupted A-tracts (this article) for p4 binding to DNA, and the drastic reduction on binding affinity of p4 Thr4Ala and Tyr33Ala mutants () indicate that the sequence-dependent characteristic of A-tracts provides an indirect readout by affecting the optimal complementarity both for amino acid–base hydrogen bonding and for precisely positioned interactions between amino acid and DNA phosphates. Hence, the stability of p4–DNA complex is a delicate balance of direct and indirect readout. The collective results of this study indicate that p4–DNA-binding stability is a consequence of p4-induced conformational modification of the DNA from the canonical B form through an indirect readout mechanism whereas the primary function of the DNA is its ability to acquire a conformation capable of enhancing positive interactions with its cognate protein. Indirect readout is less well characterized than direct readout. The affinity of protein for its DNA target by indirect readout relies on the fact that B-DNA exhibits a high degree of sequence-dependent structural variation () which includes recognition of aspects of DNA structure such as intrinsic curvature, topology of major and minor grooves, local geometry of backbone phosphates and flexibility or deformability. These mechanisms have been used to explain some aspects of the affinity of other prokaryotic transcriptional regulators for its target sequences; CAP seems to discriminate between a consensus pyrimidine–purine steps involving sequence effects on the energetic of primary-kink formation (,), while bacteriophage 343 repressor recognizes structural features on the central base pair of its target sequence (,), and water-mediated contacts are known as an important recognition tool in the trp-repressor operator system (,). The complexity of p4 interaction with its target sequence is compounded by the fact that, despite the 2-fold symmetry of the protein dimer, the protein uses pseudo-inverted repeats to interact with DNA, one monomer (monomer A) interacting with the sequence 5′-AAAAAGTT-3′ and the other (monomer B) with the sequence 5′-AAAATGTT-3′. Moreover, p4 is capable of recognizing other sequences with asymmetric inverted repeats (). We demonstrated that each inverted repeat provides different contributions to p4-binding affinity. Additionally, the p4–DNA MD simulation indicates that the hydrogen bonds of Tyr33, His10 and Lys36 with DNA are significantly more variable, both in residence time as well as bonding distance on monomer B than on monomer A, suggesting that interactions on monomer B present higher entropic stability. Since the pyrimidine–purine T/G step is more susceptible to deformation than the A/G step because it has a smaller amount of base overlap, the T/G bases of monomer B may permit a better orientation of the G + 13 for its interaction with Arg6 (Figure S3). Taking into account that DNA deformation is an important component of the driving force for p4–DNA association, the data presented here provide insights into how the role of DNA sequence may influence a directional binding for p4. We consider that: asymmetry is functionally required for p4–DNA interaction; the MD simulations suggest a net order of p4 binding to DNA sites with minor groove narrowing and curving of the helical axis; the distance from G − 13 to G + 13 is of ∼90 Å, while the distance from the Arg6 of one of the monomers to Arg6 of the other monomer of 75 Å is too short for simultaneous interaction of both p4 monomers at the inverted repeats. Therefore, we propose a zipper-binding model where one of the p4 monomers interacts first with the higher entropic stability sequence, 5′-AAAATGTT-3′ followed by local minor groove narrowing. This change in DNA conformation will allow interactions between basic residues of both monomers with the central A-tract, and the progressive bend of the DNA would permit the 5′-AAAAAGTT-3′ inverted repeat to interact with the hook of the second p4 monomer. Phage Ø29 early promoters A2c and A2b and late promoter A3 are coordinately regulated by a multimeric complex of viral proteins p4 and p6, which elicits the switch from early to late transcription repressing promoters A2c and A2b, and simultaneously activating promoter A3. In the multimeric complex, p4 dimers occupy binding sites 1 and 3, and p6 binds the sequence from sites 1 to 3 synergizing p4 binding (). Since protein p6 polymerizes from the A-track 5′-AAAAAGTT-3′ of site 1 to the A-track 5′-AAAAAGTT-3′ of site 3, a second functional implication for the site asymmetry could explain p6-mediated stabilization of p4 binding by anchoring the p4 monomer to the lower affinity A-track. This stabilization is critical for the regulation of the promoters and so required for the transition between early to late transition during Ø29 development. p p l e m e n t a r y d a t a a r e a v a i l a b l e a t N A R O n l i n e .

We show that microarray technology can provide rapid high-throughput assays for the identification of sequence-specific ss DNA–protein interactions. The majority of transcription factors recognize target sequences in duplex form. However, single-stranded regions can be induced by torsional stress of double-stranded DNA, allowing single-stranded nucleic-acid-binding proteins (SNABPs) access to their binding sites (). SNABPs have been shown to bind with high affinity, non-specifically () and specifically (), to ss DNA, which has been shown to regulate gene expression both positively and negatively (,). Gene expression can also be regulated on a translational level, by SNABP binding to mRNA (). Genome sequencing has allowed SNABPs to be identified and characterized for a range of eukaryotic and prokaryotic organisms. To understand how binding of SNABPs to ss nucleic acids regulates transcription and translation, the regions of sequence specificity must be identified. Many techniques including electrophoretic mobility shift assay (EMSA) (), nitrocellulose-binding assays (), Southwestern blotting (), phage display (), UV cross-linking () and X-ray crystallography () were developed to study sequence-specific ss DNA–protein interactions. Available techniques including, fluorescence measurements (), polymerase chain reaction (PCR), fluorescence resonance energy transfer (FRET) combined with a DNA foot-printing assay (), surface plasmon resonance (SPR) and fluorescence polarization () have all been used to study effectively specific ss DNA–protein interactions. The most frequent approach used to study the sequence specificity of DNA-binding molecules is by systematic evolution of ligands by exponential enrichment (SELEX), this method allows for the identification of sequences which bind with high affinity to the molecule of interest (). This method has been used mostly to select for double-stranded DNA molecules that bind to the target but it has been also used to screen ss DNA molecules (,). SELEX has advantages to the previous methods but still lacks in its ability for high-throughput analysis as numerous microarray experiments can be completed in a single day, thus providing a detailed analysis of binding-site recognition at an unparalleled rate. These techniques made use of non-immobilized ss DNA in liquid phase to probe ss DNA interactions with other molecules such as proteins, drugs and ligands, all of which suffered from being time-consuming, laborious, expensive and incapable of high-throughput screening. Therefore, oligonucleotides immobilized to solid supports provide an important tool for the rapid high-throughput examination of sequence-specific DNA–protein interactions. Two current studies (,) have used microarrays displaying all possible 8-mer and 10-mer DNA duplexes to study effectively the sequence-recognition of both transcription factors and small molecules. These methods illustrate the potential high-throughput use of k-mer arrays in examining the DNA–binding properties of duplex-binding molecules but leave the area of SNABP specificity unexplored. The innovative high-throughput assay described here provides a parallel screening system for identifying the specificity of SNABP binding. The major cold shock protein from (CspB) was used to develop this microarray-based assay. This protein influences transcription and translation () by binding to stretches of 6–7 nucleotides () of ss DNA with a high degree of specificity (,). We have used an oligonucleotide chip, for the identification of high-affinity 6-mer binding motifs for CspB. The chip contains all possible 4096 ss hexadeoxynucleotides incorporated onto a standardized anchor. The oligonucleotides on the array were originally designed to hybridize to folded mRNA which requires a significant spacer between the array surface and the recognition hexamer (Figure D, supplementary). The use of a competitor protein in this assay allowed the identification of high-affinity DNA-binding sites. The binding affinity of a competitor protein will limit the amount of ss DNA-binding sites available to the CspB. The competitor protein chosen was a single-stranded DNA-binding protein from the crenarchaeote (SsoSSB). SsoSSB has a molecular weight of 16 kDa and binds non-specifically to ss DNA with a binding density of 5 nt per monomer and an apparent dissociation constant () of ∼90 nM (). Thus, the competitive binding of the SsoSSB protein provided a means of identifying high-affinity consensus binding motifs for CspB by reducing non-specific and weak CspB-ss DNA binding. Oligonucleotide chips were supplied by Nyrion Ltd and contained all possible 4096 ss hexadeoxynucleotides incorporated into a general structure, 5′-NH2-C12-Spacer-AAAAAAAAAA-NNNNNNNNN-XXXXXX-3′, where N was one of four bases and X was a specific hexadeoxynucleotide. Each chip is made up of a 4 × 4 meta-grid and each of these sub grids contains 18 columns × 15 rows of spots, which are 135 μm ± 15% in diameter. Oligonucleotides were immobilized to the chip surface using standard Exiqon amino-link chemistry. All arrays were manufactured by pin spotting, according to complete standard commercial practices. This was all done under contract by MWG Biotech custom arrays. For control purposes, arrays are batch tested using a standard mRNA template and a standard QC procedure expected to give a standard signal. This standard signal serves as a positive and negative control for all arrays. MWG Biotech spot biotin on the surface of the array, which also serves as a negative standard control (generates zero signal). A mutant version of the SsoSSB protein from the crenarchaeote was constructed by changing the C-terminal glutamate residue to a cysteine (E145C mutant), allowing for the incorporation of spin labels and fluorescent probes on the C-terminal tail. This mutation minimizes the affect of labelling on DNA-binding activity as the C-terminal glutamate is not involved in ss DNA binding. The E145C mutant was constructed only as a precaution if the amine-reactive labelling methods were unsuccessful. Protein expression was induced by addition of 0.2 mM IPTG at 37°C for 3 h, after which cells were pelleted and frozen until required. Cell lysis, centrifugation and chromatography steps were carried out at 4°C. Cells (20 g) were thawed in 50 ml lysis buffer (50 mM Tris–HCl pH 7.5, 500 mM NaCl, 1 mM EDTA, 1 mM DTT) and immediately sonicated for 5 × 1 min with cooling. The lysate was centrifuged at 40 000  for 45 min. DNase I [40 µg/ml] and RNase A [10 µg/ml]) were then added to the cell lysate and incubated at room temperature for 30 min with gentle agitation. The supernatant was heated to 70°C for 30 min in a water bath, and denatured proteins were precipitated by centrifugation at 40 000  at 4°C. The supernatant was analysed by SDS–PAGE, and shown to contain recombinant SSB, which migrated as a band of ∼16 kDa as expected. The supernatant was diluted 5-fold with buffer A (50 mM Tris–HCl pH 7.5, 1 mM EDTA, 1 mM DTT) and applied to a Heparin-Sepharose (Amersham) column equilibrated with buffer A. SSB was eluted over a linear gradient comprising 0–1 M NaCl. Fractions corresponding to a distinct absorbance peak were analysed by SDS–PAGE, pooled and concentrated. A subsequent gel filtration step (HR 10/30 Superdex-200) in a buffer containing, 10 mM Tris/HCl pH 7.5, 150 mM NaCl, 1 mM EDTA and 1 mM DTT, resulted in essentially homogeneous SB, as determined by SDS-PAGE analysis. This method is an adaptation of the previously published method (). SSB was concentrated using a Viva Spin column (MWCO = 5 kDa) and quantified using both the Bradford method and the theoretical extinction coefficient, ε280 nm = 12660 M. cm. Primers B.S_CSP fwd (5′-dAGC TTA GAA GGT AAA GTA AAA TAA -3′) and B.S_CSP rev (5′-dCG TAA CGC TTC TTT AGT AAC GTT AGC-3′) were used in a PCR with plasmid DNA (pET11-CspB vector provided by Michael Wunderlich (University of Bayreuth)) containing the CspB gene. Bases were added to the primers to introduce the NdeI and BamHI restriction sites (underlined). These sites were used to clone the PCR product in NdeI-BamHI-digested pET28a vector, resulting in the plasmid pET28a_B.S_CspB. BL21 (DE3)pLysS () was transformed with pET28_B.S._CspB and transformants were grown in Luria-Bertani medium, containing 50 μg/ml kanamycin at 37°C with agitation. One-litre cultures were grown to an OD of 0.5–0.7 and IPTG (isopropyl-β-D-thiogalactosidase) was then added to a final concentration of 1 mM. Incubation was then continued for an additional 5–6 h and the cells were harvested at 8000  in a JLA-9.1000 rotor for 12 min at 10°C. Pellets were then frozen in liquid nitrogen and stored at −80°C. Cell pellets were resuspended in lysis buffer (20 mM Tris HCl pH 8.0, 500 mM NaCl, 0.1% Triton X-100, 0.1 mM phenylmethylsulphonyl fluoride (PMSF), 1 mM EDTA and protease inhibitors) to a final volume of 30 ml/5 g of cells and then lysed using a French-press. DNase I [40 µg/ml] and RNase A [10 µg/ml]) were then added to the cell lysate and incubated at room temperature for 30 min with gentle agitation. As an initial step, the fusion protein was purified using a Ni-NTA resin affinity column, as per manufacturer's instructions and then purified to homogeneity as described previously (). Briefly, to remove minor contaminants the fractions containing His-CspB were pooled and dialysed overnight into a buffer containing 20 mM Tris/HCl pH 6.8, 1 mM DTT. The solution was applied to an HR 5/5 Mono-Q (1 ml) anion exchange column. Bound protein was eluted with a NaCl-gradient ranging from 0–1 M. CspB eluted at a concentration of 250 mM NaCl. A subsequent gel filtration step (HR 10/30 Superdex-75) in a buffer containing 10 mM Tris/HCl pH 7.5 and 100 mM NaCl) resulted in visually pure His-CspB, as determined by SDS-PAGE analysis. His-CspB was concentrated using a Viva Spin column (MWCO = 3.5 kDa) and quantified using both the Bradford method and the theoretical extinction coefficient, ε280 nm = 5690 M. cm. Three hundred picomoles of each ss DNA templates were 5′-end labelled by incubating templates with 0.03 mCi of [γ–P]ATP, T4 polynucleotide kinase and T4 polynucleotide kinase buffer in a total volume of 60 μl at 37°C for 2.5 h. The reaction was stopped by heat inactivation (30 min at 65°C). Unincorporated [γ–P]ATP was removed with QIAquick Nucleotide Removal Kit (Qiagen). In the standard procedure, the contents of 1 vial (‘to label 1 mg of protein’) of cyanine 5 (Cy5) mono-functional dye the contents (Amersham) were dissolved in 50 μl of anhydrous DMSO. Proteins were dialysed into buffer B (150 mM, NaCO (pH 9.3, pH was adjusted with HPO)) and concentrated using a Viva Spin column (MWCO = 5 kDa). Typical working concentrations of proteins were 1 mg/ml, unless stated otherwise. Ten microlitres of dye/DMSO was pipetted into 200 μl protein solution under slow vortexing. After a 30-min incubation at 25°C in the dark, the reaction was terminated by the addition of 300 μl of 100 mM NaHPO (to suppress further labelling) to the sample. To separate the unbound dye, the sample was loaded onto a PD-10 column (10 ml bed of Sephadex G-25M), which had been pre-equilibrated in a buffer A (100 mM NaCl, 50 mM NaHPO and 1 mM EDTA (pH 7.5, pH adjusted with NaOH). The column was then washed with buffer A (2 × 1 ml) and the labelled protein was then eluted by adding 2 ml of water to the column. The extent of the modification was assessed using MALDI-TOF mass-spectrometry. Protein concentration was determined before and after labelling, Cy5-protein concentration was calculated as per manufacturer's instructions. Microarrays were pre-wet with phosphate-buffered saline (PBS) and 0.01% Triton X-100 and then blocked in 2% non-fat dried milk for 1 h. Blocked microarray slides were washed once with PBS and 0.1% Tween 20, and once with PBS and 0.01% Triton X-100. Protein (Cy5 labelled and unlabelled) binding to ss 25-mers (containing all possible 6-mer sequences) on a generic microchip was carried out in a hybridization chamber (Camlab, RTP/7870). Protein binding was performed in a humid chamber at 4°C with 80 μl of protein-binding reaction mix containing: 50 mM KCl, 20 mM Tris (pH 8.0), 2% (w/v) non-fat dried milk, 0.2 μg/μl bovine serum albumin (BSA) and 40 μM of test protein. Slides were covered with a siliconized cover-slip (BDH cover glass 22 × 50 mm, Cat. No. 406/0188/42, Borosilicate Glass) and incubated for 1 h at 4°C. The cover-slip was removed and the slide was washed (3x) in a slide chamber filled with PBS and 0.05% Tween-20, with PBS and 0.01% Triton X-100 (3×) and once with PBS for 3 min each. Excess water was removed from the slide surface (by flicking), which was allowed to dry before scanning. This method is an adaptation of the previously published method (). Various methods (denaturing conditions; including high temperatures, various concentrations of detergents and pH range in combination with high NaCl concentration) were used to remove bound protein from the chip surface in order to reuse the array but were all unsuccessful as they either had detrimental affects on ONs bound to the surface of the array or were unable to remove bound protein. The method was essentially the same as above except that both His-CspB and SSB proteins were added to the binding reaction mix at the specified molar ratio. The binding reaction was carried out as before. The array was then incubated for 1 h in a humid chamber at 4°C with 100 μl of diluted (1:100 in blocking buffer) Alexa 532-conjugated polyclonal antibody to His (Molecular Probes). After incubation, the array was washed (3×) with PBS and 0.05% Tween-20 and once with PBS for 3 min each. Excess water was removed from the slide surface (by flicking), which was allowed to dry before scanning. All microarray slides were scanned using an ArrayWorx microarray scanner at a range of laser settings, the highest of which produced a saturated signal for the majority of spots. The Alexa-532 (Cy3 equivalent) fluorophore was excited at 532 nm and the emission was recorded at 570 nm. The Cy5 fluorophore was excited at 633 nm and the emission was recorded at 675 nm. The data were filtered initially using a series of quality-control criteria so that only high-quality spots were used in our analysis. For each array we removed any flagged spots, these were spots that had dust flakes, scratches and irregular spots (spots that outmatched the average size). The average size of a spot is 135 μm ± 15% in size, any spot that did not correspond to this size constraint was excluded from the data. This size constraint also provided a crude method of approximating the DNA concentration of each spot, which allowed only spots with an optimum DNA concentration into the data collected. All microarray TIF images were quantified using Imagene Version 5.0 software. The extent of background fluorescence was initially determined from an array experiment using BSA. The level of background fluorescence from the spots and array surface was found to be similar. Therefore, the average fluorescence between spots on the array surface was used as the background value throughout the experiments, which was minimal in comparison to the average signal intensity. Background subtracted median intensities were calculated for each spot on the microarray and the data was normalized according to the total signal intensity, so that the average spot intensity was the same for each replicate slide (×3). The normalized data of each competitive array experiment was used to generate a list of the high-intensity sequences/spots (the highest to lowest intensity), i.e. spots which were above a threshold level of intensity (Supplementary Data, Figure Aa). High-intensity spots/sequences which occurred in all three replicates were carried forward for further data analysis, this procedure minimized the occurrence of any false positives or negatives (*False positives = spots which fluoresce highly on one array but not on all three arrays. Total = 6.5%; False negatives = spots which did not fluoresce on one array but fluoresced highly on two arrays. Total = 2%) in the overall data collection. The average intensity was calculated and the sequences were ranked accordingly. The list of sequences generated were condensed to include only the best binding sequences, these were spots that had intensity above 55% normalized fluorescence and at least 6 standard deviations away from the global mean intensity (Figure Ab, Supplementary Data). The final list contained a total of 50 high affinity-binding sequences for His-CspB. ITC experiments were carried out as described previously () with minor modifications. Four oligonucleotides were used for the experiments (B), both possibilities of the consensus-binding sequence (ITC1 and ITC2), a positive (ITC3()) and negative control (ITC). Each exothermic heat pulse (A, upper panels) corresponds to an injection of 5 μl of each oligonucleotide (100 μM) into the cell containing 5 μM CspB at 28°C. Integrated heat data (A, lower panels) constitutes a differential binding curve, which was fitted to a single-site binding model to give, the stoichiometry of binding (), binding affinity () and enthalpy of binding (ΔH) for each heptanucleotide. SSB was covalently labelled with the mono-functional dye Cyanine 5 (Cy5; Amersham). Electrophoretic mobility shift assays (EMSA) were conducted to verify that SSB retained its DNA-binding activity subsequent to labelling with Cy5. A single-stranded 25-mer (ONc:5′–dATCCTACTGATTGGCCAAGGTGCTG-3′), labelled at the 5′-end with γ-P, was used to compare the binding affinity of unlabelled and Cy5-labelled SSB. Figure Cc (Supplementary Data) shows a gel-shift experiment performed with ONc in the presence of increasing amounts of unlabelled (lanes 1–4) or labelled (lanes 5–8) protein. The similarity in the EMSA for both unlabelled and Cy5-labelled protein suggests that labelling did not significantly affect the binding affinity of the protein. The double banding seen (Figure Cc, Supplementary Data) is a result of the gradual dissociation of labelled DNA from SsoSSB. The fact that previous SSB-ss DNA-binding studies using ITC have shown that SsoSSB bound to ss DNA with a ratio of 5 nt/monmer () suggests that the double banding seen in the gel is probably a result of the EMSA technique used and the on/off-rates of SsoSSB-ss DNA complex formation. A generic chip (Figure D, Supplementary Data) was constructed, which contained all possible 4096 ss hexadeoxynucleotide sequences found in DNA and incorporated into a general construct, 5′-NH-A-N-X-3′ (X = G, C, A, or T and the stretch of 9 Ns is composed of random bases). The binding of SSB-Cy5 to the generic chip was analysed and all the spots on the array fluoresced with similar intensities, consistent with non-specific binding of SSB-Cy5 (). A ss 25-mer () (ONc: 5′–dATCCTACTGATTGGCCAAGGTGCTG-3′), labelled at the 5′-end with γ–P was used to compare the binding affinities and specificities of His-CspB and CspB. A shows a gel-shift experiment performed for ONc in the presence of decreasing amounts of His-CspB (lanes 2–6) and non-His-tagged CspB (lane 7). Lanes 3, 4 and 5 show decreasing migration patterns for the His-CspB ss DNA complex as less protein molecules bind. This is most likely a result of deviation in the affinity of CspB for specific binding sites within ONc. The data show that the addition of the His-tag does not significantly affect the binding affinity or specificity of the protein. The effect of flanking DNA at the 3′ end of the oligonucleotides, on CspB binding, was also examined by EMSA using a series of oligonucleotides that were structurally consistent with the oligonucleotides found on the array; the only difference was that a varying number of bases were added to the 3′ end. The results from these experiments show that flanking bases at the 3′ end or a lack of them did not seem to affect the binding of CspB (T1 and Figure B, Supplementary Data). A competitive EMSA was used to show that His-CspB binds more strongly than SSB to the previously reported high affinity Y-box-binding motif () (ATTGG). A 25-mer, ON1 (5′–dA--3′), which contains the Y-box-binding motif, was labelled at the 5′-end with γ–P. ON1 is similar in composition to the oligonucleotides found on the generic chip. B shows the result of a gel-shift experiment performed with ON1 in the presence of varying amounts of His-CspB and SSB proteins. An intermediate band can be seen in lane 5, corresponding to the formation of the His-CspB-SSB-ON1 complex. This band occurs only when the SSB/CspB proteins are in a molar ratio of approximately 1:1. The fluorescent signal from His-CspB bound to the chip was analysed (Figure E, Supplementary Data). About 20% of the spots had signals greater than threshold level (which was set at 40% of the maximum fluorescent signal). To eliminate weakly bound CspB, an equimolar mixture (as determined by gel-shift, B) of a competitor SSB, and His-CspB was incubated with an oligonucleotide chip and bound His-CspB was detected (C). High-intensity fluorescence spots were observed in repeated patterns on the arrays, which is indicative of selective binding affinity (). An EMSA was carried out to confirm the binding-site data generated from the oligonucleotide chip analysis. The high affinity-binding motif, GCACTT, was chosen from the data (C) to examine if the His-CspB could compete successfully with the SSB protein for this binding site. A 25-mer, ON-Microarray test (ON-Mt:5′–dAAAAAAAAAA-GCACTT-AAAAAAAAA-3′), containing the high affinity-binding motif was labelled at the 5′-end with γ–P. Figure F (Supplementary Data) shows a gel-shift experiment performed with ON-Mt in the presence of varying amounts of His-CspB and SSB proteins. An intermediate band (His-Csp-SSB-ONc complex) can be seen in lanes 3–11 for ON-MTest, indicating that the CspB protein competed with SSB for binding to the (microarray determined) high-affinity-binding site, GCACTT. The highest intensity spots identified from the microarray analysis indicate that the strongest CspB-binding sites are pyrimidine rich (Figure Ga, Supplementary Data). The high incidence of thymine bases within the high-affinity 6-mer sequences agrees with previous reports that CspB has a preference for T-rich stretches of ss DNA (). The presence of a stretch of 10 adenines in the linker region of each oligo (Figure Dd, Supplementary Data) is likely to lead to hairpin formation for T-rich sequences and may be expected to down weight the occurrence of poly T sequences. There is indeed a low intensity for TTTTTN sequences, which may in part be caused by the formation of such hairpins. Despite this effect, the averaging procedure still generates a T-rich consensus sequence. The standard motif alignment method (Genedoc) was used to align the resulting top fifty (as described in Materials and Methods) high-affinity CspB binding sequences (Figure Gb, Supplementary Data). A sequence alignment window 10 bases in length was used; only seven (coloured columns) out of those 10 positions are significantly (>40%) populated. Analysis of the relative distribution of each base within this proposed heptanucleotide-binding site gives a CspB consensus-binding sequence of (D and E). Analysis of the microarray binding results indicates that CspB can accommodate the binding of a heptanucleotide with a strong binding preference for cytosine at position 3 and thymine at positions 2, 4 and 6. This is in agreement with another recent study (), where the sequence-specific binding of heptapyrimidines to CspB was analysed by tryptophan fluorescence quenching experiments. Interestingly, the microarray results show that neither the Y-Box recognition motif (), 5′-ATTGG-3′, nor its reverse complement 5′-CCAAT-3′, bind strongly to CspB. The ATTGG sequence has recently been shown to bind with a low affinity for CspB at 15°C ( = 5.3 μM ()), which is similar to the results described here for CspB binding to sequences containing ATTGG at 4°C. Both variants of the preferential binding sequences (ITC-1 and ITC-2) were analysed by ITC and gave binding constants in the low nanomolar range (B). There is an order of magnitude difference in the s for the oligonucleotides described here and the s described previously (), for similar/identical oligonucleotides. This is likely due to a difference in temperature, buffer conditions and the method used (tryptophan fluorescence quenching). The control sequence (TTCTTTT-ITC3), used here and in the previous study allows us to compare and scale the results from the two methods. Thus, both the ITC and EMSA results confirm that the microarray assay does indeed select and identify tight binding sequences. This screening procedure could, therefore, be used as a general method for the rapid identification of high-affinity binding sites for SNABPs. The X-ray structure for CspB has been reported previously (). C shows the positively charged face of CspB, highlighting amino acid residues known to be involved in DNA binding (). Molecular modelling of CspB with the consensus oligonucleotide ITC1, using the programme WITNOTP, identified a pocket in the centre of the DNA-binding face, which provides an ideal shape and hydrogen-bonding complementarity for binding cytosine. Three hydrogen bonds are formed between the docked cytosine base and amino acid side chains Ser31, His29 and the backbone of Phe27 (C), providing specificity for cytosine over the other bases. In the recently published structure of CspB in complex with hexathymidine (), the ligand binds to two protein molecules. The nucleobase of T2, T3 and T4 make contact with one protein molecule, T5 bridges between protein molecules and T6 binds to the next protein molecule. The contacts made by hexathymidine provide the necessary scaffold for the complex to crystallize but the sequence fails to bind across the face of the CspB protein, as this sequence lacks the key cytosine nucleobase at position 3 (), which, as we have shown here, is required for optimum ss DNA docking (). SNABPs have been reported to activate transcription by binding to a specific recognition sequence upstream () or within () a promoter, resulting in activation or repression of transcription. In the present study, CspB of which is capable of binding single-stranded nucleic acids () and affects expression of over 100 genes under cold shock conditions (), was used as the test protein. Database analysis of the genome reveals that 89 copies of the consensus-binding sequence (5′-GTCTTTT-3′) exist within potential SNABP promoter regions (100 bases upstream of the ATG start for all genes), of which only 24 have an assigned function. The use of an unbiased genomic assay to identify optimal-binding sequences for SNABPs , may provide insight into their role in regulating cellular functions. The full implications of these sequences on gene expression and binding of CspB remain to be determined. p p l e m e n t a r y D a t a i s a v a i l a b l e a t N A R O n l i n e .

The discovery of RNA interference (RNAi) (), a specific gene-silencing mechanism mediated by small RNA molecules such as small interfering RNA (siRNA) () and microRNA (miRNA) (), has had a major impact on research in the life sciences, and, in recent years, great advances have been made in our understanding of these and other kinds of non-coding RNA (,). Application to basic research is exemplified by the work of Siolas . () on short hairpin RNA (shRNA) and Amarzguioui . () on siRNA. RNA is also expected to play an important role in drug discovery, and several candidate small RNA molecules have already entered clinical trials (). As a result of the increased demand for RNA for use as research reagents and potential therapeutic agents, more importance than ever is being attached to practical new methods in the chemical synthesis of RNA. From the earliest days of research on RNA synthesis, it has been appreciated that the single most demanding problem is the selection of a suitable protecting group for the 2′-hydroxyl function of the ribose of the RNA. This protecting group must be stable throughout the solid-phase synthetic cycle, yet it must be readily removable under mild conditions (). -Butyldimethylsilyl (TBDMS) () is a popular 2′-hydroxyl protecting group whose phosphoramidite is commercially available. While the TBDMS method gives RNA of reasonable purity in reasonable yield, both yield and purity are sensitive to small changes in the synthetic conditions. Furthermore, the TBDMS group is associated with relatively long coupling times () and insufficiently high coupling yields (), though improvements based on the use of more powerful activators or less bulky 2′-hydroxyl protecting groups have been proposed (). To resolve these pro-blems, three new protecting groups, bis(2-acetoxyethyl-oxy)methyl (ACE) (), triisopropylsilyloxymethyl (TOM) (), and more recently -butyldithiomethyl (DTM) (), have been developed. Though such protecting groups represent major improvements in the synthesis of RNA oligonucleotides (), however, they still leave something to be desired in their practical application. Thus, the synthesis of ACE-amidites is relatively complex () (though they are now commercially available), and automated synthesizers require special modification for use with them because of the incompatibility of glass materials with triethylamine trihydrofluoride, the 5′-desilylation reagent. TOM-protected oligonucleotides, meanwhile, are not readily amenable to routine analysis and purification by HPLC because of the hydrophobic nature of the silyl group. Finally, DTM-amidites, particularly the G amidite, are somewhat unstable in acetonitrile solution (). Chemical rather than enzymatic synthesis of RNA is desirable because it avoids the various errors and inefficiencies associated with transcription by T7 RNA polymerase (). Chemical synthesis is also desirable for the ease of incorporation of modified nucleosides, which is important for studies of RNA structure and function. For example, site-specific disulphide cross-links have been used as probes of RNA tertiary structure (). For such purposes, it is useful if the synthetic method can be extended to the efficient production of very long RNA oligomers. However, despite the advances that have been made in automated solid-phase synthesis, the production of longer RNA oligomers is still difficult and time-consuming. Previously synthesized RNA oligonucleotides up to 77 nt long have been generally restricted to tRNA (), although there is also an example of the synthesis of an 84mer RNA oligonucleotide by the TOM method (). Sometimes it is necessary to assemble shorter oligoribonucleotides to the full-length RNA by enzymatic ligation, as in the synthesis of tRNA by Ohtsuki . (). Therefore, many points remain to be resolved in the field of RNA synthesis, and because of the increased demand for RNA there is a need for a radically improved synthetic method that would allow RNA synthesis to be carried out as efficiently as DNA synthesis, scaled up to the levels now obtained in DNA synthesis, and applied to the production of very long RNA oligomers. To meet this need, we have developed an achiral 2′-hydroxyl protecting group, 2-cyanoethoxymethyl (CEM), that has low steric hindrance and that can be completely removed under mild conditions (). CEM chemistry is also compatible with standard, unmodified DNA synthesizer equipment. In the present study, we have further developed the CEM method by significantly improving the capping and coupling conditions. To verify the potential of the improved method, we have undertaken the total chemical synthesis of a very long RNA oligomer. To the best of our knowledge, there has been no report of the chemical synthesis of an RNA oligomer longer than 100 nt. Here, we describe the synthesis of a 110mer precursor-miRNA (pre-miRNA) candidate by our improved new method. In contrast to previously synthesized long RNA oligonucleotides, which have generally been characterized by biochemical or biological means alone, we have confirmed the identity of the product by physicochemical methods as well as through assessing its biological activity by measuring its gene-silencing effect. Column chromatography was performed with Wako silica gel C-200, and TLC was carried out on Merck Silica gel 60 F TLC aluminium sheets. Analytical HPLC was performed on Shimadzu HPLC systems equipped with DNAPac PA100 (4 × 250 mm, Dionex) or PLRP-S 300 Å (4.6 × 150 mm; Polymer Laboratories, Church Stretton, UK) columns, preparative reverse-phase column chromatography (DMTr-on mode) was performed with PLRP-S 300 Å resin packed in 20 × 110 mm columns, and semipreparative anion-exchange HPLC (DMTr-off mode) was performed on a DNAPac PA100 (9 × 250 mm) column. Capillary gel electrophoresis was carried out on a P/ACE MDQ Molecular Characterization System with the ssDNA 100-R Kit (Beckman Coulter). UV spectra were recorded on a Hitachi U-3210 spectrometer, and H, C, and P NMR spectra on a Bruker DRX500 or DPX300 spectrometer. Chemical shifts are reported relative to tetramethylsilane and referenced to the residual proton signal of the following deuterated solvents: CDCl (7.25 p.p.m.) or DMSO-d6 (2.50 p.p.m.) for H NMR, CDCl (77.0 p.p.m.) or DMSO-d6 (39.5 p.p.m.) for C NMR, and external 85% phosphoric acid for P NMR. Matrix-assisted laser desorption/ionization time-of-flight (MALDI-TOF) mass spectra were obtained with a Bruker Autoflex spectrometer (Bruker Daltonics, Billerica, MA). Reagents and solvents were purchased from commercial suppliers and used without further column chromatography. Anhydrous solvents were purchased from Kanto Kagaku (Tokyo, Japan), and tetrabutylammonium fluoride (TBAF) was purchased from Nacalai Tesque (Kyoto, Japan). Full experimental procedures describing the synthesis of the CEM-phosphoramidites (henceforth termed ‘CEM-amidites’) are available in Supplementary Data. The long oligomer selected for synthesis was the 110mer pre-miRNA hairpin candidate 5′-UGCUCGCUCA GCUGAUCUGU GGCUUAGGUA GUUUCAUGUU GUUGGGAUUG AGUUUUGAAC UCGGCAACAA GAAACUGCCU GAGUUACAUC AGUCGGUUUU CGUCGAGGGC-3′ (miR-196a-2 precursor; listed in miRBase) (). Using the CEM-amidites prepared as described in Supplementary Data, we synthesized the 110mer RNA on an Applied Biosystems Expedite Model 8909 nucleic acid synthesizer. In HPLC analyses and preparative purification, the following solvent systems were used: for reverse-phase HPLC, buffer A [5% acetonitrile in 50 mM triethylammonium acetate (TEAA), pH 7] and buffer B (90% acetonitrile in 50 mM TEAA, pH 7); and for anion-exchange HPLC, buffer C (10% acetonitrile in 25 mM Tris-HCl, pH 8.0) and buffer D (10% acetonitrile containing 700 mM NaClO in 25 mM Tris-HCl, pH 8.0). For preparative reverse-phase column chromatography (DMTr-on mode), the column was filled with PLRP-S 300 Å resin (50–70 μm), and elution was carried out with a gradient of buffer A and buffer B. For semipreparative anion-exchange HPLC (DMTr-off mode), elution was carried out with a gradient of buffer C and buffer D. Dialysis was carried out with a Spectra/Pore® membrane (molecular weight cut-off, 1000; Spectrum Laboratories, Laguna Hills, CA). Nuclease P1 was from Yamasa (Chiba, Japan), and alkaline phosphatase was from Promega (Madison, WI). For RNase T1 digestion, RNA (60 pmol) was incubated with 0.3 U RNase T1 (Ambion, Austin, TX) in 50 mM Tris-HCl, pH 7.5, containing 1 mM EDTA and 25 μg/ml GpA in a total reaction volume of 30 μl at 37°C for 18 h. For MazF digestion, RNA (60 pmol) was incubated with 213 U MazF (Takara, Otsu, Japan) in 20 mM sodium phosphate, pH 6.0, containing 0.05% Tween-20 in a total reaction volume of 30 μl at 37°C for 1 h. The digestion products were purified by phenol/chloroform extraction and ZipTip C18 reverse-phase microcolumns (Millipore, Bedford, MA). MALDI-TOF mass spectra were acquired in positive-ion mode with a matrix solution consisting of 10 mg/ml 3-hyroxypicolinic acid and 1 mg/ml diammonium citrate. The potential for linear acceleration was 19 kV, and the potential for reflector acceleration was 20 kV. The IS/2 potential was set at 16.6 kV. Ion extraction was delayed by 130 ns, and 240 shots were accumulated for each sample. RNA oligomer (20 ODU at 260 nm) was incubated with nuclease P1 (0.3 units) at 37°C for 24 h. Then, alkaline phosphatase (3 units) and buffer (50 mM Tris-HCl, pH 9.3, containing 1 mM MgCl, 0.1 mM ZnCl, and 1 mM spermidine, all at final concentrations) were added to give a total volume of 115 μl, and the mixture was incubated at 37°C for 2 days. The reaction mixture was then analyzed by HPLC. Reporter plasmids pHOXB-Luc and pHOXB-Luc-antisense were constructed by inserting DNA duplex into the I site of the luciferase reporter vector pGL4.13[luc2/SV40] (Promega) according to the supplier's protocol. The sequences of the inserted fragments were as follows: pHOXB-Luc sense, 5′-CTAGATTGTG CTAAGTTCTC CCAACAACAT GAAACTGCCT ATTCACGCCG TAATTT-3′; pHOXB-Luc antisense, 5′-CTAGAAATTA CGGCGTGAAT AGGCAGTTTC ATGTTGTTGG GAGAACTTAG CAGAAT-3′; pHOXB-Luc-antisense sense, 5′-CTAGAAATTA CGGCGTGAAT AGGCAGTTTC ATGTTGTTGG GAGAACTTAG CAGAAT-3′; and pHOXB-Luc-antisense antisense, 5′-CTAGATTGTG CTAAGTTCTC CCAACAACAT GAAACTGCCT ATTCACGCCG TAATTT-3′. Human embryonic kidney derived cell line G3T-hi was purchased from Takara and maintained in Dulbecco's modified Eagle's medium (Sigma) supplemented with 10% foetal bovine serum at 37°C in a humidified atmosphere. Transfections with plasmid DNA and/or RNA were performed with FuGENE 6 transfection reagent (Roche, Indianapolis, IN) according to the supplier's instructions. G3T-hi cells (1 × 10) were seeded on a 96-well plate and incubated for 20 h. Reporter plasmid DNA (10 ng) and effector RNA (30 or 100 nM each at final concentration) were transfected by means of FuGENE 6. Forty-eight hours after transfection, luciferase activity was measured with the Steady-Glo Luciferase Assay System (Promega) according to the supplier's protocol. Sense and antisense 22-nt single-stranded RNAs (Japan BioService, Saitama, Japan) were annealed in distilled water for use as mature miRNA duplex. Luminescence intensity was measured with a microplate scintillation and luminescence counter (Packard, Meriden, CT), and luciferase activity was calculated relative to the basal activity expressed by pHOXB-Luc in the absence of effector RNA. There was room for improvement in our previously reported method of synthesizing the monomer CEM-amidites (). In that method, we started with the 5′--DMTr compound and alkylated the 2′-hydroxyl group to give intermediate . However, because we could not achieve selective alkylation of the 2′-hydroxyl group, the concurrently obtained 3′--CEM derivative had to be removed by column chromatography. Therefore, we sought an improved synthetic route for the CEM-amidites that could be carried out simply on a large scale, and we established the routes shown in . In these routes, we introduce the CEM group into a 3′,5′-protected nucleoside (). With 2-cyanoethyl methylthiomethyl ether as the alkylating agent and -iodosuccinimide (NIS) as the activator, the alkylation of the uridine, cytidine, and guanosine derivatives proceeded efficiently at low temperature (−45°C). In the case of the adenosine derivative only, the efficiency of alkylation was low. Therefore, compound was produced by activating the nucleoside through methylthio derivative . Using these routes, we were able to synthesize all of the CEM-amidites with selective alkylation of the 2′-hydroxyl group with relative ease on a preparative scale. We are currently working on the further improvement of this method. The RNA was synthesized on a 0.8-μmol scale by the phosphoramidite method modified for CEM chemistry. The conditions of the synthetic cycle are shown in . Trichloroacetic acid in dichloromethane was used as the detritylation reagent. For the coupling reaction, amidites were used as 0.075 M solutions in acetonitrile, and 5-benzylmercaptotetrazole (BMT) was used as the activator instead of the 5-ethylthio-1H-tetrazole (ETT) used in the original method (). BMT was adopted as the activator after a trial of about a dozen activators in the synthesis of short oligomers. The alternative activators tried included 4,5-(dicyano)imidazole, benzimidazolium triflate, 1H-tetrazole, and pyridinium trifluoroacetate. In addition to being a very powerful activator, BMT was found to give the best coupling yields and the cleanest crude product. A mixture of phenoxyacetic anhydride (PacO) in THF and 2-dimethylaminopyridine (2-DMAP), -methylimidazole (NMI), and 2,6-lutidine in THF was used as the capping reagent to suppress the formation of 2,6-diaminopurine (2,6-DAP). [In the original method (), NMI and pyridine were used as the base catalysts.] The oxidizing agent was 0.1 M iodine in THF/pyridine/water. Commercially available controlled-pore glass (CPG) with a pore size of 2000 Å derivatized with -benzoyl-2′--TBDMS-ribocytidine as the leader nucleoside was used as the solid support instead of 500-Å or 1000-Å CPG. Although the leader nucleoside bears a different base-protecting group and 2′-hydroxyl protecting group, their removal from a single residue at the 3′ end of the oligomer presented no problem under the conditions of the CEM method. A 2000-Å resin was used because, in our experience, a large-pore resin gives better yields in the synthesis of very long oligomers. Thus, in several respects, as described above, the synthetic conditions have been improved over those used in the original method (). In a comparison of the synthesis of a mixed-base 55mer by the CEM and the TBDMS methods, both with BMT as the activator, the HPLC profiles of the crude product were quite different. The CEM profile shows a single major peak with only tiny amounts of shorter impurities (), while the TBDMS profile showed substantial amounts of shorter impurities (data not shown). Furthermore, though the CEM and TOM methods were not compared under the same conditions in the synthesis of the same sequence, published data on the TOM method shows substantial amounts of shorter impurities on capillary gel electrophoresis of the crude product [84mer; Ref. ()], whereas the CEM method showed only tiny amounts (82mer; unpublished data). While this study was in progress, another 2′-hydroxyl protecting group, 2-(4-tolylsulphonyl)ethoxymethyl (TEM), was reported by Zhou . (). So far, however, only the synthesis of a 38mer uridine homo-oligomer and mixed-base oligomers of up to 21 nucleotides has been described. In the CEM method, meanwhile, despite the loss of <5% of CEM during ammonia treatment, no chain cleavage is observed (), and no limitation in the use of the CEM method for the synthesis of RNA oligomers of up to 110 nt was noted in the present study. Reverse-phase and anion-exchange HPLC and capillary gel electrophoresis analysis of the final product showed only a single peak (A–C), confirming the purity of the RNA obtained. Polyacrylamide gel electrophoresis analysis (D) is consistent with the synthesis of a 110mer. We also confirmed the structure of the product by mass spectrometry. A MALDI-TOF mass-spectrometric analytical method is now available to confirm the molecular weight of oligonucleotides, though the analysis of oligomers longer than 50 nt is reported to be difficult (). Therefore, we carried out MALDI-TOF mass-spectrometric analysis of the partial-digestion products of MazF, an endonuclease that cleaves specifically at ACA sequences (). The highest peak corresponds to the undigested RNA oligomer, and the small difference between the calculated and observed values is consistent with a length of precisely 110 nt (). We also observed peaks corresponding to the expected 44mer, 66mer and 85mer fragments. The smallest expected fragment, a 25mer, was observed in other digests (data not shown). To check the identity of the 110mer by an independent method, we digested it with RNase T1, an endonuclease that cleaves specifically on the 3′ side of G, leaving a 3′-phosphate or 2′,3′-cyclic phosphate, and analyzed the fragments by mass spectrometry. Except for monomers and dimers, which were hard to distinguish among the noise of the spectrum, all of the expected fragments were identified (A and B and ), consistent with the synthesis of the target 110mer. RNase T1 digestion yielded pairs of fragments with 3′-phosphate or 2′,3′-cyclic phosphate, differing in molecular weight by 18. For example, fragments L and L′ (AACUCG; sequence 13, ) yielded observed molecular weights of 1939.5 (fragment L; calcd 1939.2) and 1921.5 (fragment L′; calcd 1921.2). All other fragments were similarly identified. The RNA was treated with nuclease P1 and then alkaline phosphatase, and the digestion products were analyzed by HPLC (). With the exception of 2,6-DAP noted below, no modified ribonucleosides were observed, showing that complete removal of the CEM protecting groups and base-protecting groups had been achieved. Furthermore, no dimers (2′,5′-phosphodiester-linked dinucleotides resulting from internucleotide migration or branched structures resulting from side reactions occurring during the condensation cycle) were observed. In our previously reported CEM synthetic method (), the formation of 2,6-DAP was not observed in the synthesis of oligomers up to 55 nt in length. However, with continued repetition of the capping step during the synthesis of very long oligonucleotides, the formation of 2,6-DAP due to modification of G became detectable. (When pyridine or 4-dimethylaminopyridine (4-DMAP) is used in the capping reagent, displacement of the -phosphate triester adduct on G leads to the formation of 2,6-DAP after ammonia treatment during deprotection. When 2,6-lutidine or 2-DMAP is used instead, this displacement reaction is hindered by the presence of two methyl groups or a dimethylamino group to the pyridine-ring nitrogen.) Though it was possible to completely suppress the formation of 2,6-DAP by using PacO/2-DMAP/2,6-lutidine as the capping reagent, under these conditions the efficiency of the capping reaction was slightly reduced, which is a significant disadvantage in the synthesis of very long oligomers. It was therefore decided to find compromise conditions in which very low levels of 2,6-DAP formation could be tolerated for the sake of a high capping efficiency. By adjusting the ratio of NMI to 2-DMAP in the capping reagent, we could achieve a very high capping efficiency, though not quite perfect suppression of 2,6-DAP formation. To assess the biological activity of the chemically synthesized 110mer pre-miRNA candidate, we measured its specific gene-silencing effect. Generally, small dsRNAs suppress the expression of their target genes, whose mRNA possess sequences complementary to those of the small RNAs, by causing cleavage of the mRNA or inhibiting its translation (). To determine the silencing effect of miRNAs, we constructed a reporter assay system. The sequence of the RNA synthesized was that of pre-miR-196a-2, so we used a reporter gene whose expression can be regulated by miR-196a. There is predicted to be a miR-196a target sequence in the 3′ untranslated region of human homeobox gene b-8 (), and expression has been suggested to be regulated by miR-196a (). We inserted DNA duplex corresponding to this target sequence and its flanking region into a luciferase-expressing reporter plasmid. In this system, a silencing effect by miR-196a results in a reduction in luciferase expression. Transfection with chemically synthesized 110mer RNA decreased the expression of the target gene about as effectively as 22mer mature miR-196a RNA duplex (A). The suppression of luciferase activity by 110mer RNA is considered to have resulted from the inhibition of luciferase protein expression, possibly at the transcriptional or translational level. Because pre-miRNA has to be processed to a smaller size to exert its gene-silencing effect, we conclude that our synthetic 110mer RNA was successfully processed and that the mature miRNA derived from the 110mer functioned as a silencing miRNA. Although the guide (antisense) strand of an siRNA molecule is normally specifically incorporated into the RNA-induced silencing complex (RISC), it sometimes happens that the passenger (sense) strand is incorporated, giving rise to off-target effects (,). A pre-miRNA molecule, in contrast, is expected to interact more specifically with RISC, allowing more specific incorporation of the antisense strand and exclusion of the sense strand during processing of the pre-miRNA, with resulting preferential cleavage of the target mRNA (). In the case of pre-miR-196a, we found that the silencing effect against the sense target was about twice as strong as that against the antisense target, whereas 22mer mature miR-196a showed about the same silencing effect against both targets (B). The observation of an equal silencing effect against the sense and antisense target mRNA by the 22mer is probably due to the sequences at the ends, which might prevent preferential incorporation of the guide strand. The 110mer pre-miRNA, therefore, showed a strand selectivity in the gene-silencing effect that was not shown by the mature miRNA. This observation supports the idea that the use of pre-miRNA instead of mature miRNA can be useful in gene silencing to confer selectivity of the antisense strand and to reduce the likelihood of off-target effects. h a v e d e v i s e d a n i m p r o v e d m e t h o d f o r t h e s y n t h e s i s o f C E M - a m i d i t e s t h a t i s s u i t a b l e f o r l a r g e - s c a l e s y n t h e s i s . W e h a v e a l s o i m p r o v e d t h e c o n d i t i o n s f o r t h e s o l i d - p h a s e s y n t h e s i s o f o l i g o m e r s . W i t h o u r i m p r o v e d C E M m e t h o d , i t i s p o s s i b l e t o s y n t h e s i z e v e r y l o n g o l i g o m e r s t h a t h a v e u n t i l n o w b e e n e x t r e m e l y d i f f i c u l t t o s y n t h e s i z e . W e h a v e d e m o n s t r a t e d t h e p o t e n t i a l o f t h e m e t h o d b y s y n t h e s i z i n g , f o r t h e f i r s t t i m e , a n R N A o l i g o m e r l o n g e r t h a n 1 0 0   n t . T h e 1 1 0 m e r w e s y n t h e s i z e d w a s p r o d u c e d b y t o t a l c h e m i c a l s y n t h e s i s i n a s i n g l e s y n t h e s i z e r r u n , w i t h o u t t h e n e e d f o r e n z y m a t i c l i g a t i o n o f s e p a r a t e l y s y n t h e s i z e d f r a g m e n t s . I t s s t r u c t u r e w a s c o n f i r m e d b y p h y s i c o c h e m i c a l m e t h o d s , i n c l u d i n g m a s s s p e c t r o m e t r y , a n d i t w a s a l s o s h o w n t o h a v e b i o l o g i c a l a c t i v i t y . T o o u r k n o w l e d g e , i t i s b y f a r t h e l o n g e s t c h e m i c a l l y s y n t h e s i z e d R N A w h o s e s t r u c t u r e h a s b e e n d e t e r m i n e d b y p h y s i c o c h e m i c a l m e t h o d s . W e b e l i e v e t h a t o u r n e w i m p r o v e d C E M m e t h o d i s s u f f i c i e n t l y p r a c t i c a l t o b e c o m e t h e s t a n d a r d R N A s y n t h e t i c m e t h o d , w i t h s p e c i a l a p p l i c a t i o n t o t h e s y n t h e s i s o f v e r y l o n g R N A o l i g o m e r s . p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

The transcription factor YY1 (Yin Yang 1) is a Gli–Kruppel type zinc finger protein, and can function as a repressor, activator or transcription initiator depending upon the sequence context of YY1-binding sites with respect to other regulator elements (). The protein has a DNA-binding domain at the C-terminus and other modulating domains at the N-terminus displaying repression, activation and protein/protein interaction activities (). YY1 interacts with several key transcription factors, including TBP, TAFs, TFIIB and Sp1, as well as histone-modifying complexes, such as p300, HDACs, PRMT1 and Polycomb complexes (,). Many cellular and viral genes are controlled by YY1. A recent survey estimated that ∼10% of human genes contain YY1 binding sites near their promoter regions (). Another set of studies has revealed that some of mammalian imprinted genes contain very unusual tandem arrays of YY1 binding sites in their controlling regions, suggesting potential roles in mammalian genomic imprinting (). A series of mouse mutagenesis experiments demonstrated the dosage-dependent essential roles of YY1 during mouse development as well as in cell cycle control (,). YY1 is evolutionarily well conserved throughout the vertebrate and invertebrate lineages. It has been identified in several vertebrate species (,,), and two genes very similar to YY1 are found even in flies, Pleiohomeotic (PHO) and Pho-like (PHOL) (,). PHO is one of the DNA-targeting proteins for the Polycomb complex and the phenotypes of pho-deficient mutants can be rescued by mammalian YY1 (). In mammalian genomes, two other YY1-related genes have been identified, YY2 (Yin Yang 2) and Reduced Expression 1 (REX1). YY2 is functionally very similar to YY1 (), and is a retroposed copy duplicated from YY1 based on its intronless structure and location in the intron of another X-chromosomal gene, Mbtps2 (). REX1 was independently discovered, before the identification of YY1, due to its unique expression profile: dramatic decline of expression after retinoic acid-induced differentiation of F9 murine teratocarcinoma stem cells (). Subsequently, REX1 has been mainly studied as a stem cell marker that is controlled by Oct3/4 (,). A recent comparative study, however, emphasized that REX1 is a member of the YY1 subfamily (). Despite the significant roles and evolutionary conservation of YY1-related sequences in animals, there has not been any systematic analysis of these sequences in terms of their origins, evolutionary patterns and implications for functional diversification. To address this, we have analyzed YY1-related sequences identified from genome sequences ranging from flying insects to placental mammals. We have identified two evolutionarily conserved protein domains within YY1 which were previously unrecognized. We have uncovered independent retroposition events that have been responsible for forming duplicate copies, such as PHOL from PHO in flies, and YY2 and REX1 from YY1 in placental mammals. Our analyses revealed that the zinc finger domains of YY2 and REX1 have been under different selection pressures compared to YY1. Their DNA-binding properties have evolved from YY1 by weakening DNA-binding affinity in both YY2 and REX1, and changing DNA-binding motifs in REX1. The evolution patterns of YY1 and other YY1-related genes described in the current study provide a unique paradigm for gene duplication and functional diversification. A series of database searches were conducted using the BLAST program () to obtain YY1-related sequences. Human YY1 (NP_003394.1) was first used as a query sequence to search sequence databases, including NCBI, the Genome Browser at University of California Santa Cruz and Ensembl. Later, human REX1 (NP_777560.2) and YY2 (NP_996806.1) were used to further characterize the identified YY1-related sequences from chordates, while Drosophila melanogaster PHO (NP_524630.1) and PHOL (NP_648317.1) were used for the identified insect sequences. The detailed information regarding all the YY1-related sequences described in this study is available as Supplementary Data 1 through the following website (). Multiple sequence alignments were performed with ClustalW using the following parameters: gap opening penalty = 10, gap extension penalty = 0.1 (0.2 for multiple alignment), Gonnet Protein Weight Matrix, residue specific penalties = ON, hydrophilic penalties = ON, gap separation distance = 4, end gap separation = OFF (). Sequences were edited manually in Mega3 V3.1 to remove spurious introns from some sequences (). Separate multiple alignments were performed for insects’ and chordates’ sequences. Subsequently, two phylogeny gene trees were constructed and analyzed using both the neighbor-joining and maximum parsimony methods as implemented in Mega3 V3.1 with Poisson correction and confirmed by bootstrapping 1000 iterations (). Synonymous and non-synonymous substitution rates were estimated using two different approaches: Nei–Gojobori () and Yang–Nielsen methods (). The zinc finger regions of YY1 (NM_009537.2), YY2 (NM_178266) and REX1 (NM_009556.2) were amplified from either mouse brain cDNAs or genomic DNAs by the following primer sets: YY1 (mYY1Zn5, 5′-CCAAGAACAATAGCTTGCCCTC-3′ and mYY1Zn3, 5′-TCACTGGTTGTTTTTGGCTTTAGCG-3′), YY2 (mYY2Zn5, 5′-CCAAGACCTATAGCATGCTCTC-3′ and mYY2Zn3, 5′-TTACTGGTCATTCTT GTTCTTAACATGGG-3′) and REX1 (mRexZn5, 5′-TTATCGATGCTGGAGTGTCCTCAAGC-3′ and mRexZn3, 5′-TCAGCATTTCTTCCCTGCCTTTGC-3′). The amplified products were first cloned into the pCR4-TOPO vector (Invitrogen, Carlsbad, CA, USA), and later transferred to the EcoRI site of the pGEX-4T-2 vector (Amersham Biosciences, Piscataway, NJ, USA) after sequence confirmation. The constructed vectors were transformed into BL21 (DE3) competent cells for bacterial expression (Strategen, La Jolla, CA, USA). The optimum induction of the constructs by IPTG was monitored through SDS-PAGE (Supplementary Data 4 from ). DNA-binding motif studies were conducted as described in the previous studies (,) with slight modifications. Briefly, the transformed cells were grown at 37°C in LB media (100 ml) to an optical density of 0.6 at 600 nm, and protein expression was induced with 0.4 mM IPTG for additional 3.5 h. Cells were harvested by centrifugation at 4000 for 10 min at 4°C. Lysates were prepared from the cell pellets by sonication in 6 ml of ice-cold NETN buffer (100 mM NaCl, 1 mM EDTA, 20 mM Tris–HCl, pH 8.0, 0.5% NP-40). Protein concentration in cell lysates was determined using the Bradford assay (Pierce, Rockford, IL, USA). Aliquots of 500 μg/100 μl were frozen at −80°C. Immobilized glutathione agarose (Pierce) was washed three times with 1 ml ice-cold NETN buffer and used to isolate fusion proteins by incubating 500 μg lysate with 50 μl washed agarose beads at 4°C for 30 min while rotating. The agarose beads were precipitated by centrifugation, and washed twice first with 1 ml ice-cold NTEN buffer and later with 1 ml 1× binding buffer (12 mM HEPES, pH 7.9, 60 mM KCl, 5 mM MgCl, 1 mM DTT, 0.5 mM EDTA, 0.05% NP-40, 50 μg/ml bovine serum albumin, 10% glycerol). The final pellet was resuspended in 100 μl 1× binding buffer. Randomized duplex DNAs were prepared with PCR using following oligonucleotides (10 ng of NT55, 5′-CTGTCGGAATTCGCTGACGT(N)CGTCTTATCGGATCCTACGT-3′, 0.1 μg of UpNt, 5′-CTGTCGGAATTCGCTGACGT-3′ and 0.1 μg of DwNt, 5′-ACGTAGGAT CCGATAAGACG-3′ as a template and primers for extension reaction, respectively). Duplex DNAs were labeled 10 μCi [α-P] dATP for the easy chase of the bound DNAs with the PCR reaction containing 5 U of i-StarTaq DNA polymerase (Intron Biotech), 0.2 mM each of dGTP, dTTP and dCTP and 10 μM dATP. PCR was performed for 25 cycles (95°C 30 s; 65°C 1 min; 72°C 1 min). The labeled DNAs were allowed to bind to the fusion protein immobilized on the agarose beads at room temperature for 30 min with rotation. The bound DNAs were washed three times with 1 ml of 1× binding buffer, eluted by phenol: chloroform extraction, and finally precipitated ethanol. The eluted DNAs were amplified again with the same conditions described earlier for another round of DNA-binding. The following PCRs were performed only for 10 cycles. After five rounds of DNA-binding and amplification (Supplementary Data 4), the DNAs were subcloned into pCR4-TOPO vector (Invitrogen). For each fusion protein, 40–60 clones were purified and sequenced. The identified DNA motifs for each fusion protein were further analyzed with gel shift assays (Gel shift Assay System, Promega, Madison, WI, USA). About 10–20 μg of each fusion protein was used for each experiment with the [γ-P] ATP-labeled duplex probes prepared from the following oligonucleotides: CSE2-A, 5′-CCCACCCACCTGGGAATGAAAG-3′, and CSE2-B, 5′-CTTTCATTAAAGATGGCGCCCAGGTGGGTGGG-3′; 2a-A, 5′-CCCACCCACCTGGGAATGAAAG-3′, and 2a-B, 5′-CTTTCATTAAAGATGGCACCCAGGTGGGTGGG-3′; 2b-A, 5′-CCCACCCACCTGGGAATGAAAG-3′, and 2b-B, 5′-CTTTCATTAAAGATGGTGCCCAGGTGGGTGGG-3′; Probe1-A, 5′-GATAAGACGCGGAACGTCAGCG-3′, and Probe1-B, 5′-CGCTGACGTTCCAAATGGCTGCCGCGTCTTATC-3′; Probe2-A, 5′-GATAAGACGAGGCCCACGTCAGCG-3′, and Probe2-B, 5′-CGCTGACGTGGGCCTCAAAATGGCTGCCGTCTTATC-3′; Probe3-A, 5′-GATAAGACGCGGAACGTCAGCG-3′, and Probe3-B, 5′-CGCTGACGTTCCTAATGGCTGCCGCGTCTTATC-3′; Probe4-A, 5′-GATAAGACGAGGCCCACGTCAGCG-3′, and Probe4-B, 5′-CGCTGACGTGGGCCTCATAATGGCCGTCTTATC-3′; Probe5-A, 5′-GATAAGACGAGGCCCACGTCAGCG-3′, and Probe5-B, 5′-CGCTGACGTGGGCCTCAAAATGGCCGTCTTATC-3′; Probe6-A, 5′-GATAAGACGAGGCCCACGTCAGCG-3′, and Probe6-B, 5′-CGCTGACGTGGGCCTCAAAATGGCTGTCTTATC-3′; Probe7-A, 5′-GATAAGACGAGGCCCACGTCAGCG-3′, and Probe7-B, 5′-CGCTGACGTGGGCCTCAAAATGGCGGTCTTATC-3′. To monitor our gel shift assays, we also performed a set of control experiments using endogenous YY1 from HeLa nuclear extracts (Promega). The protein sequence of human YY1 (GenBank accession no. NP_0034941, 414 amino acid long) was used to search databases to identify YY1-related sequences from all available genome sequences. One YY1 homolog, known as PHO, was identified from each of the flying insects, including mosquitoes, honeybees, beetles and 10 different species of flies. In flies, a similar sequence, known as PHOL, was identified from each of the 10 different fly species. This totals to 23 different YY1-related sequences from insects. Database searches identified 39 different YY1-related sequences in chordates, ranging from urochordates (sea squirts) to placental mammals: one each from sea squirts and purple sea urchins, six from fish, one from frog, one from chicken, 29 from mammals. In fish, two copies of YY1 sequences were identified from each of three sequenced genomes, zebrafish, pufferfish and spotted pufferfish whereas three copies of YY1-related sequences were identified from each placental mammal. Database searches have identified a total of 62 YY1-related sequences. Based on sequence similarity, these are categorized into five groups: the PHO and PHOL groups from flying insects, the YY1 group from vertebrates, and the YY2 and REX1 groups from placental mammals (). Individual sequences and other related information are available through the following website (). Comparison of the amino acid sequences derived from the YY1-related sequences identified three evolutionarily conserved protein domains (). These include two domains in the middle of the protein (amino acid position 203–226 and 250–281 in the human YY1, respectively), and one DNA-binding zinc finger domain at the C-terminus (aa 298–414). The two domains in the middle, Domains I and II, are located within the region previously known as the Spacer between several N-terminal domains and the C-terminal DNA-binding domain (). These two domains are found in the YY1-related sequences of most, but not all, vertebrates and insects. In flies, only Domain I is found in both PHO and PHOL sequences. In placental mammals, two domains are found in YY2, but only Domain II is found in the REX1 sequences. However, the zinc finger domain is found in all the YY1-related sequences with high levels of sequence conservation, ranging from 66 to 100% similarity. The relative positions of these three protein domains are also conserved among all the identified sequences. The conservation of these three domains in the YY1-related sequences suggests that these three domains constitute the original domain structure of the YY1 protein. Several lineages have more than one copy of YY1-related sequences, including flies, fish and placental mammals. Two copies of YY1-related sequences, PHO and PHOL, are found in all the fly species examined to date while only one copy, PHO, is found in the other flying insects. This suggests a gene duplication unique to the fly lineage. According to the results of phylogenetic analyses (A), the topology of the two gene trees corresponding to the PHO and PHOL groups in flies is very similar to that of the known species tree of the fruitfly genus , indicating that this gene duplication predates the radiation of all fly species. The PHO sequences of the other flying insects show slightly greater levels of sequence similarity to the PHO rather than PHOL sequences in flies, suggesting that PHO is the original sequence that gave rise to the duplicated copy PHOL. This is further confirmed by the different exon structures of PHO and PHOL (A). The coding region of PHO is split into five exons, and a similar split exon structure is also found in the PHO of other insects, such as beetles and honeybees. In contrast, the entire coding region of PHOL is located within one exon, an intronless structure of its coding region. This intronless genomic structure is usually observed in the sequences that have been duplicated through an RNA-mediated mechanism, retroposition, by which processed mRNAs are reverse-transcribed and transposed to other genomic loci without introns in germ cells (). These data therefore indicate that PHOL has been duplicated from PHO through retroposition. The two copies of YY1 sequences found in the fish lineage show an almost identical sequence and exon structure to each other (data not shown). Chromosome-wide duplications are known to have been prevalent at the early stage of the fish genome evolution (,). Therefore, the two copies of YY1 present in each fish genome are thought to be another outcome of this chromosome-wide duplication event. In contrast, the two additional copies in placental mammals, YY2 and REX1, show quite different evolution patterns. First, like PHOL, the coding regions of both YY2 and REX1 are also located within one exon while the YY1 genes of all vertebrates show a very similar split exon structure with five coding exons (B). This suggests that both YY2 and REX1 were also duplicated from YY1 by retroposition. The detection of YY2 and REX1 exclusively in placental mammals further suggests relatively recent formation of these two copies during mammalian evolution with the estimated time being about 60–100 million years ago. In mammals, both YY2 and REX1 are transcribed and maintain their Open Reading Frames (ORFs), confirming the functionality of these two retroposed copies. Second, despite this recent origin, inter-species sequence divergence levels of YY2 and REX1 are much greater than those of YY1, as reflected on the phylogenetic tree shown in B. Very low levels of sequence divergence are observed between all the YY1 sequences of different vertebrates whereas each sequence from the YY2 and REX1 groups exhibits average 20% divergence between different species. This indicates relaxation of evolutionary constraints on both the YY2 and REX1 genes. As compared to REX1, YY2 displays greater levels of similarity to YY1 in terms of its overall sequence and protein domain structure, suggesting that the retroposition of YY2 may have occurred in more recent times than that of REX1. Pairwise sequence comparison also revealed that both YY2 and REX1 share higher sequence identity with YY1 than each other (Supplementary Data 3), suggesting that both REX1 and YY2 have been independently derived from YY1. The presence of two conserved domains, Domains I and II, in YY2 also supports the idea that YY2 has been derived from YY1, not from REX1, since REX1 has only Domain I. Overall, exon structure and sequence conservation levels suggest that the two retroposed copies, YY2 and REX1, have been under different levels of functional constraints than the original gene, YY1. All the YY1-related sequences show very unusual levels of sequence conservation in the DNA-binding domain of the predicted proteins (). The zinc finger domains of PHO and PHOL from all of the different fly species share 5 and 18 amino acid differences, respectively, as compared to those of vertebrate YY1. The zinc finger domains of the other flying insects, however, show an almost identical sequence to those of vertebrate YY1. Thus, the observed amino acid differences in flies represent the substitutions that had occurred in the fly lineage. Apparently, the overall consensus sequence of flying insects’ PHO is still identical to that of vertebrate YY1. Similarly, the zinc finger domains of vertebrates’ YY1 also do not show any shared substitution except for one or two species-specific amino acid changes. Thus, YY1 is believed to have maintained its DNA-binding domain without any amino acid changes in the past 600 million year period, representing one of the most extreme cases for functional selection imposed on an eukaryote gene. As described earlier, YY2 and REX1 have been under different levels of evolutionary constraints since their formation in placental mammals. This is in stark contrast to the extreme conservation of YY1. The zinc finger domains of different species’ YY2 protein show an average of 6–11 amino acid differences as compared to that of YY1 (). None of these changes are shared among different mammals, indicating that these changes represent independent substitutions that occurred in each species. Similarly, the zinc finger domains of different species’ REX1 proteins also show an average of 11–20 amino acid differences between each other, implying a slightly higher level of relaxation of evolutionary constraint on REX1. As compared to vertebrate YY1, however, the zinc finger domains of all REX1 sequences share 8 amino acid substitutions (). These substitutions represent the changes that occurred and were fixed before the radiation of eutherian mammals. The sudden fixation of these substitutions might be an evolutionary remnant suggesting positive selection that might have occurred in the early stages of REX1 evolution, although our analyses point toward purifying selection with relaxed constraints for the REX1 evolution (Supplementary Data 5). Interestingly, most of these changes are localized within Fingers 1 and 4, and are also non-conservative amino acid substitutions from the original amino acid residues of YY1. In particular, the amino acid change T398N in Finger 4 is localized within the region known to contact directly with the bases of target DNAs (). Therefore, this change along with other amino acid substitutions in REX1 may have a functional outcome possibly allowing REX1 to bind to DNA motifs divergent from the YY1 DNA-binding motif. Similarly, the amino acid substitutions within YY2 also appear to be slightly more frequent in Fingers 1 and 4, suggesting the presence of different selection pressures on each zinc finger. However, none of YY2 changes appear to be located within critical regions for its DNA binding, predicting no major difference between the DNA binding motifs between YY1 and YY2. We have further investigated the functional consequences of different selection pressures imposed on the zinc finger domains of YY1, YY2 and REX1 by characterizing their DNA-binding motifs. For this experiment, the zinc finger domain of each protein was subcloned into the downstream region of the GST protein, expressed as part of a fusion protein in bacteria, fixed on agarose beads, and finally we allowed them to bind to duplex DNAs derived from randomized oligonucleotide sequences. After five rounds of selection, the bound DNAs were subcloned and sequenced (). In the case of YY1, 20 of 34 bound DNAs contain DNA motifs that have either a perfect match or 1 base difference from the known YY1 consensus sequence. All of the remaining 14 bound DNAs still show an almost identical sequence as YY1 but have an average of two base differences from YY1. Our approach used only the zinc finger domain of YY1, but most of the bound DNAs are identical to the known consensus sequence of YY1. This confirms the modular nature of the zinc finger domain of YY1, and subsequently the feasibility of this approach. In the case of YY2, 16 of 46 sequences contain DNA motifs similar to the YY1 consensus sequence. As with the YY1 fusion protein, the remaining sequences also contain a motif similar to YY1 with two base differences, confirming our initial prediction: there is no major difference between YY1 and YY2 motifs. Interestingly, however, most of the YY2-bound sequences have more than two binding motifs within the randomized portion of each sequence. About half of the bound sequences show two motifs in an opposite orientation, with the other half in the same orientation. In contrast to YY2, the DNAs bound by REX1 seem to be slightly different from those bound by either YY1 or YY2. The sequences bound by REX1 can be divided into two groups. These two groups can be represented by two slightly different consensus sequences (): Type 1 (5′-AGCCATTA-3′) and Type 2 (5′-GGCCATTA-3′). The consensus sequences of these two groups differ by the presence (or absence) of three bases (GGC) at the 5′-side. These two consensus sequences also show one unique difference at their 3′-side final position: all the DNAs bound by REX1 contain A instead of T. This is consistent with the amino acid change detected in the critical DNA binding region of REX1, T388N in . Despite these changes, the core sequences of the YY2 and REX1 binding motifs are still the same as that of YY1 (5′-CCAT-3′), suggesting that the conservation of the two fingers, and , may be responsible for maintaining a similar core target motif among these three genes. The DNA-binding motifs of YY1, YY2 and REX1 were further analyzed using gel shift assays (). In the case of YY1, we have used the same set of duplex oligonucleotides used in a previous study to demonstrate the subtle but unique property of YY1, methylation-sensitive DNA-binding (). As expected, the DNA-binding domain, as part of the GST-YY1 fusion protein, showed an almost identical pattern of DNA binding as endogenous YY1 protein (A). The GST-YY1 protein is methylation-sensitive: methylation on the upper strand is inhibitory to the binding (A, Lanes 1–4). One base change in this CpG site, either CpA or TpG, somewhat reduced the affinity of the YY1 binding, but still allowed YY1 binding to these probes (A, Lanes 5–6). The DNA-binding domain of YY2 also showed a similar pattern of DNA binding: methylation-sensitive binding and subtle effects by single base changes caused by the CpG site (B). However, the DNA-binding affinity of YY2 is much weaker than YY1 based on the results derived from our control experiments for gel shift assays (Supplementary Data 4). We have also tested some of the DNAs that contain two motifs within the randomized portion of the target DNAs (B). We did not observe any difference in binding between the duplex DNAs with two binding motifs versus single binding motif (data not shown). Overall, the DNA-binding patterns of YY1 and YY2 appear to be similar except for the fact that the binding affinity of YY2 is much weaker than YY1, consistent with the observed relaxation of evolutionary constraint on the DNA-binding domain of YY2. Several sets of gel shift assays were performed for the identified DNA-binding motifs of REX1 (Upper panel in C). The REX1 and YY1 fusion proteins were individually allowed to bind to seven duplex probes. These include three consensus motifs, the consensus of YY1 (Probe 7), the consensus of REX1 Type 1 (Probe 3) and Type 2 (Probe 4). We have also included four other probes containing one or two base variations from the three consensus motifs to further dissect the binding specificity of REX1 and YY1. The REX1 protein bound to the four probes containing REX1 motifs (Probes 1–4), but not to the YY1 or related probes (Probes 5–7). On the other hand, the binding of the YY1 protein to the REX1 probes was detected but very marginal compared to its binding to the YY1 or related probes (Probes 5–7). This indicates the different binding specificity between the YY1 and REX1 proteins. This different binding specificity is originated from three key differences found in the REX1 binding motifs as compared to the YY1 binding motifs. First, the REX1 motifs have A instead of T at the 8th position of the YY1 consensus (CGCCATNT). This change reduced dramatically the binding affinity of the YY1 protein, but increased the binding affinity of the REX1 protein (Probe 4 versus Probe 5). Second, the REX1 motifs do not show any base preference at the 9th position of the YY1 consensus (CGCCATNT). Interestingly, the T base at this position reduced slightly the binding affinity of the REX1 protein, but is required for the binding of the YY1 protein (Probe 1 versus Probe 2). Third, one of the REX1 motifs contains additional three bases (5′-GGC-3′) at the 5′-side of its sequence. The addition of these three bases reduced the binding affinity of the YY1 protein, but increased the affinity of the REX1 protein (Probe 2 versus Probe 6). The significance of these key differences was further demonstrated by competition assays using three representative probes (Lower panel in C). Overall, these data clearly demonstrate the different binding specificity between the YY1 and REX1 proteins, and also prove that the two identified motifs, Types 1 and 2, represent bona fide DNA-binding motifs for REX1. The positions of these three critical base differences in the surrounding regions of the core motif (5′-CCAT-3′) are consistent with an observed evolution pattern (), differential selection pressures on each of the four zinc finger units of the REX1 protein. In the current study, we have analyzed all the YY1-related sequences identified from genome sequences of invertebrates and vertebrates. We have identified two other protein domains, besides the zinc finger domain, that are conserved throughout all the YY1 and YY1-related sequences. Our analyses also confirmed that independent retroposition events have been responsible for forming duplicated copies, such as PHOL from PHO in flies, and YY2 and REX1 from YY1 in placental mammals. The zinc finger domains of YY2 and REX1 have been under different selection pressures than YY1, and consequently their DNA-binding properties have evolved from those of YY1 by weakening DNA-binding affinity in YY2 and REX1, and changing DNA-binding motifs in REX1. The evolution patterns of YY1 and other YY1-related proteins appear to be unique in several regards, as discussed subsequently. Besides the zinc finger domain, two other protein domains, Domains I and II, are evolutionarily well conserved throughout all the YY1-related sequences ranging from flying insects to mammals (). The conservation of these two domains is somewhat less obvious within the sequences of flies, but the detection of these domains within the PHO sequences of honeybees and beetles undoubtedly indicates that these two domains are part of the original domains of YY1. Database searches with these two domains did not find any proteins other than YY1 or YY1-related sequences, suggesting that these two domains are unique to YY1-related sequences (data not shown). According to previous studies analyzing protein–protein interactions, the Spacer region, a relatively large region of YY1 (aa 201–298 in human YY1) encompassing these two domains, is responsible for the interaction with the viral oncoprotein E1A and the p53-interacting partner Hdm2 (,). It should be interesting to test whether YY2 and REX1 also interact with the above two proteins. Nevertheless, the functional roles played by these two domains are predicted to be essential for YY1 functions based on their conservation in most of the YY1-related sequences. There are several key transcription factors with similar evolutionary ages as YY1, such as Sp1 and the E2F family of proteins. These transcription factors have increased their gene copy numbers along with the increase of complexity and genome size of animals (,), but the duplication of these genes has been mainly driven by DNA-mediated mechanisms involving the entire genomic fragments surrounding individual genes (,). That is, in the Sp1 and E2F families, the whole gene structure has been duplicated with exons, introns and promoters intact. In the case of YY1, however, retroposition has been the primary mechanism for its duplication: PHOL duplication from PHO, and YY2 and REX1 duplications from YY1 (), which is quite different from the general duplication mode observed in other key transcription factors. A gene copy duplicated through retroposition is subject to transcriptional controls different from those of its original gene due to its random insertions at other genomic regions. As an outcome, the duplicate copy tends to show different expression patterns compared with its original gene. Consistently, both YY2 and REX1 also display expression patterns quite different from that of YY1. As compared to the ubiquitous expression patterns of YY1, YY2 shows more germ cell-specific expression patterns (), and REX1 exhibits stem cell-specific expression (). It is still puzzling why YY1 duplication has been driven by retroposition, but different expression patterns resulting from this duplication mode may have been one major factor contributing to the success of YY1 duplications in placental mammals. The evolutionary patterns observed with YY2 and REX1 are quite different from that of YY1 (). YY1 shows high levels of sequence conservation throughout its coding region. In particular, the zinc finger domain of YY1 has maintained its amino acid sequence without any changes in the past 600-million year period, implying that the YY1 homologs, insect PHO and vertebrate YY1, may still bind to similar DNA motifs. This turns out to be the case based on DNA-binding motif studies (,). In contrast, the zinc finger domains of YY2 and REX1 show much higher levels of inter-species sequence divergence, suggesting relaxed constraints on their DNA-binding domains. Consequently, both YY2 and REX1 display much weaker DNA-binding affinity than YY1 ( and Supplementary Data 4). The loosened DNA-binding affinities of YY2 and REX1 may have allowed these duplicates to bind to slightly different binding motifs, as seen in REX1 (), and subsequently to bind to new sets of downstream genes. Together different expression patterns, loosened affinities and different DNA-binding motifs may have contributed to the functional diversification of the two duplicates, YY2 and REX1, in the mammalian lineage. Successful gene duplication is still regarded as a rare evolutionary event (), which is further supported by the single-copy status of YY1 in the majority of animal lineages. Then, what could be the main reason(s) underlying the sudden formation of two YY1 duplicates in placental mammals? This may be indirectly answered by observations drawn from other gene duplicates in mammals. For instance, DNMT3L is a member of the DNA methyltransferase family, which is found only in mammals (). Yet, DNMT3L has been found to be involved in genomic imprinting (), a gene dosage control mechanism unique to placental mammals (). CTCFL (or BORIS), a mammal-specific duplicate of the vertebrate insulator protein CTCF (), might be involved in establishing the gametic imprinting mark of DNA methylation for H19 during germ cell development (). In both cases, gene duplicates appear to play specific roles in mammal lineage-specific novelties, such as genomic imprinting and epigenetic modification. These two duplicates, interestingly, share some similarities with the YY1 duplicates, YY2 and REX1, such as recent formation, rapid evolution, lineage-specific conservation in mammals and germ cell-specific expression (). Furthermore, recent studies suggest that several imprinted domains may be controlled by YY1 or related transcription factors (,). This entices the speculation that both YY2 and REX1 may be also involved in novel placental mammal-specific functions, such as genomic imprinting. This idea needs to be tested, but the evolutionary patterns presented in this study clearly indicate the tight linkage of both YY2 and REX1 to the biology of placental mammals. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Field-deployable biosensors require more rapid and sensitive, single-step identification methods. However, efforts to enhance assay rapidity, sensitivity and simplicity can result in an increase in false positives or in false negatives (). Such false positives and negatives can have immense impact in biosensing for medical and biowarfare applications as even rare occurrences can have disastrous consequences. A paradigm shift in understanding and designing for the specificity–sensitivity trade-off is absolutely essential to developing field-deployable biosensors experiencing few to no false positives and negatives. Molecular beacons (MBs) are one probe methodology used to move towards more rapid, single-step genomic sensors (,). A fluorescent label is attached to one end of a polynucleotide and a quencher is attached to the other. Complementary base-pairs near the label and quencher cause a hairpin-like structure, placing the fluorophore and quencher in proximity. This hairpin opens in the presence of the target producing an increase in fluorescence. The proximity of the quencher to the fluorophore can result in reductions of fluorescent intensity of up to 98% (). The perceived efficiency can further be adjusted by altering the stem strength (length of the stem) which affects the number of beacons in the open state in the absence of the target. Accordingly, one trade-off a MB experiences is with regard to its stem strength, low strength limits fluorescent increase upon hybridization whereas high strength limits kinetics of hybridization (). Regardless of the detection platform or strategy, the majority of biosensors incorporate molecular recognition through a biological affinity interaction. A biosensor cannot be more accurate than this interaction (). This interaction is used for one or more functions that include identifying the presence of a given analyte, determining changes in expression level and quantifying the agent (). Specificity and sensitivity in biosensor research often refer to the ability of the sensor to eliminate false positives and negatives respectively for one or more of the foregoing objectives. Unfortunately, there is usually a trade-off between specificity and sensitivity (,) as shown in . In support of the limitations in improving biosensor accuracy, the numbers tell a compelling story. By way of identification of the presence of specific species, Peplies . tested six strains of bacteria with a 1% rate of false positives and 41% rate of false negatives (). Diagnostic PCR, although rarely having false-negatives owing to its extreme sensitivity, experiences a reported rate of false-positives between 9 and 57% (). Detectors monitoring expression level are worse having a 10–30% false negative rate for samples the sensitivity threshold and a false positive rate of 10% (). While this rate of false positives and negatives may be damaging for phenotypic or other biological exploration, even one error can prove lethal in clinical diagnostics and could result in loss of life (false negative) or economic harm (false positive) in homeland security applications. In order to overcome the deficiency with these reagents and assays, we have developed a new class of reagent that performs with heightened affinity and with greater selectivity compared to single oligomer reagents. We adapted the principles of cooperativity abundantly described in cell targeting () to combat the specificity–sensitivity trade-off. Cooperativity has already been shown to enhance SNP detection and assay sensitivity (,). Here, we combine principles of cooperativity with a label-free hairpin in what we term tentacle probes (TPs). We discuss the effects of cooperativity on sensitivity, specificity and kinetics, demonstrating an increase in each without sacrificing the others. These faster, more sensitive and more specific probes may offer a number of advantages in many of the diagnostic applications where MBs have already been applied. In this study, we derive a mathematical model predicting the cooperative behavior of TPs. We test theoretical descriptions of increased kinetics and specificity over MBs and demonstrate these results without a loss in sensitivity. By so doing, we introduce the first class of reagents to our knowledge to be originated as a mathematical solution to problems associated with molecular recognition in biosensors. Four target sequences were synthesized representing the wild-type target (WT), a single nucleotide polymorphism (SNP) in the capture region of the TP (SNP), a SNP in the detection region (SNP) and a SNP in both regions (SNP). All probes and target sequences were suspended in TE buffer (10 mM Tris-HCl, 1 mM EDTA, pH 7.0) with 0.18 M NaCl and 0.1% SDS and are summarized in . Fluorescent data was plotted against time and the rate constants were fit using the kinetic equation for polynucleotide reactions in an excess of target by minimizing the sum of square errors. The rate constants from each of three trials were averaged and plotted against stem strength with 95% confidence intervals. The fluorescence ratio of specific (WT) to nonspecific (SNP) targets was calculated for the 50-nM target concentration and plotted versus temperature. The TP and MB with the largest peaks were selected as the optimum probes after verifying first that their kinetic rate constants were each above 1000 M s. Melting curves were fit to an adaptation of models used by others in describing MB thermodynamics (,). These models were enhanced by the addition of a cooperative stem and the possibility of more configurations (). Cooperative models are derived from collision theory, where the rate of the first binding event is proportional to the product of the reagent concentrations: The second binding event follows the same model, but with an enhanced local probe concentration,  = 1 molecule/(volume swept out by linker length × Avogadro's number), reacting with the newly formed complex, , and with an adjusted cooperative reaction probability, , reflecting enthalpic and entropic penalties: Assuming no cross-reactions, the following equilibrium constants for TPs can be derived: Fluorescence as a function of temperature can be adapted from models for MBs (,) by: Alpha, beta and gamma refer to characteristic fluorescence of bound probes, closed probes (subscript cl) and random coil probes (subscript op) respectively. For T ≫ P, which eliminates calculation of quadratics: is fit first by measuring fluorescence as a function of temperature for beacons with no target and minimizing sum of square errors as performed by Bonnet . and Tsourkas . (,): Next is fit on probes with no capture probe (e.g. on an MB): The thermodynamic parameters necessary for calculating were estimated with Mfold () for 0.18 M NaCl (about the same sodium concentration as 1 × SSC). Using the Mfold estimates provided, theoretical curves that diverged at ∼61°C for differing concentrations of target with TP 5, while adjusting the predicted entropy from −0.4989 to −0.494 kcal mol K moved the divergence to ∼65°C forming a more accurate visual fit to the data. Therefore the latter value was used for all remaining tests. Finally, was fit to the fluorescent curves of three different dilutions of target mixed with TPs using the original equation. All parameters necessary for calculations are summarized in . The best fit enthalpies and entropies were used to calculate the equilibrium constants which in turn were used to calculate the amount of analyte bound to the detection probe and producing fluorescence as a function of temperature in an excess of target for MBs for specific (subscript s) and nonspecific (subscript ns) analyte: And for TP: These equations were then matched with normalized fluorescent data in order to verify accuracy of thermodynamic parameters in predictions at the lowest detectable binding levels and to identify trends in binding below the level of detection. Thermodynamic values provided from best fits often fit high-level binding tightly at the expense of fitting low-level binding data due to inequalities arising from the sum of square errors. A manual adjustment to the best fit (e.g. ±0.6 kcal mol for enthalpy and ±0.0016 kcal mol K for entropy) provided a fit that more thoroughly represented both high and low binding data. Once the parameters provided a perfect visual fit to low-level binding as well as high-level binding, they were kept constant in order to make predictions. The Stratagene Mx4000 plate reader was used to read the fluorescence of 1-μM nine-base stem TP and 1-μM five-base stem MB (both probes chosen as described in ) in WT and SNP targets at concentrations of 0, 2, 10, 20, 100 nM, 1 and 10 μM. Higher concentrations of 100 μM and 1 mM SNP were also used where detection limits had not yet been established. Fluorescence was read at equilibrium for both probe types. The reading was performed at 60°C for the TP and 55°C for the MB. Three replicates of each type were performed. The predicted binding curves as a function of target concentration were generated using best fit thermodynamic parameters for MBs: And for TP: These predictions were compared to experimental data plotted with 95% confidence intervals. Before plotting, data was normalized by subtracting the background fluorescence measured at 0 nM target concentration and dividing by the maximum intensity experienced at 100 μM WT target at room temperature. The background fluorescence level plotted was the average plus one standard deviation of the signals of experiments run below and including the highest concentration that could not be statistically confirmed as having fluorescence greater than the preceding concentrations (-test,  > 0.05). TPs have measured rate constants ranging from 100 to 200 times larger than their MB counterparts (). In comparison with literature, the TP rates are several times faster than those reported for standard linear DNA-probe-binding kinetics and stand in contrast to the MB rates which are slightly slower than literature values (). The faster rates of TP over linear probes can be explained by the presence of two probes, a capture and a detection probe, providing for increased probability of reaction. In addition, these probes are longer than the linear probes used by Tsourkas . (). The MB's slower reaction than what was reported in literature can be explained by their longer stems and the fact that the reactions were monitored at room temperature instead of 37°C. Although molecular beacons are not expected to perform well at room temperature, this setting was selected to better contrast TPs with MBs. The rate constants indicate that TPs react in ∼1% of the time that MBs with the same stem strength require. The room temperature rates of TPs are still >10-fold faster than those reported by Tsourkas . using MBs with short stems and higher reaction temperatures (). Melting curves were used to extract thermodynamic constants (, ). As was expected, the enthalpy of the stem increased with stem length, whereas the enthalpy of the hybridization reaction decreased. The enthalpic and entropic penalties also decreased with increasing stem strength, as did the kinetic rate constants. The only unexpected finding was that for the shorter stems (five and six bases), the stem melting temperature varied greatly between TP and MB. Since these two probe types were constructed identically including choice of dyes and even the attachment of a PEG linker, the only other explanation for this difference in melting temperatures appeared to be that the proximity of the capture probe allowed for some interaction with the stem-loop structure. Tsourkas . have performed the most extensive work on thermodynamics in MBs (). Their data for MB enthalpies of reaction ranged from −80 to −221 kcal mol and entropies of reaction ranged from −0.21 to −0.61 kcal mol K. While these numbers appear similar to what was recorded in our experiments, it should be noted that our beacons were larger, which should produce larger enthalpies, and had stronger stems, which should result in lower net enthalpies for probe–target interactions. Therefore a direct comparison cannot be made. However, we feel confident in the accuracy of the thermodynamic parameters recorded inasmuch as adding the energetic losses due to the stem to the probe enthalpy (−Δ + Δ) produces consistent values near the predicted enthalpies for binding to linear probes of −175.3 kcal mol and similarly for the predicted entropies of −0.4936 kcal mol K (). The capture probe parameters were calculated on Mfold. The predicted enthalpy and entropy of reaction were −175.8 kcal mol and −0.4989 kcal mol K, respectively (). It should be noted that the design criterion for both the detection probe and capture probe affinities was a melting temperature 10°C above the desired reaction temperature. The optimal TP stem strength was designed to have a melting temperature 30°C above the reaction temperature. In practice, in order to achieve specificity, we had to increase the reaction temperature such that the predicted melting temperatures for the detection and capture probes were 5°C below the reaction temperature, whereas the predicted stem melting temperature was only 15°C above the reaction temperature. Once the parameters were fitted, they could be used to calculate binding curves and be double checked for accuracy by comparing theoretical curves against normalized data in a semi-log plot (). These predictions reveal a slightly different binding pattern for TP than for MB. TP exhibit a slight bend in the binding curves ∼70°C. This is mathematically due to the melting of the capture probe, causing an instant loss in the cooperative interaction and consequent signal. Optimal candidates were selected by looking at the specific to nonspecific signal ratio for each stem at 50 nM probe and target concentrations (data not shown). The optimal candidates were then verified with the kinetic results in order to assure that the kinetics were ideal as well. Since TP 9 had the greatest signal differentiation and its kinetics were still much faster than the fastest MB, it was selected as the optimal TP. MB 5 was the fastest and most selective of the MBs, so it was chosen as the best MB. Negligible differences exist between the melting curves of WT and SNP and between SNP and SNP for both the TP and MB (, data not shown for MB). This is most likely due to the relatively high affinity of the capture probes. By utilizing a capture probe which does not respond to SNPs, the location of a SNP can be pinpointed to the region of the detection probe. However, if greater selectivity is required, it may be possible to design the capture probe to respond to SNPs by reducing the probe length. This also indicates that the design of the capture region is probably not as significant as the choice of the detection probe. Because of the cooperative interaction of TP, their melting curves do not change significantly over a wide range of concentrations (). A reaction temperature can be chosen for TPs (vertical line in A) such that there is 100% accuracy in SNP determination based merely on whether or not a signal is detected. This would be ideal for real-time PCR in SNP discrimination. Additional mismatches, such as are common in organism detection, should allow for greater resolution. In contrast, the melting temperatures of MB change with concentration making it difficult to find a reaction temperature which allows for discrimination over a wide range of concentrations. We performed an experiment at the optimum SNP resolution temperatures for both TP (60°C) and MB (55°C). Theoretical predictions were graphed alongside the experimental results for binding as a function of target concentration (). The theoretical predictions appeared to describe experimental data well. TP isotherms reveal wild-type detection limit of 15.4 nM and no SNP detection at concentrations tested up to 1 mM. The model predictions indicate that binding to mutant targets possessing a SNP will never be sufficient to cause a signal above the background resulting in false positives regardless of how high the concentration is. In contrast, mutant targets resulted in false-positive signals for MB at concentrations above 3.88 μM, 154 times greater than the detection limit for the wild-type target of 22.7 nM. The ratio of specific to nonspecific detection limits for the TP was tested in excess of 53 200 and is predicted to increase indefinitely. In contrast with , TP specificity was improved without sacrificing sensitivity. Due to the thermodynamic principles that govern molecular interactions, it is extremely difficult to develop an assay that increases specificity without sacrificing sensitivity. Likewise, sensitivity is not easily increased without a loss in specificity. We have developed a new class of reagents that does not improve specificity or sensitivity via a trade-off, but that improves overall assay accuracy as indicated by the ratio of detection limits by utilizing cooperativity. The difficulty in increasing sensitivity or specificity without sacrificing the other lies in the fact that efforts to address assay performance are often directed at instrumentation and buffers. For example, the temperature is raised to increase specificity or lowered to increase sensitivity. Salt concentrations are raised and lowered to optimize binding. Detection volumes are decreased and filter sets are optimized to increase sensitivity (,). However none of these methods addresses the fundamental issue of how affinity reagents recognize and bind an analyte. Altering instrumentation does not affect thermodynamic parameters governing binding and thus only affects the threshold of detection. Altering the threshold merely results in a sensitivity–specificity trade-off as demonstrated by receiver operating curve analysis (). Raising the temperature or decreasing salt concentrations lowers the equilibrium constant for both specific and nonspecific binding, leading to greater specificity at the expense of sensitivity. Likewise decreasing the temperature or increasing salt concentrations increases the equilibrium constant for specific and nonspecific binding, resulting in increased sensitivity at the expense of specificity. None of these methods decreases the nonspecific equilibrium constant while maintaining or increasing the specific equilibrium constant. Accordingly, trade-offs in specificity and sensitivity can be achieved but the ability of the affinity reagent to discriminate between specific and nonspecific targets remains largely unchanged. TPs represent a novel innovation inasmuch as they are mathematically engineered to manipulate thermodynamic principles that govern affinity reagent recognition of a biomolecule. Collision theory dictates that a second binding event from a molecule already bound to a substrate and held at an enhanced local concentration will typically occur at a faster rate than it would in free solution. This principle was used to develop mathematical models that predicted both enhanced kinetics and specificity of cooperative probes. For a homovalent probe pair with no penalty term, the effective equilibrium constant reduces to  = (2 +  · ), which is identical to antibody theory as originally derived by Crothers and Metzger () and as embodied by Kaufman and Jain (). Others have also used this approximation for modeling bivalent interactions and have experimentally confirmed its accuracy (). However, we are the first to our knowledge to use this model to examine the specificity–sensitivity trade-off in biosensors. These derivations of the equilibrium constant apply a different approach from that which has already been done for multivalent systems (,,,). However, the result confirms thermodynamic estimates where the avidity is equal to the sum of the free energies of each individual reaction plus an interaction effect, which is typically an entropic penalty (,). It might be expected in cooperative interactions that an SNP would not be as easily recognized due to an overall higher binding affinity. However, in our experiments, the bivalent accuracy was >345-fold greater than the monovalent accuracy (the ratio of concentrations at which the signal is greater than the detection limit, from ). This added enhancement to selectivity in our experiments is due to the unique structure of the TPs. By utilizing only one probe in the pair as the detection probe, cooperativity can be designed to allow theoretically asymptotic separation of specific and nonspecific detection limits. For analyte concentrations greater than probe concentrations, the capture probe has a sufficiently high affinity to present the detection probe with a constant local concentration of bound target. This allows for detection that is independent of increasing concentrations, effectively reducing the mechanism in to the top two reaction states. With no shift in the melting curve for increasing concentrations, SNPs can be identified with greater confidence using TPs than with conventional probes. For analyte concentrations below the probe concentration, melting curves were observed to shift with increasing analyte concentrations (data not shown). We believe the reason that concentration independence occurs only for target concentrations above the probe concentration is due to target saturation of the detection-probe-binding site. For relatively low target concentrations, the detection-probe-binding sites are not saturated and can continue to bind more analyte as it is added to solution, presenting a variable concentration to the detection probe. However, once the capture probes have been saturated with analyte, a constant local concentration is presented to the detection probes, causing concentration independence of reaction. Having biphasic concentration dependence is an advantage of TPs. Quantification of analyte can be performed for anlayte concentrations below the probe concentration. However, once analyte concentrations exceed probe concentrations entering into the nonlinear region of binding where quantification is no longer possible, melting curves cease to be affected by concentration and specificity is rendered concentration independent. As the analyte concentrations continue to increase, it might be thought that the unreacted hairpins in the detection probe would be forced to bind, such that each TP would be bound to two target molecules (one on the capture probe and one on the detection probe). However, as observed in the TPs with even the weakest stem, the melting curves still remained immobile with increasing concentrations (data not shown). The analogous MB, on the other hand, continued to show increases in fluorescence with increasing analyte concentrations (), demonstrating that the hairpin strength was not the cause of failure to fluoresce in TPs. Rather, the lack of an increase in fluorescence with increasing analyte concentrations in TPs demonstrates that a second molecule is not bound to the free detection probe. We believe that the mechanism preventing a second binding event to the detection probe at high analyte concentrations may involve the affinity between the detection probe and the captured analyte. While the affinity is not enough to keep the captured analyte firmly bound to the detection probe, we believe it is sufficient to keep the captured analyte close, continually binding and releasing it, thereby precluding a binding event from the bulk solution. Taken together, the constant local concentration and the ability to preclude binding of more than one molecule, these mathematically designed assets give TPs an accuracy in SNP discrimination that is matched by no other probe set to our knowledge. MBs, on the other hand, are known to possess greater specificity than linear probes due to the competition from stem formation (). Discrimination between wild type and mismatched targets is expected to increase with stem strength (). However, utility of stronger stems in MBs is compromised by reduced sensitivity from lower affinity and by slower kinetics. By combining the unique secondary structure of MBs with the idea of cooperative probes, we have created a label-free, highly specific reagent that has greater specificity than a MB, but that reacts up to 200 times faster. Again TPs show no sign of a trade-off. This may make the use of TPs in room temperature assays where MB reaction rates are extremely slow or in continuous flow detection where reagents passing by only have a short time to react an ideal candidate over MB. The sensitivity loss from stronger stems in MBs has been studied by Tsourkas . (). The loss in fluorescence experienced by MBs is replaced by gains in TPs with identical stem strengths because of the lower free energy available from cooperative binding. We collected thermodynamic parameters in order to compare with the results of Tsourkas . In order to get their thermodynamic parameters, they fit a line through the fluorescence at the melting temperature of six different concentrations of target. We deviated from this method because we felt it was statistically more relevant to use the entire set of fluorescent data for each concentration rather than a single point. Also, their method did not directly apply to fitting thermodynamic parameters for TPs. In addition to a melting curve which does not shift significantly with target concentration, reveals a difference between TP and regular probes in the amount of specific and nonspecific binding resulting in fluorescence as a function of temperature. Data points are plotted to where fluorescent levels are no longer distinguishable from the background. Since the actual data could not be observed at low amounts of binding, the fitted thermodynamic parameters were used to calculate binding as a function of temperature beyond where data could be collected. Since the models and fitted parameters account for the majority of the TP behavior, and because these models have been derived for other applications as previously discussed, we feel justified in their use to examine the thermodynamic patterns of cooperative binding. As a further justification of the use of the models in describing TP behavior, the models were used to predict binding as a function of concentration (). Theoretical predictions estimated that a target molecule possessing a SNP in the detection region would never be detected by TPs regardless of target concentration, virtually eliminating false positives. We were able to test this hypothesis to 1 mM SNP concentrations with no false detections. We note that the models were sufficiently accurate to predict this trend. We also note that the models predict the ability to concentrate nonspecific analyte to levels that are not physically possible without false positives. This is supported by the fluorescent signals of TP 9 at 60°C, which seem to be completely independent of target concentration (). The fact that we were unable to create SNP analyte concentrations sufficiently high for TP detection without sacrificing the sensitivity of wild-type detection illustrates the point of this article. This stands in stark contrast to any other probe set of which we are aware. By creating cooperative probes, we have been able to effectively reduce false positives without sacrificing sensitivity. This high level of specificity may find utility in a number of research and diagnostic applications. Real-time PCR, for example, is often required to perform SNP discrimination. While comparative analysis of a multiplexed assay often yields good results in a laboratory, field tests do not always have the same options of sample purity or multiplexing and those running the tests do not always have the capacity to interpret the results. Rather many field test units resolve the complexity of analysis by resorting to a signal/no signal report. Few probes possess the capacity to distinguish SNPs by the presence or absence of a signal. This problem is compounded in homeland security where a number of select agents like and have near neighbors with large portions of similar genetic content (,). TPs may again provide an answer to this real-world need. Although TPs have shown benefits in several areas, there is still room for improvement. This set of experiments only examined the effect of stem length on binding. Other options for adaptation include linker length and composition, capture and detection probe affinities, the distance between capture and detection regions on the target polynucleotide, and the interactions of each of these features. Given the large enthalpic and entropic penalties, there may be a great deal of room for improvement, potentially producing a much sharper slope in the melting curves. This in turn could lead to higher signal-to-noise ratios and greater detection limit ratios, while maintaining the incredible specificity. Six years ago, Iqbal . called for the creation of new affinity reagents as the most important means of improving biosensor accuracy (). We have created a new class of label-free affinity reagents, TPs. These special beacons manipulate thermodynamic principles in order to increase kinetics up to 200-fold and molecular accuracy in SNP detection by at least 345-fold, with predicted enhancements of near infinite improvement. Because of their cooperative interaction, they truly allow significant enhancements in sensitivity, specificity and kinetics without a trade-off.

Friedreich ataxia (FRDA) is the most common inherited ataxia (). The ataxia, which is recessively inherited, is relentlessly progressive with patients frequently becoming wheelchair-bound in their early teens. Hypertrophic cardiomyopathy is a common co-morbid feature of FRDA, with congestive heart failure being a frequent cause of death. The most common cause of FRDA is the expansion of a GAA•TTC-repeat (henceforth referred to as the repeats) in intron 1 of the frataxin gene (). Unaffected individuals have at least one allele with 8–33 repeats, while most individuals with FRDA have 90 or more repeats in both alleles (). These expanded alleles produce less mature mRNA than alleles in the normal range (,). This results in a deficiency of frataxin, an essential protein thought to be involved in mitochondrial iron metabolism (). The repeats form an intrinsic block to RNA polymerase (,), suggesting one potential mechanism for the mRNA deficit. However, the repeats may affect transcription in other ways. For example, the long stretches of tandem repeats present in normal human centromeres are associated with transcriptionally silent chromatin () as are the genes containing the expanded triplet repeat tracts responsible for Fragile X syndrome and congenital myotonic dystrophy (). In addition, transgenes containing long GAA•TTC-tracts become associated with heterochromatin when inserted into arbitrary locations in the mouse genome (). It is thus possible that repeat-mediated chromatin changes also contribute to the reduction in transcript in FRDA. In principle, changes in chromatin modifications in non-coding sequence could affect transcription by affecting RNA polymerase II elongation or by changing the accessibility of this region to regulatory factors important for transcription initiation (). Since the first intron of many genes contains regulatory sequences important for gene expression () and the sequence adjacent to the repeat is the region of intron 1 most likely to be affected by repeat expansion, we examined the contributions of this region to promoter activity. We also examined the chromatin modifications and DNA methylation status of this region in cells from unaffected individuals and individuals with FRDA. We have identified a region that is important for maximal gene expression and shown that repeat expansion leads to changes in DNA methylation and chromatin organization in this region. Our data also suggests how these changes may lead to the frataxin deficit responsible for FRDA. Primate sequences in the region of intron 1 immediately 5′ of the repeat were aligned using the online version of the Multalin algorithm [(); ]. The repeated DNA elements in this region were identified using the Repeat Masker track generated using the RepBase database (). This region was also analyzed for potential transcription factor binding sites using MatInspector and Matrix Family Library Version 4.1 (). Lymphoblasts from individuals without FRDA (GM06895, GM06865, GM06891 and GM09145), and those with FRDA (GM04079, GM15850, GM16207, GM16209 and GM16243) were obtained from the Coriell Cell Repository (Camden, NJ). Lymphoblasts were grown in RPMI 1640 medium supplemented with 10% fetal calf serum under standard conditions (Invitrogen Life Technologies Inc., Carlsbad, CA). Genomic DNA was prepared from these cell lines using standard procedures. The repeat numbers in normal individuals were determined as described earlier (,). DNA obtained from fresh blood of affected and unaffected individuals was a kind gift of Dr Ed Grabczyk, Louisiana State University. The plasmid construct expressing the M.eco72I methylase was a kind gift of Fermentas AB (Vilnius, Lithuania). The primers and double-stranded oligonucleotides used were synthesized by Integrated DNA Technologies (Coralville, IA) and are listed in . Genomic DNA from cell lines of four unaffected and four affected individuals was bisulphite modified according to standard procedures except that the bisulphite treatment was carried out overnight at 55°C. The methylation status of the promoter and intron 1 were examined. The promoter was analyzed using the primers Me74-F and Me956-R. The region 5′ of the GAA repeat in the gene was amplified by PCR using the primers Me1212-F and Me1930-R. The PCR fragments were gel purified, digested with XhoI and EcoRI, cloned into pBS SK+ (Stratagene, La Jolla, CA, USA) and sequenced. Five clones from each individual were sequenced. The data for each population was pooled and the frequency with which each residue was methylated was then plotted along with the standard deviation. The primer PrF was used as the forward primer for all constructs. This primer has an MluI site at its 5′ end and primes 1118 bp upstream of the start of translation. Each of the reverse primers contained the splice acceptor site from the 3′ end of intron 1 and an NcoI site and were designed so that the coding sequence in exon 1 would be translated in frame with the luciferase open reading frame from pGL3-Basic (Promega, Madison, WI, USA). After PCR amplification, the gel-purified fragment was digested with MluI and NcoI and cloned into pGL3-Basic. The numbers in the construct names are based on the location of the reverse primer in the chromosome 9 sequence from the May 2004 freeze of the human genome in the UCSC database with the numbering starting at 68 800 000. Thus the 3′ end of the sequence in construct 80940 corresponds to base 68 880 940 in the chromosome 9 sequence. 81423 contains 1006 bases from the 5′ end of intron 1 terminating at position 68 881 423, and 81658 contains the first 1241 bases of intron 1 and terminates at position 68 881 658. E-box/Mt+ and E-box/Mt− clones were generated using reverse primers that contain either the intron E-box/Mt sequence or the sequence immediately downstream of this region. The resultant clones differ only by eight bases. The second C in the E-box core sequence 5′-CACGTG-3′ can be methylated by M.eco72I, a DNA methyltransferase from RFL72 (). Methylation was carried out as previously described (), and the extent of methylation checked by digestion with PmlI, a restriction enzyme whose cleavage is blocked by M.eco72I methylation. Reporter assays were carried out using mouse C2C12 myoblasts as previously described (). To assess any differences in the RNA stability of these constructs, cells were treated 24 h post-transfection with 5 µg/ml actinomycin D. RNA and protein samples were collected after 8 h. The luciferase protein levels were determined as previously described (). The luciferase mRNA levels were determined by real-time PCR as described below. The RNA and protein values after 8 h of actinomycin treatment were calculated as the percentage of the values obtained at 0 h. A double-stranded oligonucleotide E-box was radiolabeled using [γ-P] ATP (MP Biomedicals, Solon, OH) and T4 polynucleotide kinase (New England Biolabs, Beverly, MA, USA) according to standard procedures. Nuclear extracts were prepared from C2C12 cells using NE-PER Nuclear Extraction Reagents (Pierce, Rockford, IL, USA) according to the supplier's instructions. Binding was carried out at 20°C for 30 min in 30 μl of reaction buffer containing 5 µg of nuclear extract, 25 mM HEPES (pH 7.5), 100 mM NaCl, 1 mM EDTA and 4 mM DTT with 0.005 pmol of labeled probe. A 1000-fold molar excess of oligonucleotides containing a consensus E-box, E-boxcon, the E-box and E-boxΔ were used as competitors in this assay. The samples were subjected to electrophoresis on 5% native polyacrylamide (60:1, acrylamide:bisacrylamide) gels. Total RNA was isolated from the cell lines using Trizol (Invitrogen Life Technologies Inc.) and reverse transcribed using SuperScript™ III RT First strand synthesis system for RT–PCR (Invitrogen Life Technologies Inc.), as per the manufacturer's instructions. Real-time PCRs for measuring the endogenous and 18S RNA were carried out using an ABI 7500 FAST PCR system (Applied Biosystems, Foster City, CA, USA) and appropriate Taqman probe primer mixes (Applied Biosystems). Quantitation of luciferase mRNA was carried out using Power SYBR® Green PCR mix (Applied Biosystems) and the primer pair -Luc-F1 and -Luc-R2 which amplifies across the intron between the Frataxin exon 1 and the luciferase coding sequence. The anti-dimethyl-Histone H3 (Lys9) (H3K9me2) antibody (Cat. no. 07-441) was purchased from Upstate (Charlottesville, VA, USA). The rabbit preimmune serum was purchased from Santa Cruz Biotechnology, Inc. (Santa Cruz, CA, USA). The ChIP assay kit from Upstate was used according to the manufacturer's instructions with slight modifications. Briefly, cells were cross-linked in RPMI medium 1640 with 1% formaldehyde at 23°C for 5 min. The cells were washed in PBS and lysed in a buffer containing 1% SDS, 10 mM EDTA and 50 mM Tris–HCl pH 8.0. The chromatin from lysed cells was sonicated to lengths between 200 and 1000 bp using a Bioruptor sonicator (Diagenode, Philadelphia, PA, USA) and the cell debris was removed by centrifugation. The sonicated lysates were precleared and incubated with either preimmune serum or the anti-H3K9Me2 antibody at 4°C overnight. The immunocomplexes were recovered by Protein A agarose, washed and eluted in 1% SDS, 0.1 M NaHCO and the crosslinking reversed by incubation at 65°C for 4 h in 200 mM NaCl. The samples were phenol: chloroform extracted and ethanol precipitated in the presence of Pelletpaint (EMD Biosciences, San Diego, CA, USA). The amount of DNA immunoprecipitated was determined using quantitative real-time PCR using Power SYBR® green PCR master mix (Applied Biosystems). The PCR primers used were chip-QF1 and chip-QR1. The ChIP experiments were performed in triplicate and each PCR reaction was done in duplicate. The immunoprecipitated DNA was normalized to the amount of input DNA and plotted as the fold enrichment over the preimmune serum (IgG). xref fig #text

The challenges presented by biological weapons, global health-care issues and emerging diseases of natural origin lend urgency to the development of rapid, field-deployable pathogen detection and diagnostic tools (,). Ideally, to be of general field utility, a diagnostic device must be capable of sensitive and specific pathogen detection while retaining simplicity of use and independence from complex laboratory instrumentation (). Additional challenges are presented by the need to screen samples for multiple pathogenic or toxic agents, a characteristic highly desirable in cases where commonalities in early symptom presentation confound differential diagnoses. While nucleic acid-based assays for pathogen detection and identification offer sensitivity, specificity and resolution, they are relatively elaborate and often costly, limiting their utility for point-of-care diagnostics and deployment under field conditions where a supporting laboratory infrastructure is limited or absent. Reliance upon polymerase chain reaction (PCR) and fluorescent detection of amplified nucleic acids has contributed significantly to the complexity and cost of nucleic acid diagnostics (,). Retaining assay sensitivity while circumventing requirements for thermocyclers and fluorescence detection hardware remains a significant challenge. The recent advent of DNA microarray technology has promised to increase the information capacity of nucleic acid diagnostics and enable the highly multiplexed detection of genetic signatures (). The potential of DNA microarrays to detect, in parallel, large panels of distinct nucleic acid sequences has proven to be a powerful technique for many laboratory applications (). Nonetheless, the reliance of this technology on costly instrumentation for high-resolution fluorescence signal transduction severely limits the utility of microarrays for field applications where a laboratory infrastructure is limited or unavailable. Additionally, the long hybridization incubations required for microarray assays increase sample-to-answer times beyond what would be acceptable for a rapid screening assay. Though microarray hybridization times as short as 300 and 500 s have been reported (,), such methods employ designs that remain reliant upon fluorescent detection and supporting instrumentation and do not address the need for low-cost, easily manufactured devices that can be used in the absence of laboratory infrastructures. In contrast to DNA-based assays, immunoassays have found widespread acceptance in low-cost, easily used formats, perhaps the most notable of which is the chromatographic lateral flow immunoassay (). Lateral flow assays, also known as hand-held assays or dipstick assays, are used for a broad range of applications where rapid antigen detection is required in an easily used, low-cost format. Expanding the domain of lateral flow chromatography to nucleic acid detection, a number of recent reports have described lateral flow detection of PCR products using a variety of capture and detection schemes (). Unfortunately, the utility of lateral flow detection in the context of a PCR-based assay is severely limited by the fact that reliance on thermocycling hardware largely negates the potential benefit of the otherwise highly simplified lateral flow platform. Additionally, a PCR-based approach to lateral flow detection necessitates each PCR reaction be subjected to post-amplification manipulations required to generate single-stranded products for hybridization-based detection. Recent work has sought to alleviate reliance on PCR through employing isothermal nucleic acid amplification schemes or direct detection of unamplified genetic material. Enabled by the use of up-converting phosphor reporters, unamplified DNA sequence has been detected using a lateral flow assay format (). Up-converting phosphor technology, while sensitive, remains dependent upon the hardware required to detect phosphor emission (). The use of simple colorimetric detection schemes that circumvent the requirements for complex instrumentation require an upstream amplification strategy to attain suitable sensitivity. Isothermal nucleic acid amplification coupled with lateral flow detection has been reported for assays making use of cycling probe technology [CPT, ()], recombinase polymerase amplification [RPA, ()] and nucleic acid sequence-based amplification [NASBA, ()]. While the work by Fong . () and Piepenburg . () made use of a lateral flow immuno-assay for DNA detection, the RNA targets amplified by NASBA in the work from Baeumner's group () were detected using a lateral flow system enabled by the use of liposome encapsulated dye and a sandwich hybridization assay similar to that reported by Rule . (). While shown to display nanomolar sensitivity, the reported dye encapsulating liposome-based methods require additional washing steps and the liposomes are relatively labile, must be custom synthesized, and stored under stabilizing hydrated conditions (). To develop more capable nucleic acid detection methods that offer many of the advantages of microarray technology yet retain the simplicity of lateral flow-based platforms, we have developed a microarray-based lateral flow technology. Using an assay based on the nonsense mutation in the gene of , that is absent in the near phylogenetic neighbors and (,), we illustrate the utility of the lateral flow microarray (LFM) approach for sensitive detection and discrimination of closely related microbial signatures when present as minority sequences in complex nucleic acid mixtures. The results demonstrate that LFMs, making use of stable detection reagents suitable for dry storage, can be used to detect as little as 250 amol analyte within 2 min of sample addition. The miniaturization of lateral flow detection decreases reagent consumption and sample-to-answer times while increasing the potential information capacity of the platform to enable the development of highly multiplexed nucleic acid detection assays. Total RNA was isolated from strain Sterne 7702 and strain HD 621 () using a previously reported protocol (). Purified RNA was quantified by measuring OD and evaluated by gel electrophoresis. 3 × 10 cells were used for RNA isolation typically yielding 50–75 μg of total RNA. NASBA () primers, plc-P1 and plc-P2, were designed to amplify a fragment of the locus from . Primer sequences used for NASBA reactions are provided in , the T7 promoter sequence is italicized in plc-P1. Plc-P1 hybridizes to the transcript such that the 3′-end of the primer forms a base pair with the previously reported polymorphism strictly associated with (,). The NASBA P2 primer, plc-P2, is located such that the amplified RNA resulting from NASBA is 179 bases in length (A). Previously reported -based real-time PCR assays (,) have made use of an alternate upstream primer that generates a 83 bp product but may be poorly suited for NASBA given the optimal NASBA product size of 120–250 bases (). NASBA reactions were prepared according to the manufacturer's instructions using the NucliSens Basic kit (Biomerieux) and primers plc-P1 and plc-P2 at 0.4 μM each. Amounts of total cellular bacterial RNA were varied, as indicated, between 0 and 2 ng. Sterne 7702 was used as a test strain and strain HD 621 was employed as a negative control. One microgram of human total cellular RNA isolated from HeLa S3 cells (Stratagene) was included in all NASBA reactions to provide a complex RNA background consistent with the composition of human diagnostic samples. Following a 60-min incubation at 41°C, NASBA reaction products were detected by using a LFM. LFMs were printed using a NanoPlotter 2.0 (GeSim, mbH, Dresden, Germany) non-contact picoliter deposition system equipped with NanoTips (GeSim). Unless otherwise indicated, LFMs were patterned with 400 μM solutions of oligonucleotide in HO containing a 1:50 dilution of Ponceau S (P7767, Sigma) as a tracking dye. A lateral flow compatible nitrocellulose membrane (HiFlow 135, Millipore) was used as the LFM substrate. Following oligonucleotide deposition, nitrocellulose membranes were air dried and exposed to 5000 μJ UV in a StrataLinker (Stratagene). The resulting membrane sheets were cut into 3-mm wide, 30-mm long strips, which were either used directly with buffer-suspended dyed microspheres or assembled with conjugate release pads into a custom plastic housing. Housings were fabricated from polycarbonate sheet cut using a CO laser (VersaLaser VL-300, Universal Laser Systems, Inc., Scottsdale, AZ, USA). A gasket of 500-μm thickness was used to generate an internal chamber of sufficient size to accommodate the LFM substrate and the conjugate release pad. The small sample volumes used obviated the need for sample and downstream absorbent pads, the function of which was supplied by the conjugate release pad and unpatterned regions of the LFM substrate, respectively. Conjugate release pads were made by impregnating glass fiber conjugate pad (GFCP203000, Millipore) with dyed microspheres covalently conjugated to R-57-76-3TN in 1% SDS. Conjugate release pads measuring 3.5 mm × 4.5 mm were doped with ∼8 × 10 oligonucleotide conjugated dyed microspheres. Microsphere saturated release pads were allowed to air dry under ambient conditions prior to assembly with LFM membranes. provides capture and detection oligonucleotide sequences, their binding sites within the amplicon are depicted in B. Amine modification and a T spacer sequence were included on the 3′-end of detection oligonucleotide R-57-76-3TN to allow covalent cross-linking to dyed microspheres and to facilitate hybridization in lateral flow sandwich assays, respectively. SPHERO™ carboxyl-polystyrene 0.35-μm blue microspheres (Spherotech) were covalently conjugated to amino modified oligonucleotide R-57-76-3TN using the coupling agent 1-etyl-3-(3-dimethylaminopropyl-diimide HCl) (EDAC, Pierce) under conditions adapted from Spiro . (). Briefly, 1.1 × 10 microspheres were suspended in 100 mM 2-(-morpholino)ethanesulfonic acid pH 4.5 (MES, Sigma). Indicated amounts of oligonucleotide were introduced to MES suspended microspheres, vortexed and incubated in the presence of 0.5 mg/ml EDAC. Reactions were protected from light in aluminum foil wrapped tubes and incubated at room temperature for 30 min followed by the introduction of additional EDAC to bring the final EDAC concentration to 1 mg/ml. Incubation was continued for an additional 30 min after which beads were washed once with 1 ml 0.02% tween-20 (Sigma) and twice with 0.5 ml 0.1% SDS (Fisher Scientific). Beads were re-suspended in 0.5 ml DNAase/RNAase free HO. Bead suspensions were assessed for aggregation by phase-contrast light microscopy using a Zeiss IM135 inverted microscope. A DNA oligonucleotide, dnaR89, composed of sequence derived from a region of the gene of , as shown in B, was used to provide a readily available and quantifiable target for LFM assay development and optimization. The sequence of this synthetic target is provided in . Additionally, a full-length synthetic target RNA was generated. This RNA, referred to here as plcRivt, was used to confirm that reaction conditions established with dnaR89 were also suitable for the detection of NASBA reaction products. Synthesis of plcRivt was accomplished by using plc-P1 and plc-P2 primers in PCR reactions containing 20 ng of Stern strain 7702 genomic DNA. PCR reactions using Platinum PCR Supermix (Invitrogen) were conducted for 40 cycles of 94°C for 30 s, 60°C for 30 s and 72°C for 1 min following an initial 2-min incubation at 94°C. The resulting amplicon was subjected to purification using QIAquick PCR clean-up spin-columns (QIAGEN) and subsequently used to program an transcription reaction using the T7 AmpliScribe kit (EpiCentre). The transcription reaction product was subjected to treatment with RNase free DNase I (Ambion) and purified using a RNeasy column (QIAGEN). The resulting RNA was quantified by measuring the OD. plcRivt is predicted to be identical in sequence to the NASBA product generated from total cellular RNA using plc-P1 and plc-P2. For time course studies, LFM assays were conducted with sample buffer containing 0.1% w/v suspended dyed microspheres (∼4 × 10 particles) in running buffer. Lateral flow was recorded using a digital video recorder (DCR-PC1, Sony). Video frames were collected for quantification using iMovie (Apple Computer). Feature intensity was quantified for time course studies and some optimization experiments using uncalibrated optical density in ImageJ (). For better reproduction contrast, LFM images used for figures were cropped and modified by applying the Auto Contrast function in Photoshop CS2. No other modifications were applied. Oligonucleotides for hybridization sandwich assays were designed to detect NASBA amplified mRNA or synthetic targets based on relevant subregions of the sequence. Oligonucleotides immobilized on the lateral flow substrate are referred to here as capture oligonucleotides while those conjugated to dyed microspheres for signal generation are referred to here as detection oligonucleotides. Supported large-pore nitrocellulose membranes were patterned with varying concentrations of capture oligonucleotides using a NanoPlotter 2.0 robotic positioning system and NanoTip piezoelectronically actuated micropipets. Oligonucleotide dnaR89 was printed on LFM substrates as a positive hybridization control as this oligonucleotide carries sequence complementary to bead coupled detection oligonucleotide R-57-76-3TN. Negative hybridization control oligonucleotides included the reverse complement of capture oligonucleotide R-24-43 (F-24-43), the detection probe sequence T spacer (R-57-76-3N) and a unrelated sequence complementary to a region of the locus (FT-S18). By ejecting droplets from the micropipet at a distance of 500 μm from the nitrocellulose substrate, microarray feature sizes of ∼200 μm could be generated. In contrast to contact microarray printing methods, this approach preserves the fragile pore structure of the membrane required for microsphere-based detection. Patterned nitrocellulose sheets were cut into 3-mm wide strips and then assembled with conjugate release pads in a custom designed plastic housing. An example of the resulting device is shown in A. Hybridization-mediated capture of analyte at the cognate capture element of the microarray and non-overlapping hybridization to dyed microsphere conjugated detection oligonucleotide generates a colorimetric signal arising from an increased local concentration of dyed microsphere particles. In the absence of hybridization, microspheres are sufficiently dispersed that additional washing steps are not required to reduce background signal levels. A schematic representation of the hybridization sandwich assay scheme is depicted in B. LFMs were fabricated using varying concentrations of capture oligonucleotide to determine optimum printing concentrations. Following lateral flow of 25 fmol dnaR89 in 4× SSC, 5% formamide, 1.4% Triton X-100, 0.1% SDS containing 0.1% R-57-76-3TN coupled microspheres LFMs were scanned on a flatbed scanner and the resulting images quantified. For all capture sequences examined, 400-μM oligonucleotide-printing concentrations provided the most favorable signal intensity (A). Standard hybridization conditions employed for these and other characterization studies were determined through an iterative set of optimization experiments that examined the effects of ionic strength, formamide concentration and detection oligonucleotide to bead cross-linking ratios. Given the profound impact ionic strength has on the stringency of DNA hybridization, microsphere dispersion and lateral flow characteristics, SSC concentration was varied from 1 × to 9 × and assay performance evaluated by densitometry of LFMs following hybridization sandwich assays conducted using 25 fmol of the synthetic target dnaR89 or ∼200 fmol of plcRivt (,). B summarizes the results of SSC concentration optimization experiments. Near optimal signal intensity was obtained for both dnaR89 and plcRivt at SSC concentrations between 2× and 7×. For use in standard LFM running buffer 4× SSC was selected as it provided sensitive hybridization-based detection of derived sequences, good capillary lateral flow characteristics, and favorable microsphere dispersion and release-pad liberation. Formamide is known to reduce the melting temperature of DNA and RNA duplexes and may facilitate capture and detection probe accessibility to binding sites within the target through a destabilization of analyte secondary structure (). To determine the optimum concentration of formamide in LFM running buffer, a series of LFM experiments were conducted at varying formamide concentrations using both dnaR89 and plcRivt. 10 μl of 4× SSC, 1.4% Triton X-100 and 0.1% SDS containing 25 fmol dnaR89 or ∼200 fmol plcRivt and varying concentrations of formamide, as indicated in C, were subjected to LFM analysis and the resulting hybridization signals quantified by densitometry. These experiments revealed a slight but reproducible increase in signal intensity at 5% formamide. All subsequent studies presented here were performed using 4× SSC, 1.4% Triton X-100, 0.1% SDS and 5% formamide. Given that higher stock concentrations of synthetic oligonucleotide dnaR89 could be obtained, which allowed high confidence quantification of this synthetic target relative to what could be achieved with comparatively dilute solutions of the transcription product plcRivt, subsequent LFM characterization studies made use of dnaR89. The similarity of buffer optima displayed by dnaR89 and by plcRivt synthetic targets supported the assertion that dnaR89 could be used as an accurate proxy for the performance of LFM assays for NASBA product detection. Others have reported similar findings concluding that appropriately designed DNA oligonucleotides can be used as synthetic targets for the development of assays ultimately used for NASBA product detection (). Therefore, subsequent LFM assay optimization and characterization was conducted using dnaR89. To determine the optimum ratios for cross-linking detection oligonucleotides to dyed polystyrene microspheres, we examined populations of beads coupled to oligonucleotide at varying ratios. The 3′-amine modified detection oligonucleotide R-57-76-3TN was covalently linked to polystyrene dyed microspheres using EDAC. The resulting bead/oligonucleotide complexes were evaluated for their ability to mediate detection of dnaR89 in a hybridization sandwich assay. Coupling reactions using a 2.2 × 10:1 oligonucleotide to bead ratio were found to provide optimum signal as determined by densitometry (D). The detection oligonucleotide R-57-76-3TN carried a 3′-spacer region consisting of 18 T residues to increase the accessibility of bead bound oligonucleotides for hybridization. R-57-76-3N, which carried the same analyte complementary sequence as R-57-76-3TN but without the T spacer, was found to exhibit significantly reduced hybridization to dnaR89 consistent with prior reports that a poly(dT) spacer sequence increases hybridization efficiency to solid-phase coupled oligonucleotides (data not shown) (,). T spacers were not incorporated into LFM immobilized capture oligonucleotides as they were found to be dispensable for hybridization. To determine the relative performance of hybridization sandwich assays making use of capture oligonucleotides with complementarity to different locations of the target sequence, three capture oligonucleotides were synthesized and compared using sandwich assays employing detection oligonucleotide R-57-76-3TN coupled dyed microspheres. R-77-96 was designed to participate in base stacking with R-57-76-3TN when hybridized to the target. Base stacking has been reported to stabilize hybridization and allow efficient capture of dilute nucleic acids by hybridization (). The binding sites for the three capture oligonucleotides examined (R-77-96, R-36-55 and R-24-43) are illustrated in B. Varying quantities of synthetic target dnaR89, between 0 and 200 fmol, were used for these studies. depicts LFM membranes following detection of the indicated amounts of target oligonucleotide dnaR89. LFMs carried dnaR89, which hybridizes directly to the microsphere conjugated detection probe, as a positive hybridization control. Positive control features were printed as the left most element of each LFM row to assist in feature identification. Negative hybridization controls, F24-43 and FT-S18, were based on the reverse complement of R-77-96 and an unrelated derived sequence, respectively. Additionally, to confirm that no carryover contamination occurred during printing, HO containing Ponceau S was printed on LFM substrates between positive control and capture oligonucleotide deposition. No signal was detectable in either hybridization negative controls or HO negative control microarray elements. Background corrected signal intensity was determined from LFM images using GenePix Pro 6.0 microarray data extraction software. The results, presented in A, reveal R77-96 produces significantly higher hybridization signals than R-36-55 or R-24-43 for all examined quantities of dnaR89, suggesting a significant contribution of base stacking effects to LFM hybridization sandwich assay sensitivity. To define the detection limit of the LFM assay, a one-tailed -test was used to determine quantities of dnaR89 that generated signal intensities significantly above 0 amol negative controls. Signals generated at R-77-96 capture features with 250 amol and greater quantities of dnaR89 were significantly >0 amol dnaR89 controls ( < 0.05, = 6). By the same criterion, 1 fmol dnaR89 detection limits were obtained for both R-24-43 and R-36-55 ( < 0.05, = 6). B depicts the performance of LFM detection over 0 to 2500 amol dnaR89 range using the R-77-96/R-57-76-3TN capture/detection probes. LFM detection exhibited excellent linearity, = 0.989, over this 10-fold range of target molecules. While capture probe R-24-43 exhibited less sensitivity than R-77-96, this capture probe displayed excellent signal linearity between 2.5 and 100 fmol dnaR89, = 0.968 (C). These findings demonstrate that the LFM capacity to display multiple capture sequences can be used to simultaneously provide sensitive detection and extend assay linearity through the use of capture probes with differing hybridization characteristics. The small sample volumes used for LFM detection and the reduced surface area traversed during capillary lateral flow significantly reduces detection times for the LFM relative to traditional lateral flow devices. To quantitatively present the speed of LFM nucleic acid hybridization-based detection, we used digital video to follow hybridization sandwich assay-mediated detection of synthetic target molecule dnaR89. Individual frames were isolated from video datasets and quantified for relative signal intensity over the course of capillary lateral flow across the LFM substrate. The resulting signal data was plotted versus time in seconds as shown in . For time measurements, t was defined as the time when the sample front reached the first row of LFM features. Signal was detectable for 1000-fmol target in 2 s following sample transport across R-77-96 capture elements. Within 4 s 100 fmol dnaR89 was detectable while 10 fmol was clearly detectable by 30 s as defined by the earliest time point at which 90% of the pixels composing the R-77-96 microarray features were greater than one standard deviation above background. Lateral flow transport of the 10-μl sample was complete by 120 s. Prior reports have described a single nucleotide polymorphism (SNP) present in but not close phylogenetic near neighbors including and (,). This SNP has been used as the basis for a sensitive and highly discriminatory real-time PCR assay for (). To determine the utility of LFM technology for detecting minority nucleic acids in complex samples, NASBA primers were designed to amplify the allele of . P1 and P2 primer sequences, plc-P1 and plc-P2, used for NASBA amplification are provided in and their binding positions illustrated in . Varying amounts of total cellular RNA isolated from or 2 ng of HD 621 RNA as a negative control were introduced to 1 μg of total human cellular RNA isolated from HeLa S3 cells. The resulting mixtures were subjected to NASBA amplification using plc-P1 and plc-P2 primers. Human RNA was included in NASBA amplification reactions to approximate the nucleic acid complexity expected in human diagnostic specimens. An aliquot of 2 μl of NASBA reaction mixture was removed after a 60-min incubation at 41°C, mixed with 8 μl of LFM running buffer and assayed for amplicon by LFM. Dyed microspheres cross-linked to R-57-76-3TN were used for detection of NASBA amplicons captured on LFMs carrying R-77-96. Data from these studies are presented in . Following 60 min of NASBA amplification, as little as 0.5 pg for total cellular RNA could be detected in a background matrix of 1 μg of human total RNA. These studies closely approximate the conditions expected for complex human diagnostic samples and reveal the capacity of the LFM platform to specifically detect NASBA reaction products generated from mixed samples where the target sequence is a minority species. While the number of mRNA copies in a cell has not been determined, an estimate of LFM assay sensitivity, in terms of cells, can be calculated based on total RNA yields. Total RNA yields from vegetative were in the range of ∼167–250 fg RNA/cell. Using this value, an estimate of LFM sensitivity corresponds to the detection of approximately to 2–3 cells. Lateral flow detection of DNA or RNA amplification reaction products provides one means of simplifying nucleic acid detection. Indeed, a lateral flow platform may offer many significant advantages for employing nucleic acid assays under conditions where a fully equipped molecular biology laboratory infrastructure is not available or desirable. Such situations would include resource poor settings, point-of-care, battlefield deployments and scenarios where first responders must quickly determine the threat presented by an unknown substance. To date, however, lateral flow devices have predominantly been fabricated using one or a few capture lines thus limiting the information capacity of the device to one or a few analytes (,). As a step toward higher information-content lateral flow nucleic acid detection, we have developed nitrocellulose-patterning methods that enable microarray feature density to be attained on lateral flow compatible substrates. Making use of a non-contact peizo actuated picoliter deposition system, we have patterned lateral flow compatible nitrocellulose membranes with features similar in size and spacing to those typically found on spotted glass microarrays. The sensitivity of lateral flow nucleic acid detection methods previously reported in the literature has been in the order of 1 fmol (). We find that the LFM platform provides rapid detection of as little as 250 amol of target using a low-cost and widely available flatbed scanner, a standard personal computer system and a commercially available microarray data extraction suit or free image analysis software. This detection limit is similar to the sensitivity reported for fluorescence and chemiluminescence microarray detection strategies (,). While the LFM implementation reported here exhibits excellent linearity ( = 0.989), the linear dynamic range is less than that commonly associated with fluorescence-based detection. Alternative LFM detection schemes and the use of capture probes of differing hybridization characteristics should enable greater dynamic range while retaining the simplicity of the LFM approach. Indeed, examining the signals generated by R-77-96 and by R-24-43, the effective linear range of the LFM assay extends over a 400-fold range of target from 250 to 100 fmol (B and C). The information density of the LFM offers the capacity for additional capture probes of varying hybridization potential to be included that should allow this dynamic range to be extended further. The uniformity of sample flow exhibited by the LFM suggests that larger capture probe sets can be accommodated without complications arising from physical factors. For example, concentrations of analyte 40-fold above the linear range of R-77-96 did not adversely impact the linearity of R-24-43 signal at LFM elements situated directly downstream (with respect to sample flow) of R-77-96 capture features (B and C). Only at artificially high microsphere capture densities, such as those produced by the positive control hybridizations in , are signal gradients observed as a function of physical location on the LFM, presumably due to physical occlusion of membrane pores by high local accumulations of microspheres. LFMs offer several advantages arising directly from the miniaturization of the system without sacrificing detection sensitivity. While traditional lateral flow assays make use of sample volumes in the order of hundreds of microliters to milliliters, the miniaturization approach we have developed reduces sample volume to 10 μl. This reduced sample volume significantly decreases the consumption of reagents required for amplification. Here we have made use of 2 μl of a NASBA reaction diluted to 10 µl in running buffer. By reducing standard NASBA reaction volumes from 20 to 2 μl, a one order of magnitude reduction in enzyme consumption is realized. It should also be noted that other amplification schemes, such as those that make use of microfluidic systems or lab-on-a-chip technologies, could be integrated with a miniaturized lateral flow-based detection system to provide a rapid and cost-effective means of detecting analytes. A further benefit of miniaturization is the time required to detect analyte following introduction of amplified material to the LFM. While the procedures used here employed NASBA amplification and traditional RNA isolation protocols requiring ∼90 min to complete, more recent advances in nucleic acid preparation and amplification have reported significant reduction in sample processing times (). As amplification protocols become more rapid, the speed with which amplicons can be detected, without complex optical systems and fluorescent detection, becomes critical to realizing the potential of these technologies. The LFM methods described here detect nucleic acid analytes in less than 2 min. Given that 250 amol is equivalent to 1.5 × 10 molecules, efficient amplification methods that offer 10-fold amplification, widely cited amplification levels for PCR- and NASBA-based techniques (,), would theoretically enable the detection of single-copy targets by LFM following amplification. Future systems that couple advanced amplification technologies and compatible streamlined nucleic acid preparation modalities with rapid LFM detection will allow significant decreases in sample-to-answer times without costly or complex instrumentation.

Meganucleases are sequence-specific enzymes which recognize large (12–45 bp) DNA target sites. These enzymes are often encoded by introns or inteins behaving as mobile genetic elements. They recognize sites that usually correspond to intron-free or intein-free genes, where they produce a DNA double-strand break (DSB). Eventually, DSB repair by homologous recombination with an intron- or intein-containing gene results in the insertion of the intron or intein where DSB occurred in specific loci in living cells (). Meganucleases are used to stimulate homologous recombination in the vicinity of their target sequences in cultured cells and plants (). These results present new perspectives for genome engineering in a wide range of applications, such as the correction of mutations linked with monogenic inherited diseases or the bypass of risks due to the randomly inserted transgenes used in current gene therapy approaches (). The use of meganuclease-induced recombination has long been limited by the repertoire of natural meganucleases. In nature, meganucleases are essentially represented by homing endonucleases (HEs), a family of endonucleases encoded by mobile genetic elements, whose function is to initiate DSB-induced recombination events in a process referred to as homing (). Several hundreds of HEs have been identified in bacteria, eukaryotes and archaea (); however, the probability of finding a HE cleavage site in a chosen gene is low. Thus, the making of artificial meganucleases with custom-made substrate specificity is an intense area of research (). Lately, zinc-finger DNA-binding domains () could be fused with the catalytic domain of the FokI endonuclease, to induce recombination in various cell types, including human lymphoid cells (). However, these chimeric proteins showed high toxicity in cells (,), probably due to a low level of specificity. Given their biological function and their exquisite specificity, HEs represent ideal scaffolds to engineer the substrate specificity of proteins that cleave or recombine DNA (). Sequence homology has been used to classify HEs into four families, the largest one having the conserved LAGLIDADG sequence motif (). HEs with only one such motif, such as I-CreI (), function as homodimers. In contrast, larger HEs containing two motifs, such as I-SceI () or I-DmoI (), are single-chain proteins. The 3D structures solved for several LAGLIDADG endonucleases (, , , ) indicate that these proteins adopt a similar active conformation as homodimers or as monomers with two separate domains. The LAGLIDADG motifs form structurally conserved α-helices tightly packed at the center of the interdomain or intermonomer interface (,, , ). On either side of the LAGLIDADG α-helices, a four-stranded β-sheet provides a DNA-binding interface that drives the interaction of the protein with each one of the half sites of the target DNA sequence (,). The last acidic residue of the LAGLIDADG motif participates in the DNA cleavage by a metal-dependent mechanism of phosphodiester hydrolysis (). The I-CreI structure has been solved without DNA before. In this study, the protein crystallized with only one monomer in the asymmetric unit (). We have solved the structure of the I-CreI dimer without DNA; its comparison with the DNA-bound crystal structure () depicts a different conformation for the C-terminal loop which connects the last two helices α5 and α6. These two structural elements (C-terminal loop and α6 helix) are hereafter referred to as the C-loop and the C-helix. The sequence of the C-terminal loop is highly conserved among dimeric LAGLIDADG HEs and it is involved in contacts with the DNA backbone (,). To unravel the function of this region in the endonuclease mechanism of I-CreI, we designed trimmed, double and single mutants in this area. Binding and cleavage experiments illustrate the important role of the residues located at the C-terminal loop in DNA binding and cleavage. Even though different regions that define novel target DNA specificities have been identified in the I-CreI scaffold (), this work reveals a novel site essential for binding and cleavage which can also be engineered to generate novel specificities. All the constructions used in this work are based on I-CreI pET24-d(+) plasmid used in (). Protein expression and purification were performed as in (). Site-directed mutagenesis was performed using the Quickchange XL site-directed mutagenesis kit from STRATAGENE (). All the mutations were checked by DNA sequencing (data not shown). An initial screening for I-CreI crystallization conditions was performed in 96-well plates by vapor-diffusion methods using the Hampton crystal screening using drops containing 1 μl protein solution (7 mg/ml in 20 mM HEPES pH 7.5) and 1 μl precipitant solution equilibrated against 50 μl of reservoir solution at 20°C. Crystals were obtained under several conditions (Crystal Screen 1 conditions 10, 22, 33, 40, 41 and Crystal Screen 2 condition 32). Crystal was made by hanging-drop vapor-diffusion methods using VDX plates; optimization experiments led to the following conditions for crystallization: 1 μl protein at 7 mg/ml in 20 mM HEPES pH 7.5 and 1 μl precipitating buffer containing 20% PEG 4000, 0.1 M HEPES pH 7.5, 10% Iso-propanol, 10% ethylene glycol and 0.01 M magnesium acetate equilibrated against 500 μl precipitating buffer at 20°C. Rod-shaped crystals grown in 4–8 days and were directly collected and frozen in liquid nitrogen. All data were collected at cryogenic temperatures using synchrotron radiation at 100 K. I-CreI crystals were mounted and cryoprotected. The data sets were collected using synchrotron radiation at the ID14-4 beamline at the ESRF (Grenoble), and at the PX beamline at the SLS (Villigen). Diffraction data were recorded on an ADSC-Q4 or Mar225 CCD detectors depending on the beamline. The best data set () was collected using a Δ  1 and a wavelength of 0.97 Å. Processing and scaling were accomplished with HKL2000 (). Statistics for the crystallographic data are summarized in . The structure was solved using the molecular replacement method as implemented in the program MOLREP (). The search model was based on a polyalanine backbone derived from the PDB entry 1GZ9. The coordinates from the DNA were deleted in the search model. A refined 2Fo–Fc map showed clear and contiguous electron density for the protein backbone and for many of the side-chains. ARP/wARP and REFMAC5 were applied for automatic model building and refinement to 2.0 Å (). The coordinates have been deposited in the PDB (2O7M). Data were acquired with a Jasco 810 model dichrograph, previously calibrated with -10-camphorsulfonic acid, and equipped with a Jasco Peltier thermoelectric temperature controller CDF-426S. Experiments were performed in phosphate buffer saline (PBS, 137 mM NaCl, 10 mM NaHPO·2HO, 2.7 mM KCl, 2 mM KHPO, pH 7.4) at 1°C/min intervals. The protein concentration was 10 μM. The ellipticity at 222 nm was followed from 5 to 95°C in a 2 mm Hellma 110-QS cell. Sedimentation equilibrium experiments were performed at 20°C in an Optima XL-A (Beckman-Coulte, Inc.) analytical ultracentrifuge equipped with UV–visible optics, using an An50Ti rotor, with 3 mm double-sector centerpieces of Epon charcoal. Protein concentration was 200 μM in PBS buffer. Short column (23 µl), low-speed sedimentation equilibrium was performed at three successive speeds (11 000, 13 000 and 15 000 r.p.m.), the system was assumed to be at equilibrium when successive scans overlaid and the equilibrium scans were obtained at a wavelength of 280 nm. The base-line signal was measured after high-speed centrifugation (5 h at 42 000 r.p.m.). Whole-cell apparent molecular weight of the protein was obtained using the program EQASSOC (). The partial specific volume of I-CreI was 0.7436 ml/g at 20°C, calculated from the amino acid composition with the program SEDNTERP (downloaded from the RASMB server) (). The sedimentation velocity experiment was carried out in an XL-A analytical ultracentrifuge (Beckman-Coulte, Inc.) at 42 000 r.p.m. and 20°C, using an An50Ti rotor and 1.2 mm double-sector centerpieces. Absorbance scans were taken at 280 nm. The protein concentration was 50 μM in PBS. The sedimentation coefficients were calculated by continuous distribution c(s) Lamm equation model () as implemented in the SEDFIT program. These experimental sedimentation values were corrected to standard conditions to get the corresponding values using the SEDNTERP program (). Further hydrodynamic analysis (i.e. calculation of frictional coefficient ratio) was performed with the SEDFIT program to obtain de c(M) distribution (). NMR spectra were recorded at 25°C in a Bruker AVANCE 600 spectrometer equipped with a cryoprobe. Protein samples were 500 μM in PBS buffer plus 5% HO. 2,2-Dimethyl-2-silapentane-5-sulfonate sodium salt (DSS) was used as internal proton chemical shift reference. Band-shift assays were performed in 10 mM Tris-HCl pH 8, 50 mM NaCl, 10 mM CaCl or MgCl, 1 mM DTT incubated 1 h at room temperature using 5 μM (0.0793 μg/μl) 6-FAM duplex and 20 μM (0.463 μg/μl) protein. After incubation, the samples were subjected to electrophoresis using a 15% acrylamide-TBE gel. Titration curves were performed with 500 nM 6-FAM DNA duplex and 0–10 000 nM protein (monomer concentration). Assuming that one molecule of duplex DNA binds to one molecule of I-CreI protein dimer, the dissociation constants () () were determined by data fitting (Origin, Microcal) to the equation: [DNA–P] = [DNA–P] X (( + [DNA] + [P] – sqrt(( + [DNA] + [P]) – 4 × [DNA] × [P])))/(2X[DNA]), where [DNA–P] is the concentration of the DNA–protein complex, [DNA–P] is the maximum possible concentration of complex, [DNA] is the total concentration of 6-FAM DNA duplex (500 nM), [P] is the total concentration of the protein (dimer protein concentration). The value of [DNA–P] was calculated from the reduction in the intensity of the lower band in the gel (free DNA) as the intensity of the shifted band (bound DNA) was smaller than the loss of intensity in the lower band, due to fluorescence quenching or dampening by the bound protein (gels are provided in Supplementary Data). The adjustable parameters during the fitting were [DNA–P] and . Cleavage assays were performed at 37°C in 10 mM Tris-HCl (pH 8), 50 mM NaCl, 10 mM MgCl and 1 mM DTT. The amount of enzyme and target were: 100 ng for the XmnI-linearized DNA substrate (pGEM-T Easy 10AAA_5GTC_P) (,) and 0.25–120 ng dilutions for I-CreI and helix mutant proteins, in 25 μl final volume reaction. The linearized target plasmid has 3 kb and after cleavage yields two smaller bands of 2 and 1 kb. Reactions were stopped after 1 h by the addition of 5 μl of 45% glycerol, 95 mM EDTA (pH 8), 1.5% (w/v) SDS, 1.5 mg/ml proteinase K and 0.048% (w/v) bromophenol blue (6× buffer stop), incubated at 37°C for 30 min and electrophoresed in a 1% agarose gel. The gels were stained using SYBR Safe DNA gel staining (INVITROGEN) and the intensity of the bands observed upon UV light illumination were quantified with the ImageJ software (). The percentage of cleavage was calculated with the following equation: percentage cleavage = 100 X ( + )/( +  + ), where , and are the intensities of the 1-, 2- or 3-kb bands. The cleavage rate was calculated using the enzyme concentration needed to cut 50% of the target DNA () (). The 10AAA_5GTC_P 24-bp target sequence (TCAAAACGTCGTACGACGTTTTGA) is a palindrome of a half-site of the natural I-CreI target (TCAAAACGTCGTGAGACAGTTTGG). 10AAA_5GTC_P is cleaved as efficiently as the I-CreI natural target and in both yeast and mammalian cells. The palindromic targets, derived from 10AAA_5GTC_P, were cloned as previously described () into the reporter vectors: the yeast pFL39-ADH-LACURAZ (using the Gateway protocol, Invitrogen), containing a I-SceI target site as a control. Yeast reporter vectors were transformed into strain FYBL2-7B ( a, , , , ). For screening homodimer mutants, we used the protocol described previously (). Briefly, mating was performed using a colony gridder (QpixII, Genetix). Mutants were gridded on nylon filters covering YPD plates, using a high gridding density (∼20 spots/cm). A second gridding process was performed on the same filters to spot a second layer consisting of 64 or 75 different reporter-harboring yeast strains for each variant. Membranes were placed on solid agar YPD-rich medium, and incubated at 30°C for one night, to allow mating. Next, filters were transferred to synthetic medium, lacking leucine and tryptophan, with galactose (1%) as a carbon source, and incubated for 5 days at 37°C (30°C for I-CreI), to select for diploids carrying the expression and target vectors. After 5 days, filters were placed on solid agarose medium with 0.02% X-Gal in 0.5 M sodium phosphate buffer, pH 7.0, 0.1% SDS, 6% dimethyl formamide (DMF), 7 mM β-mercaptoethanol, 1% agarose, and incubated at 37°C to monitor β-galactosidase activity. Results were analyzed by scanning and quantification was performed using proprietary software. The structure of the I-CreI was solved by molecular replacement and refined to 2.0 Å resolution. The dimer without DNA allowed us to observe the protein conformational changes upon DNA binding after comparison with the protein–DNA complex (PDB code 1G9Z) (a). The most striking differences could be observed in the conformation of the C-terminal region. Whereas in the DNA-bound structure the C-helix and the C-loop are aligned with the DNA, in the unbound structure both elements are located on top of the cavity where the DNA binds, suggesting that the loop and the C-helix could work as a lock opening and closing the DNA-binding groove. This region was not observed in a previous structure of I-CreI with only one monomer in the asymmetric unit (). Besides, the C-terminal domain of I-CreI is well conserved among other members of its family () indicating its important role in this meganuclease group working mechanism (b). A detailed view of the protein–DNA interactions in the C-terminal area showed that Ser138, Lys139, Lys142 and Thr143 at the SKTRKT motif are involved in hydrogen bonds with the DNA backbone () (c). The position of these residues is completely different in the unbound DNA state (d), indicating that a conformational change is needed to bind the nucleic acid. Although these interactions were described before () and the amino acids are conserved, there is little information about their role during meganuclease action (,). To unravel the role of the C-terminal region of I-CreI, a series of trimmed, double and single mutants were designed based on the structural differences between the bound and unbound DNA structures. The two truncated mutants were designed to clarify the role of the C-loop and the C-helix. I-CreI Δ1 (amino acids 1–137) lacked both the C-loop and the C-helix whereas I-CreI Δ2 (amino acids 1–144) contained the C-loop. Based on the contacts with the DNA backbone in the SKTRKT motif, the double mutants I-CreI AM (S138A and K139M) and I-CreI GG (K142G and T143G) were produced, as well as their single variants I-CreI S138A, I-CreI K139M, I-CreI K142G and I-CreI T143G. The S138A and K139M mutations were designed to change the polar character of the corresponding side chains with minimal alteration of their sizes. On the other hand, The K142G and T143G mutants were designed to maintain the polarity and the flexibility of the loop, whereas the size of the side chain was reduced. To demonstrate that the effect in meganuclease activity was due to the mutations and not to structural changes, we analyzed their structure stability and oligomerization state. Thermal denaturation followed by circular dichroism of all the mutants showed cooperative, sigmoidal transitions similar to that of the wild type (Supplementary Data Figure 1a). Even though the denaturation is irreversible and causes protein precipitation, this result is consistent with the presence of a folded tertiary structure, although with small changes in stability as indicated by the narrow range of apparent mid point temperature values. This tertiary structure is the same for all the mutants and the WT as shown by the same set of dispersed signals observed in (Supplementary Data Figure 1b) their 1D H NMR spectra. It is well known that the I-CreI family of meganucleases binds DNA as homodimers, therefore to analyze the oligomerization state of the mutants they were subjected to analytical ultracentrifugation. The experiment showed that all the mutants behaved as dimers independently of the mutation (Supplementary Data Figure 1c). Altogether, these experiments indicate that the mutants are folded and conserve the I-CreI scaffold involved in meganuclease activity. Electrophoretic mobility shift assays (EMSAS) in the presence of Mg and Ca were used to analyze qualitatively the behavior of the C-terminal mutants in DNA binding (). Whereas the presence of Ca allows DNA binding, Mg is indispensable to bind and cleave DNA (). Even though the binding capability of I-CreI was abolished in the Δ1 mutant, the Δ2 one was able to bind the labeled DNA probe demonstrating that the C-loop is essential in DNA binding while the C-helix is not. In addition, binding was detected in the presence of both cations as in the wild-type I-CreI. To define the distinct properties of each site in the C-loop, the single mutants were assayed by EMSA in the same conditions. The binding of the labeled probe to the single mutants is less affected than to the double ones; however, they displayed differences depending on the cation present in the assay. Whereas a strong dependence of Mg could be observed in the Ser138–Lys139 site, the single mutants in the Lys142–Thr143 site could bind DNA notwithstanding the cation present in the mobility assay. These differences could be observed quantitatively after the measurement of the dissociation constants for all the different mutants in the presence of Ca (). The affinity of the wild-type is similar to that measured by others (), however all the mutants presented affinities that are reduced by a factor of 24 to 6000, the effect being larger for the double mutants as compared to the corresponding single ones, indicating a synergy between the two residues in each region for DNA binding. The analysis of the different mutants in the DNA-binding assays has clear implications for DNA cleavage activity, consequently an examination of their cleavage properties on a wild-type DNA sequence was carried out. displays a graph representing the percentage of cleavage against the amount of HE (the original gels are available as Supplementary Data). The mutants can be divided into two groups based on the comparison of their cleavage properties to the wild-type HE; the first is composed of the truncated mutants I-CreI Δ1 and I-CreI Δ2 and the double mutants I-CreI AM and I-CreI GG, whereas the single mutants I-CreI S138A, I-CreI K139M, I-CreI K142G and I-CreI T143G form the second. Members of the first group displayed a reduced cleavage activity when compared to the wild-type I-CreI ( and ). Although I-CreI Δ1 and I-CreI GG cleavage properties are abolished or severely affected, I-CreI Δ2 and I-CreI AM showed a reduced activity () that is increased when higher HE amounts are used (). However, the cleavage properties of the single mutants that composed the second group are similar to the wild type ( and ). These results indicate that the trimmed and double mutants whose DNA binding is abolished or affected do not cleave DNA or they need higher amounts of HE to cleave the plasmid. It is noteworthy that the I-CreI Δ2 mutant which conserves the wild-type amino acids in the C-loop but lacks the C-helix, even though its cleavage activity is reduced ( and ), is able to cleave the target, while the Δ1 is not, indicating that the C-loop is essential for cleavage but not the C-helix. On the other hand, the single mutants depict an activity similar to the wild type. To confirm our results , we performed cleavage assays with all the mutants as in (), including the I-CreI and I-CreI D75N proteins, as well with the 128 possible 5NNN_P and 10NNN_P targets (c). The use of the D75N mutation decreases the toxicity of I-CreI in overexpression experiments, and the wild type and its D75N mutant display similar activities and levels of specificity (). Cleavage patterns were similar to those previously found with I-CreI and I-CreI D75N (). None of the trimmed or double mutants, whose binding and cleavage activities were affected by the mutations , displayed activity on any of these targets. As an example, the I-CreI AM and I-CreI GG are shown (c, upper panel). Interestingly, the four single mutants cleave the 5GTC_P target (). Whereas I-CreI K139M is able to cleave 5GGG_P, 5TAA_P and 5TTC_P targets, the I-CreI S138A cleaves 5GTG_P, 5GTT_P, 5GCT_P and 5GCC_P targets displaying a very similar profile to that of I-CreI D75N (c). In contrast, all the single mutants showed activity on 10AAG_P and 10AAT_P target, in addition to the wild-type 10AAA_P. Lower levels of cleavage could also be observed with these four mutants with 10TCG_P and 10AAC_P. In addition, the I-CreI K139M mutant was also able to cleave seven additional targets, as it can be observed in c. The profile of the I-CreI K139M mutant is very similar to that of I-CreI (without its toxicity) while the three other single mutants are closer to I-CreI D75N (). It is noteworthy that I-CreI S138A, the only position of the four mutated residues that makes contacts with the 5NNN_P region of the DNA backbone (), shows a 5NNN_P cleavage profile similar to I-CreI D75N. Modifying the substrate specificity of DNA-binding proteins by mutagenesis and screening/selection is a difficult task. This is even harder in the case of HEs whose main characteristic is their large DNA recognition sites. Several laboratories have relied on a semi-rational approach to limit the mutant libraries to be handled, choosing a small set of relevant residues based on structural data. This set is generally composed of residues that in the I-CreI DNA complex structure make direct contacts with the homing site. However, the comparison of the C-terminal region conformations in the I-CreI and the I-CreI DNA complex suggested that this region could be involved in a lock mechanism with clear implications in DNA binding and cleavage (). In this article, we have dissected the role of the C-terminal region of I-CreI in the HE mechanism. After comparison between the I-CreI structures in different DNA-bound states, we identified two regions in the SKTRKT motif involved in DNA binding. To address the role of these amino acids in the endonuclease mechanism of I-CreI, we designed trimmed mutants in the C-terminal area and double and single mutants in the SKTRKT motif. The finding of these two regions demonstrates that the C-loop plays an essential role in DNA binding () and this function is performed in a concerted manner between the two amino acids in the site because the double mutants were preferentially affected in cleavage, both and . Remarkably, the mutations K142G and T143G bind the DNA probe with higher affinity than the S138A and K139M ones (). This difference between the two sites is probably due to their distinct proximities to the residues involved in DNA cleavage and metal binding. The well-conserved D137 (b) has been proposed to contribute to the organization of the water shell around the scissile phosphate in the metal-binding site (). Thereby, mutations in the surrounding area such as S138A and K139M, where residues that can promote hydrogen bonds are suppressed, could disturb the water shell in the metal-binding site, turning it more selective towards the physiological cation. This is not the case for the K142G and T143G mutations located ∼23 Å away from the metal-binding site, too far to promote changes in the solvatation in that area. Even though the I-CreI AM and the I-CreI Δ2 showed reduced cleavage activity ( and ), they did not show any activity suggesting that this residual activity was due to the large amount of protein used during the assay. The yeast cleavage assay confirmed that double and trimmed mutations abolished cleavage activity whereas the single mutants displayed a clear activity (c). In addition, our results indicate that protein regions lacking base-specific homing site contacts could play an important role not only in binding and cleavage, but also perhaps in target specificity (c). New specificities could be achieved by the creation of new base-specific contacts or the release of negative interactions. If wild-type I-CreI makes energetically non-favorable contacts with a particular 10NNN_5NNN_P site, then a mutant endonuclease could interact better with that site simply by eliminating the unfavorable interactions; this seems to be the explanation for the results in c. Therefore, the similar cleavage profile exhibited by the I-CreI S138A mutant and the D75N could arise from the release of energetically non-favorable contacts present in wild-type I-CreI. The use of meganuclease-induced recombination has long been limited by the repertoire of natural meganucleases. For genome-engineering applications, the major advantage of HEs is their exquisite specificity, a feature that becomes essential when engaging into therapeutic applications. Thus, the production of artificial endonucleases with custom specificities is an intense area of research. The finding that specificity of the target DNA could be controlled by amino acids lacking base-specific homing site contacts with the DNA complicates the production of intelligent molecular scalpels that recognize and substitute certain DNA sequences; however, this knowledge can be used to improve even more the specificity of the engineered enzymes. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e

In site-directed spin labeling (SDSL), a nitroxide containing a stable, unpaired electron is attached to a specific site within a macromolecule (). This yields a simple electron spin system that allows for a range of problem-targeted studies. Information on the local environment at the labeled site is derived by analyzing the electron paramagnetic resonance (EPR) spectrum of the nitroxide. SDSL has matured as a structural tool in protein studies, and has been remarkably successful in tackling systems that are difficult to study using methods such as X-ray crystallography and NMR spectroscopy (). In SDSL studies of nucleic acids, structural and dynamic information at the level of an individual nucleotide can be obtained via analyzing the EPR spectrum of a singly labeled nitroxide (see recent review ()). Recent studies also reported distance measurements ranging from 5 to 70 Å between a pair of nitroxides attached to DNA (,) and RNA (). The ability of SDSL to measure nanometer distances renders it an attractive tool for global structural mapping of nucleic acids. The key step in SDSL is attaching the nitroxide probe to a specific site of the biomolecule under investigation. This is achieved by placing a reactive functional group, such as a thiol or an amino group, at the intended labeling site, followed by a reaction with a nitroxide derivative (,). In proteins, generally the amino acid at the intended labeling site is mutated to cysteine, which provides a reactive thiol for nitroxide attachment (). Because normal nucleic acid functional groups are not reactive under physiological conditions, reactive thiol or amino groups have to be introduced exogenously in the form of modified nucleotides. Currently, solid phase chemical synthesis is the method of choice for placing the modified nucleotide at a specific internal location within a strand of DNA or RNA. Methods have been reported to attach nitroxides at specific backbone (), sugar (,) and base () positions of modified nucleotides introduced via chemical synthesis. The limitation of these methods is the size restrictions posted by chemical synthesis (∼100 nucleotide (nt) long for DNA, and ∼50 nt for RNA). Methods have also been reported for attaching nitroxides to RNA terminus (,). Shin and co-workers have reported a method for attaching nitroxides to the 5′ terminus of a RNA synthesized by the T7 RNA polymerase (). In this method, a modified nucleotide, guanosine 5′-O-(3-thiomonophosphate) (GMPS), is added together with the unmodified nucleoside triphosphates (NTPs). T7 RNA polymerase will utilize GTP or GMPS at the 5′ terminus of the RNA transcript, but will incorporate GTP at the subsequent internal positions. With a large excess of GMPS to GTP (3:1), a majority of the RNA transcripts will possess a 5′ phosphorothioate instead of a 5′ triphosphate, and the phosphorothioate group can be subsequently reacted with a thiol-reactive nitroxide derivative. The modified transcription method is restricted to RNAs that can be transcribed . The labeling yield depends on the fraction of RNA transcripts containing the 5′-phosphorothioate, which cannot be 100% due to the limit imposed by GMPS to GTP ratio. For labeling at 3′ terminus, there was a report on tRNA (), which utilizes periodate to oxidize the 3′-ribose, followed by subsequent nitroxide attachment. However, this method cannot be applied directly to DNA. Work reported here presents a new approach to attach nitroxides at the 5′-terminus of DNA and RNA molecules. This method, which was named the 5ps labeling scheme (), utilizes T4 polynucleotide kinase (PNK) to transfer the γ-phosphorothioate group of adenosine-5′-O-(3-thiotriphosphate) (ATPγS) to the 5′ terminus of DNA or RNA. Following reaction with an iodomethyl derivative of nitroxide, a stable nitroxide-nucleic acid adduct was formed. The labeling method is applicable to DNA and RNA molecules that are either chemically synthesized or obtained from cellular sources. The resulting covalently linked nitroxides were able to report the structural state of the parent molecule. These results demonstrate a novel, facile method for attaching nitroxides to the 5′-terminus of arbitrary nucleic acids. In principle, the 5ps method is applicable to other external probes such as fluorophores and photocrosslinkers. This will aid in a wide range of biophysical and biochemical studies of nucleic acids and protein/nucleic acid complexes. The model DNA, designated as DNA1, is 13-nt and has a sequence of 5′ CTCAAGGCAAGCT. The model RNA, designated as RNA1, is 16-nt and has a sequence of 5′ rCrGrGrUrArCrUrUrUrUrGrUrGrUrGrG. Both DNA1 and RNA1 contain a 5′-hydroxyl group and were obtained via solid-phase chemical synthesis (Integrated DNA Technology, Coralville, IA). DNA1 was used in subsequent reactions without further purification. RNA1 was purified using anion-exchange HPLC (). The phenylalanine-specific transfer RNA (tRNA) from Brewer's yeast (Sigma-Aldrich, St. Louis, MO) was used as a model of a large and naturally occurring RNA. Prior to the reactions, the 5′ phosphate group of tRNA was removed using calf intestinal alkaline phosphatase (CIAP) following a procedure previously described (). For a large-scale reaction (∼200–400 picomoles), the reaction mixture (400 μl) contained 0.025 unit of CIAP per picomole of nucleic acid in a 1× CIAP buffer (50 mM Tris–HCl, pH 9.3, 1 mM MgCl, 0.1 mM ZnCl and 1 mM spermidine). After incubation at 37°C for 30 min, another 0.025 unit of CIAP per picomole of nucleic acids was added, and incubation was continued for another 30 min. The mixture was then subjected to phenol extraction to remove the proteins, and the nucleic acids were recovered using ethanol precipitation. Nucleic acid concentrations were determined according to absorbance at 260 nm. The extinction coefficients used for DNA1, RNA1 and tRNA were 123 600, 155 800 and 760 000 M cm, respectively. To covalently attach a 5′-phosphorothioate group, DNA or RNA containing a 5′-hydroxyl terminus was treated with T4 polynucleotide kinase (PNK, New England Biolabs, Ipswich, MA) in the presence of ATPγS (VWR, USA). A typical reaction mixture (50 μl) contained 1× PNK buffer (70 mM Tris–HCl pH 7.6, 10 mM MgCl and 5 mM dithiothreitol), 10–100 μM nucleic acid, a 5:1 molar ratio of ATPγS to nucleic acids, and 10 units of PNK per nanomole of nucleic acids. Reactions were carried out at 37°C overnight. Immediately following ATPγS kinasing reaction, the solution was diluted 4-fold with a buffer containing 100 mM sodium phosphate (pH 8.0) and 20% acetonitrile (v/v). This mixture was reacted with a thiol-reactive nitroxide derivative, 3-iodomethyl-1-oxyl-2,2,5,5-tetramethyloyrroline (100–200 mM), which was freshly prepared as described (,). The reaction was allowed to proceed at room temperature in the dark, with reaction times ranging from 3 to 48 h. To remove the excess nitroxide from the short DNA1 and RNA1 oligonucleotides, the reaction mixtures were first subjected to anion-exchange HPLC using a PA-100 column (4 × 250 mm, Dionex Inc., Sunnyville, CA) (,). Nucleic acids were eluted using a low salt stationary phase (buffer A: 1 mM NaClO, 20 mM Tris–HCl, pH 6.8 and 20% v/v acetonitrile) and a high salt mobile phase (buffer B: 400 mM NaClO, 20 mM Tris–HCl, pH 6.8 and 20% v/v acetonitrile). Samples were detected via absorbance at 260 nm. The HPLC fractions were then desalted using a G-25 Sephadex column (22.5 × 1.2 cm) with water as the running media. The desalted samples were lyophilized and stored at −20°C. For tRNA labeling, the reaction mixture containing the nitroxide labeled tRNA (R-tRNA) was purified using a 6% denaturing acrylamide gel. The appropriate R-tRNA band was cut and eluted in 10 mM MOPS, pH 6.5 and 5 mM EDTA for 2 h. A subsequent ethanol precipitation was carried out to remove the remaining trace of unattached nitroxides and to recover R-tRNA. Desalted nucleic acid samples (∼2.0 μM in 5 mM sodium phosphate buffer, pH 8) were mixed with a matrix solution (8:1 ratio of 3-hydroxypicolinic acid/diammonium citrate) in a 1:2 ratio. Droplets of the sample mixture (∼1.50 μl) were spotted onto a sample plate and air-dried. Spectra were acquired on a Voyager-DE STR system (Applied Biosystems, Foster City, CA) in the range of 500–10 000 Dalton (Da). X-band EPR spectra were acquired on a Bruker EMX system following procedures previously described (). Briefly, samples of 5 μl were loaded into 0.84 mm O.D. capillaries (VitroCom Inc., Mountain Lakes, NJ) that were sealed on one end. All spectra were acquired at room temperature using a 2 mW incident microwave power and a 100 Gauss scan width. The 100 kHz field modulation amplitude and time constant of the detector were optimized to provide maximum signal to noise ratio with no line broadening. For each sample, individual spectra from multiple scans were averaged, and the resulting spectrum was corrected for baseline and normalized with respect to the total number of spins. To test whether T4 PNK accepts ATPγS as a substrate, the chemically synthesized DNA1 oligonucleotide, which contains a 5′-hydroxyl group, was kinased in the presence of either ATP or ATPγS. The reaction mixtures were analyzed by anion-exchange HPLC using a column that discriminates against charge as well as polarity. The product from ATP kinasing, p-DNA1, eluted later than the original DNA1 oligonucleotide (A), which is consistent with the increased negative charge upon converting the 5′-hydroxyl to the 5′-phosphate. When the reaction mixture from the ATPγS kinasing was analyzed, the product eluted later than p-DNA1 (A). This difference in the elution profiles can be accounted for by the fact that a 5′-phosphorothioate group has the same (–2) charge as that of a 5′-phosphate group but has a different polarity. This indicates successful production of the ps-DNA1 that contains a 5′-phosphorothioate. The identity of the ps-DNA1 was further confirmed by MALDI-TOF mass spectrometry. The measured average mass of ps-DNA1 was determined to be 4038.1 ± 1.1 Da based on multiple measurements (data not shown). This is consistent with the expected mass of 4037.6 Da for the corresponding sequence with a 5′ phosphorothioate. Based on the area under the curve corresponding to the respective species in the HPLC elution profile, close to 100% of DNA1 was converted to ps-DNA1 (A). Experiments on synthetic RNA oligonucleotides also demonstrated efficient 5′-phosphorothioate attachment (). To demonstrate nitroxide attachment at the 5′-phosphorothioate, ps-DNA1 was reacted with a nitroxide derivative, which is designated as R (). In initial experiments, attempts were made to purify ps-DNA1 from the excess ATPγS that is present in the kinasing reaction. However, during the purification process, the 5′-phosphorothioate group is hydrolyzed and converted to a 5′-phosphate (data not shown), resulting in a loss of reactive moiety and preventing subsequent reaction with the nitroxide derivatives. To overcome the hydrolysis problem, nitroxide attachment was carried out without removing the ATPγS. Excess nitroxide derivatives were added directly to the mixture containing the ps-DNA1, PNK and ATPγS, and the labeling reaction was allowed to proceed at room temperature overnight. When the products were analyzed by anion-exchange HPLC, the major fraction, designated as R-DNA1, was eluted earlier than the ps-DNA1 (B). R-DNA1 migrated differently from p-DNA1, and thus, is not the result of hydrolysis. Instead, R-DNA1 is a nitroxide labeled product, which loses one negative charge upon derivatization of the 5′-phosphorothioate group, and therefore migrates faster on an anion-exchange column. Similar phenomena have been previously observed in reactions of R with internal phosphorothioate groups (,). Based on the HPLC result (B), the nitroxide labeling efficiency is close to 100%. In addition, the average mass of R-DNA1 was determined to be 4192.0 ± 1.8 Da (C). This is in excellent agreement with the theoretical value of 4190.8 Da for a nitroxide labeled DNA1. Further proof of successful nitroxide labeling of DNA1 came from X-band EPR spectroscopy (). The aqueous spectrum of R-DNA1 showed three sharp peaks with asymmetric amplitude, and is clearly different from that of the free nitroxide, which shows three peaks with equal amplitudes (compare A and B). As an independent estimation of the overall efficiency of nitroxide labeling, the total number of spins (nitroxides) was determined by double-integration of the observed R-DNA1 spectrum. The corresponding DNA concentration was determined by UV-Vis absorption measurements. In the measurement, the spin concentration was determined to be 8.7 μM, and the corresponding DNA concentration was 10.5 μM. This yielded an 83% overall nitroxide labeling efficiency. We note that spin counting requires double integration of the EPR spectrum that is very sensitive to the baseline, and the number of spins determined has a certain degree of uncertainty. The 83% labeling efficiency determined via spin counting is consistent with the HPLC data (B). The R-DNA1 spectrum is characteristic of a nitroxide undergoing fast, isotropic rotation (B). In this motional regime, the effective rotational correlation time, τ, of the nitroxide can be estimated as (,): For the single-stranded R-DNA1 (B), the calculation yielded a τ of 0.19 ns for the labeled nitroxide. This short τ is due to the fast tumbling of the relatively small DNA1 molecule (4.2 kDa), the conformational flexibility of the single stranded DNA, and the flexible nature of the bonds connecting the nitroxide to the DNA. One would expect spectral changes upon formation of DNA duplex, which increases the size of the molecule and reduces the overall tumbling rate. This was indeed the case. Upon addition of a DNA that complements the DNA1 sequence, line broadening was observed in the resulting EPR spectrum (C), and a τ (for double-stranded DNA) of 0.61 ns was measured. Both the line broadening and the increase in τ indicate reduction in nitroxide motion. The change in the observed nitroxide motion reports slower uniform tumbling of the DNA duplex, as well as possible reduction in the motions of individual DNA nucleotide upon base pairing. The data indicate that the nitroxide attached via the 5ps scheme is capable of reporting duplex formation in DNA. The labeling scheme described above does not distinguish a deoxyribonucleotide . a ribonucleotide, and should be equally applicable to RNA molecules. To demonstrate this, the scheme was first applied to a chemically synthesized RNA, designated as RNA1. Following the procedures described for DNA modification and labeling, a phosphorothioate group was incorporated at the 5′ terminus of purified RNA1, and nitroxide labeling was carried out. Anion-exchange HPLC showed that ps-RNA1, which is the product of ATPγS kinasing, eluted later than RNA1 (A). The nitroxide labeled product, R-RNA1, eluted earlier than that of ps-RNA1 (B). This indicates successful nitroxide labeling. The HPLC data again indicate ∼100% efficiency in the overall nitroxide labeling (B). Mass spectrometry yielded an average mass of 5338.7 ± 0.3 Da for R-RNA1 (Supplemental Figure S1), which is very close to the expected value of 5339.2 Da. We also note that purifying the RNA before applying the 5ps scheme is not required, as the method is also demonstrated on a crude, chemically synthesized RNA (Supplemental Figure S2). The aqueous X-band EPR spectrum of R-RNA1 also shows three asymmetric peaks that indicate that the nitroxide is undergoing fast, isotropic rotation (C). The τ was estimated to be 0.45 ns. Based on a measurement of the spin concentration vs. the RNA concentration (14.1 μM vs. 15.7 μM), the nitroxide labeling efficiency is 90%. Upon duplex formation of R-RNA1 with its complementary strand (D), line broadening of the EPR spectrum was observed, and a τ of 0.65 ns was obtained. This indicates reduced nitroxide motion upon RNA duplex formation. We note that τ is different from τ, while τ and τ are similar. This may reflect variations of nitroxide motion in single-stranded oligonucleotides that have little defined structures. Minor peaks are observed in the HPLC traces of ps-RNA1 and R-RNA1 (). These likely represent RNA degradation products. By integrating areas under the respective peaks, it is estimated that the minor peaks account for 36% of the total population in the R-RNA1 trace (B). This represents degradations that occurred in both the kinasing and the nitroxide labeling steps. Most of the degradation occurred in the kinasing step (26% from the ps-RNA1 trace, A), where the RNA is subjected to overnight incubation at 37 °C in the presence of MgCl. After the 5ps labeling method was successfully demonstrated on a chemically synthesized RNA, it was applied to a naturally occurring RNA, the tRNA (A). As the tRNA is 76-nt long and contains a large number of modified nucleotides () (A), it cannot be spin labeled at the 5′-terminus using any method reported previously, and serves as a good example to demonstrate the unique power of the 5ps labeling method. In the experiment, the naturally present 5′ phosphate group on tRNA was first removed. The dephosphorylated tRNA was then subjected to ATPγS kinasing, followed by coupling with the thiol reactive nitroxide derivative (see Methods). The labeled tRNA (R-tRNA) was purified via denaturing gel electrophoresis. Based on results from the gels, >75% of the tRNA remains intact after kinasing and nitroxide labeling (Supplemental Figure S3). The degree of tRNA degradation is 18–25%, which is similar to that of the small synthetic RNA1 (). Overall, the data indicate that the 5ps method preserves the integrity of the parent RNA molecule. B shows the aqueous EPR spectrum of the gel purified R-tRNA obtained in the absence of added salt, which represents the unfolded state of tRNA. Compared to the spectra of short synthetic oligonucleotides (B and C), the tRNA spectrum has broader centerline and much smaller amplitude for the high-field peak (B). This reflects reduced nitroxide mobility, and is consistent with a nitroxide attached to the high molecular weight tRNA (∼23 kDa). Control experiments showed no EPR signal if the ATPγS kinasing step is omitted. This negates the possibility of the nitroxide reacting with the modified nucleotides present in tRNA, or that the nitroxide simply binds non-covalently to the tRNA. Upon addition of 5 mM MgCl, tRNA tertiary folding occurs, and clear changes in the R-tRNA spectrum were observed (B, ‘folded tRNA’): the centerline was further broadened, and extra features were observed at both the low-field and high-field regions (indicated by arrows in B). These report a drastic decrease in nitroxide mobility. The presence of 5 mM Mg favors the formation of the helical stem at the 5′/3′ terminus, as well as tertiary folding of tRNA. The nitroxide spectral changes reports the change in the local conformation at the 5′-terminus of tRNA upon its folding. Unlike the small synthetic RNA oligonucleotides, it is difficult to estimate the tRNA labeling efficiency. R-tRNA and tRNA are difficult to distinguish using chromatography or gel electrophoresis methods. Spin counting is problematic, as double integration of the R-tRNA spectrum, which is in a different motional regime than that of the small molecule standard, suffers severe baseline interference. Nonetheless, data presented above indicate that a nitroxide is successfully attached to tRNA, and the label can report tRNA folding. The data presented above demonstrated a facile 5ps labeling method that allows efficient attachment of a nitroxide probe at the 5′ terminus of nucleic acids. The 5ps method is applicable to both chemically synthesized and naturally occurring DNA and RNA, and is not limited by the size of the target molecule. The resulting nitroxide tag provides an informative probe for the growing field of nucleic acid SDSL, where information at the level of individual nucleotide can be derived via analysis of nitroxide dynamics as well as quantitative distance measurements. A major advantage of the 5ps labeling method is its applicability to naturally occurring nucleic acid species, as demonstrated by the results on the tRNA (). These naturally occurring species may include a large number of modifications that are essential for their function (A). These modifications are introduced via complex cellular machineries, and are extremely difficult, if not impossible, to synthesize using solid phase chemical synthesis or via simplified enzymatic methods such as RNA transcriptions. For these reasons, the 5ps labeling method is the only available means to attach a nitroxide at the 5′ terminus of these naturally occurring nucleic acid molecules. Compared to a previously reported method that utilizes periodate oxidation to attach nitroxides to the 3′-terminus of tRNA (), the 5ps method is application to both RNA and DNA, and thus is applicable to a wide array of important systems, such as methylated DNA (), ribosomal RNA (), spliceosomal RNA () and non-coding RNA (). Another advantage of the 5ps method is that it does not have a limitation regarding the size of the target molecule. In most of the other labeling schemes, the target molecule needs to be produced via solid phase chemical synthesis. For molecules as large as the tRNA, solid phase chemical synthesis remains to be challenging and costly. Even for smaller molecules that are easy to synthesize chemically, the 5ps method provides an attractive labeling approach, as it is carried out post-synthetically and provides a high degree of flexibility in the execution of the experiments. A limitation in the 5ps method is that the label is restricted to only one position—the 5′ terminus. Methods have been reported, such as circular permutation (), which allow one to engineer the system so that the terminus is positioned near the site of interests without diminishing function. This can be combined with the 5ps method to ‘walk’ the nitroxide label throughout the molecule. In addition, the 5ps method may not be appropriate for molecules that contain naturally occurring thiol groups (e.g. thiouridines). For such molecules, nitroxides can be directly attached to the naturally occurring thiol groups. The key step in the 5ps labeling method is to introduce the reactive 5′-phosphorothioate group. This is achieved via the T4 polynucleotide kinase, which is a workhorse in modern molecular biology. A single atom change between ATPγS and ATP only marginally affected the function of T4 PNK, and γ-phosphorothioate group can be transferred with high efficiency. Along this line of reasoning, one might wonder whether a nitroxide derivatized ATPγS might also be taken up as a substrate by T4 PNK? This was indeed tested during the development of the 5ps method, but unfortunately the answer is no. While ATPγS was coupled successfully with the iodomethly derivative of nitroxide, the purified nitroxide/ATPγS showed no reactivity when used as the substrate in the kinase reaction (data not shown). Studies reported here utilized an iodomethyl-derivative of the nitroxide to react with the 5′-phoshporothioate group. Other thiol-reactive nitroxide derivatives, such as methanethiosulfonate () and iodoacetamide (), should also be compatible. Furthermore, in principle the 5ps labeling scheme is not limited to nitroxide probes, but is applicable to other probes, such as fluorophores and photocrosslinkers. It provides a general approach for site-specific attachment of external probes in nucleic acids. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Bisulphite sequencing is a technique that is widely used for analysis of the methylation status of mammalian DNA (). The method allows cytosine and 5-methylcytosine to be distinguished, because of the selective deamination of unmethylated cytosine to uracil following sodium bisulphite treatment. Consequently, after conversion, unmethylated regions of DNA contain no cytosine. The dsDNA product obtained after subsequent strand-specific PCR amplification is abnormal in two respects: it contains only three nucleotide types in each strand (A,G,T versus A,C,T), and each strand has an excess of one nucleotide (T or A, respectively). Analysis of bisulphite-treated DNA can be performed either by directly sequencing the PCR products or by cloning the product and sequencing a number of independent clones. Both methods have their respective advantages and disadvantages, and the final choice depends on the exact aim of the experiment. The experimental differences that arise between the two approaches fall into four main areas: the template characteristics and primer design; the way in which the base-calling software interprets the trace data; the cost and speed of the analysis and the loss of methylation phase information (see the Discussion Section). When these modified DNAs are used as templates for automated sequencing by the Sanger dideoxy method, problematic aberrant base-calling can sometimes occur. Miscalling of the DNA sequence by some automated sequencers is the result of the adaptive algorithms, that are used by the analysis software in order to achieve extended reads. Such programs track the signal intensity of each dye and artificially amplify the signal when it is low for an extended period. The absence either of a C or of a G signal, in bisulphite sequencing of unmethylated DNA, triggers this adaptive response, which progressively amplifies the weak background signal and eventually inserts spurious C or G residues into the sequence. A further, non-experimental difficulty that frequently hinders bisulphite sequencing projects relates to editing and alignment of sequences. The bisulphite-treated DNA strands are no longer complementary, and neither strand is a perfect match to the original sequence (unless completely methylated). While it is possible to align these sequences to a reference sequence that has been bisulphite-modified , this process becomes problematic if the trace sequences come from a mixture of clones whose inserts originate from different daughter strands. In addition, in regions of extended runs of a single nucleotide the alignment may go out of register. Consequently, individual sequence reads often have to be manually corrected, which can be a tedious error-prone process when multiple clones are being analysed. To address these problems, a number of software programs have been developed that facilitate individual steps in the bisulphite sequencing process. These include primer design (,), bisulphite-treated sequence alignment (,) and generation of graphical or text-based outputs (). None of these programs offers an integrated solution for routine use; the individual programs are variously controlled from the command line, a graphical user interface (GUI) or a web browser. Their combined use requires a degree of computer literacy somewhat beyond that of the average biomedical scientist, which is a significant consideration now that bisulphite sequencing has been widely adopted into routine use in genetics laboratories. Also, while command-line programs are powerful and flexible (if well documented) they sometimes do not provide a facility for user intervention and data editing, which may be required for resolving anomalous results. A well-designed GUI allows easier interaction, since data can be entered and then repeatedly reanalysed and edited using different parameters, with the results readily visible. As a practical solution for the genetics laboratory, we have therefore developed CpGviewer, a well-documented GUI-based program. It can use as input either multiple plain text files or sequence electropherograms (without adaptive peak height adjustment), and then align them to a reference sequence. The methylation status of each CpG dinucleotide is determined automatically and the results displayed as an interactive grid, within which each column represents one CpG dinucleotide found in the reference sequence. Interactive editing is facilitated by the fact that each cell in a row provides a direct link to the underlying sequence data surrounding the corresponding CpG dinucleotide, aligned to the reference sequence. The program also allows the user to view the entire electropherogram of each file. It offers the ability to save the exported summary methylation data in a variety of graphical formats for publication purposes. It also has a facility for assisting in the selection of primers for amplification of bisulphite-treated DNA. Programming was done using Microsoft Visual Studio 2005 using the Visual Basic language. The program has been tested only on Microsoft Windows XP, and requires the .NET framework 2.0 to be installed. A sodium bisulphite DNA modification protocol was used as described previously (). The example data were derived from bisulphite sequencing of the differentially methylated region (DMR) in the imprinted ZAC (PLAGL1) tumour-suppressor gene (). The reference sequence corresponds to nucleotides 48434376–48433838 of the Chromosome 6 assembly, accession NT_025741.14. PCR of bisulphite-modified DNA was carried out using the following primers, which amplify both methylated and unmethylated sequences: Zac9: dCCCAACCRTATCTAAATCAAAACT; and Zac1: dGTGTTTAGGATAGTGTTTGGTT. PCR conditions were as follows: denaturation at 94°C for 3 min followed by 35 cycles of 94°C, 30 s; 58°C, 30 s; 72°C, 30 s and a final extension step at 72°C for 3 min. PCR products were gel-extracted, purified using the Geneclean II kit (MP Biomedicals, Solon, OH, USA) and ligated into pGEM-T Easy vector (Promega, Madison, WI, USA) according to the manufacturer's protocol. Ligations were transformed into DH5α cells and DNA extracted from recombinant clones using the Qiaprep Spin miniprep kit (Qiagen, Valencia, CA, USA). For analysis on the MegaBace500 sequencer (GE Healthcare, Amersham, UK), miniprep DNA was sequenced using the DYEnamic ET Terminator kit according to the manufacturer's protocol. For analysis on the ABI3130xl sequencer, sequencing reactions were prepared using the BigDye Terminator v3.1 Cycle Sequencing kit (Applied Biosystems, Foster City, CA, USA). Plasmid clones were sequenced using standard M13 forward and reverse primers. Since the program is designed to use as input either raw or analysed trace data from a number of different sources, it first extracts the trace data from each file type, and converts them into a common format that is used by its internal basecaller. If the data originate from an unprocessed file, the program automatically adjusts the baseline intensities of each trace and then corrects for spectral overlap between different dyes and different dye motilities. (However, since Staden files can originate from a variety of sources, it is assumed that .SCF data have already been corrected for differences in dye migration and spectral overlay.) Once the data are in the correct format, the base caller detects nucleotide peak positions by differential analysis of the individual traces. The typical peak to peak distance is determined and the data reanalysed to detect missed and false peaks. The sequence is then trimmed to remove poor quality sequence as determined by peak spacing. Although CpGviewer can be used perfectly well with previously analysed sequence data, we wrote this non-adaptive basecalling routine in order to circumvent major difficulties that certain adaptive routines (notably the Cimmaron basecaller supplied with MegaBace instruments) have when analysing sequences derived from unmethylated templates. The alignment is created using a local extension algorithm similar to that used in BLAST (). An array of overlapping DNA fragments 10 bp in length is created from the bisulphite-modified reference sequence. (This fragment length can be changed via  >  > .) These fragments are each mapped to regions on the query sequence that have an identical sequence. These sequences are then extended at both ends, the extension being terminated when three of the last five bases do not match between the reference and query sequences. The extended sequences are then sorted to remove duplicate alignments. The remaining alignments are then linked together, such that the largest fragment is linked to the start and end of the reference sequence using the remaining extended fragments, keeping the number and size of gaps to a minimum. Where gaps of unaligned sequences are created these are then aligned to each other using a pairwise alignment algorithm (). Each alignment is given a quality score that is the percentage of the alignment's total length derived from the extended fragments. While this alignment is performed on the converted reference sequence, the original reference sequence is also manipulated in the same way, and is displayed unmodified in the screen alignments, since this allows the user to identify specific CpG dinucleotides more easily than if traces are displayed aligned against the converted sequence. DNA sequences can be loaded into the program in a number of different electropherogram file formats. However, some knowledge of the behaviour of the individual sequencing instrument is advisable. Thus, for the MegaBace family of sequencers (GE Healthcare, Amersham, UK), *.esd files are best avoided, because of the adaptive baseline manipulation performed in generating them; instead, data are imported from unprocessed *.rsd files. Similarly, it is possible to use the raw sequence data in a *.AB1 file, rather than the processed data. During the analysis, the program examines the peak spacing of the traces and uses this information to trim the nucleotide sequence at its ends. However, internal regions of low sequence quality are not discarded. Rather, we have chosen to detect and discard low quality regions and vector DNA sequences later, when aligning the sequences to the reference sequence. The sequence alignment can be viewed either in the form of an interactive grid () or as a web page, containing ‘snapshot’ images of the local trace and alignment around each CpG in the reference sequence. These views each allow verification of the quality of the sequence and its alignment, around each CpG dinucleotide. The web page view can also be saved to disk, and so viewed independently of the source data. The interactive view comprises a window containing a grid of colour-coded squares (). Each column represents one CpG dinucleotide in the reference sequence and each row an individual experimental DNA sequence. Clicking on an individual cell displays its underlying data. The blue cells at the head of each column contain information on the position of that CpG dinucleotide in the reference sequence, while those at the start of each row identify the file name containing that sequence. For the remaining cells, black is used to highlight methylated CpG dinucleotides. Unmethylated CpG dinucleotides can either be displayed in white, or CpA and TpG can be shown in two distinguishable colours (pink and green). (In general, only one of CpA and TpG is applicable for a single experiment, but the two-colour display enables identification of any anomalous results.) Where a CpG dinucleotide in the reference sequence has been aligned to a sequence other than CpG, TpG or CpA its cell is yellow; this implies either the presence of a SNP at that position, or a sequencing or PCR artefact. Cells that correspond to positions that lie outside the alignment of an individual sequence trace are shaded grey. When a blue border square is left-clicked, the caption below the grid displays either (i) for a row, the source file name and the aligned sequence's percentage of dC and dG content or (ii) for a column, information about the position of that CpG within the reference sequence. Similarly, left-clicking a square within the main body of the grid identifies the individual CpG, its methylation status and its file name (). Right-clicking on a blue row square will display that file's electropherogram image, allowing the quality of the sequence to be observed (). Right-clicking on a main grid cell produces an output that depends on the origin of the sequence file: if the sequence was loaded as text, the sequence alignment is displayed with the chosen CpG position capitalized and marked by two asterisks (). If the sequence was generated from an electropherogram, right-clicking displays the local sequence trace with the corresponding part of the alignment (). This trace image display can be disabled by selecting  > . Using the CpG grid display window, it is also possible to generate a consensus sequence interactively. Selecting  >  adds a new row at the bottom of the grid. The cells in this row are initially grey, but change to match whichever colour is chosen by clicking on any cell in the same column. Once the consensus row is completed, the underlying sequence can be saved via  >  . This feature may be useful for various purposes. For example, if multiple sequence runs are performed on a single cloned product, the consensus row can be used to assemble the most accurate sequence of that clone. Alternatively, if each row represents a separate clone, a sequence can be built that represents the population consensus methylation state, e.g. for a specific tissue or patient. Also, since consensus sequences can be saved and then reloaded as text input files, the feature can be used consecutively to generate clone sequences and then a population consensus. For an imprinted CpG island, where the methylation patterns fall into two clearly distinguishable groups, a separate consensus can be created and saved for each allele. Most bisulphite sequencing projects require some additional editing of the alignment. Therefore, the true status of mismatched CpGs that are highlighted in yellow or grey can be manually assigned after inspecting the sequence data. To do this, the  >  option is first selected. A left-click on a square then allows the user to choose any of the possible alternative states of this CpG dinucleotide. (This choice is overlaid on the original grid square as a slightly smaller square of the new colour.) A right-click while in this mode brings up (as described earlier) the electropherogram or sequence alignment display window; the user can inspect this and then click on the window to assign the correct status (). The  >  command allows the grid display to be exported to a variety of graphics file formats. In addition to the square pattern used for the interactive grid, two ‘lollipop’ styles as commonly used in publications are available, either with fixed spacing or scaled to show the relative positions of each CpG within the sequence (). This exported graphic can also be generated either with the edited or the original cell data ( > ). As stated earlier, methylation analysis can be performed either by direct sequencing of PCR products or by sequencing a number of cloned PCR products. Each approach has its merits. At first sight, direct PCR sequencing appears a much quicker and cheaper option than cloning and then sequencing multiple copies of the product. However, it often requires significant effort to optimize the initial PCR conditions so as to eliminate template contamination with spurious amplification products. This is probably a consequence of the reduced sequence complexity of bisulphite-treated DNA. Also, if the genomic DNA displays two divergent methylation states (as is the case for differentially methylated regions (DMRs) of imprinted genes), the bisulphite PCR product contains two very different template sequences. These highly divergent sequences may then demonstrate quite disparate electrophoretic mobilities, (e.g. due to one sequence having a greater tendency to form a stable secondary structure). This can lead to superimposed ‘staggered’ electropherograms representing methylated and unmethylated alleles, which are hard to read. Neither of these technical issues is a problem when sequencing cloned products; since each clone contains just one sequence, differences in mobility do not arise, while any spuriously amplified PCR products can also be readily identified and discarded. CpGviewer is therefore primarily aimed at projects in which multiple cloned bisulphite-PCR sequences are being compared. Others, in contrast, have taken computational approaches that aim at robust quantitative analysis of directly sequenced bisulphite-PCR products (). The latter approach is necessary for high-throughput ‘epigenomic’ studies. For smaller studies focussed on one or two genomic regions, however, the cost and time incurred in template optimization for direct sequencing may be greater than that of cloning and sequencing the products. Furthermore, direct sequencing can never reveal the full information content present in bisulphite-PCR products derived from a diploid genome or from a mixed population of cells. For example, if a tumour DNA appears to show a methylation level of 50%, this could represent randomly distributed methylation of ∼50% of each CpG, or the existence of a mixture of two cell populations, one completely methylated and one completely unmethylated. Similar considerations apply to the interpretation of direct sequencing of DMRs of imprinted genes, where the two alleles may differ markedly in methylation status. Sometimes it is wise also to compare the results of direct sequencing with cloning and sequencing, thus permitting the detection of cloning biases that occur at some DMRs. Both direct and clonal bisulphite sequencing approaches can also sometimes be affected by base-calling problems. Recent advances in sequencing technology have been directed primarily at increased read length with the undesired side effect that the computational approaches used by some base callers can hamper the analysis of DNA with an unusual base composition. As described above, adaptive amplification of the C (or G) baseline can be induced by the deficiency of C (or G) residues in templates derived from bisulphite-modified DNA. Even if this effect (which is much more pronounced when methylated CpGs are absent from the starting material) is not severe enough to cause extra bases to be called, it also reduces the sequence quality score. We have found this effect to be particularly problematic when sequencing unmethylated DNA sources on the MegaBace500 instrument; this was the principal reason for including a non-adaptive base-calling algorithm for analysing raw sequence traces that does not adjust baseline intensity according to perceived sequence context. For the related reason alluded to above, we have also chosen not to discard sequence traces based on their quality scores. While such quality scores are very important in the construction of sequence contigs, they are of limited value when aligning multiple independent test sequences against a known reference sequence. In any bisulphite sequencing project, once the sequences have been base called, they must be aligned against the reference sequence. Again, the commonly used local and global alignment tools are not well suited to this task, both because the sister strands of bisulphite-treated template DNA no longer match the original sequence, and because variability in methylation status means that individual product molecules vary in sequence. We have therefore used an algorithm that can align DNA sequences irrespective of the sequence direction, methylation status and completeness of cytosine deamination. As stated earlier, it is at this stage of analysis that sequences are discarded, if they cannot be aligned to the reference. This approach allows us to shift the experimental quality control away from analysis of fluorescence peak heights, onto the comparison of the base-called sequence with a reference sequence (which it should align perfectly to, and against which discrepancies can be rapidly eliminated interactively). The final convenience offered by this program is the production of a graphical representation of the methylation status of all the experimental sequences. Graphical representations of this type are commonly used when publishing the results of bisulphite sequencing studies. Our program automates this tedious and error-prone data collation step, with user input limited to interactive editing and generation of a final consensus sequence (if desired). It is a simple matter at this stage to identify sequences with poorer quality reads, by the presence of yellow cells. Rapid display of the trace data corresponding to these cells allows the rapid validation or discarding of individual results. In conclusion, analysis of the methylation status of DNA by the sequencing of cloned sodium bisulphite-treated DNA products is a well-established and important laboratory technique that has for some time been hindered by a lack of easy to use data analysis software. We have made a number of design decisions in developing this program, that focus on the special requirements of this methodology. By discarding unnecessary or inappropriate steps that create bottlenecks in the analytical process, we have found that the time required to complete a methylation study can be dramatically reduced.

Posttranscriptional RNA modifications appear to be present in all organisms. At present, 107 different types of nucleoside modifications have been established, and 91 of them are found in tRNA (), (). One of the most prevalent modified nucleosides found in tRNA is 5-methyluridine (mU or rT), and in it occurs once in every tRNA species. The enzyme responsible for this modification in is encoded by the gene (). Although mU is present at position 54 in the TΨC-loop in almost all tRNAs from bacteria and eukarya, its absence induces only a minor growth defect (,). The TrmA enzyme belongs to a family of methyltransferases that catalyses methyl group transfer from S-adenosyl--methionine (SAM) to position 5 of the heterocyclic base of uridine (U) at position 54 of the tRNA. At present, this family of methyltransferases includes 67 proteins from 42 species, and is listed as COG2265 (lusters of rthologous roups (COGs) (). The biochemical function of the TrmA, Trm2p, RumA and RumB proteins of COG2265 is known. Both TrmA of and Trm2p of the budding yeast catalyse the formation of mU54 in all tRNA species, except for the yeast initiator tRNA (,). The RumA and RumB from synthesize mU1939 and mU747 in 23S rRNA, respectively (,). Ten different conserved motifs (I-X) are present in the Rossman fold MTases (), although not all MTases contain all of these motifs. Alignment of the four mU-forming enzymes TrmA, Trm2p, RumA and RumB reveals six of the ten conserved motifs (Motifs I, II, IV, VI, VIII and X, ). Formation of the mU54 by TrmA involves a covalent intermediate between the tRNA and a nucleophilic C324 in the enzyme (). The SH group of C324 reacts with the 6-carbon of U54 in tRNA, producing a nucleophilic centre at the 5-carbon of the U54 (enol or enolate; compound 2 in ). The methyl group from SAM is transferred to the 5-carbon of U54 (compound 3). Following a β-elimination, mU54 and the free enzyme (compound 4) are produced. The release of TrmA from the tRNA requires a general base, which has not been identified for TrmA. The U54 is buried in the tRNA through stacking between G53 and Ψ55, and is also involved in a reverse Hoogsteen hydrogen bond with A58. Therefore, prior to catalysis, TrmA must open the T-loop in order to gain access to U54, perhaps by disrupting the hydrogen bonds between the D- and TΨC-arms, which would also disrupt the U54–A58 interaction. This conformational change of the TΨC-loop occurs before the formation of the C324–U54 covalent adduct (,). A ‘flip-out’ mechanism similar to that shown for the RumA enzyme is most likely to occur (). The RumA catalyses the formation of the mU at position 1939 in 23 S rRNA. Its 3D structure has been determined, alone () and in complex with RNA and S-adenosyl--homocysteine (SAH), the product of the SAM cofactor following transfer of the methyl group to the RNA (). The catalytic C389 of RumA is present in motif VI as is the catalytic C324 of TrmA (). Based on the crystal structure and enzymatic assays of mutant RumA proteins, roles for several additional amino acids in the active site were proposed (). The F263 (F188 in TrmA) and Q265 (Q190 in TrmA), present in motif X, are important for the U1939 recognition. The D363 (D299 in TrmA) binds to SAH, Q265 and U1939. Amino acid R366 (R302 in TrmA) is also involved in the U1939 binding. The E424 in motif VIII (E358 in TrmA) acts either as the general base releasing the peptide from the enzyme and/or stabilizing the enolate intermediate. Purification of the TrmA from revealed that not only the native 42-kDa polypeptide was obtained but also that the native TrmA is associated with RNA (). The RNA is bound covalently to the enzyme, forming either a 54-kDa complex containing a piece of the 3'-end of 16S rRNA, or a 62-kDa complex containing a subset of undermodified tRNAs (). The latter complex was suggested to be the TrmA-tRNA intermediate during the formation of mU54 in tRNA (intermediate 2 in ). The reason for the presence of the TrmA-16S rRNA linkage is not understood. Thus, in logarithmically growing cells, the enzyme is present in three forms: a 42-kDa native form, a 54-kDa TrmA-rRNA complex and a 62-kDa TrmA-tRNA complex. Here, we have analysed the formation of mU54 in tRNA and the formation of the TrmA-tRNA intermediate in several mutants. The mutants were isolated for their inability to make mU54 in tRNAs () and by mutagenesis. The analysis was made in exponentially growing cells having the mutated gene in its normal location on the chromosome and with normal levels of the enzyme, SAM and the various tRNA species. Based on the recent findings on the action of the RumA protein (see above), we discuss the role which various amino acids might have in the formation of the TrmA-tRNA intermediate and mU54 in tRNA. We also compare the role of these amino acids with the corresponding amino acids of RumA. Our results suggest that the conserved amino acids in the TrmA protein, most likely, have similar roles as in RumA. Therefore, the structure of TrmA in the regions important for catalysis is predicted to be similar to that present in RumA. The surprisingly high level of the 62-kDa TrmA-tRNA intermediate found in exponentially growing cells is also discussed, and is suggested to be caused by the suboptimal concentration of SAM, which is required for the resolution of this intermediate. strains and plasmids used are listed in . LB medium () was used for growth of bacteria. When required, carbenicillin and chloramphenicol were used at concentrations of 50 and 15 μg/ml, respectively. Procedures for DNA digestions, agarose gel electrophoresis, DNA ligation and transformation of competent cells were performed essentially as described earlier (). The PCR amplification was performed using Taq DNA polymerase (Boehringer Mannheim GmbH, Mannheim, Germany) using the buffers supplied with the enzymes. Routinely, 5 pmol of the appropriate primers and ∼100 ng template DNA were added to the reaction mixture. Alternatively, the gene was amplified from cell suspensions using the PuReTaq Ready-To-Go™ PCR Beads (Amersham, UK) and purified by the PCR Kleen Spin Kit (Biorad). The PCR products were visualized by running 1% agarose gels, staining with ethidium bromide and exposition to the UV light. mutagenesis to obtain the C324A substitution in the TrmA protein was done on the pGP100 plasmid () using QuickChange™ site-directed mutagenesis kit (Stratagene, La Jolla, CA, US) according to the instructions of the manufacturer. The mutated gene was moved to the pDM4 suicide plasmid () which subsequently was transformed into the strain MW100. Resulting duplications were resolved by growing cells in the presence of 5% sucrose. The duplications are resolved since the sucrose is toxic for containing the pDM4 with gene (). Presence of the mutation corresponding to the mutant C324A TrmA was verified by sequencing. Alternatively, in order to mutate the other codons in the gene, a kanamycin resistance cassette from the plasmid pKD4 was placed between codons 107 and 108 of the gene by linear transformation into strain MW100 (). Since gene is close to the gene on the chromosome, this strain was used as a template in a PCR, where one of the primers was homologous to the kanamycin resistance cassette, and the other contained a desired mutation in the gene. The resulting product containing the resistance gene and the mutation was then transformed into strain MW100 carrying the pKD46 plasmid coding for the λ Red recombinase. The transformants were screened for the desired mutations by sequencing. Column-purified PCR fragments were used to sequence mutations in the gene. Sequencing was mainly performed with a BigDye Ready Reaction Kit (Perkin Elmer) sequencing premix in an ABI Prism 377 DNA sequencer. The sequences were analysed using the nucleotide BLAST at the National Center for Biotechnology Information (). Different mutants were grown in LB medium at 37°C to ∼4 × 10 cells/ml and harvested. Transfer RNA was prepared as described previously () and degraded to nucleosides with P1 nuclease followed by treatment with alkaline phosphatase (). The hydrolysate was analysed by high-performance liquid chromatography (HPLC) () on a Supelcosil LC column (Supelco) with Waters HPLC system. Alternatively, the hydrolysates were run on a Develosil 5µ RP-AQUEOUS C30 column (Phenomenex) with an identical gradient. The level of mU modification was normalized to the absorbance of tA at 254 nm. The relative amounts of mU in each mutant varied ± 15% in different runs. The detection limit was calculated by comparing the area of a small clearly visible peak to the area of tA. Bacteria were grown in 10 ml LB broth to a density of ∼4 × 10 cells/ml. The cells were disrupted by sonication, and cell debris was removed by centrifugation. Twenty micrograms of protein from the supernatant was separated by 12% SDS-PAGE. Separated proteins were blotted onto a Hybond-C™ membrane (Amersham Life Science, UK) essentially as described by () and immunodetection was performed using the ECL-PLUS western blotting kit (Amersham Life Science, UK). Primary antibodies, specific for the mU54-methyltransferase, were a kind gift from D. Santi (San Francisco, CA, US). Bands were scanned using a Fluor-S™ MultiImager (Biorad, Hercules, CA, US) and quantified using the Quantity One® software. The relative intensities of the TrmA proteins varied ± 15% in different western blots. We suggest that several additional bands appearing on the western blots is cross-reacting material since they are present in an strain deleted for the gene and in the mutant containing no detectable TrmA protein (see Results). BLAST program () was used to search for the gene and protein sequences, mainly at NCBI. The TrmA protein family was analysed using the cluster of orthologous groups at . Sequences were aligned using Multalin program () (), and the alignments were manipulated manually using the Genedoc program (). Four proteins of the TrmA family with known biochemical function were aligned using the Multalin program. They display several well-established motifs typical to the Rossman-fold-like SAM-dependent methyltransferases () (). The tRNA(mU54)methyltransferase (Trm2p) has a long N-terminal extension, which is absent in the enzymes. The 23 S rRNA(mU)methyltransferases RumA and RumB are characterized by the presence of an [FeS] cluster-binding motif (C81, C87, C90 and C162, RumA nomenclature). The presence of such a cluster was experimentally demonstrated for RumA (), but it is not clear whether it is present in RumB. No [FeS] cluster-binding motif is present in TrmA and Trm2p. Further, TRAM, a predicted RNA-binding domain, is present in the N-terminus of RumA and of Trm2p, but is lacking in the TrmA and RumB proteins. Several mutants were randomly isolated as being deficient in mU54 in their tRNA (). The amino acid changes in these mutants have been established by sequencing and are marked by ‘*’ in and the corresponding strains listed in . Relative amounts of the native 42-kDa TrmA enzyme and the 62-kDa TrmA-tRNA intermediate were determined in the various mutants by western blot analysis. From the same cultures, the level of mU54 in tRNA was also measured by HPLC analysis. All cultures were grown in LB medium at 37°C and harvested at a cell density of 4 × 10 cells/ml to ensure that the cells were in the exponential growth phase. The results are presented in . The mutant has a W202C amino acid exchange in TrmA, but also a silent mutation in a codon corresponding to H340 at the C-terminus. This mutant has only 15% of the wild-type level of the mU54 in tRNA, and no detectable native TrmA or TrmA-tRNA intermediate. Apparently, the residual level of enzyme present in the cell is sufficient to catalyse the formation of some mU54 in tRNA, but the enzyme is quickly degraded before or during protein extraction for western blot analysis. The G220D amino acid substitution in the putative SAM-binding site in the mutant results in almost no mU54 in tRNA and no 62-kDa TrmA-tRNA intermediate. The mutant has a W132C substitution, which leads to an increased level of the 62-kDa TrmA-tRNA intermediate and a 66% decreased level of mU54, suggesting that this alteration in TrmA decreases the resolution of the intermediate, resulting in less formation of mU54. Two other mutants, and , contain G360D and G358K substitutions, respectively, located at the extreme C-terminus of the TrmA. These mutants do not accumulate the 62-kDa TrmA-tRNA intermediate, and have a very much reduced level of mU54. Since the total level of TrmA was reduced to about 55% of that found in the wild type, these alterations probably also reduce the stability of the TrmA protein. In order to test which additional amino acids could be important for the formation of the covalent tRNA-TrmA complex and the formation of mU54 in tRNA, several additional alleles of the gene in strain MW100 were created (). Motif X of the TrmA protein contains two conserved residues, F188 and Q190, predicted to have an important role for the uracil recognition of the RNA(mU)methyltransferases (corresponding to the F263 and Q265 in RumA, see above). We have constructed mutants and , which have the F188A and Q190A alterations in TrmA, respectively. Although the combined level of the native TrmA and the TrmA-tRNA intermediate was only slightly reduced to 84% of the wild-type level in the F188A mutant, the ratio between these two forms was about the same as in wild-type cells. However, the level of mU54 in tRNA was only 22% of that found in wild-type strain. By contrast, a Q190A substitution, which is located close to the F188, leads to almost no formation of the TrmA-tRNA intermediate, although the total level of the TrmA proteins were about the same as in the (F188A) mutant (77%). The mU54 level was reduced to 14% of wild-type level in the Q190A mutant. Apparently, the Q190A alteration, but not the F188A alteration, affects the step resulting in the formation of the TrmA-tRNA intermediate, which in turn is pivotal for the formation of mU54 in tRNA according to the model of catalysis (). The D363 and R366 residues in the motif IV of RumA were proposed to bind the U1939 of 23 S rRNA (,,). To test the significance of the corresponding residues in TrmA (D299 and R302, respectively), alterations in these residues were obtained by site-directed mutagenesis, resulting in the chromosomal (D299A) and (R302A) mutants (corresponding to D363 and R366 in RumA). In the (D299A) mutant, the level of the 62-kDa tRNA-TrmA intermediate is significantly reduced, and accordingly, the level of mU54 in tRNA is almost undetectable (). In the construct, the level of the 62-kDa TrmA-tRNA intermediate is significantly reduced, but the amount of the mU54 in tRNA is only slightly reduced to 87% of the wild-type level. Apparently, these two amino acids, being close to each other, have a very different impact on the activity of the TrmA . Amino acid C324 was demonstrated to be the catalytic amino acid residue (). We have therefore created the chromosomal allele encoding the C324A substitution in TrmA by directed mutagenesis. As expected, this mutant does not form the 62-kDa tRNA-TrmA complex, and contains no detectable mU54 in the tRNA (). Note that the level of the native TrmA was also reduced, indicating that this enzyme is less stable when it is not able to bind to tRNA. We show here how various amino acid alterations of the tRNA(mU54)methyltransferase (TrmA) affect the synthezsis of mU54 in tRNA, and how the relative levels of the 42-kDa native form of TrmA and the 62-kDa TrmA-tRNA intermediate were affected . Amino acid substitutions of F188, Q190, G220, D299, R302, C324 and E358, which are conserved in the four biochemically characterized RNA(mU)methyltransferases (), reduce the formation of the covalent 62-kDa tRNA-TrmA intermediate and/or the enzymatic activity, as shown by the reduced level of mU54 in tRNA. Also, the substitutions of W132 or W202, conserved among the bacterial tRNA(mU54)-methyltransferases but not among the other RNA(mU)methyltransferases, reduce the synthesis of mU54 in tRNA. Moreover, the W202 is important for the stability of TrmA, and the W132C alteration resulted in increased accumulation of the 62-kDa TrmA-tRNA intermediate as compared to the wild-type level of this intermediate. Our results suggest that the structural elements important for activity of TrmA are similar to those of RumA, which is responsible for the formation of the same methylated nucleoside in position 1939 of 23S rRNA, and for which the 3D structure of the catalytic centre is known. Here, we have analysed the formation of the TrmA-tRNA intermediate and the mU54 in tRNA in several mutants. Our analysis was performed in true conditions, since the TrmA protein was expressed from the gene located at its normal position on the chromosome. Thus, unlike the approach, our method compares the enzymatic activity of the wild-type and mutant tRNA(mU54)methyltransferases at physiologically normal conditions. In addition, the relative level of the mutant forms of the TrmA enzyme and formation of the 62-kDa TrmA-tRNA reaction intermediate were monitored using the same batch of cells. We believe, therefore, that the results obtained by such analysis reflect the kinetics of mU54 formation in tRNA in normal conditions. We discuss below the influence of various amino acids in the TrmA protein on the formation of mU54 in tRNA in relation to those important for the synthesis of mU1939 in 23S rRNA catalysed by RumA protein as suggested by an analysis of the 3D structure of this enzyme. The crystal structure of the RumA protein was determined alone () and in complex with RNA and the inhibitor SAH (). Although the functions of several conserved amino acids in RumA were proposed (see Introduction), the enzymatic activities of mutant RumA proteins were experimentally tested only for alterations of Q265 (corresponding to Q190 in TrmA), D363 (D299 in TrmA) and E424 (E358 in TrmA) (). Here, we have tested the enzymatic activity of the corresponding amino acids in TrmA, as well as several other amino acids judged to be important for catalysis. Amino acid C324 of the TrmA is the catalytic amino acid residue (). The corresponding C389 of the RumA is covalently attached to carbon-6 of U1939 via a thioether linkage in the RumA-23S rRNA co-crystal (), A). The C324A TrmA mutant is unable to form the covalent TrmA-tRNA intermediate and mU54 in tRNA (), which confirms the crucial role played in catalysis by the corresponding cysteines in RumA and TrmA, respectively. Moreover, the absence of the 62-kDa TrmA-tRNA complex in this mutant is consistent with our suggestion that this complex is the postulated intermediate compound 2 or 2a (). The F188A amino acid substitution in motif X of TrmA did not affect the level of the 42-kDa native TrmA or of the 62-kDa TrmA-tRNA intermediate but severely reduced the synthesis of mU54 in tRNA (). The corresponding F263 of RumA forms an edge-to-face aromatic interaction with the uracil ring and is itself held by the sugar-phosphate backbone of U1939 and the homocysteine moiety of SAH [, ()]. Since the F188A mutation does not affect the formation of the 62-kDa TrmA-tRNA intermediate (), it suggests that F188 (F263 in RumA) is not important for the positioning of the uracil ring and the formation of the intermediates 2 and 2a (). Instead, it may be important for the proper positioning of SAM for the methylation of the U54. In contrast, the Q190A substitution in TrmA affects both the formation of the covalent 62-kDa TrmA-tRNA complex and the synthesis of mU54 in tRNA (). Q265 (Q190 in TrmA) in RumA forms hydrogen bonds with N3 and with O4 of U1939 (), and is also involved in the binding of SAM. The Q265A mutant of RumA displays an 830-fold lower specific activity compared to the wild-type enzyme (). These and our results suggest that Q190 (Q265 of RumA) is the primary uracil-recognizing residue, and is important for positioning of the U target prior to the nucleophilic attack by the catalytic cysteine. The D299A alteration in motif IV of TrmA leads to a severely reduced formation of the TrmA-tRNA intermediate and absence of mU54 in tRNA (). The corresponding D363A substitution in RumA also results in complete loss of the RumA activity (). According to the 3D structure of RumA, D363 makes two H-bonds with O4 of U1939 and one H-bond with SAH (B). We propose that D299 has a similar role in TrmA. The guanidine group of R366 in the motif IV of RumA hydrogen bonds to O2′ and O3′ of the ribose of U1939 (B), which suggests an important role for this amino acid residue in RumA. However, the R302A (R366 in RumA) alteration in TrmA only slightly reduced the formation of mU54, although the formation of the TrmA-tRNA intermediate was reduced by about 2-fold (). Apparently, the role of R302 in the formation of mU54 is only a minor one, and the lower level of the TrmA-tRNA intermediate in the R302A mutant is likely due to reduced stability during cell extraction for western blot analysis. The E424 of RumA was demonstrated to be a general base for proton abstraction and ß-elimination, since intermediate 3 () accumulated in the presence of SAM in a reaction catalysed by an E424Q mutant (). One would expect that in the absence of SAM, such a mutant enzyme should catalyse the formation of intermediate 2 and/or 2a. However, this is not the case because the E424Q alteration probably affects the relative stabilities of the reaction intermediates 2 and 2a (), possibly by changing the local electrostatic environment. The corresponding E358K mutant of TrmA has a very low level of the TrmA-tRNA reaction intermediate, and the cellular level of mU54 is only 9% of the wild-type strain. We therefore suggest that E358 is the general base for proton abstraction and ß-elimination in TrmA. As in the case of RumA, substitution of E358 probably makes the reaction intermediates 2 and 2a unstable. The (G220D) mutation that alters the putative SAM-binding site and thereby abolishing the formation of mU54, also abolished the formation of the 62-kDa TrmA-tRNA complex (), which we suggest to be the reaction intermediate 2 and/or 2a (see above). Accumulation of this reaction intermediate(s) occurs in the absence of SAM (), while one would expect its accumulation in the mutant deficient in SAM binding. However, this is not the case. Apparently the G220D alteration, which probably changes the structure of the SAM-binding domain, also blocks the formation of the covalent intermediate 2 and/or 2a, similarly to the E358K mutant (and E424Q mutant of RumA). About 30–45% of the wild-type TrmA protein in is covalently bound to undermodified tRNA as the 62-kDa TrmA-tRNA complex [ and ; ()], which represents the intermediates 2 and 2a (). The resolution of intermediate 2 requires SAM (). Therefore, presence of this intermediate at such high level in exponentially growing wild-type cells may indicate that the concentration of SAM is suboptimal in relation to the of TrmA, which is between 12.5 () and 17 μM (). When is growing exponentially in LB at 37°C (i.e. the same conditions as used by us), the intracellular concentration of SAM is between 1 μM () and 13 μM () assuming a cell volume of ∼2 × 10 l (). Thus, the estimated intracellular level of SAM is below the required concentration for attaining for TrmA to resolve the 62-kDa TrmA-tRNA intermediate. This may explain the relatively high level of such intermediate in wild-type cells. Interestingly, the cellular level of the dimethylallyl pyrophosphate, the cofactor for the production of another tRNA modification, iA37, is limited. This results in a reduced level of iA37 if the demand for dimethylallyl pyrophosphate is increased in other areas of metabolism in which it also participates (). Since some hypomodification results in less efficient translation, these cases exemplify links between metabolism and translation, and may constitute a regulatory device for their co-ordination (,). Some of the mutants were isolated by the classical genetic approach of random screening for the mutants unable to form mU54 in tRNA (). Although such an approach resulted in alterations in well-established sequence motifs ( and ) and their function could be predicted by the ‘sequence–structure–function’ approach, some of the isolated mutants ( and ) have alterations in unexpected positions, and their role in the formation of mU54 cannot be easily explained at present. However, when a 3D structure of TrmA is available, their role in the synthesis of mU54 should be apparent and validate the structure. These results demonstrate the usefulness of an unbiased genetic approach to elucidate the role of certain amino acids in the protein in addition to the ‘sequence–structure–function’ approach which requires detailed knowledge of the protein structure. In summary, our results suggest that several conserved amino acids in the C-terminal domain are important for catalysis in both TrmA and RumA proteins. In addition, the G428D or the C521A substitutions (corresponding to the G220D and C324A of TrmA, respectively) completely inactivated Trm2p (). The TrmA protein lacks the N-terminal ‘TRAM’ domain that is present in Trm2p and is involved in the 23S rRNA binding in RumA [; ()]. The RNA substrate recognition should therefore be different between TrmA and the two other RNA(mU)methyltransferases, even though the catalytic domains seem to share extensive similarity. Our results are consistent with the theory of the modular evolution of RNA-modifying enzymes (), which suggests that the specificity of the enzymatic reaction is achieved by combining different (predicted) RNA-binding domains with different catalytic domains.

The papillomaviruses (PVs) are small DNA viruses that infect epithelial cells and cause warts, but some viruses are linked to cancers in humans (). PVs are also important model organisms for gene control and replication in mammalian cells, as their genomes are regulated similarly. The viral transcription factor E2 is the master regulator of the PV chromosome, controlling both transcription and replication. Tethering E2 to DNA via the C-terminal DNA binding and dimerization domain (DBD) is essential for both these functions, and the N-terminal trans-activation domain (TAD) is linked to the DBD via a flexible hinge (). The role of E2 in replication is to recruit the viral initiator protein E1 to the viral origin of replication (; see A). The critical E2 TAD-E1 helicase domain interaction (,) is conserved in bovine papillomavirus (BPV) and human papillomaviruses (HPV), and its molecular details have been revealed in the crystal structure of HPV 18 E1 in complex with the E2 TAD (). The molecular events of the initiation of viral replication are becoming increasingly well understood. The first step in replication is the formation of an E1E2– complex on the E1 binding site (BS) and E2 BS12 (A). This pre-initiation complex contains a dimer of E1 and a dimer of E2 (). In a reaction that requires ATP, E2 is displaced and more molecules of E1 are recruited to . This initial E1– complex forms the nucleus of a higher order E1 initiator complex that melts the DNA. The action of the distal E2 BS11 is to promote further recruitment of E1 molecules to the precursor of the -melting complex, but E2 itself is not required directly for -melting activity (,). templates with only a distal E2 binding site also support plasmid replication. , the initial recruitment of E1 to is also via an E1E2–-like complex, mediated exclusively through the E2 TAD-E1 helicase domain interaction, but this complex forms less efficient compared to templates with proximal E2 BS12 (). Despite the emerging details of initiator complex assembly, the underlying mechanisms of cell-cycle regulation of BPV replication initiation remain unclear. One important general question is how differential control of processes like transcription or replication is achieved by a limited array of regulatory proteins. This is particularly relevant in BPV where a single protein, E2, governs the entire genome. One mechanism is to regulate DNA-binding site occupancy, for example by DNA sequence affinity. However, controlling protein activity through redox sensitive thiols in DNA-binding domains (DBDs) has also emerged as a fundamental mechanism of regulation. Transcription factors whose DNA-binding domains are known to be regulated by thiols include AP-1, NF-kB, SP-1 but also the papillomavirus E2 DBD (). In the case of E2, Cys340 has been identified as the critical residue, sensitive to oxidizing agents and sulphydryl modifying reagents such as -ethylmaleimide. However, as in most cases, the underlying mechanisms of redox regulation, in particular the chemical species involved, are not known in molecular detail (). At the same time, many cellular processes, including proliferation, differentiation, senescence and apoptosis, are redox-regulated, and proteins other than transcription factors have functional thiols (). Here we describe the crystal structure of the BPV E2 N-terminal TAD where we were surprised to find a novel dimerization interface stabilized by a disulphide bond. Our complementary studies show that this interface forms between TADs within a preformed E2 dimer which is itself stabilized by a tight interaction between the C-terminal DBD domains (). We also demonstrate that TAD–TAD dimerization inhibits the TAD–E1 interaction, in agreement with the observation that the TAD–TAD dimerization interface buries part of the surface involved in the interaction with E1. Taken together, the data suggest that in the case of BPV the E2 TAD interaction with E1 is redox regulated. Furthermore, the reactive Cys57 residue of the TAD is more sensitive to oxidation than Cys340 that regulates DNA-binding activity, indicating that the TAD dimerization reaction is a significant means of regulating BPV E2 activity. To our knowledge, this is the first demonstration that the association of activation domains with their targets can be redox regulated, and evidence supporting the hypothesis that mammalian DNA replication may come under redox control. A truncated form of BPV-1 E2 comprising the N-terminal transactivation domain (TAD; amino acids 1–209 of 410) was expressed using pET11c in BL21 (DE3). This construct includes the entire TAD, and terminates at the beginning of the E2 hinge region. Growth and expression was at 18°C for 8 h after reaching an OD ∼0.8. Frozen cells were lysed in 50 mM Tris–HCl pH 8.0 (4°C), 0.2 M NaCl, 5 mM EDTA, 20% w/v sucrose, 10 mM DTT and 1 mM PMSF by lysozyme treatment (0.5 μg ml), and sonication. The cleared lysates were adjusted to 0.6 M NaCl, nucleic acids removed with polyethylenamine P, and protein precipitated with 30% w/v (NH)SO. The TAD was purified by gel filtration (Sephacryl S-100, 25 mM Tris–HCl pH 8.0 (4°C), 0.25 M NaCl, 0.1 mM EDTA, 5% v/v glycerol, 2 mM DTT and 0.1 mM PMSF), anion exchange [Source Q, 0–0.25 M NaCl in 25 mM Tris–HCl pH 8.9 (4°C), 0.1 mM EDTA, 2.5 mM DTT, 10% v/v glycerol and 0.1 mM PMSF] and hydrophobic interaction chromatography [Source-phenyl, 1.25–0.25 M (NH)SO in 25 mM NaPhosphate pH 7.5, 0.1 mM EDTA, 2 mM DTT, 10% v/v glycerol and 0.1 mM PMSF]. The purified E2 TAD was dialysed against 20 mM Tris–HCl pH 8.0 (4°C), 0.3 M NaCl, 0.15 mM EDTA, 10% v/v glycerol, 2 mM DTT and 0.1 mM PMSF, concentrated and stored at −80°C. E2 and GCN4E2 proteins were purified as previously described (), except for an additional gel filtration step with 1 or 10 mM DTT. Mutations in the TAD were generated by overlapping primer extension. Crystals were grown using hanging drop vapour diffusion by mixing 1 μl of 15 mg ml protein solution, containing 10 mM Tris–HCl pH 8.5 and 0.3 M NaCl, with 1 μl of precipitant containing 0.1 M Tris–HCl pH 8.5, 0.3 M NaCl and 18–22% tertiary butanol. For R172A the reservoir also contained 2 mM DTT. Crystals of both wild-type and R172A mutant were transferred into a cryoprotectant solution containing 60% -butanol, 0.3 M NaCl and 10 mM Tris pH 8.5. The X-ray data were collected at ESRF and processed using DENZO and SCALEPACK (). Crystallographic calculations were performed using the CCP4 suite of programs (). The initial structure (wild-type protein) was solved by molecular replacement using the structure of HPV16 E2 TAD () as a search model, where there is 36% sequence identity between BPV E2 and the HPV E2 TAD segment used (residues 1–188). Refinement was performed using REFMAC () and model rebuilding was carried out using X-Autofit () implemented in Quanta (Accelrys). Statistics of the X-ray data and final refined models are shown in . E2 TAD proteins were incubated at 1 mg ml (4°C for 14–16 h) in 20 mM Tris–HCl pH 8.0, 0.15 M NaCl, 5% v/v glycerol, 0.01% NP40, 0.1 mM PMSF) with 20 or 0.05 mM DTT after buffer exchange with a G25 microspin column (Amersham Bioscience). Proteins were treated with 30 mM -ethylmaleimide (NEM, from a 200 mM stock in ethanol) for 10 min at room temperature before heating to 95°C (5 min) and loading on a standard SDS–PAGE gel (12%, 29:1 acrylamide:-acrylamide). Full length E2 proteins were analysed on 8% gels (29:1 acrylamide:-acrylamide). Analytical gel filtration was performed on a Superdex 75 column (Amersham Bioscience). Gradient sedimentation of E2 (120–150 μg) was performed on 20–40% glycerol gradients spun for 16 h at 237 000 × , at 4°C (Beckmann SW55 rotor). The buffer was as described earlier, except 100 mM NaCl. Sedimentation profiles were analysed by SDS–PAGE, after treating protein samples with -ethylmaleimide as described earlier, and densitometry (digitized images, Kodak 1D 3.5.4 software). GST-E1 was purified as previously described (). GST was purified on glutathione-sepharose followed by gel filtration. Ten pmol of GST-E1 or GST were bound to 10 μl of glutathione-sepharose before washing and binding of 1 pmol of E2 in 200 μl reaction (20 mM NaPhosphate pH 7.2, 135 mM NaCl, 10% glycerol, 0.1% NP40, 0.1 mg ml BSA, 1 mM PMSF and 1 mM DTT) for 30 min. PKA-tagged E2 proteins (MGH) were labelled using PKA (Novagen) and [P]γATP, and free ATP was removed using a G25 column. Binding reactions were washed in binding buffer, and recovered proteins analysed by SDS–PAGE after treatment with NEM. Gels were analysed by phosphorimaging (Fuji FLA3000, image guage V3.3 software). Probes and binding reactions with E1 and E2 have been described before (). Briefly, the BPV sequence cloned into the pUC19 vector is TCAAGTAAAGACTATGTATTTTTTCCCAGTGAATACAC, the distal (BS11) and proximal (BS12) E2 binding sites are shown in bold and the E1 BS underlined. The sequence of the GCN4 binding site that replaces distal BS 11 is ATGACTCAT. Reactions that assayed E1E2– formation were performed in the absence of ATP, while reactions that assayed stimulation of E1– formation by GCN4E2 from a distal binding site contained 5 mM ATP (,). The E2BS9 probe was made by annealing two oligonucleotides, CCGGGAAGTCGAAC and CCGGGTTCGACTTC, and labelling with [αP]dCTP using Klenow exo (NEB), followed by a chase with 40 μM dCTP/dGTP. Gels were exposed to phosphorimager screens. E2 binding reactions were analysed on 6% (29:1 acrylamide:-acrylamide) gels and E1E2–-binding reactions on 5% (79:1 acrylamide:-acrylamide gels), both with 0.25 × TBE buffer. E1– complexes were resolved on agarose gels (1% TAE running buffer), after cross-linking with glutaraldehyde. Potassium permanganate footprinting was performed as described (). We determined the crystal structure of the BPV TAD at 2.8 Å resolution. Notably, the crystals formed only in the absence of DTT. In the structure, two L-shaped monomers of the TAD are arranged in a manner resembling a handshake to form a dimer stabilized by a disulphide bond formed between Cys57 residues of the two monomers, B. The dimer has overall dimensions of 45 × 60 × 65 Å, and a substantial ∼2500 Å of surface area buried per monomer. Two contact areas of roughly equal size form, one between the N-terminal halves (N1) where the disulphide linkage resides and another between the C-terminal halves (N2) of two TADs. A pair of hydrogen bonds is formed in N1, between the side chains of Gln12 and Arg58 (not shown), and an ion pair is formed in N2 by the side chains of Arg172 and Asp175, circled and shown on the right of B. The total number of direct inter-subunit hydrogen bonds, 2.4/1000 Å, is significantly lower than the average value of 7/1000 Å of contact area observed in dimeric proteins (), notwithstanding the more polar interface of E2, with 44% of interface atoms polar compared to the average value of 35% observed in stable dimers. Thus, the dimer appears to be largely stabilized by the disulphide linkage between the two monomers. We also obtained crystals for R172A mutant TAD under the same conditions but with 2 mM DTT and solved the structure to 2.35 Å resolution. The crystals belong to the space group P3 with one molecule per asymmetric unit, inconsistent with any oligomeric arrangement. Comparison of the R172A structure with the structure of the wild-type protein shows that the mutant has an identical fold with the overall r.m.s. difference of 0.76 Å calculated over the Cα atoms. Most significant differences are in a few surface loops that adapt to new crystal contacts (C). The data indicate that the inability of R172A to form the dimer is exclusively due to the reducing conditions and the R172A substitution. To determine if the E2 TAD dimer forms in solution Cys57 was exploited as a natural cross-linking group. Recombinant C57A, R172A and D175A substituted proteins were purified from . As negative controls we mutated residues R37 and Q80, involved in formation of the different HPV16 E2 TAD dimerization interface (), to alanine. The CD spectra of all mutant proteins tested were similar to wild-type, demonstrating that their overall protein fold was conserved (not shown). In SDS–PAGE, all proteins were monomeric (23.86 kDa), under reducing conditions (20 mM DTT, A). However, when the wild-type protein was incubated under non-reducing conditions (50 μM DTT), and treated with -ethylmaleimide (NEM) to alkylate-free cysteine residues prior to electrophoresis, a dimer formed (lane 8). C57A and R172A were almost completely impaired for dimer formation, while D175A was significantly dimerization defective. R37A and Q80A behaved like wild-type, as did a double mutant R37A/Q80A (not shown). Dimerization under native conditions was confirmed by gel filtration. The wild-type TAD was monomeric under reducing conditions, but a dimeric peak formed under non-reducing conditions (B top left and right). C57A, R172A and D175A mutant proteins eluted as a monomer under reducing or non-reducing conditions, except D175A where a small proportion of dimer was evident under non-reducing conditions, as in A, lane 11. The dependence of stable dimerization on C57 cross-linking is consistent with the paucity of direct hydrogen bonds at the dimer interface (B). Furthermore, there are other cysteine residues on the surface of the protein that have their side chains exposed, such as C5 and C160. However, these two cysteines do not result in TAD cross-linking under non-reducing conditions. These data therefore support the crystallographic data demonstrating the importance of Cys57 in domain N1 and residues R172 and D175 in N2 for specific dimer formation. To confirm redox-dependent dimerization in full length E2, E2 and mutants were purified by ion-exchange chromatography followed by gel filtration in 10 or 1 mM DTT and assayed for dimerization by SDS–PAGE after NEM treatment (C). With 10 mM DTT, wild-type E2 was monomeric under denaturing conditions (lane 1). However, in the presence of 1 mM DTT a significant proportion of the protein formed cross-linked dimers (lane 2), that were dependent on Cys57 (lane 3). E2 R172A and D175A were also defective for redox-dependent dimerization (lanes 4 and 5), but not the double mutant R37A/Q80A (lane 6). E2 forms stable dimers via the C-terminal DBD. The TAD interaction could occur either within a pre-formed dimer or between two E2 dimers, resulting in native molecular masses of ∼90.8 and 181.6 kDa, respectively, as illustrated in D. When glycerol gradient sedimentation followed by SDS–PAGE was performed in the presence of 1 mM DTT, E, we only observed protein peaks between the 67 and 158 kDa markers, and an overall sedimentation profile that was similar to the protein assayed in 10 mM DTT (data not shown). The sedimentation profiles of the reduced and oxidized forms of E2 were almost coincidental, with native masses of 84 and 94 kDa, respectively. This difference is inconsistent with tetramer formation, demonstrating that the TAD–TAD interaction is intra-dimeric. Comparison of the TAD dimer with the structure of the HPV E1–E2 complex (), predicts that TAD dimerization would interfere with the E1–E2 interaction for two reasons, as illustrated in A. First, TAD dimerization buries more than half of the protein surface area used in the E1–E2 interaction. Secondly, E2 TAD dimerization would prevent its interaction with E1, as E1 occupies the same space as residue segments 4–12 and 30–81 of the second E2 subunit within the E2 TAD dimer, as shown on A. Here, the TAD subunit in blue is interacting with E1, while the subunit in red is seen to overlap with E1 (yellow surface). To test this we purified, in the presence of 1 mM DTT, E2 and E2 C57A that were tagged with a protein kinase A (PKA) recognition sequence for P-labelling and used them in GST-pulldown assays. B lane 1 shows input wild-type protein (E2s-sE2 dimer and E2 indicated). Only the non-cross-linked form of E2 bound GST-E1 immobilized on glutathione Sepharose (lane 2), while the C57A mutant retained full functionality in the E1 binding assay (lane 5). The proteins did not bind to GST alone (lanes 3 and 6). The BPV E1–E2 protein–protein interaction can therefore be effectively redox regulated. When binding of E2 to a high-affinity site was analysed by gel-shift assay, after equilibrating proteins in high (10 mM) or low (0.6 mM) DTT (A), ∼70% of the complex that formed with wild-type E2 moved with increased mobility in the gel (E2-DNA, lanes 5–7), compared to high DTT conditions (lanes 2–4). This form likely represents the internally cross-linked species, as it did not form with C57A, R172A (lanes 8–13) or D175A (not shown). Since the difference in charge and mass between the E2-DNA and E2-DNA species is minimal, a probable explanation for the difference in gel-mobility is a change in molecular shape. This is consistent with the observation that the E2 protein sediments at a slightly slower rate than the reduced species in glycerol gradient sedimentation experiments (E), as would be predicted for a more compact spherical structure. Probe binding was also reduced by 30–40%, since E2 DNA binding is itself redox sensitive (). We next asked if the oxidative effects on E2 DNA binding are reversible. As shown in B, when E2 was pre-incubated at very low DTT concentration (<25 μM), DNA binding was abolished (lane 3 compared to 2, 10 mM DTT), but could be restored by supplementing binding reactions with increasing concentrations of DTT (lanes 4–7). DNA-binding activity was first recovered in the form of the cross-linked TAD dimer (E2-DNA, not seen with C57A, lanes 10–13), which was converted to the non-cross-linked form at higher DTT concentrations. Together with the results in A, these experiments demonstrate that Cys57 is a reversible and more reactive redox centre than C340 in the DBD (). E1E2– formation was then analysed by gel-shift using an probe (A), with the E1 binding site and E2 BS12, under reducing conditions (10 mM DTT) and low DTT conditions (0.6 mM). E2 DNA binding is itself redox sensitive and DNA binding is abolished at very low DTT concentrations. At 0.6 mM DTT, sufficient DNA-binding activity is retained, while ∼70% of the observed E2 binding activity is in the form of the fast migrating E2-DNA species, as described earlier (A and B). With 10 mM DTT wild-type E2 bound DNA (A, lane 2). When the concentration of E1 was increased, the E1E2– complex formed and the amount of detectable E2–DNA complex diminished (compare lanes 2–5). The minor species migrating between the E2 and E1E2– band shifts most probably correspond to intermediates in the E1–E2 assembly. With low DTT concentrations (0.6 mM), consistent with the results described in A, a mixture of complexes corresponding to the oxidized and reduced forms of E2 formed (E2–DNA and E2–DNA, lane 6), but only the non-cross-linked form (E2–DNA) was observed with C57A (compare lanes 6 and 10). However, in the presence of E1 at low DTT concentrations significant differences were observed between wild-type E2 and C57A: with wild-type E2 formation of E1E2– was reduced compared to C57A (lanes 7–9, compared to 11–13), while the E2-DNA-binding activity appeared similar for both proteins (lane 6 compared to 10). When the products of E2-DNA binding in these reactions are also compared, it is clear that for the wild-type protein (0.6 mM DTT) the amount of E2DNA complex that forms does not change when E1 is added (lanes 7–9 compared to 11–13), but the amount of non-cross-linked complex (E2-DNA) diminishes. Comparing the ratios of the E2 to E1E2– band-shifts indicate that conditions that favour E2–DNA complex formation are inhibitory for E1E2– formation. Furthermore, E2 C57A assayed in 0.6 mM DTT was as competent as the wild-type protein assayed at high DTT concentration for E1E2– formation (lanes 3–5 compared to 11–13), suggesting that the redox-dependent change in E2-DNA-binding affinity has a minimal effect on E1 recruitment to . This is consistent with the E2 TAD–E1 HD interaction being a major component of cooperative E1–E2 DNA binding (). E1 does not bind the probe without E2 (lane 14), and E1 binding alone is relatively insensitive to DTT concentration over the range tested (10 to 0.05 mM, see later). We next asked if E1E2– formation could be restored after complete oxidation of the proteins. The results shown in B demonstrate that pre-incubation in 25 μM DTT abolishes E2-DNA binding (lane 6 compared to 3, as demonstrated in B) and E1E2– complex formation (lane 7 compared to 4). However, supplementing binding reactions with increasing concentrations of DTT restored E1E2– complex formation (lanes 8–10), indicating that the proteins are not intrinsically susceptible to irreversible oxidative inactivation. Our biochemical studies therefore demonstrated that C57-dependent oxidative dimerization of the E2 TAD, within a stable E2 dimer otherwise dimerized by the tight DBD interaction (), can regulate E1 recruitment to , the first event in replication initiation. In BPV a second high-affinity E2 binding site (E2 BS11) is positioned 33 bp upstream of the E1–E2 BS12 binding site arrangement, as depicted in A. A distal E2 site alone can drive replication in transient assays (), from an E1E2–-like complex (). However, its most likely role in viral replication is to promote formation of a replication active -melting complex from the primary E1– complex derived from E1E2– (). In the experiments described earlier in A, a direct assessment of the effects of TAD cross-linking on E1E2– pre-initiator complex formation is complicated by the redox-sensitive DNA-binding component of E2. However, it has been demonstrated that when E2 functions from a distal position, there is no requirement for the E2 DBD, in marked contrast to initiator complex assembly from proximal BS12 where there is an obligatory requirement for a specific E1 DBD–E2 DBD interaction (). In the former case, targeting of the E2 TAD to DNA can be achieved with a heterologous DBD, and a chimaeric E2 protein where the E2 DBD is replaced with that of the yeast transcription factor GCN4 (GCN4E2) is active in transient replication assays (). Since DNA binding of GCN4 is not redox-sensitive, we were able to test directly the effect of oxidative TAD dimerization on formation of an active E1– initiator complex. Accordingly, reactions were assembled with purified recombinant E1, E2GCN4 and a template with a distal GCN4 binding site, with the ATP (5 mM) required for formation of an E1– DNA melting complex. Stimulation of E1– complex formation by GCN4E2 proteins was measured by gel-shift analysis, and -DNA melting with the potassium permanganate footprinting assay. In A, purified GCN4E2 (GE2) and the C57A mutant (GE2C57A) were assayed in parallel for E1– complex formation and DNA melting under reducing conditions (2 mM DTT). The results of the gel-shift assay are shown above the results of the potassium permanganate assay; the lane numbers in each case correspond to the same reaction. In the gel-shift shown earlier (A), GCN4E2 and the C57A mutant demonstrated little difference in their ability to promote E1– complex formation at low E1 concentration (lanes 5–8 and 10–13 compared to lane 4 with E1 alone). Likewise, in the potassium permanganate assay shown below, both proteins promoted -melting to similar extents, over the A/T-rich region and in the proximal E2 BS (lanes 5–8 and 10–13 compared to lane 4). However, we note that high concentrations of the chimaeric proteins appeared inhibitory for melting, which is not normally observed with wild-type E2 (,). At low DTT concentrations (0.05 mM), B, wild-type GCN4E2 was significantly impaired in its ability to stimulate E1 complex formation (lanes 5–8 and 10–13 compared to 4), while the C57A mutant functioned like the proteins assayed at 2 mM DTT. The impaired ability of wild-type to stimulate E1– complex formation was reflected in the potassium permanganate melting assay, shown below. Here, the extent of -melting promoted by the highest concentrations of GCN4E2 was comparable to those observed at the lowest concentrations of the C57A mutant (lanes 8 and 10). Oxidative TAD dimerization therefore impinges on all levels of E2-dependent replication initiation complex assembly. We have characterized structurally and functionally a dimeric form of the BPV E2 transactivation domain. A remarkable feature of this novel BPV E2 TAD dimer is that it is stabilized by only a few direct inter-subunit hydrogen-bonding interactions and a salt bridge resulting in a relatively loose contact. However, under non-reducing conditions the two subunits covalently cross-link by a reversible disulphide bond, thus stabilizing the dimer interface. In a pre-formed E2 dimer stabilized via the tight interaction of the DBDs, two TAD domains linked via flexible hinges are always in close proximity. The flexible or ‘open’ nature of the TAD interface would allow the interaction with regulatory partners, but upon disulphide bond formation a ‘closed’ dimeric form with reduced surface area for contact would result. This type of post-translational modification is reminiscent of the modes of regulation of the redox sensors Yap1 () and OxyR () that function in micro-organisms to regulate the response to environmental redox changes. This structure therefore has the hallmarks of a regulatory sulphydryl switch. It is noteworthy that the BPV TAD dimer is distinctly different in its contacts from those seen in the HPV 16 E2 TAD (), and that in HPV 11 the species is apparently monomeric (). However, at the amino-acid level TAD sequences exhibit a great degree of variability (), with only 32, 33 and 36% sequence identity between BPV E2 TAD and the corresponding domains in HPV18, HPV16 and HPV 11, respectively (). Examination of the amino-acid sequences of E2 proteins from various PVs does show that a cysteine at position 57 is found in BPV type 1 and 2 that cause skin warts in cattle. Other papillomaviruses with known sequence do not have a cysteine in the equivalent position, indicating that the disulphide bond observed in BPV will not form and that replication control will not be achieved using an identical redox sensor to the one we describe. However, it is possible that similar dimers can form in other PVs but are stabilized by different post-translational events. It is also possible that the varying types of oligomerization observed in PV E2 structures reflect the somewhat different regulatory roles played by E2 in the various viral sub-groups. For example, the arrangement of E1 and E2 binding sites in a tandem array at (A) is unique to the fibropapillomaviruses, and the significance of the E1E2– complex that therefore forms is not clear. These viruses also generate distinctive lesions that have a fibroblastic as well as epithelial component, and are large with a high viral burden characteristic of vigorous replication. The Cys57-induced dimerization of E2 in BPV may thus reflect host-specific adaptation of the virus selected by evolution against other modifications. Our findings suggest that oxidative TAD dimerization could have a role in controlling BPV replication initiation by masking and unmasking a binding surface for interaction with the initiator protein E1. E2 may act as a redox sensor where determinants in the TAD and DBD regulate initiator complex assembly, as modelled in . In , A and we demonstrate that disulphide bond formation abolishes the E1–E2 protein–protein interaction and replication initiation complex formation , consistent with the structural analysis of the E1E2 complex (). Furthermore, we demonstrate in that Cys57 is a more sensitive redox centre than Cys340, the accepted determinant of redox-dependent E2-DNA binding (). This therefore implies a significant role for TAD dimerization in regulating BPV E2 activity. The intracellular redox potential is known to fluctuate in the cell-cycle, with a pro-oxidative state being established at the G1/S boundary, which lasts until mitosis (,,). Furthermore, cell-cycle arrest during oxidative stress may be through a Cdc25C redox response (). It has therefore been hypothesized that critical cell-cycle events are regulated by redox switches (), and our data on BPV E1/E2 proteins provides experimental evidence indicating that replication initiation reactions may be under such control. S-phase replication of BPV occurs by a ‘random choice’ mechanism rather than each replicon replicating only once (). The redox mechanism of control is not necessarily inconsistent with the random choice mode, since E1–E2 dissociation occurs on initiation () and would be required before TAD cross-linking could occur. In the pro-oxidative environment of S-phase E2 TAD cross-linking could prevent further initiation until E2 reduction restores the E1–E2 interaction. This mechanism may function, with the assistance of the redox control of E2-DNA binding () and cyclin-dependent kinase activity (), as part of the licensing system that ensures the stable propagation of the BPV genome. Redox control may negatively regulate initiation, while cyclin activity provides an initiating signal confining replication to S-phase. Papillomavirus transcription and replication are also both tightly coupled to epithelial differentiation. A gradient of protein oxidation exists across the dermis, with the stratum corneum exposed to an oxidative environment (). In the outer layers of the dermis BPV switches from a Cairns-type to a rolling-circle mode of replication that may not require E2 (). Oxidative inactivation of E2 could be the mechanism whereby replication changes from an E1–E2 regulated to an E1-only unregulated mode that drives high copy number viral amplification. The process of characterizing redox switches directly in cultured cells is inherently complex and has thus far not been achieved in mammalian cells, where the redox status of cells is difficult to measure, control or correlate to the situation. A verification of our proposals therefore awaits the development of systems to achieve just this, and may reveal additional regulatory roles for E2 redox-dependent dimerization. However, our structure-function study with BPV should provoke renewed interest in this area, and may lead to the discovery of similar redox sensors that function in the mammalian cell-cycle or epithelial differentiation.

Nuclear receptors form a superfamily of ligand-inducible transcription factors that is characterized by two conserved domains, the DNA-binding domain (DBD) composed of two C4 type zinc fingers, and the ligand-binding domain (LBD), which also contains a dimerization interface (,). Nuclear receptors can bind DNA as homo- and/or heterodimers, and recognize response elements arranged as direct repeats, palindromes or inverted palindromes of conserved motifs (). Each motif is bound by the DBD of a single monomer, the two zinc fingers of the DBD combining into a single structural fold with a DNA recognition helix and variable dimerization interfaces (,). Consensus estrogen response elements (EREs, A), which are palindromes of PuGGTCA motifs with a three base-pair spacer (), are bound by estrogen receptors (ERs) with highest affinity . Perfect or imperfect EREs present at promoter-proximal locations () or, as revealed by genome-wide screens, at large distances from the transcriptional start site of estrogen-regulated genes (,), are bound by ERα and mediate regulation of estrogen target genes. Other steroid receptors, including the androgen receptor (AR) and glucocorticoid receptor (GR), also bind palindromes with a three base-pair spacer, but the repeated motifs are PuGNACA sequences (,). Non-steroid receptors also recognize PuGGTCA or related motifs, but these motifs are arranged as direct repeats or everted repeats with variable spacing. The affinity and selectivity of nuclear receptors for single PuGGTCA motifs is generally low, but can be increased by receptor-specific recognition of additional 5′ flanking bases (,). Thus, nuclear receptors have achieved selectivity in DNA recognition while interacting with only two main types of motifs. The fact that few variations have been observed in the base-contacting amino acids of the 48 human nuclear receptors () suggests that this type of protein–DNA recognition has been conserved throughout evolution, possibly because it affords the most favorable combination of affinity and selectivity. Interestingly, however, nuclear receptor homologs identified in offer a considerably wider variety of amino acid composition in the DNA recognition helix. Numerous mutations have also been described in the DBD of some nuclear receptors such as VDR and AR, but few changes in DNA-binding patterns have been reported for these mutant receptors (). The first clues to the molecular basis of specific DNA recognition by steroid receptors were provided by mutagenesis experiments of the ER and GR, which led to the identification of amino acids responsible for discrimination between the two types of recognition motifs bound by these receptors. Exchanging three amino acids in the ER DBD for the corresponding ones in the GR DBD resulted in a receptor mutant capable of transactivating glucocorticoid target promoters (). The converse experiment also demonstrated that amino acids at the same three positions (P box, B, underlined residues) in the GR were crucial for discriminating between glucocorticoid response elements (GREs) and EREs (,). In addition, a loop in the second zinc finger was found to be responsible for specific recognition of two motifs arranged as palindromes with 3 bp spacing (). Crystallographic analyses of complexes between the ER or GR DBDs and their response elements have uncovered the amino acid–base-pair interactions established by these two receptors (,). Two residues of the P box, V at the third position in the GR and E at the first position in the ER (E203), contact the central discriminating bases in the ERE and GRE motifs (C). In addition, an invariant K residue located further C-terminal in the DNA recognition helix (K210 in ERα) binds the two central bases on the opposite strand with respect to bases contacted by E203 in the ERα–ERE complex, but does not participate in contacts in the GR–GRE complex (C). Other bases common to the ERE and GRE are contacted by amino acids conserved in ER and GR (K206 and R211 in ERα). In addition, these interactions are buttressed by a complex network of amino acid–amino acid interactions and amino acid–phosphate interactions. The specificity of response motif recognition by steroid receptors is thus determined by a small number of specific interactions established by 3–4 amino acids. As a consequence, it may be expected that changing the identity of interacting amino acids in the DNA recognition helix would alter the selectivity of receptor–DNA interaction as can be achieved with artificial C2H2 zinc fingers (). A simple model for amino acid–base interactions within the structural framework of the steroid receptor DBD has been proposed (). This model relies on chemical rules for possible pairing of amino acid side chains and DNA bases through hydrogen bonding or hydrophobic interactions, and also incorporates stereochemical contraints specific to steroid receptors, based on the position of the DNA recognition helix with respect to the major groove of the DNA. Small, medium or large chains may thus be prefered depending on the position of the interacting amino acids in the DNA recognition helix with respect to the bases (see D for possible interactions involving amino acid at position ). Study of a spontaneous mutation in the first amino acid of the P box of the AR has revealed changes in DNA-binding specificity compatible with the predictions of this model (). Replacement of G at the first P box position in the AR DBD by R resulted in a mutant that could only bind a subset of the PuGNACA motifs normally bound by this receptor, and this according to the chemical preferences of the R residue with respect to the base at the third position of the motif. However, it remains unclear whether novel types of DNA specificities can be achieved through rational design of nuclear receptor mutants, especially in the case of the ERs, which appear to have more complex determinants of motif recognition than other steroid receptors. The purpose of this study was to examine whether mutating ERα residues that interact with the two central base pairs of EREs either separately or in pairs could generate artificial nuclear receptor DBDs with novel DNA-binding specificities. The bacterial expression vector pET3.1-HE81 containing the ERα DBD has been described previously (). The bacterial expression vector pET3.1-ERβ-DBD was constructed by PCR amplification of a cDNA fragment corresponding to amino acids 140–246 of ERβ and subcloning between the KpnI and XhoI sites of a pET3.1, a pET3 (Novagen, San Diego, CA, USA) derivative modified by insertion of a linker containing the KpnI and XhoI sites (). All ERα DBDs mutants were constructed by insertion between the KpnI and XhoI sites of the pET3.1 vector of a fragment obtained by site-directed mutagenesis of pET3.1-HE81 using PCR amplification. The wild-type ERα and ERβ expression vectors pSG5-HEG0, pCMV-SPORT-ERβ have been described previously (,). Expression vectors for ERαE203A, ERαE203N, ERαE203H, ERαE203R, ERαK210A, ERαE203A-K210A, ERαE203N-K210A, ERαE203H-K210A, ERαE203R-K210A have been constructed by substitution of the fragment between the KpnI and XhoI sites of pSG1-HE80 () by a fragment containing the mutation excised from the corresponding pET3.1 vector. The tk-CAT reporter plasmids containing one copy of the consensus ERE or of palindromes containing base replaments (pAT-tkCAT, pCT-tkCAT) were derived from the pBL-CAT8+ reporter vecteur () by insertion of double stranded oligonucleotides containing the response elements flanked with BamHI-BglII sites at the BamHI site upstream of the thymidine kinase promoter. The TATA-CAT reporter vectors containing two copies of the consensus ERE or of the CT palindrome were prepared by substitution of the three EREs in pERE3-TATA-CAT () by double-stranded oligonucleotides containing two tandem response elements and flanked by BamHI and BglII sites. BL21 DE3 cells were transformed with pET3.1 expression vectors containing the cDNAs for the DBDs of ERα, or of mutants of ERα or ERβ and expression was induced in exponentially growing cultures by IPTG (0.5 mM final) for 2 h. Whole bacterial extracts were prepared by sonication in extraction buffer (Tris-HCl pH 7.4, 25 mM; EDTA pH 8.0, 0.1 mM; NaCl 400 mM; glycerol 10%; DTT 1 mM; PMSF 10 mM and protease inhibitors) followed by centrifugation (at 10 000 for 15 min) of lysates. To determine the levels of expression of ER DBDs, aliquots (1 ml) were taken from each culture before induction with IPTG. Bacteria were isolated by centrifugation and resuspended in M9 medium containing each amino acid except methionine and cysteine (0.01% weight/volume each). Rifampicin was added (200 μg/ml final) to inhibit bacterial RNA polymerase and expression of the T7 polymerase was induced with IPTG (0.5 mM final) for 30 min. [S]-methionine was then added and bacteria were further incubated at 37°C for 5 min, collected by centrifugation, resuspended in Laemmli buffer and boiled for 5 min. Labeled proteins were analyzed by electrophoresis on 12% polyacrylamide–SDS gel and visualized by fluorography. HeLa cells were maintained in DMEM supplemented with 5% fetal bovine serum (FBS). Cells were switched 3 days before transient transfection to medium without phenol red containing 5% FBS pretreated with activated charcoal to remove traces of hormones. For gel shift assays, cells were transiently transfected with ER expression vectors (5 μg completed to 15 μg with carrier DNA in 10 cm dish) using the calcium phosphate method (). For CAT assays, HeLa cells were electroporated (10 cells, 0.24 kV, 950 μF in a Bio-Rad Gene Pulser II apparatus) with varying amounts of expression vectors for wt ERα or for different ERα mutants, 4 μg of pCMV-βGal, and 4 μg of tk-CAT reporter vectors containing single copies of different palindromic response elements. DNA mixes were completed to 80 μg with salmon sperm DNA in a final volume of 100 μl of NaCl 210 mM. Electroporated cells were plated in duplicate for parallel immunoblot and CAT assays. Two days post-transfection, HeLa cells were treated for 1 h with 25 nM estradiol and whole cell extracts were prepared by three cycles of freeze-thawing in extraction buffer (Tris HCl pH 7.6, 20 mM; glycerol 20%; KCl 500 mM; DTT 1 mM; EDTA 0.1 mM; PMSF 10 mM and protease inhibitors) followed by centrifugation for 10 min at 13 000 . Cell extracts were diluted to a final KCl concentration of 125 mM. Samples were preincubated with 2 μg poly(dIdC) for 15 min on ice before addition of [P]-labeled double-stranded oligonucleotide probes (300 000 c.p.m. per sample). The consensus ERE used for gel retardation assays is derived from the Xenopus vitellogenin A2 gene. Reactions were incubated at 25°C for 15 min then loading buffer (0.1% bromophenol blue, 60% glycerol) was added (1/5 V/V). Complexes were separated by electrophoresis on 5% polyacrylamide gels in 0.25× TBE (45 mM Tris–HCl, 45 mM boric acid and 1 mM EDTA) and visualized by autoradiography. Amounts of bound and free probe were quantified using a Phosphorimaging screen and the Quantity One software from Bio-Rad. Immediately after seeding, cells were treated with estradiol (25 nM) or vehicle (EtOH) for 40 h. Cells were then harvested and extracts were prepared by three cycles of freeze–thawing in CAT extraction buffer (Tris HCl pH8.0, 0.25 M; DTT 1 mM and protease inhibitors). CAT assays were performed and standardized over β-galactosidase activity as described (). Protein concentrations in whole cell extracts prepared for gel shift or CAT assays were estimated using a Bradford assay. Proteins (50 μg) in Laemmli buffer were separated by electrophoresis on 8% polyacrylamide–SDS gels and transferred onto a PVDF membrane. Blots were incubated in blocking solution (PBS 1X, Tween 20 0.05%, BSA 3%) for 1 h and probed with anti-ERα mouse monoclonal antibody B10 (obtained from Prof. P. Chambon) at dilution 1:5000. Membranes were then washed and incubated with a horseradish peroxidase-coupled secondary antibody and visualized with an ECL detection kit. Modeling was performed interactively, using the InsightII/Discover package (Version 2000, Accelrys Inc., San Diego, CA, USA). The X-ray structure of the ER DBD bound to DNA () was used as a starting conformation. Each model was submitted to unrestrained energy minimization using the AMBER forcefield () until an energy minimum was reached. The presence or absence of particular pair-wise amino acid–base interactions in the final structure was treated as a possibility or impossibility to form a particular interaction in a given structural context. Distance measurements between atoms were performed with InsightII tools using a Silicon Graphics O2 computer. Estrogen receptors bind with high affinity the consensus palindromic EREs consisting of two PuGGTCA motifs separated by a three base-pair spacer (A). Although natural response elements are often imperfect palindromes (,), base-pair replacements usually result in a loss of affinity (,). To verify that binding patterns are mainly derived from the DBDs of ERs, we have compared the effect of symmetric substitutions at each position of the consensus ERE on binding by full-length ERs transiently expressed in HeLa cells (ERα and ERβ, A and B, respectively) or by the ERα or ERβ DBD (ERα DBD and ERβDBD, C and D, respectively). The wild-type ERE was bound by either the full-length ERα receptor or the corresponding DBD with the highest affinity (note the higher degree of free probe depletion with wt ERE, A and C, lanes 19). Consistent with our previous observations that the DBDs of steroid receptors are sufficient to discriminate between EREs and GREs (), ER DBDs selectively bound the array of probes in a pattern similar to that of the full-length receptors, although all probes were less efficiently bound with isolated DBDs. This is consistent with a loss of affinity, but not specificity, resulting from the absence of the strong dimerization interface in the LBD (). In addition, all replacements introduced in both arms of the ERE reduced binding to the same extent for ERα and ERβ (compare panels A and B, and C and D), which share a high degree of conservation in their DBDs [90% in region C as defined in ()]. Although the expression levels of transiently expressed full-length ERα and ERβ could not be compared, note that the DBDs were expressed to similar levels as assessed by [S]Met incorporation (data not shown). The DNA recognition helix formed by the C-terminal part of the first zinc finger of ERα contains several basic amino acids involved in contacts with bases in the target motifs (B, residues in bold; underlined residues are part of the P box). Nucleotides G (−5) and G (+2) interact with residues at position 5 (K206) and 10 (R211) of the DNA recognition helix, respectively (C). In addition, K210 (position 9 in the DNA recognition helix) interacts with both G (−4) and T (−3) through direct and water-mediated contacts (C). Thus, basic residues are involved in recognition of all positions in the ERE except −/+6 and −/+1, which are not directly contacted in the crystal structure, although both display a preference for purines ( A–D, lanes 1–3 and 16–18). These direct interactions involving basic amino acids conform to general chemical rules, with G and T, which present only negatively charged groups in the major groove, being preferred over A, which presents both a positively and a negatively charged group, and C, which contains only a positively charged group, leading to unfavorable electrostatic interactions. Accordingly, replacement of G by C in a single motif at position −4 or −5 was sufficient to abolish binding (A, lanes 5 and 8). Of note however, the order of the preferred bases is not identical for each contacting basic amino acid. Most noticeably, replacement in a single motif of G at position +2 by T was sufficient to abolish binding (A, lanes 13). Molecular modeling suggests that lack of binding to T at position +2 results from steric hindrance due to the methyl group of T + 2, which prevents productive interaction between the amines of R211 and O4 of T + 2 (B). The pattern of recognition of bound response elements carrying replacements at positions −4 (G > A > T > C) and −3 (T > G > A > C) also differed from that predicted from charge preference due to interaction with basic amino acid K210 (G > T > A > C). Because E203 interacts with bases at positions +3 and +4, i.e. on the other DNA strand, in the crystal structure (), we compared the roles of both K210 and E203 in modulating patterns of base recognition. While E203 has an opposite and complementary type of chemical selectivity for bases (C > A > T > G) compared to K210, the complex mode of recognition of the two central base pairs by two amino acids may be expected to result in increased steric constraints. At position −4, replacement in a single motif by T drastically reduced binding (A, lane 9) and replacement in both motifs of the ERE completely eliminated complex formation (A–D, lanes 9) and transactivation [data not shown, see also ()]. This contrasts with the capacity of other nuclear receptors such as RAR, RXR and VDR to bind to PuGTTCA motifs [(,); see also () for a review]. Energy minimization indicates that movement of K210 to avoid the methyl group of T-4 prevents interaction with DNA (C). In addition, E203 is not capable of interacting with the amino group of A at position +4 (not shown). At position −3, G was less favorable than T (compare lanes 12 and 19 in A and A–D), contrary to what is observed at the other positions recognized by basic residues (A). Modeling indicates that C + 3 would create packing problems with E203, preventing interaction of this amino acid with C + 4, and K210 is too distant to interact with G−3 (not shown). Thus, our experimental and modeling data suggest that both E203 and K210 contribute to the selectivity of response element recognition with respect to the two central bases of the ERE. To better analyze the respective roles of E203 and K210 in determining the specificity and affinity of ER interaction with response elements, we further characterized the ER-binding specificity with respect to the two central base pairs of the palindrome motifs using a panel of 16 probes representing all combinations of these central bases introduced in both arms of the palindrome. Full-length ERs and isolated DBDs bound with high affinity only to the GT (response elements are designated by bases found on the minus strand at position −4/−3) combination found in the wt ERE and, with lower affinity, to the GG element (A–D). Note that complexes formed on element CC with full-length receptors (lane 6) were also observed with extracts from cells transfected with the parental expression vector, and that the smears observed in some lanes with bacterially expressed ER DBDs also appeared with extracts from bacteria transformed with the parental expression vector (not shown). The very high selectivity of ERs for the two central base pairs in their response elements differs from that of the GR or AR, whose DBDs could both recognize all four PuGNACA-containing palindromes [() and data not shown]. Crystallographic analysis of the GR DBD indicates that contacts are established only with the base at position +3 through a V residue (position 6 in the DNA-binding helix). E203 is replaced by a G residue in GR and AR, and the K residue corresponding to K210 in ERα does not contribute to GRE recognition. The higher selectivity of ER interaction with response elements could therefore result either from the fact that E203 interacts with both adjacent +4/+3 bases, and/or from the contribution of K210 in recognizing the −4/−3 bases on the opposite strand of DNA. To investigate the respective roles of these two amino acids to the affinity and/or selectivity of binding, we replaced either E203 or K210 by A residues in ERα and examined complex formation with the panel of 16 probes corresponding to all possible variants at the two central positions of the repeated motifs. Both mutants still bound with highest efficiency to the consensus ERE (A, lanes 12), although the intensity of the retarded band was ∼5-to 10-fold lower than with the wt ERα (compare A, lanes 12 and 17; see also B). However, the K210A mutation was active in transactivation assays using a reporter gene containing the consensus ERE (C). Intriguingly, peak transcriptional activity was ∼60% higher for K210A than for the wt receptor, suggesting that K210 plays a negative role in transcriptional activation. Mutant E203A bound with reduced efficacy to the consensus ERE compared to the wt ERα (A and ), and formed a weak complex with the PuGATCA probe (AT, A, lane 4), which was not detected with the K210A mutation. Finally, the double mutant K210A–E203A bound very weakly to the consensus ERE probe, and also formed a detectable complex with the AT probe (A, lanes 4, 8, 12 and 16). Binding of E203A to the AT probe suggests that the main role in restricting binding to this element in the wt receptor is played by E203 rather than K210. The absence of high affinity complexes observed with the K210A mutant suggests that the role of K210 in the selectivity of response element recognition is redundant with the role played by E203. A simple DNA recognition model for steroid receptors has been proposed previously (), based on both the general rules of chemical compatibility between amino acids and base pairs, and stereochemical constraints due to the position of the DNA-binding helix across the major groove as derived from the crystal structures of the ER and GR DBDs (D for predicted interactions at amino acid position 2). Note that this model is based on a one amino acid–one base interaction relationship, and in particular does not take into account recognition of the two central bases by both K210 and E203 (C, dotted lines), considering only the role of E203 with C + 4 (C, bold line). Our finding that E203 and K210 cooperate for selective binding to GT palindromes incited us to test the predictions of this model for replacement of E203 by other amino acids. Replacement of E by N is predicted to switch recognition from C at position +4 to A (TN elements), and replacement by H or R should lead to specific recognition of a G at this position (CN elements, D). These mutations were introduced in ERα full length and in the isolated DBD. Similar expression levels were obtained for all full-length mutants and the wild-type receptor as determined by western analysis (data not shown). Whole cell extracts containing the mutant receptors E203N and E203H formed complexes mainly with the wt ERE (GT probe, A and B, lanes 13 and D and E, lanes 12). While binding to the wt ERE was not abolished, the patterns of probe recognition were altered with both mutants, which bound to the AT element (A and B, lanes 5 and D and E, lanes 4). These interactions are transcriptionally productive, as demonstrated by transient cotransfection of CAT reporter vectors containing the corresponding response elements cloned upstream of the thymidine kinase promoter with expression vectors for the different mutants (B, ERE-tk-CAT and AT-tk-CAT reporter vector). Titration curves were performed using varying concentrations of transfected plamids, and protein levels were measured by western analysis (A). Surprisingly, saturation was reached at identical protein concentrations for all receptors on either response element, indicating that the difference in efficiency of complex formation is not observable in our reporter assay (data not shown). This is possibly due to cooperative effects with other transcription factors bound to the promoter, or to chromatin organization. Peak transactivation levels were similar for the wt and mutant receptors on the consensus ERE, reflecting intact transcription activation properties for the mutants (B). On the other hand, mutant E203H, but not E203N, had ∼6-fold lower peak levels of transactivation on the AT element than on the consensus ERE (B; peak levels on the two response elements were obtained at the same protein concentration). The differential levels of maximal transcriptional activation by the E203H mutant on the two response elements may be related to the previously reported allosteric effect of the DNA sequence on ERα transcriptional activation properties (,). The observed DNA-binding specificity of these mutants do not correspond to predictions based on the proposed model, as no stable binding was observed with TN motifs for E203N (A, lanes 14–17), or CN motifs with E203H (B, lanes 6–9). Accordingly, the TT and CT elements did not confer estrogen responsiveness to the tk promoter in the presence of these mutants (data not shown). Note that no specific complexes were observed with the E203R mutant on any of the PuGNNCA probes (C and F). Lack of binding to the consensus ERE and lack of transactivation on an ERE-tk-CAT reporter (C) suggest that R at this position has a destabilizing effect that is stronger than the absence of side chain (A mutation, see ). Replacement of E203 by R in the structure of the ER–ERE complex reveals that the R side chain is too bulky to fit in the major groove of DNA, and that the amino groups exert repulsive effects with the positively charged groups of C + 4 and C + 5 and steric conflict with C + 4 (C and data not shown). In addition, neither binding to CN elements nor transactivation from reporter vectors containing the CT palindrome (C, CT-tk-CAT reporter vector) was obtained, indicating again that the predicted switch in specificity has not occured. These results indicate that a model based on a one amino acid—one base-pair relationship is not an accurate description of the interaction between the ER and its target response element, and suggest that K210 plays a role in modulating the DNA-binding specificity of receptors carrying mutations at position 203. Contrary to the total absence of binding observed with the E203R mutant of ERα, mutation of the corresponding amino acid in the AR (G577) to R led to a different pattern of response element binding than that of the wt receptor. While wt AR bound all four PuGNACA elements, the G577R mutant bound PuGGACA elements very weakly, but interacted with the three other PuGNACA elements, with a preference for the element containing a G at position +4 (PuGACA). In the context of the GR DBD, and presumably also the AR DBD, the K residue corresponding to K210 (position 9 in the DNA recognition helix) does not bind DNA, but is involved in an interaction with E at position 13 in the DNA recognition helix (B). Thus, we investigated whether mutation K210A may facilitate association of the E203 mutants to novel binding sites and/or reduce binding to the consensus ERE. The resulting E203N/K210A and E203H/K210A double mutants had a much reduced binding to this probe (), confirming the role of K210 in binding to the consensus element in the absence of a residue at position 203 recognizing C + 4. However, these mutants still did not form stable complexes with probes containing bases predicted to interact with the N or H residue at position 203 (TN, CN, respectively, see ). In addition to complexes with AT palindromes already observed with the single mutants, the strongest complexes detected with E203N/K210A and E203H/K210A double mutants were with the AC element (). On the other hand, the double mutant E203R-K210A gained binding to the CT element (), as could be predicted from chemical compatibility between R at position 203 and G + 4 (D). This mutant also bound to the AC element (). Thus, the specificity of base recognition expected from replacing the E203 residue by R was revealed in the presence of the K210A mutation, although an additional unpredicted complex was also formed with comparable efficiency. As the level of complex formation on the CT element was much lower than that of the wt receptor on the consensus ERE, we examined whether the interaction between the E203R-K210A mutant and the CT probe is transcriptionally productive. Whereas no estradiol-induced transcription could be detected with wt, K210A or E203R on a CT-tk-CAT reporter vector, the double mutant E203R-K210A gave rise to a detectable increase in estradiol-stimulated transcription (C). The double mutant, like the single mutant E203R, was not active on ERE-tk-CAT, confirming the switch in specificity. The complete switch in DNA-binding specificity of the double mutant was confirmed using reporter vectors containing tandem response elements. No transactivation of the reporter vector containing EREs was observed in the presence of the double mutant after estrogen treatment, while peak stimulation of the reporter vector containing CT elements was comparable to that of wt ERα on the reporter containing consensus EREs (D). Although C2H2 zinc fingers can be tailored to bind virtually any DNA sequence, nuclear receptors have not demonstrated similar flexibility. A possible reason for success in the rational design of artificial C2H2 zinc finger proteins is the modular structure of this type of DBDs, each finger recognizing three base pairs and multiple zinc fingers extending the length of bound DNA. On the other hand, steroid receptors bind DNA as dimers recognizing palindromic response elements. As dimerization contributes to the affinity of DNA binding, binding to non-palindromic sequences would likely be of low affinity. Nevertheless, the question remains whether steroid receptors can bind other DNA elements than their two natural cognate response elements. A previously proposed model for prediction of specific amino acid–base interactions by the steroid receptor DNA-binding helix () incorporates both chemical rules governing amino acid–base interactions and stereochemical constraints resulting from the position of the DNA recognition helix in the major groove of response element as determined by crystallographic studies. However, this model relies on one amino acid—one base-pair relationships and ignores some of the contacts described in the ER–ERE cocrystal structure. In particular, E203 recognizes not only the base at position +4 (C, bold line), but also that at position +3 (dotted line), and bases on the opposite strand make contacts with K210. On the other hand, the lysine residue corresponding to K210 in the context of the GR does not bind DNA, and its role in binding of ERs to EREs has not been experimentally confirmed. Contrary to the GR and AR, which recognize motifs with variable base composition at position −4 conforming to the PuGNACA consensus (), ERs display a high degree of specificity for bases at position −4/−3. GT was the only element bound with high affinity, while GG was tolerated when introduced in only one motif, such as in the ERE found in the promoter of the rabbit uteroglobin gene (), but reduced binding efficacy dramatically when introduced in both motifs. This binding pattern is consistent with chemical preferences of E203 for C or A at position +4 and +3, and of K210 for G or T at position −4 and −3, with restrictions imposed by steric constraints with either amino acids. Mutagenesis of each amino acid indicated that both contribute to recognition of the two central base pairs in terms of binding efficiency. E203 also plays a specific role in restricting binding to AT elements, since mutation E203A, but not K210A, allowed formation of complexes. Molecular modeling indicates that this is due to steric conflicts between the carbonyl group of E203 and the methyl group of thymine +4. The contribution of K210 to ERE binding likely explains the fact that E203N and E203H mutants still interacted with high affinity with the consensus response elements (GT). Interactions with the AT element, which were not predicted by the proposed model, were observed in gel shift and transactivation assays with both E203N and E203H mutants, and likely result from removal of negative constraints exerted by E203 on T + 4, as observed also in the E203A mutant. Our modeling indicates that the methyl group of thymine can be accomodated by side chain rearrangement when E203 is replaced by N or H (A and B). On the other hand, lack of binding of the E203R mutant to all tested elements and of transcriptional activity on the consensus ERE and CT element can be explained by conflicts in charge preference between R203 and K210 for recognition of the same base pairs. The role of K210 in preventing a specificity switch by mutations at position 203 is demonstrated by the observation that the double mutant E203R-K210A could bind the CT element, as predicted by base compatibility between R and G + 4, whereas this interaction was not observed with the single mutation E203R. As noted above, the K residue in AR at the corresponding position does not contact DNA, explaining the capacity of the AR mutant with an R at the position equivalent to 203 to interact with G + 4. Another difference between the two receptors is the contribution of V at position 6 in the DNA recognition helix of AR (B) in complex stabilization. This may explain the relatively low affinity of the E203R-K210A mutant for the CT element. However, this interaction was transcriptionally productive. Surprisingly, similar levels of transactivation were obtained with the mutant receptor on the CT element as with the wt receptor on the consensus element at comparable protein concentrations along a range of amounts of transfected plamids, failing to reveal a different efficiency of reporter DNA saturation (data not shown). Titration curves of mutant E203N and E203H on the consensus ERE and AT element also failed to reveal a shift in protein concentrations necessary to reach peak transcriptional activity. This indicates that our assay may not discriminate between binding sites in the range of affinities that is observable in gel shift assay (within 30-fold of wt ERα-consensus ERE affinity) and may be due to synergism with other transcription factors for binding to our reporter genes A higher stringency of versus assays is supported by the observation that half-palindromes, which are not bound in our gel shift assay, are enriched, albeit to a much lower degree than high-affinity EREs, in chromatin fragments bound by ERα in a genome-wide analysis (). On the other hand, our transactivation data confirms the total loss of interaction between the E203R-K210A and the consensus ERE and thus the switch in specificity toward the CT element. Binding of the E203R-K210A to some unpredicted sites, i.e the AC element, reflects the limitations of simple models in fully accounting for the complex interactions between these residues and the two central base pairs. In addition, the expected binding of the E203H-K210A mutant to the CN palindromes and of the E203N-K210A mutant to the TN elements were not observed. Replacement of E203 by N or H reveals that both residues are too short to interact with charged groups at position 6 or 7 of A (+4) or G (+4), respectively (A and B). Finally, mutants E203H-K210A or E203N-K210A still bound the consensus ERE (GT element), contrary to what was observed with mutation E203R-K210A. The total abrogation of binding to the consensus ERE appears due to charge and steric conflict of R with C + 4 (C). Together, our results indicate that simple chemical and stereochemical rules cannot predict accurately the changes in the selectivity of ER–DNA interactions induced by specific mutations in the two central base pairs. A clear limitation is the need to incorporate the contribution of several amino acids to recognition of the same base pairs, and the role of one amino acid in recognizing two adjacent bases. The combined effects of E203 and K210 in interacting with the same bases is apparent both at the level of charge and steric constraints, resulting in the tighter DNA-binding specificity for the two central base pairs observed for ER versus other steroid receptors. Further, steric constraints play an important role in preventing potential interactions. Additional experiments will also be necessary to determine whether the effect of amino acid replacement at other positions in the ERα DNA-binding helix, which are involved in simpler one residue–one base interactions, is more easily predictable. It remains possible that other receptors may be more amenable to the rational design of mutant receptors with altered DNA-binding specificity, due to differences in composition of the DNA-binding helix and/or in the mode of DBD dimerization. In this respect, it will be of interest to investigate the DNA-binding specificity of non-classical receptors, which contain widely diverging combinations of amino acids in their DNA recognition helix. Finally, combinatorial approaches as performed for C2H2 zinc fingers (,) may help clarify how amino acids of the DNA-binding helix cooperate toward the establishment of novel DNA-binding specificities.

RNA molecules fold into defined structures that are critical for their biological functions. During RNA evolution, the structure is much more conserved than the sequence (,). The sequence variations that contribute to differences between species are those that preserve the structure and function of the RNA molecule. An important model for studying RNA evolution is the ribosomal RNA (rRNA). The ribosome is a large complex of both RNA and protein, but it is the RNA component that catalyzes one of the most fundamental and most highly conserved biochemical activities: protein synthesis (). Some universally conserved regions of the rRNA might date back to the RNA world, a hypothetical stage of evolution in which RNA performed all major biochemical reactions (). In particular, the peptidtyl transferase center, which catalyzes peptide bond synthesis, has been independently recovered by artificial selection from random-sequence pools (), suggesting that it would have been relatively easy to ‘discover’ after the evolution of RNA (). The rRNA is present in all extant species and presumably dates back to the earliest forms of life. It thus reflects the evolutionary history of life itself, and can be used to establish the evolutionary relationships between all species on earth (). Because reconstruction of phylogeny depends on the evolutionary model that is assumed, it is important to understand how rRNA actually evolves. The most widely accepted model of rRNA evolution is a ‘rates across sites’ model, in which a multiple sequence alignment is used to assign rates of evolution to each position in the rRNA (). Secondary structure is expected to influence evolutionary rates primarily through compensatory mutations in stems. Because stems are assumed to be largely structural, any substitution of one base pair for another should typically be acceptable. In contrast, unpaired regions are thought to depend more specifically on their sequence. For example, tetraloops fall into only a few families (). This view was promoted by the paradoxical finding that most of the highly conserved regions, i.e. regions with no or small variability at the sequence level, in the bacterial small subunit (SSU) rRNA were in unpaired, rather than in paired, regions (). This finding suggested the then-revolutionary view that base pairing is a weak constraint on sequence compared to other influences on the sequence near the active site of the ribosome. This idea is further supported by two additional observations: it is often possible to experimentally swap one base pair for another while preserving function, and paired regions change faster than unpaired regions when the GC content of each region is plotted against total GC content (). The assumption that RNA evolution is composed predominantly of compensatory mutations in paired regions suggests that specific rate matrices should be used to describe paired regions for evolutionary studies. RNA violates the assumption of site independence that underlies many evolutionary models, because maintaining base pairing requires the bases at two interacting sites to change in a correlated fashion. Currently, many models of RNA evolution incorporate the nonindependence of sites in paired regions by allowing correlated mutations (, ), including noncanonical base-pair interactions represented as isostericity matrices (). The special treatment of paired regions is a more accurate model of RNA evolution than using a single four-state rate matrix. However, these models could potentially be refined further with detailed knowledge about the rates of change in different unpaired regions (hairpin loops, bulges, and multi-helix junctions) and in different taxonomic groups. Although the standard model of fast-evolving stems is widely accepted (,,), there are three good reasons to believe that the paired–unpaired dichotomy provides a limited view of RNA evolution. First, although many base pairs in many molecules can be changed experimentally without disrupting function, the same is true for unpaired regions. For example, replacing large or poorly structured loops with tetraloops is commonly performed to improve crystallization of RNAs [see for example ()]. Accordingly, it is unclear whether, on average, changes in stems can be tolerated more often than changes in unpaired regions. Second, the early observation that many highly conserved bases in rRNA are unpaired () need not imply that most unpaired bases in rRNA are highly conserved. For example, the conservation maps from the comparative RNA web site () show that 44 and 35% of the nucleotide positions in bacteria and eukaryotes, respectively (both large subunit (LSU) and SSU) are conserved in more than 98% of the sequences in the alignment. Of these more than 98% conserved positions, only 50–54% are unpaired. Because there are more paired positions than unpaired positions in the rRNA, on average about 50% of the unpaired positions and 30% of the paired positions are highly conserved (more than 98%). The other half of the unpaired positions are thus free to evolve at higher rates. (Note that only positions that are present in at least 95% of the sequences are counted, excluding about 8% of the positions in the bacterial model and about 30% in the eukaryotic model, and that differences in the definition of ‘highly conserved’ can change the figures substantially.) Third, we recently showed that even random sequences that have never been exposed to selection show different rates of change in the GC contents of paired and unpaired regions as the GC content of the whole molecule changes, suggesting that different bases have different intrinsic propensities for base pairing (). Consequently, the paradigm introduced by Muto and Osawa for detecting selection as a different response to changes to GC content in different parts of the molecule, which works well for coding regions (,), is not valid for rRNA. The aim of this article is to test the commonly accepted hypotheses that compensatory mutations in paired regions quantitatively dominate RNA evolution, and that the unpaired regions form a single category that can be treated as homogeneous. Specifically, we address the following questions: There are many ribosomal sequences and structural models available, allowing a detailed analysis of evolutionary rates. The first complete rRNA sequences for both the SSU and LSU were determined for shortly after the Sanger sequencing method became available (,). Today, a wealth of aligned sequence data is available. The European rRNA database (RDB) () and the comparative RNA web site (CRW) () provide alignments containing up to several hundred LSU sequences per phylogenetic domain (about 400 bacterial LSU sequences), and thousands of SSU sequences (about 12 000 bacterial SSU sequences in the RDB). Soon after the first full-length rRNA sequences were determined, the first covariation-based secondary structure models were developed (10,11,40–44). These models predicted the secondary structure in terms of Watson–Crick and G–U wobble base pairs. As the amount of sequence data has increased, the structural models have repeatedly been refined. Over time, they have matured into complex models that also incorporate non-standard base pairs and tertiary interactions (2,45–49). These models are available on the CRW (). The RDB () provides a similar, independently developed, set of structural models. These models were originally derived by comparing 14 SSU rRNA sequences and surveying existing structural models (), and have been successively refined (). In general, the available secondary structure models are of high quality. The bacterial secondary structure model is especially well established and is consistent with chemical experiments () and crystal structures of the ribosome (). The eukaryotic structural model has been accurately determined for the more conserved regions, but the structure of some of the variable regions is still disputed (). Thus far, there is no crystal structure to resolve these controversial regions. In this study, we used sequence and structure information from three sources: the European RDB (), the CRW (), and the RCSB Protein Data Bank (,). provides details about the model organisms, alignments, sequence accession numbers, and crystal structures used. RNA secondary structure is a collection of base pairs, interspersed with unpaired bases. Base pairs can either be nested or non-nested. Two base pairs, one between positions and and the other between positions ′ and ′ (where < < ′ < ′ and <′) are nested if either <′ < ′ < or < < ′ < ′. Pseudoknots are non-nested base pairs between a loop of one stem and residues outside that stem (). RNA secondary structures can be decomposed into distinct structural classes. A fully nested structure without pseudoknots can be represented as a tree, and thus each position can be classified into either stem, loop, bulge, junction, end or flexible (). In this study, we did not remove pseudoknots from the structural models, which required us to combine some of the structural classes for simplicity. We distinguished stem, loop, bulge and ‘junction/other’ [essentially the same as in ()]. The ‘junction/other’ category includes the categories junction, end and flexible in the fully nested structures, and pseudoknotted regions. Most bases in this class are from multi-helix junctions. We used two types of ‘variability maps.’ In these maps, variability is calculated from a large alignment of rRNA sequences (separated by phylogenetic domain and subunit) and superimposed onto a structural model. First, we used the RDB variability as calculated by the substitution rate calibration method (,), available from the European rRNA database (). Second, we used the CRW secondary structure conservation maps, provided on the comparative RNA web site (), where conservation is calculated based on the nucleotide distribution at a particular alignment position. The substitution rate calibration method classifies each position as one of six rate categories (seven in more recent publications). Sites that are absent in 75% or more of the sequences in the alignment are considered too variable to be classified and are excluded from the analysis. On the CRW conservation diagrams, only four rate categories are distinguished, ranging from more than 98% conserved to less than 80% conserved. For a base to be classified in one of the four categories, it has to be present in at least 95% of the sequences in the alignment. The CRW conservation method thus uses stricter requirements for classifying residues. The two different measures of conservation agree well (average = 0.824 when using a sliding window); see Supplementary and for the strength of association between the two measures. Both CRW conservation and RDB variability data are available for SSU/LSU as bacterial model and as eukaryotic model. The structural models from both sources are not exactly the same, but share on average 80.5% of their base pairs (see Supplementary Data). We used the variability maps to assign each position to a rate category. For the pairwise comparisons, we used the sequences, structural models and a high-quality alignment from the CRW web site (all downloaded in June 2006). Initially, we did a large-scale comparison within each phylogenetic domain and subunit, where we compared all sequences for which there was a structural model and an entry in the CRW alignment. When looking over the whole range of diversity, many positions were ignored because of conflicting structural information. Focusing on an individual lineage reduced the structural differences, because the species were more closely related and thus less structural changes had occurred since the time of divergence. Not all sequences for which a structural model was available had an exact match in the alignment. For these groups, we aligned the sequences with MUSCLE () and inserted the gaps into the corresponding structural classifications. These alignments were of high quality, because the species were closely related (see Supplementary Data). Since there were at most very small differences between the data calculated from the CRW alignment or from the MUSCLE alignment, we reported the results for the largest data set. We performed structural calculations using the PDB files 1GIX and 1GIY, corresponding to the crystal structures of the ribosomal subunits from solved at 5.5 Å. These files provide the 3D coordinates for the phosphorus atom in each residue. It has previously been shown that the average variability of residues increases with distance from the center of the ribosome (), but it is unclear that the geometric center is the correct reference point. We calculated distances between the P atom of each residue and the following locations: the distance from the peptidyl transferase center (‘PTC’), defined as the P atom of residue A2451; the distance from the tRNA path (‘path’), defined as the distance from the closest P atom of any of the three tRNAs or any of the two mRNA codons included in the crystal structure; the distance from the closest protein (‘protein’), defined as the distance from the closest C-α atom of any protein attached to the same subunit; the distance from the subunit interface (‘interface’), defined as the distance from the closest P atom in the other subunit, as well as the distance from the center (‘center’), defined as the distance from average of the coordinates of all P atoms in the ribosome (including SSU, LSU and 5S rRNA). We then correlated each of these distances for each residue with evolutionary rate and structural category, as calculated above. f i n d t h a t s t r u c t u r a l c a t e g o r i e s i n t h e r i b o s o m a l R N A e v o l v e a t d i f f e r e n t r a t e s , a n d t h a t t h e s e r a t e s v a r y a c r o s s p h y l o g e n e t i c d o m a i n s . A l t h o u g h i t i s t r u e t h a t h i g h l y c o n s e r v e d r e g i o n s t e n d t o b e u n p a i r e d , t h e c o n v e r s e , t h a t u n p a i r e d r e g i o n s a r e m o r e c o n s e r v e d , i s n o t a l w a y s t r u e ( a l t h o u g h i t i s w i d e l y a s s u m e d ) . We have demonstrated that different structural elements change at different rates in different lineages. In bacteria and archaea, we observe the generally accepted pattern of fast-evolving stems. However, this pattern differs markedly in eukaryotes, where hairpin loops actually evolve considerably faster than stems do. This result is not primarily due to insertions and deletions in non-conserved surface loops in the eukaryotes, because it persists when these regions are excluded from the analysis. The different types of unpaired regions always behave differently from one another, underscoring the importance of moving beyond the paired–unpaired dichotomy in studies of evolutionary rates in rRNA. To minimize the effects of errors in the structural models, the alignments, and the rate inference procedure, we used several complementary methods that agreed well with one another. The general trends we identified are supported by existing conservation maps and secondary structure models calculated by two different research groups (RDB and CRW), and by direct inference of the amount of change in each structural category from pairs of sequences. We verified that the choice of whether to include or exclude gaps in calculations of evolutionary distance, and use of either automated MUSCLE alignments or hand-curated alignments from CRW, produced similar results. No matter which metric is used to measure the substitution rates, hairpin loops evolve substantially faster than stems in the eukaryotic lineage, and these results hold both over short and long evolutionary distances. There is a small but significant effect of the distribution of structural elements throughout the ribosome: for example, the region around the PTC is largely made up of junctions, whereas the subunit interface and the regions near proteins (subject to the limits of the 5.5 Å resolution of the crystal structure) are largely made up of stems. We believe that differences in evolutionary rate between structural categories are not due to these differences in distribution because we can calculate the distribution of rates in each structural category that would be expected if distance from functionally important regions were the only factor, and these distributions of rates do not match. Thus, the differences in rates are likely to be meaningful and are not simply an artifact of the composition of the most conserved regions. The distribution of structural categories in the ribosome was influenced more strongly by proximity to defined structural features, such as the PTC and the tRNA path, than by proximity to the geometric center. These results suggest that the factors driving the distribution of structural elements within the ribosome are primarily adaptive rather than consequences of, say, the physics of helix packing. However, the results contrast strikingly with proteins, in which hydrophobic residues preferentially assort themselves into the core of the molecule. Thus, secondary structure (and, presumably, nucleotide composition) is likely to be a poor guide to predicting whether a particular region of the rRNA is buried or surface exposed. Relative rates of evolution of different structural categories, especially the ratio of changes in stems to loops, differ drastically in different lineages. These results suggest that the influence of each structural category on the rate of evolution is not universally consistent, diminishing the plausibility of using differences in rates in different regions to infer properties of the secondary structure. However, the results do suggest that models of rRNA evolution that are specific to particular lineages will be important for making the best alignments and phylogenies. For example, the knowledge that loops evolve rapidly in eukaryotes would lead us to give changes in these regions of the sequence less weight for phylogenetic inference. With the vast number of sequences now flooding the databases (∼300 000 SSU sequences deposited in the Ribosomal Database Project as of this writing, and pyrosequencing able to produce 100 000–300 000 sequence fragments in a single 4-h run), detailed models of specific groups of organisms will become increasingly feasible. Outliers from the general pattern of rRNA evolution may suggest interesting biology. For example, the Actinobacteria appear to resemble the eukaryotic pattern more than the general bacterial pattern. It is possible that ecological factors such as multicellularity, or molecular features such as linear rather than circular chromosomes, in this lineage () cause them to resemble eukaryotes more than other bacteria in factors influencing rRNA evolution. This group has relatively high GC content, contrasting with the low GC content in eukaryotes overall, suggesting that differences in base composition are not the main factor. Similarly, in the archeal SSU, about 5% of the comparisons (when counting only point mutations) or 10% (when adding indels), do not support the general conclusion that stems evolve faster on average outside the eukaryotes. Loops evolve faster than stems in comparisons between and , , or , and between and . and are very similar in this respect. is thought to be among the deepest diverging aerobic archaea, which may suggest some convergence with the eukaryotic pattern. This work has several implications for future analyses. For example, when constructing phylogenetic trees, different models of RNA evolution should be adopted (provided that sufficient sequences are available to infer the parameters robustly). These models should be both specific for structural categories, including treating the different types of unpaired regions separately; they should also be specific for particular phylogenetic groups. For example, the general substitution model for bacteria does not fit the Actinobacteria well. Similarly, methods for comparing microbial communities, such as (), are based on diversity in an rRNA alignment. These methods may be improved by adding masks that weight more or less variable regions differently. Weighting by structural category may be an important first step for relatively unconserved regions. The differences in the distributions of different structural categories appear to be driven primarily by proximity to functional features in the ribosome, rather than assorting by geometric configuration such as the ribosome center. This observation, combined with the lineage specificity of the rates of evolution of the different structural categories, suggest that the findings outlined here are likely to vary by lineage rather than reflecting universal characteristics of RNA evolution. Interestingly, the model that most change in functional RNAs comes from compensatory mutations in stems is not universally true. In this context, we eagerly await the availability of the structure of a eukaryotic ribosome for comparison with the results presented here for the bacterial ribosome. We conclude that rates of evolution in different lineages and structural features of the rRNA show an unexpectedly rich and complex pattern, and that better understanding of this pattern will refine the results of a wide range of studies. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

ϕC31 integrase and several of its relatives are being widely used for precise engineering of complex genomes () and are emerging as promising new tools for gene therapy (). In addition to being highly portable ϕC31 integrase is, unlike other recombinases used for genome manipulation such as Cre and Flp, unidirectional (,,). In nature phage integrases are required for recombination of the phage genome with the host chromosome either to establish or exit from the lysogenic state. For integration the host-encoded site undergoes a conservative and reciprocal recombination with the phage site to form the hybrid product sites, and . During induction into the lytic cycle, the phage genome excises and this reaction normally requires integrase and an accessory protein Xis (,). Phage-encoded integrases can belong to the tyrosine or the serine recombinase families (). Both families of proteins act by binding to their cognate substrates and bringing the DNAs together in a synapse. Recombination is initiated by cleaving DNA strands, which undergo strand exchange to form recombinant products and these are then released (). While the mechanism of phage λ integrase, a tyrosine recombinase, is well understood (,,), the mechanism of action of integrases such as ϕC31 integrase that belong to the serine recombinase family, is less clear. All serine recombinases have a conserved catalytic domain required for DNA cleavage and rejoining (,). The resolvase/invertases also have a small (∼60 amino acids; aa) C-terminal DNA-binding domain (). The serine integrases, some transposases and the staphylococcal cassette recombinases (Ccr proteins; required for the movement of methicillin resistance gene in MRSA) are so-called large serine recombinases as they have extensive C-terminal domains (∼300–500 aa in length; 20). Sequence alignments of these large serine recombinases indicate that they are an extremely diverse family. Experiments with ϕC31 integrase, mycobacteriophage Bxb1 integrase and TnpX transposase suggest that the recombination mechanism used by the large serine recombinases resembles that of the well-studied resolvase/invertases (,,). DNA cleavage occurs at a 2 bp crossover sequence to form a staggered break and a transient covalent phosphoserine bond between the recessed 5′ ends and the recombinase is formed (,). Strand exchange most likely occurs by rotation of two recombinase subunits bound to half sites relative to the other two subunits (). Rejoining of the products is dependent on the complementarity of the DNA sequence at the staggered breaks; if there is a mismatch at this sequence, joining of the products is severely inhibited but iteration of strand exchange results in changes in the topology of the substrates (,,). Divergence from the resolvase paradigm by the serine integrases occurs in the nature of substrate recognition and the formation of the synapse. The pairs of recombination sites used by the serine integrases have different sequences; for example, the ϕC31 and sites share 39% sequence identity [, ()]. The recombination sites are generally short, ∼50 bp (,,,). The minimal sites for ϕC31 integrase have been defined as a 39 bp site and a 34 bp site (). Under conditions ϕC31 integrase converts ∼80% of and to products in the absence of accessory proteins and there are no restrictions on the topology of the substrates (,,). Moreover, in these reactions, ϕC31 integrase is catalytically inert on all other combinations of substrates including and (). Hatfull and colleagues have shown that Bxb1 integrase has similar properties and they have gone on to show that Bxb1 integrase binds to its substrates as a dimer (,). The synapse is therefore likely to contain a tetramer of integrase subunits (). A major focus in our lab has been to understand why ϕC31 integrase can only recombine and . We have shown previously that integrase cannot synapse pairs of recombination sites other than with indicating that the formation of the synapse is the major block to excision (). We and others have proposed that integrase adopts specific conformations when bound to and sites that enable the formation of a synapse, but when bound to and disable or destabilize the synapse (,,). In this model, the interactions between integrase and and are central to the formation of the synaptic interface. Some clues as to the preferred sequences of and have been obtained previously through studies that have characterized the substrates used by integrase when one of the cognate sites is not present (,,). Pseudo- sites in the bacterial host, and other actinomycetes show a strong preference for certain bases [, (,)]. Similarly, pseudo- sites have been characterized in mammalian genomes and these also show base specific preferences (). Many of the bases that are conserved in the pseudo- and pseudo- sites are also conserved between and (). To examine the integrase– interaction in more detail, the minimal site was subjected to mutagenesis and the activities of the mutants assayed in recombination and binding assays. Recombination defective mutants that could still bind to integrase with affinities not dissimilar to the wild-type site were found to be blocked either at synapsis or at DNA cleavage. The most likely explanation is that there are two separate recognition events that occur between integrase and the site. The first event results in a protein–protein interface that enables synapsis and the second post-synapsis event results in activation of DNA cleavage. strains DH5 and DS941 were used as general cloning hosts and were grown in LB or 2xYT (). transformation, plasmid preparations and DNA manipulation were performed as described previously (). Plasmids pRT600 and pRT700 were constructed previously by insertion of annealed oligonucleotides RM1/RM2 containing (51 bp) and RM3/RM4 (50 bp) sites inserted into pGEM7 cut with EcoRI and Csp45I (). For this work, the site from pRT700 was excised with BamHI and EcoRI and inserted into BamHI and EcoRI cut pSP72 to form pRT702. Plasmids containing mutant sites at all positions except for −/+3, −/+8 and −/+12 were constructed as for pRT600; annealed oligonucleotides (see Supplementary Data, Table S1) were inserted into pGEM7 cut with EcoRI and Csp45I. Plasmids containing mutations at −/+3, −/+8 and −/+12 were constructed differently; PCR amplification using primers containing a randomized base at positions 3, 8 or 12 (Supplementary Data, Table S2) resulted in fragments that could be inserted into pGEM7 and these were then sequenced to determine the nature of the mutations. To create the double mutants with mutations at symmetrical positions, fragments containing the two single mutations were spliced together using the unique StyI site in the centre of the site. All the plasmids containing the mutant sites were subjected to confirmation by sequencing. Standard recombination assays between two attachment sites located on two separate plasmids were performed as described previously. Plasmids (100 ng each) containing (or the mutant s) and were mixed with 18 μl of recombination buffer (10 mM Tris pH 7.5, 1 mM EDTA pH 8, 100 mM NaCl, 5 mM DTT, 5 mM spermidine, 4.5% glycerol and 0.5 mg/ml bovine serum albumin) and ϕC31 integrase was added to the recombination reaction to final concentrations 0, 441, 110, 55, 27 or 14 nM unless otherwise stated. Reactions were incubated at 30°C for 1 h unless otherwise stated and terminated by incubation at 65°C for 10 min. After addition of an equal volume of 2× restriction buffer (20 mM Tris-HCl pH 7.9, 100 mM NaCl, 20 mM MgCl, 2 mM DTT) the plasmids were treated with HindIII restriction endonuclease (37°C for 2 h) and the fragments were separated by electrophoresis through 0.8% agarose gels in 1× TBE buffer (100 V). HindIII linearizes the substrates containing (or mutant and to give DNA molecules of 3035 and 2491 bp, respectively. The recombination product is a cointegrate of the two substrate plasmids and is cut by HindIII into two fragments; 5435 bp containing and 91 bp containing . Only the fragment is detected routinely after electrophoresis. Recombination reactions were also performed using a plasmid encoding the site, pRT702, and annealed oligonucleotides containing the sequence or its mutant derivatives (the ‘oligo-plasmid’ assay; see Table S2 in the Supplementary Data for the sequences of the oligos) (). pRT702 (100 ng) was mixed with 4.5 ng of annealed oligonucleotides encoding a mutant site and 18 μl of recombination buffer. ϕC31 integrase (1 μl) was added to give final concentrations as described above (i.e. 0, 441, 110, 55, 27 and 14 nM) and the reactions were incubated at 30°C for 1 h. The recombination reactions were terminated by heat inactivating the samples at 65°C for 10 min and the products of recombination were analysed on 0.8% agarose gels in 1× TBE buffer (100 V). The products of recombination were identified as linear DNAs (2546 bp). DNA affinity and synapse assays were performed as described previously (). DNA fragments for radioactive labelling were prepared by digestion of pRT600 (encoding wt ), pRT700 (encoding ), or plasmids containing cloned annealed oligonucleotide pairs encoding mutant sites with HindIII and XhoI restriction enzymes. The 72 bp fragments containing the sites were separated on 4% agarose gel (Nusieve agarose) and then purified using gel extraction columns (QIAGEN) as per the manufacturer's protocol. The concentration of the purified fragment was determined on 4% agarose gels following which the fragments were end-labelled using DNA polymerase I large (Klenow) fragment in the presence of [−P]dCTP (as described previously in Sambrook . ()). Unless otherwise stated, binding affinity assays were performed with 1.0 ng labelled probe in binding buffer (20 mM Tris-HCl pH 8.0, 0.1 mM EDTA, 50 mM KCl, 5% glycerol), 1 μg sonicated salmon sperm DNA and integrase added to final concentrations of 0, 351, 87, 43, 21 and 10 nM. Reactions with no integrase contained 1 μg BSA. Reactions were incubated at 30°C for 30 min prior to electrophoresis following which the reaction mix were separated on 5% non denaturing 0.5× TBE polyacrylamide gels in 0.5× TBE running buffer (200 V, 5 W for 2 h). For radioactive recombination assays, unlabelled ‘partner’ fragments prepared by PCR amplification of pRT600 and pRT700 with SP6 and T7 primers containing either (193 bp) or wt or mutant (194 bp) were added to the radiolabelled attachment site in the presence of integrase. Complexes containing either the uncleaved synapse, the cleaved intermediates with integrase bound covalently to the sites, and integrase bound to the labelled substrate and products were observed by non-denaturing PAGE as described previously (). These assays were performed using 1.5 ng of labelled probe (72 bp), 20 ng of the unlabelled fragment containing a ‘partner’ attachment site and 66 nM integrase in binding buffer. Unless otherwise stated, all reactions were incubated for a period of 2 h at 30°C prior to electrophoresis on 5% non-denaturing polyacrylamide gels (200V, 5W for 2 h). Wild-type ϕC31 and S12A integrase were purified as described previously (). Integrase concentration was assayed using a method based on the dye-binding procedure of Bradford () employing the BioRad protein assay solution, and bovine serum albumin as a standard. The minimal site, according to Groth ., () is 34 bp with the crossover 5′TT (abbreviated to XO) at the centre (). Footprinting confirmed that integrase binds either side of the crossover site in all the attachment sites and, as integrase is a dimer in solution it probably binds as a dimer (). Moreover, we have shown that integrase binds to and in a functionally symmetrical manner. Thus in order to maximize any phenotype arising from mutations in we generated a set of doubly mutated sites with base pair changes at symmetrical positions with respect to the crossover sequence. To aid in the description of the positions of mutations, the base pairs in the minimal and sites were annotated with either a negative number when they lie to the left of the crossover dinucleotide sequence (5′TT) i.e. B or P arm to use the λ terminology) or positive when it lies to the right of the crossover (B′ or P′ arm); the numbers count upwards as the position extends away from the crossover (). Thus mutations in a double mutant involving the two base pairs adjacent to the crossover is at −/+1 and mutations at the next position moving outwards are at −/+2, etc. Mutations were chosen that would introduce sequence symmetry at the desired position. Thus each double mutant was designed to contain one of the four bases, A, T, C or G, at position on the B arm and at on the B′ arm its complement, T, A, G or C, respectively, was inserted. The choice of mutation was made on the basis that the introduced bases had to be different from those present in both arms of and preferably also different to those seen in the pseudo- sites (). For example, position 15 is a T in the B arm and a C in the B′ arm and the pseudo- sites have a G in the B arm and a C or A in the B′ arm. T-15C:C+15G and the T-15A:C+15T contain changes at −/+15 on the B and the B′ arm to base pairs that are different from both the wt and the pseudo-site sequences and should be functionally the same mutation in both arms. For most positions at least two mutant forms were made but for some sites (positions 4, 7, 16) only one option was available. Other positions where only a single mutant form was made are at 14, 17 and 18. Except for mutations at positions −/+3, −/+8 and −/+12 the activities of the double substituted sites was first assayed using annealed oligonucleotides. Oligonucleotides containing the double substitutions were purified by PAGE, annealed and used in an oligo-plasmid recombination assay (). In this assay, a supercoiled plasmid containing was mixed with the oligonucleotide containing or one of the mutant forms and various concentrations of integrase. The extent of linearization of the plasmid indicated the extent of recombination and this was assayed after separation of the DNA in an agarose gel. A control reaction using the wild-type site was performed in every assay so that the activities could be compared under identical conditions. The lowest integrase concentration at which recombination could be observed was scored (Figure S1 and ). Many of the mutant sites showed little or only 2-fold change in activity compared to the wild-type site. These sites were changed at −/+1, −/+4, −/+5, −/+7, −/+10, −/+11 and −/+13 (, ). The remaining mutants showed defective or partially defective activity ranging from 4-fold less active than wild type to apparently inactive. Oligos encoding sites C-2G:G+2C, C-2A:G+2T, G-6A:C+6T, (G-6T:C+6A, G-9T:C+9A, G-9A:C-9T, G-9C:C-9G, T-15C:C+15G, G-16T:G+15A and G-18C:A+18G were cloned into pGEM7 (Promega) so that the activities of the mutant sites could be verified by a standard recombination assay using both sites residing on plasmids. Only one of the mutant sites that was partially defective (at position −/+14) was not represented in the cloned mutant site collection; this site was instead subjected to single site substitutions (see later). T-15A:C+15T was not cloned as a plasmid containing another mutant at −/+15 (T-15C:C+15G) with the same activity was quickly obtained. A plasmid encoding G-6C:C+6G was not obtained due to technical difficulties. Plasmids containing mutations in C-3T:G+3A, C-3G:G+3C, C-3A:G+3T, G-8T:C+8A, G-8C:G+8C, C-12A:G+12T and C-12T:G+12A were obtained by PCR mutagenesis as described in the Material and Methods section. The relative activities of the double substitution mutants were estimated compared to a standard reaction with wild-type ( and and ). As for the oligo-plasmid assay the activity of each mutant site was scored as the concentration of integrase required to observe recombinants in an agarose gel stained with ethidium bromide (). The relative activities compared to the wild-type site are summarized graphically (). Three of the mutant sites were very defective for recombination and these contained substitutions at −/+2, −/+15 and −/+16. In all cases no recombination was observed in either the oligo-plasmid or the standard assay using these double substituted sites ( and Figure S1). Recombination was just detectable with containing substituted −/+18 in the plasmid assay with 351 nM integrase (). The low activity of the −/+18 double mutant was surprising given that this position is outside the minimal site defined previously by Groth . (). The nature of the mutations made small differences to activity in only a few mutants. The −/+12 mutant containing the double transversion C-12A:G+12T was only just active with 87 nM integrase while the −/+12 mutant containing the transitions C-12T:G+12A was active with 43 nM integrase (). The G-6T:C+6A transversions had similar activity to wild-type but another −/+6 mutant, containing transitions (G-6A:C+6T) was 2-to 4-fold less active than (). All of the mutant sites described in this section that were cloned into plasmids were used to test whether they would recombine with or but no activity was detected in any case. Thus none of these mutant sites had any detectable gain-of-function. The effects of mutations at positions −/+2, −/+14, −/+15, −/+16 and −/+18 were studied further. Oligonucleotides were synthesized that had single mutations at either the position in the B arm or in the position in the B′ arm. Recombination was performed with the oligo-plasmid assay and with the standard recombination assay using the sites cloned into pGEM7. The sites containing the single mutations C-2G and G+2C regained much of the activity of the wild-type site suggesting that a correct interaction on one or other side of the crossover at this position is sufficient for recombination (). Similarly the single mutation at –18 or +18 also regained some activity compared to wild-type (Figure S1). Single mutations at the 15 and 16 positions behaved differently. Mutants at –15 or –16 had much greater effects on recombination than the mutants at +15 or +16. The single mutations C+15G and G+16A regained some activity compared to the double mutants T-15C:C+15G and G-16T:G+16A whereas the single mutants at T-15C and G-16T did not (). A similar difference, but less so, was also observed at position 14 where the left B arm was more sensitive to mutation than the right B′ arm (). To test this further we experimented with partially symmetrical sites. The B arm of that included the region from –12 to –18 was replaced with the +12 to +18 sequence from the B′ side [2R (−12 to −18)] and vice versa, [2L (+12 to +18)]. The 2L (+12 to +18) site was as active as the wild-type site whereas the 2R (−12 to −18) site was inactive (). These data indicate that the sequence in the left arm of plays a major role in function and its loss removes all activity. A mutant site RL, with the straight swap of the B arm sequence between −12 and −18 with the B′ arm sequence at +12 to +18 was inactive (Figure S1) indicating that whatever positive role the −12 to −18 sequence plays in function, it is not acting independently of other sequences in the site. This mutational analysis of showed that double mutations at three positions −/+2, −/+15 −/+16 and the single mutants at –15 and –16 were particularly defective for recombination. We have shown previously that it is possible to assay several intermediate steps in recombination i.e. DNA binding, formation of the synapse and cleavage of the DNA to form the covalent intermediate in which integrase is covalently bound to its cleaved substrate (). The mutant sites were used first in affinity assays with integrase. As seen previously integrase bound to the wild-type site with an affinity of ∼60 nM (,). Most of the mutant sites bound with a similar affinity to the wild-type site including the severely recombination defective sites C-2A:G+2T and G-16T:G+16A (, ). The mutant T-15C:C+15G had a slightly lower affinity for integrase (∼128 nM) but this loss of affinity was abolished in the single mutant at −15 (T-15C) which was still defective in recombination (, and ). Differences in binding affinities by integrase for mutant sites C-2A:G+2T, G-16T:G+16A, T-15C and G-16T cannot therefore account for the defectiveness of these sites in recombination. Mutations involving position 18 from the crossover dinucleotide showed an ∼3-fold lower affinity for integrase than wild-type which could contribute to the observed decrease in recombination activity (). It seems likely that sites with mutations at −/+2, −/+15, −/+16 were blocked elsewhere in the recombination pathway. The formation of both the cleaved covalent intermediate and the synapse can be observed in a recombination assay using a radiolabelled or site, a cold partner site and integrase (). These assays are performed in a buffer that is sub-optimal for recombination (binding buffer) that enriches for synaptic complexes and the cleaved covalent complex compared with standard recombination conditions (). was labelled with [α−P] dCTP, mixed with cold wild-type or mutant sites and integrase and run in a non-denaturing PAGE gel (A, left panel). Compared to wild-type , sites with −/+2 changes (C-2G:G+2C and C-2A:G+2T) showed an accumulation of synapse with almost undetectable cleaved covalent complex or product formed (A, left panel). Treatment of the recombination intermediates with the protease, subtilisin showed a small amount of cleaved probe with C-2G:G+2C but this was undetectable with C-2A:G+2T (B). Subtilisin treatment of reactions containing wild-type clearly revealed the two recombination products and but these were not visible with C-2A:G+2T and barely visible with C-2G:G+2C (B). As seen in the recombination assay reverting one of the two mutations in C-2G:G+2C back to the wild-type sequence was sufficient to regain activity similar to the wild type site (C-2G or G+2C in A, left panel). The catalytically inactive integrase mutant (S12A) was able to bind to C-2G:G+2C, C-2A:G+2T, C-2G and G+2C normally to form a synaptic complex indistinguishable from the wild type site (A, right panel). Experiments in which the labelled probes were or the −/+2 mutant derivatives and unlabelled was used to supershift the complexes showed similar results, i.e. very little cleavage of the −/+2 mutant was observed (Supplementary Data—Figure S2). These data indicate that sites with a double substitution at −/+ 2 are able to generate a stable synapse but are severely defective in cleavage of the substrates. The sites with mutations at −/+15 and −/+16 that were defective in the standard recombination assay did not appear to be defective in the radioactive assay to detect intermediates (A, left panel). The amounts of cleaved covalent intermediate and shifted / products were indistinguishable from the reaction with the wild-type site (A, left panel). The only observable difference was in the amount of synapse, which was reduced in the reactions with the most defective sites i.e. T-15C:C+15G, T-15G, G-16T:G+16A and G-16T and the appearance of some free product. These differences were also observed when the sites were labelled and incubated with integrase and cold (Supplementary Data—Figure S2). When the S12A catalytically inactive integrase mutant was used, there was a small reduction in accumulation of the synaptic complex with T-15C:C+15G, T-15G, G-16T:G+16A and G-16T compared with wild-type (A, right panel). The inconsistency whereby mutants T-15C:C+15G and G-16T:G+16A were inactive in the recombination assay but active in the assay for intermediates was addressed. As the standard recombination assay is performed over 1 h and the synapse assay is over 2 h, time courses were performed for each assay. In the standard recombination assay, products were observed in 2 and 3 h with mutants T-15C:C+15G and G-16T:G+16A (A). Using the more sensitive radioactive assay, cleaved substrates, T-15C:C+15G and G-16T:G+16A, and their recombinant products started to appear at 30 min of incubation and accumulated further over the next 30 min whereas with the wild-type , most of the substrate had been converted to intermediates or products at 15 min. Even after 60 min, less in the presence of T-15C:C+15G or G-16T:G+16A was converted to product compared to in the presence of wild-type (B). Thus both the mutants T-15C:C+15G and G-16T:G+16A could undergo recombination but the reaction is considerably slower than that for the wild-type site. As there is a consistently reduced level of synaptic complex observed with these mutant sites, it is likely that changes in at −/+15 and −/+16 both result in an unstable synapse that explains the slow rate of recombination. We reasoned that altered recombination conditions might partially suppress the defect in T-15C:C+15G and G-16T:G+16A by stabilizing the putative protein–protein interface. Recombination was observed when the NaCl concentration was increased to 500 mM or 1 M in recombination buffer (A). However, increasing the concentration of NaCl did not increase the amount of synaptic complex observed with these mutant sites (B). Indeed at 1 M NaCl there was a reduction in the level of synapse observed with the mutants at −/+15 and −/+16 and a slight reduction in the affinity for the site by integrase (B and ). The interactions between ϕC31 integrase and its attachment sites are critical in determining the directionality of recombination. integrase only recombines and to form the hybrid products, and . We have shown previously that, , integrase selectively brings and together to form the synapse and no other combination of sites forms a stable synapse under these conditions (). These observations have led to the proposal that integrase adopts specific conformations when bound to or that permit formation of the protein:protein interface required for stable synapsis (,). Here we showed that mutations in can significantly affect the ability of integrase to form a stable synapse or to cleave the substrates. These perturbations in the reaction are likely to be due to the absence of important interactions between integrase and and could be indicative of ‘non-permissive’ conformations of integrase that block recombination at these different stages. The mutations at −/+2 in showed a failure to cleave the DNA but these substrates could still form a stable synapse (). These data show clearly that there is a post-synaptic activation step required for recombination by ϕC31 integrase. This activation step depends on an interaction that has been disrupted in the −/+2 mutants, C-2G:G+2C or C-2A:G+2T. The nature of the interaction is not known but could be a specific base-pair contact or a DNA conformation that is recognized. The block in DNA cleavage occurred in both the mutant sites themselves and in the wild-type sites ( and Figure S2). This behaviour is consistent with concerted DNA cleavage in the reaction with wild-type recombination sites. Possibly the block in cleavage in reactions containing the −/+2 mutant sites could be due to failure to undergo a conformational change in the whole synaptic complex which would normally lead to cleavage. Alternatively the DNA conformation of the mutant sites prevents the catalytic sites gaining access to the scissile phosphate. As reversion of just one of the bases from the double mutant back to the wild type was sufficient to regain most of the activity it would seem that activation only requires a ‘correct’ interaction at one half-site of . The mutants T-15C:C+15G and G-16T:G+16A were able to recombine but at a slow rate compared to wild-type (). There was a consistent reduction in the amount of synapse observed during recombination with these mutants suggesting that the synaptic complex was unstable ( and ). Raising the concentration of NaCl partially suppressed the defect in recombination with the −/+15 and −/+16 mutants but it is not clear which step was affected by NaCl (). The stability of the synapse did not increase in the presence of a higher concentration of NaCl, if anything the binding affinity and the level of synapse was reduced at 1 M NaCl (). Despite this, suppression was still observed suggesting that high NaCl activates or stabilizes an event later in the recombination pathway. The single point mutants at –15 and –16 were sufficient to severely affect recombination while mutations at +15 or +16 had a lesser effect (, , ). Thus the single mutations at positions –15 and –16 accounted for most of the defect in the −/+15 and −/+16 double mutants. These data argue that there could be a specific interaction between the B arm and integrase that contributes significantly to the activity of the site. The partial symmetrization of the sites (with either the sequence from the B arm [2L (+12 to +18)] or with that from the B′ arm [2R (−12 to −18)]; , panel H) showed that the B arm was indeed more active than the B′ arm. However, it is known from previous work that the and sites act with integrase in a functionally symmetrical manner as integrase does not control the relative orientation of the sites when they come together at synapsis (). Thus the interactions by each subunit of integrase bound to each arm of are not independent of each other and we propose that a specific integrase conformation that results from the –15, −16 interactions in the B arm is communicated through both subunits. These conclusions can be combined with information from other large serine recombinases and the resolvases to generate a model that focuses on substrate recognition and formation of the synapse by integrase (adapted from that published previously for Bxb1 integrase, 26; ). In the resolvases, the DNA is contacted in the minor groove in the centre of each binding site and through specific contacts in the major groove towards the outer flank of the site via the C-terminal DNA binding domain (,). The geometry of DNA-binding is such that the C-terminal domain of resolvase extends around the DNA and contacts on the opposite side of the DNA to the catalytic serine (). As in Bxb1 and TnpX, ϕC31 integrase has a proteolytically sensitive site between the N and C terminal domains (K152, unpublished data)(,,). Moreover, the C-terminal domains of Bxb1 and TnpX have been shown previously to be capable of binding specifically to DNA (,,). Thus we propose that the C-terminal domains interact with the outer flanks of the sites, that these interactions determine the conformations of integrase bound to each site and therefore whether they are compatible for synapsis. In this information is ‘read’ at least in part from –15 and –16 where disruption of this interaction disables the ability of integrase to form a stable synapse (). The model predicts that there is communication between the putative DNA-binding motifs in the C-terminal domain and the regions of integrase that generate the protein-protein interface for synapsis. We currently envisage this communication as an allosteric switch mediated by conformational changes. In γδ resolvase, the synaptic interface is located at the DNA distal surface of the catalytic domain and it is likely that the serine integrases use the equivalent of this interface for synapsis, although it is possible that the C-terminal domain may also have a role in synapsis. After synapsis an activation step is required for DNA cleavage and in this depends on the base pairs at position −/+2. In only one of the −/+2 bases needs to be wild type for activity and this can be either on the B or B′ arm. Given the proximity of −/+2 to the scissile phosphate, position 2 is more likely to interact with the catalytic domain than with the C-terminal domain. As in resolvase there may be significant conformation changes that occur with activation of recombination (,). The data presented here provides a source of information that could be used for the design of alternative sites for genome engineering using ϕC31 integrase. Indeed we have already used this information to create a non-methylatable site for use in vertebrate cell lines (). In this site, , the CpG steps have been replaced with bases that we have shown here were neutral with respect to recombination activity. It is noteworthy that the positions in that are critical for recombination other than position 2 i.e. positions 15, 16 were not highlighted as being particularly preferred in the pseudo- sites (). However all of the pseudo- sequences have either the wild-type C at −2 or a wild-type G at +2 () (,). It is also noticeable that the regions where there are most identities between and are not particularly sensitive to mutation (). A plausible explanation for both these observations is that the 24 bp of site DNA from about position 11 to the crossover site is a core sequence that integrase binds to specifically. We propose that the role of positions 15 and 16 in revealed by this study is to greatly enhance the efficiency of the reaction and effectively discriminate between the pseudo-sites and the cognate site. It is envisaged that this data and a similar analysis with will enable an understanding of the optimal sequences to target for continued application of ϕC31 integrase. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

The spindle checkpoint functions to prevent premature anaphase entry until all chromosomes have completely aligned at the metaphase plate. BubR1 is a key protein mediating spindle-checkpoint activation during which it is phosphorylated. A loss of spindle-checkpoint function inevitably results in chromosomal instability and aneuploidy (,). Given that aneuploidy is prevalent in many types of cancers, it is believed that spindle-checkpoint failure may be at least partly responsible for the development of cancer (). Mouse genetics study showed also that haplo-insufficiency of BubR1 resulted in enhanced genomic instability and development of lung and colon cancer (,). In synchronized cells, expression of the BUB1B gene, which encodes BubR1, is undetectable in G1 but it peaks in G2/M (,). This cell cycle dependent expression explains the tissue distribution and the abundance of BubR1 mRNA in cells with a high mitotic index (). In this work, which represents the first report concerning the regulation of BUB1B gene expression, we localized the promoter of this gene to positions −464/−107 relative to the translation initiation codon. We found that the BUB1B promoter contains three positively -acting motifs: one Inr-like motif and two elements recognized by the hStaf/ZNF143 transcription factor. This factor can up-regulate the BUB1B promoter activity, and chromatin immunoprecipitation assays demonstrated that the endogenous hStaf/ZNF143 is bound to the BUB1B promoter . The ZNF143 protein is the human ortholog of Staf, the Xenopus selenocysteine tRNA gene transcription factor (,). The human Staf (hStaf/ZNF143) is a seven C2-H2 zinc finger protein capable of enhancing transcription of the tRNA but also of the snRNA and snRNA-type genes (,). hStaf/ZNF143 can also stimulate transcription from an mRNA-type pol II promoter (,,). To date, only seven protein-coding genes have been described to be regulated by hStaf/ZNF143: the cytosolic chaperonin containing t-complex polypeptide 1 (TCP1) (); the interferon regulatory factor (IRF3) (); the neuronal nitric-oxide synthase (NOS1) (); the transaldolase (TALDO1) (); the aldehyde reductase (AKR1A1) (); the mitochondrial ribosomal protein S11 (MRPS11) () and the synaptobrevin-like 1 (SYBL1) (). In addition, vertebrates contain the ZNF76 protein which constitutes a ZNF143 paralog (,). ZNF76 and ZNF143 are basically considered to play the same role even though their relative expression levels differ in various tissues (). However, recent results suggested that ZNF76 functions as a transcriptional repressor through its interaction with TBP and that sumoylation modulates its transcriptional properties (). Very recently, genome-wide analysis led us to identify 1175 hStaf/ZNF143-binding sites distributed in 938 mammalian promoters in protein-coding genes. By extrapolating these values to the full sizes of the genomes, we can infer the existence of at least 2500 Staf-binding sites (SBS) distributed in 2000 promoters. This large number suggests that the SBS constitutes one of the most widespread transcription factor binding sites in mammalian promoters (). In G1/S, the transcriptional repression of many genes such as CDC25C, CDC2, CCNA2 (coding for cyclin A), PLK1 (polo-like kinase) and RB6K (rabkinesine 6) is regulated by two repressor elements known as CDE (cell cycle dependent element) and CHR (cell cycle genes homology region). In these genes, mutation of the CDE and CHR elements allowed elevated transcription during G1 and the consequent loss of cell cycle regulated expression (). In the present study, we demonstrate that the cell cycle regulation of BUB1B gene transcription is also achieved through the presence of two elements homologous to the repressor elements CDE and CHR. The human BUB1B promoter fragment −1185/−31 was PCR amplified from human genomic DNA using direct and reverse primers incorporating SacI and BamHI sites, respectively. The amplified product was cloned directly at the 5′-end to the luciferase reporter gene into the SacI/BamHI digested pFLASH I vector (SynapSys). The 5′-end truncated derivatives of the −1185/−31 fragment (−864/−31, −585/−31, −464/−31, −314/−31, −305/−31, −236/−31 and −107/−31) were PCR amplified from construct −1185/−31 and ligated to SacI/BamHI cut pFLASH I. Mutant constructs were generated by using the QuickChange II XL site-directed mutagenesis kit (Stratagene). All constructs were verified by automated DNA sequencing. Constructs −464/−31 mCHR contain the −176ATTTGAA-170 to −176CGGGTCC-170 substitution, −464/−31 mCDE contain the −166TGGCGG-161 to −166GTTATT-161 substitution. The expression vectors pPAC-ZNF76 and pPAC-ZNF143 containing the human ZNF76 and ZNF143 cDNAs were described in (). The oligonucleotide sequences used in this study are available on request. COS-7 cells were transfected by the calcium phosphate co-precipitation procedure with 1 μg of reporter construct, 0.5 μg of pCH110 plasmid as the internal control, and carrier DNA to bring up the total DNA content to 10 μg/plate. SL2 cells were transfected as described in () with 25, 100 or 500 ng of pPAC-hStaf/ZNF143 or pPac-ZNF76, and 200 ng of pACH110 as internal control. After 48 h, cells were lysed and the β-galactosidase activity was measured as previously described (). The luciferase assay was performed as recommended by the manufacturer (Promega). The luciferase activity was normalized to the β-galactosidase activity. Each transfection experiment was done in triplicate. The hStaf/ZNF143 DNA-binding domain was produced using the glutathione S-transferase (GST) gene fusion system. Briefly, the DNA containing the Staf zinc finger coding region between A264 and E472 was excised from E10 () as a HindIII/EcoRV fragment, blunted and inserted in the direct orientation into the SmaI site of pGEX-2TK (GE Healthcare). The fusion protein was purified using glutathione-sepharose beads. The glutathione S-transferase moiety was cleaved with thrombin. Full-length hStaf/ZNF143 was synthesized by coupled transcription-translation with the TnT system (Promega) as described in (). Fifty microliter reactions were programmed with 1 μg of pSK(-)-ZNF143 (). Gel retardation assays were performed essentially as described in () with 20 fmol of the labeled probe in the presence of either the hStaf/ZNF143 DBD (1, 2, 4 and 20 pmol), 2.5 and 5 μl of programmed lysate or 10 μg of HeLa cells nuclear extracts. The various probes containing the wild-type and mutant versions of the SBS in the BUB1B promoter were generated by PCR amplification of regions −342/−196, −342/−265, −281/−196 using P-labeled oligonucleotides. The rabbit polyclonal antipeptide antibody against a C-terminal epitope of the Staf () was used for ChIP as essentially described in (,). Purified DNA was analyzed by PCR with the test primer pair TAAGTGTTCCTCGCTCGGCTCAGA and CTCAGAGCACCCCCTTCCTTCTTC specific for the BUB1B promoter and complementary to positions –427/–404 and +12/+35, respectively. The BUB1B control primers CCACTGTGGGGTGCTGATGTCTGG and CGGGATGCGGGGGTTGC hybridized to sequences 2647–2623 and 2439–2422 bp, respectively, upstream of the BUB1B ATG initiation codon. The human tRNA gene test primer pair hybridized to sequences −391/−365 and −205/−181 of the human tRNA gene promoter. The human tRNA control primer pair recognizes sequences located at 2555/2530 and 2346/2321 bp upstream of the human tRNA gene. Cycling parameters were 95°C for 3 min, 35 cycles at 95°C for 30 s, 55–65°C (depending on each primer pair) for 30 s, 72°C for 30 s and 72°C for 5 min. Transiently transfected COS-7 cells were arrested at the G1/S boundary by a single or double thymidine block. To block them at the M phase, cells were treated with nocodazole. In brief, for thymidine block and nocodazole treatment, thymidine or nocodazole were added 18 h after transfection to 2 mM or 0.17 mM final concentration, respectively. After 22 h incubation, cells were harvested and used for luciferase and flow cytometry assays. For the double thymidine block, thymidine was added to 2 mM as the first block. After a 16 h incubation, cells were washed twice with PBS and incubated in complete growth media for an additional 8 h. Thymidine was then added as the second block for 16 h. Subsequently, cells were washed twice with PBS, and complete growth media was added to release them from the block. This time point was set as 0. Cells were harvested at various time points and used for luciferase and flow cytometry assays. For flow cytometry, cells were briefly trypsinized, pelleted by centrifugation at 200 , resuspended in PBS and fixed in 75% ethanol. After centrifugation and rehydration in PBS for 15 min at room temperature, cells were pelleted and resuspended in 1 ml of staining buffer containing propidium iodide (0.5 mg/ml), 0.1% Triton-X100, 0.1 mM EDTA and RNase A (25 mg/ml). Cell cycle distribution was determined by analyzing their DNA content on a Becton Dickinson FACScalibur flow cytometer. A search at the database of transcription start sites (DBTSS) (,) revealed that transcription of the BUB1B gene is directed from multiple transcription start sites (TSS) located in a 75-bp long region. In this report, promoter numbering starts from the first nucleotide of the translation initiation codon. The TSS region is located between positions −198 and −124, with a major transcription start site at −178 (A). To identify the regions responsible for transcriptional regulation of BUB1B, we transiently transfected COS-7 cells with several luciferase reporter constructs containing progressively deleted 5′-flanking regions of the BUB1B gene and then measured the luciferase activity of the resulting cell extracts (A and B). The parental construct −1185/−31 contains the region −198/−124 covering the transcription start sites. A deletion from −1185 to −586 resulted in ∼5-fold decrease of luciferase activity (B, compare constructs −864/−31 and −585/−31 with −1185/−31). Further deletions from −584 to −306 did not significantly affect the transcriptional activity (B, compare the activity of constructs −464/−31, −314/−31 and −305/−31 with construct −585/−31). An additional deletion to −237 resulted in ∼10-fold decrease of activity. Strikingly, a further deletion to −107 created a construct unable to drive transcription of the reporter gene (B, compare −236/−31, −107/−31 and the empty vector Luc with −1185/−31). We concluded from the luciferase assays that transcription of the BUB1B gene was positively regulated by two regions, one located between −1185/−585 and the other between −305/−107. Sequence comparison and analysis of upstream regions of the BUB1B gene that contain orthologs in the mouse, rat and dog genomes, revealed the presence of transposable elements in regions homologous to −1185/−585 in the four genomes. shows the occurrence of two short interspersed nuclear elements (SINE). In addition, a long-terminal repeat sequence (LTR) is present upstream of the mouse and rat BUB1B genes. Computational analysis of the −1185/−585 region with the Matinspector software () did not reveal the presence of interspecies conserved sequences for transcription factor binding sites. Upon computational analysis of the first 464 bp of the BUB1B promoter, no TATA box could be found in the vicinity of the transcription start site. The main transcription start site at −178 is likely to function as an initiator region (Inr) since the sequence TTAAATT located at positions −180 to −174 is very similar to the Inr consensus sequence YYAN(T/A)YY (). No downstream promoter element (DPE) was found at proximity of the main transcription start site. Interestingly, however, two consensus binding sites for the transcription factor hStaf/ZNF143 (,,) were found in the region −305/−107 which positively regulates BUB1B gene expression. They are called hereafter SBS for simplicity's sake. The first SBS1, located at positions −305/−288, and the second SBS2 at positions −256/−239, were found to be interspecies conserved at 72 and 88%, respectively (A). In B, the 18-bp SBS1 and SBS2 were aligned with the sequence of the human tRNA SBS () and with the consensus sequences determined by binding site selection (,). Immediately, upstream of the two SBS is found an interspecies conserved 7 bp ACTACAA motif (A), which does not correspond to any binding site for known transcription factors. Furthermore, A shows that the −305/−107 region contains blocks of high sequence identity to the CHR (positions −174/−170) and the CDE (positions −165/−161) (see also A). These elements are known to be involved in the cell cycle regulated transcriptional repression of many genes (). In the promoter of these genes, the CDE is generally adjacent to or 1–5 bp upstream of the CHR element (A). This contrasts with the BUB1B promoter where the CDE homologous motif precedes the CHR by 4 bp. Finally, the upstream 5′-flanking region of the BUB1B gene contains one CpG island extending from −415 to −50. In the first place, gel retardation assays were performed to determine whether hStaf/ZNF143 does bind the BUB1B promoter. To do this, the P-labeled DNA fragment (−342/−196) encompassing the two putative SBS (probe I in A) was incubated with increasing amounts of the purified hStaf/ZNF143 DNA-binding domain (hStaf/ZNF143 DBD). A shows that the hStaf/ZNF143 DBD bound with high yield to the BUB1B promoter. Increasing amounts of the protein generated two retarded complexes (C1 and C2 in A, compare lane 1 with lanes 2–5). The C1 and C2 complexes were specific because they were competed out by an excess of unlabeled SBS of the Xenopus tRNA gene (A, lane 6) but not of an unrelated oligonucleotide (A, lane 7). We next examined the binding capacities of the hStaf/ZNF143 DBD to a BUB1B promoter carrying alterations in the SBS. Three mutant versions of the BUB1B promoter were engineered. In mSBS1, the CCCA sequence at positions 3–6 of SBS1 was replaced by AAAC; the same mutation was introduced in the SBS2 sequence to yield mSBS2. The mSBS1-2 construct combined both mutations simultaneously. It appeared that formation of the retarded complexes was strictly dependent on the SBS integrity. Indeed, whereas mSBS1 and mSBS2 enabled formation of one single retarded complex only (B, compare lanes 5–7), the simultaneous presence of both mutations in mSBS1-2 completely abrogated DBD binding (B, lane 8). The binding to wild-type or mutant BUB1B promoter of the full-length hStaf/ZNF143, produced from programmed rabbit reticulocyte lysate, was further evaluated. As observed with the DBD, increasing amounts of protein generated the characteristic C1 and C2 complexes (C, lanes 2 and 3) which are specific since they disappear in the presence of an excess of the wt SBS but not with an unspecific competitor (C, lanes 4 and 5). This binding pattern is strictly dependent on the SBS integrity because the combined presence of the mSBS1 and mSBS2 mutation totally abrogated hStaf/ZNF143 binding (C, lane 7). To ask whether hStaf/ZNF143 is expressed in HeLa cell, we performed gel retardation assays with HeLa cell nuclear extracts and probes II and III containing SBS1 and SBS2, respectively. In such an experiment, we expect finding one single complex with each of the probe. This was effectively the case, as shown in D (lanes 2 and 10). The specificity of binding was attested by the competition obtained with an unlabeled hStaf/ZNF143 consensus oligonucleotide (D, lanes 3, 4, 11) but not with an unrelated unlabeled oligonucleotide (D, lanes 5, 6, 12). The presence of hStaf/ZNF143 in the complex was assessed by the displacement of the complexes observed with an anti-hStaf/ZNF143 (D, lanes 7 and 13) but not with a pre-immune antibody (D, lanes 8 and 14). To further validate these findings, the association of hStaf/ZNF143 to the human BUB1B promoter was investigated with the chromatin immunoprecipitation assay (ChIP). Chromatin, formaldehyde cross-linked with sheared DNA 0.5–1 kbp in length, was prepared from HeLa cells and incubated with an antipeptide antibody directed against the C-terminal part of hStaf/ZNF143 () or with a control pre-immune antibody. The recovered DNA was analyzed by semi-quantitative PCR with primers spanning the SBS of the BUB1B promoter (test sequence) or a region located 2.5 kbp upstream of it (control sequence). The analysis was performed with two dilutions of the DNA obtained from anti-hStaf/ZNF143 and pre-immune ChIP. We also tested a serial dilution of the input material to demonstrate that the PCR was quantitative within a linear range of concentration. A specific signal, absent with the pre-immune antibody, was obtained with the DNA immunoprecipitated with the hStaf/ZNF143 antibody (A, compare lanes 1, 2 and 3, 4 of the test). In contrast, no specific signal could be obtained with the control primer pair (B, lanes 1–4 in the control). As an additional control, the same DNA samples were used to show the binding of hStaf/ZNF143 to the human tRNA gene promoter, known to be targeted by hStaf/ZNF143 () (B). As expected, a specific signal was obtained only for the PCR reaction using the DNA immunoprecipitated with hStaf/ZNF143 and performed with the test primer pair (compare lanes 2 and 3 of the test and control reactions). Collectively, these results demonstrate the presence of two SBS and the association of hStaf/ZNF143 to the BUB1B promoter . The effect of the debilitating mutations mSBS1, mSBS2 and mSBS1-2 was assessed by introducing them into construct −464/−31 (C). The mutant constructs were then transfected into COS-7 cells and promoter activities were reported by the luciferase activity. Mutation of the SBS1 and SBS2 resulted in a slight decrease to 78 and 61% of the wild-type level, respectively (C). The simultaneous mutation, however, induced a much more pronounced effect since the activity dropped to 19% of the wt level (C), indicating that SBS1 and SBS2 are of prime importance to BUB1B promoter activity. We also observed that the TTAAATT sequence (positions −180 to −174) is very similar to the Inr consensus sequence YYAN(T/A)YY (). To test the functional importance of this interspecies conserved motif, the TTAAATT sequence was changed to TTCCCTT. The activity of the luciferase reporter decreased to 49% of the wild-type level (mInr in C). From this data, we conclude that the SBS are functional and that an Inr-like motif lies in the BUB1B promoter. To further confirm the role of hStaf/ZNF143 in BUB1B gene transcription and to investigate whether ZNF76, the human paralog of hStaf/ZNF143, can functionally interact with the BUB1B promoter, transient transfection experiments were performed with SL2 cells. This model offers the advantage to lack many of the homologs to vertebrate transcription factors, in particular ZNF143 and ZNF76 (,). Thus, effectors for ZNF143 (pPac-hStaf/ZNF143), ZNF76 (pPac-ZNF76) () and the empty vector pPac under the control of SL2-specific promoter, were co-transfected along with the luciferase reporter gene under the control of the wild-type −436/−31 BUB1B promoter. The effect on promoter activity of varying the intracellular level of ZNF143 or ZNF76 was assessed by introduction of increasing amounts of the expression vectors (A). Expression from the pPac-hStaf/ZNF143 or pPac-ZNF76 expression vectors was confirmed by gel retardation assays with SL2 extracts and probe II containing SBS1 (A). No binding was observed with extracts from untransfected or empty-pPac transfected SL2 cells (B, compare lane 1 with lanes 2 and 3). In contrast, increasing the amount of expression vectors in the transfection led to higher yield of retarded complexes (B, compare lane 1 with lanes 4–6 and 7–9). As ZNF143 and ZNF76 recognize a same DNA motif with identical affinities (), the yield of the complexes is a gauge of the expression of hStaf/ZNF143 and ZNF76 in SL2 cells (B, lanes 4–9). In these assays, the luciferase activity in cells extracts was normalized to the amount of the effector proteins. ZNF143 and ZNF76 exhibited a dose dependent transactivation effect on the −464/−31 promoter, reaching 24-fold for ZNF143 and 8-fold for ZNF76 compared to the empty vector pPac (A). We concluded that ZNF76 and ZNF143 can mediate the transcriptional activation of the BUB1B promoter in SL2 cells. We next analyzed the cell cycle dependent transcription of BUB1B by assessing the luciferase activities of the −464/−31 construct transfected into COS-7 cells that had been synchronized in G1/S by a double thymidine block. After the cells were released from the block, the luciferase activities at the indicated time points were measured and normalized (A). Most the cells were in G1/S at the start of the experiment. As the cells entered in the S phase, the −464/−31 associated luciferase activity increased with an optimum ∼9–13 h after the release when most of the cells were in G2/M. In contrast, similar experiments using a 3′-truncated promoter containing only the two SBS (construct −464/−196) resulted in a significant increase in reporter activity in G1/S which remained high throughout the cell cycle (A). To test whether the binding of hStaf/ZNF143 to the SBS of the BUB1B promoter is cell cycle dependent, we performed gel retardation assays using probe II (A) containing the SBS1 and nuclear extracts from synchronized COS-7 cells. B shows that the binding of hStaf/ZNF143 was similar with cells in the G1/S and G2/M cell cycle. In the light of these findings, we speculated that hStaf/ZNF143 did not play a central role in the G2/M-specific transcription of BUB1B and that another mechanism must be involved in the G2/M specificity. Indeed, transcription of the TCP1 (), IRF3 (), TALDO1 (), MRPS11 () and SYBL1 genes () is controlled by hStaf/ZNF143 but is not cell cycle dependent (). Furthermore, expression of the ZNF143 gene is known not to be cell cycle regulated (). The G2/M specific transcription of many genes such as CDC25C, CDC2, CCNA2, PLK1 and RB6K (,,,) is regulated by a tandem of repressor elements, the cell cycle dependent element (CDE) and the cell-cycle genes homology region (CHR). We found that two sets of sequences at positions −174/−170 and −165/−161 bear a strikingly high similarity to the CHR and CDE consensus sequences (A). The difference resides in their relative organization, the CHR motif lying upstream of the CDE in the BUB1B promoter (A). These observations suggest that the two putative -elements can also function as a G1/S-specific repressor, as previously reported for the G2/M-specific genes. To test the hypothesis, we introduced mutations in the putative CHR and CDE motifs (mCHR and mCDE in B). After transfection into COS-7 cells, the luciferase activity was measured in cells arrested in G1/S with thymidine or in G2/M phase transition by nocodazole (C). Drug treatment resulted in the synchronization of at least 85% of the cells in the various phases as determined by propidium iodine staining and flow cytometry (data not shown). The luciferase activity in extracts of cells transfected with the wild-type −464/−31 and arrested in G1/S was 2.2-fold lower than that from cells arrested in G2/M (C and D). In contrast, transfection of the CHR and CDE mutants of the BUB1B promoter-luciferase constructs abolished cell cycle periodicity and resulted in a 2.8- and 3.1-fold enhanced transcription activity relative to the wild-type level in G1/S arrested cells (C and D). In G2/M arrested cells transfected with the CHR and CDE mutants, however, the measured activities were similar to the wild-type (C and D). These results strongly suggest that CDE and CHR act as G1/S-specific repressor elements in the BUB1B promoter and are essential for the cell cycle expression of this gene. The BUB1B gene, which is highly expressed in cells with a high mitotic index, exhibits a cell cycle dependent expression with an undetectable transcription in G1 and a gene expression peak in G2/M (,). In the present study, we investigated the transcriptional regulation mechanism of the human BUB1B gene. We identified that regions −1185/−585 and −305/−107 in the promoter are involved in the transcriptional regulation of the BUB1 gene. Region −1185 to −585, which performs a positive regulation on BUB1B gene expression, contains two transposable elements. Sequence comparison of the regions homologous to −1185/−585 in other mammalian genomes did not reveal the presence of conserved transcription factor binding sites. This suggests that the positive regulation is performed by species-specific elements. Similar to other cell cycle regulated genes, the BUB1B promoter does not possess a TATA box and, as a consequence, gives rise to many transcription start sites. We identified five interspecies conserved -acting elements in the 5′-flanking region −305 to −107, and a trans-acting factor involved in the control of BUB1B gene expression. Three -acting motifs positively regulate transcription and the other two constitute a cell cycle dependent transcriptional repressor. The motif surrounding the major transcription start site is likely to function as an initiator. The other two positively -acting motifs, SBS1 and SBS2, are specifically recognized by hStaf/ZNF143 which itself was found associated to the BUB1B promoter in HeLa cells. Mutations of SBS1 and SBS2, singly or in combination, led to a sharp decrease of reporter activity. Analysis of the BUB1B promoter activity in insect cells demonstrated that hStaf/ZNF143 and ZNF76 can reconstitute BUB1B promoter activity. Although Staf was originally identified as the transcription factor regulating tRNA gene transcription, it also controls snRNA, snRNA-type and mRNA gene expression (,,,). The present study extends the role of hStaf/ZNF143 to BUB1B gene transcription. The two identified SBS1 and SBS2 sites, distant by 31 bp, were found associated to the 7 bp ACTACAA motif which lies immediately 5′ to the SBS. This motif, under our conditions, is apparently not involved in transcription activity of the BUB1B promoter since we observed that the 5′ deletion eliminating the ACTACAA sequence did not affect promoter activity. Scrutiny of the sequences flanking the SBS in snRNA and snRNA-type genes did not reveal the presence of an ACTACAA-associated motif. In contrast, it is found in the SBS characterized in the SYBL1 promoter () and in 58% of the SBS that we recently identified by a genome scale analysis (). The finding that the SBS are essential for expression of the BUB1B gene raises the interesting question of whether hStaf/ZNF143, the main activating factor of the Staf family, is involved alone in transcription activation of the BUB1B gene. Alternatively, could ZNF76 also play a role in the expression of this gene? Although the ChIP assay was performed with antibodies recognizing specifically hStaf/ZNF143, this does not exclude the possibility that ZNF76 can also be involved in BUB1B expression. The other two -acting motifs that we identified are localized in the region covering the transcription start site and we showed that they function as cell cycle dependent repressor elements. They harbor high identity with the consensus sequences of the CDE (G/CGCGG) and CHR (TTGAA) elements identified in the cell cycle regulated promoter genes such as CDC25, CDC2, CCNA2 (coding for cyclin A), PLK1 (polo-like kinase) and RB6K (rabkinesine 6) (). These two elements are known to induce repression of transcription and we established that their mutation led to almost complete impairment of the cell cycle dependent transcription activity of the BUB1B promoter. Worth of note, however, the tandem repressor element is organized in the CHR-CDE configuration in the BUB1B gene whereas the arrangement CDE-CHR occurs in all the other identified repressors. Careful inspection of the promoter sequences of other genes known to be regulated with a tandem of CDE and CHR repressor elements pointed to the presence of SBS in the promoters of the RBL2 (p130 protein), PLK1 (polo-like kinase protein) and BIRC5 (survivin protein) genes (,,,) (). Studies on genes that are cell cycle regulated by the CDE-CHR tandem of repressor elements (CDC25C, CDC2 and aurora A) revealed that ubiquitously expressed transcription factors such as Sp1, NF-Y and E4TF1 act positively by binding to target sequences upstream of the CDE-CHR motifs (,,). The cell cycle dependent repression of upstream activators via the CDE and CHR elements has been established as the major regulatory mechanism (,). As shown by genomic footprinting on these promoters, both CDE and CHR elements are protected in a periodic fashion, suggesting involvement of a specific factor. A binding activity termed CDF-1 has been identified and proposed to interact with both CDE and CHR in the CDC25C gene promoter (). In a further work, a factor (CHF) interacting specifically with CHR has been isolated in the CCNA2 gene promoter, encoding cyclin A (). In the situation described for the CDC25C promoter, it was proposed that CDF-1 presumably functions by specifically repressing the NF-Y mediated activation (). However, no further characterization of the CDF-1 and CHF complexes was undertaken. Recently, a very interesting mechanism was reported to involve the CDE and CHR elements in chromatin remodeling of the CCNA2 promoter. In this case, the CDE-CHR tandem which is needed for repression in quiescent cells, was also demonstrated to be required for organizing the chromatin structure specific for the inactive promoter (). However, this particular nucleosome organization becomes disorganized as a consequence of mutations in the CDE-CHR repressor element. In the case of the BUB1B promoter, we have clearly demonstrated the functionality of the CDE-CHR tandem as a repressor element, but the precise underlying mechanism remains to be investigated.

Serpins are the largest and most widely distributed family of protease inhibitors (). The native fold of inhibitory serpins is metastable and conformationally labile (). Following interaction with a target protease, the serpin molecule is cleaved within the reactive centre loop (RCL) and the molecule switches to a more stable ‘cleaved' conformation; typically, the for cleaved serpins is greater than 120°C, compared with less than 60°C for the native state (). This conformational rearrangement results in the protease being trapped, at the acyl intermediate stage of the catalytic cycle, in a distorted conformation (; ). The serpin conformational change is commonly termed the stressed-to-relaxed (S to R) transition (; , ) and involves a change in topology, with the RCL forming an additional, fourth β-strand () in the A β-sheet in the cleaved state. Certain serpins are able to undergo the S to R transition in the absence of RCL cleavage to form the ‘latent' conformation, which represents the most stable monomeric conformation of the serpin chain (; ; ). Most notably, plasminogen activator inhibitor-1 (PAI-1) folds to the native state, but in the absence of the cofactor vitronectin spontaneously converts to the latent inactive state (). This represents an elegant mechanism that controls the inhibitory activity of this serpin. It is unclear why serpins fold to a native metastable state and do not fold to the latent state. Similarly, the molecular mechanism of spontaneous conformational change of disease-linked variants of serpins remains to be fully understood (). We have begun to investigate this problem by studying a group of serpins from thermophilic prokaryotes. These molecules are able to function as normal inhibitory serpins, but have developed strategies to fold and function at high temperatures (; ). Here, we investigated the structure of the serpin, tengpin, from the extremophilic prokaryote (). Tengpin contains a serpin domain preceded by a 56-amino-acid amino-terminal region of unknown structure and function ( online; ). Attempts to express full-length material were unsuccessful and resulted in small amounts of insoluble material; therefore, we initially expressed two constructs: tengpinΔ51 that represents the serpin domain alone—that is, lacking the N-terminal region—and tengpinΔ31 that includes 20 amino acids of the N terminus. Bioinformatic analysis suggested that tengpin would be expected to function as an authentic protease inhibitor (). Inhibitory data showed that tengpinΔ31 was an effective inhibitor of the chymotrypsin-like protease human leucocyte elastase (SI=2.1, =1.35 × 10 M s) and formed the SDS-stable complex typical of native serpins ( online). By contrast, we were unable to measure any inhibitory activity of tengpinΔ51 against a range of target proteases (data not shown). Furthermore, biophysical studies showed that tengpinΔ51 did not undergo thermal denaturation even in the presence of a denaturant, suggesting that it is in the latent conformation. Thus, the N-terminal region seems to have a crucial role in the folding and inhibitory activity of tengpin. To understand the structural basis of these contrasting characteristics, we determined the crystal structures of the tengpinΔ31 and tengpinΔ51 constructs. The 2.7 Å crystal structure of tengpinΔ31 shows two molecules in the asymmetric unit. Both molecules are essentially identical with the exception of minor differences in the RCL, indicating the flexibility of this region ( online). The overall structure of the molecules adopts a ‘partly inserted' native serpin conformation (), in which a gap in the β-sheet hydrogen bonding at the top of strands s3A and s5A allows the insertion of two residues of the RCL (). By contrast, the 1.6 Å crystal structure of tengpinΔ51 shows that the molecule adopts the latent, inactive conformation (; ). The structure of tengpinΔ31 shows that the N terminus adopts an extended conformation spanning the D- and E-helices, followed by a short β-strand (residues 38–41) termed S1A (). With the exception of the first four residues of the N terminus, all residues are resolved in the electron density ( online). N-terminal to s1A′, Q37 and A38 pack against the F-helix and form hydrogen bonds with D182. In total, residues 37–56 of tengpinΔ31 make 20 hydrogen bonds and 37 van der Waals contacts, and bury 170 Å accessible surface area at the interface with the serpin domain ( online; ). A structural comparison of the native and latent conformations of tengpin shows that strands s3A, s2A and s1A, together with the E- and F-helix, shift to accommodate the RCL as a fourth strand in the A β-sheet ( online). Furthermore, substantial conformational change in strands s3C and s4C is apparent as a result of the transition to the latent state and the repositioning of s1C ( online). Together, these structural data explain the lack of inhibitory activity of tengpinΔ51 and indicate that the additional 20 amino acids at the N terminus of tengpinΔ31 have a crucial role in maintaining the metastable native state. We investigated whether the N terminus of tengpin is required for initial folding to the native state or to maintain the serpin in a native conformation. Equilibrium refolding of tengpinΔ31 and tengpinΔ51 shows a two-state transition with midpoints centred around 1 M guanidinium thiocyanate (). Critically, both of the refolded tengpin constructs—rΔ51 and rΔ31—were able to inhibit target proteases (). However, monitoring the inhibitory activity over time at 37°C showed that rΔ51 rapidly and spontaneously lost inhibitory activity (∼5 h; ). By contrast, minimal loss of inhibitory activity was observed for rΔ31 (>800 h; ). Together, these data suggest that the N terminus of tengpin is not required for initial folding to the native state, but is required to stabilize and maintain the native conformation, thus preventing the subsequent folding of the molecule into the inactive latent conformation. Next, we investigated whether the N-terminal region in isolation could perform the same function and stabilize the native state of tengpinΔ51. We refolded denatured tengpinΔ51 in the presence of a peptide, corresponding to residues 39–51 of the N-terminal region (Ac-ANLMDRIKANPVS), and monitored the inhibitory activity of the refolded material. Our data showed that, although approximately 30% of tengpinΔ51 lost activity, the peptide was able to maintain the native state of approximately 70% of tengpinΔ51 (). It is unclear why 30% of the refolded material was not stabilized by the peptide. We suggest that this is most probably the result of competition between the rate of peptide binding and the rate of conformational change during folding or the transition to the latent state. However, these data strongly indicate that the peptide forms an analogous interaction with the body of the molecule, which prevents tengpinΔ51 from undergoing the transition to the latent conformation. To define crucial interactions made by the N terminus that are important for tengpin metastability, we subjected the N terminus to a combination of truncation and mutagenesis. We constructed nine tengpin mutants and investigated whether each construct adopted the native or latent conformation (). We used three criteria to determine the conformation: (i) we were able to distinguish between native and latent material by using phenyl-epharose chromatography (tengpinΔ31 and tengpinΔ51 eluted at approximately 1.1 M and approximately 0.35 M of ammonium sulphate, respectively; online); (ii) native tengpin unfolds completely in 6 M guanidine hydrochloride, but latent tengpin does not; and (iii) all proteins were tested for inhibitory activity and ability to form SDS-stable complexes with elastase, and the half-life of each variant was calculated. Initially we truncated the molecule from the N terminus; these data showed that it was possible to remove the N-terminal sequence up to but not including amino acid N40 (tengpinΔ39) and form a stable native conformation (tengpinΔ39 =594 h; ; ). Examination of the structure shows that N40, L41 and M42 make substantial interactions with the body of the serpin (). Indeed, mutation of any one of these three residues in tengpinΔ39 resulted in more rapid formation of the inactive, latent state ( of the native state 25–30 h; ; ). The side chain of D169 forms a hydrogen bond with the ND2 atom of N40 (; online); therefore, we were able to define further interactions made by N40 by generating D169A. Analysis of the mutant protein tengpinΔ39 shows that this variant adopts the native conformation (=526 h; ; ). Together, these data define the minimum contacts required to prevent native tengpin folding to the latent conformation. Furthermore, it is possible to estimate the additional interactions made between the body of the serpin and the N terminus in tengpinΔ39 (native) in comparison with those present in tengpinΔ39 (latent): these comprise one hydrogen bond and three van der Waals interactions. Mutating L41 and M42 to alanine, thus truncating the side chains of these residues, would be predicted to result in the loss of two and four van der Waals interactions, respectively. Together the three residues form a cap protecting the underlying hydrophobic core that includes I162, L159 and I170. A structural comparison of the native and latent state shows that, following the S to R transition, strand s1A of tengpinΔ51 adopts a position similar to the region occupied by the N terminus in tengpinΔ31. In particular, I170 moves to partly cover I162 and L159 (). We were interested in probing further the role of the N-terminal region in stabilizing the native state. We reasoned that mutations of hydrophobic residues ordinarily covered by the N terminus might abrogate the requirement to undergo transition to the latent state. Thus, we mutated the three hydrophobic residues (I162, L159 and I170) contacted by L41 and M42 of the N terminus. The mutations were made in tengpinΔ51 and all three residues were changed to the polar, uncharged residue glutamine. Interestingly, these mutations did not abolish conformational change, although the native state of tengpinΔ51 was substantially stabilized (=60 h) in comparison with tengpinΔ51 (=5 h) (). Our data show that, despite its extremophilic source, the serpin domain of tengpin readily undergoes conformational change to the latent state. We have shown that the N terminus of tengpin functions to trap the serpin domain in the native metastable state and to prevent the spontaneous transition to the latent conformation. The function of the N terminus of tengpin is thus strikingly similar to the role of the plasma protein vitronectin in stabilizing the metastable state of the mammalian PAI-1 (). Furthermore, similarly to the crucial residues 40–42 of the N terminus of tengpin, structural studies have shown that the somatomedin B domain of vitronectin binds to strand s1A of serpins (; ). However, the N terminus of tengpin does not adopt the same fold as that of the somatomedin B domain of vitronectin, and therefore this is consistent with convergent, rather than divergent evolution. Studies on mammalian serpins have shown that numerous mutations causing conformational disease (or serpinopathies; ) localize on or around a mobile ‘trigger point' in the central portion of the molecule, commonly termed the shutter region (see ). Our structural, mutagenetic and biophysical data extend these studies and show that in tenpgin exposure of a relatively small hydrophobic patch on the surface of the serpin domain, approximately 20 Å from the centre of the shutter, seems sufficient to promote conformational rearrangement. Consistent with this hypothesis, mutations in the hydrophobic patch can at least partly compensate for the lack of the N-terminal region and slow the transition to the latent state. Furthermore, we show that it is possible to stabilize the native state in the absence of the N terminus studies by using an exogenous peptide. Together, our research supports the therapeutic strategies that aim to prevent conformational change in mammalian serpins by targeting hydrophobic cavities in the mobile region of the molecule (). The genomic DNA of was obtained from the Beijing Genomic Institute (Chinese Academy of Sciences, China; ). Details of the cloning, mutagenesis, expression, purification, peptide-binding studies, kinetic characterization, stability measurements and crystallization of tengpin are given in the online. Data were collected from cryo-cooled crystals at 100 K at the BIOCARS and IMCA-CAT beamlines at the Advanced Photon Source (Chicago, IL, USA). Structure elucidation was carried out using , unless stated otherwise. The structure of tengpinΔ51 was determined by molecular replacement using AMORE () and the structure of native thermopin (1SNG) as a search model. The structure of tengpinΔ31 was determined by molecular replacement using PHASER () and an ensemble search probe consisting of structurally aligned molecules of thermopin (1SNG; ) and tengpinΔ51. Tengpin contains two molecules in the asymmetric unit. The limited resolution of the data necessitated the use of strict non-crystallographic symmetry restraints throughout refinement; however, by using the as a guide, we were able to model small differences between the RCL of each molecule by loosening restraints in this region. Structure refinement and building proceeded using the suite, REFMAC () and O (). Final refinement statistics () for tengpinΔ31 and tengpinΔ51 are /=26.0/21.2% and /=25.1/21.3%, respectively. Structures were superimposed using the program MUSTANG (). Accessible surface areas were calculated using the CCP4 program AREAIMOL. Figures were produced using PYMOL (Delano Scientific Pty Ltd, San Diego, CA, USA). Coordinates have been deposited in the RCSB Protein Data Bank (; identifiers 2PEE and 2PEF). is available at online ().

In the last few years, there have been significant findings that have helped to understand the molecular mechanisms of eukaryotic DNA replication; however, the identity of the complex that unwinds DNA has remained elusive. Several lines of evidence provide support for the idea that the minichromosome maintenance mutant (MCM)2–7 hexamer constitutes part of the replicative helicase (; ). Interestingly, the purified MCM2–7 complex does not show helicase activity , whereas a subcomplex of Mcm4, Mcm6 and Mcm7 presents modest activity with low processivity (). Therefore, it seems that other factors have important roles in the initiation and elongation processes of DNA replication, working as replicative helicase cofactors. Cell-division cycle protein (Cdc)45 is one of these crucial factors that participates in both initiation and elongation (; ). Cdc45 interacts with several DNA replication proteins, including origin recognition complex subunit 2 (Orc2), MCM2–7, Replication Protein A (RPA), DNA polymerases (; ), synthetic lethality with dpb11-1 () and Mcm10 (). In addition, antibodies against Cdc45 disrupt DNA unwinding in a replication assay carried out in cell-free extracts (). Recently, it has been reported that phosphorylation of Mcm4 by the S-phase promoting kinase Cdc7-Dbf4 (Dumb bell former 4) facilitates the formation of a stable Cdc45–MCM complex at the origins of replication (). The interaction between MCM2–7 and Cdc45 is maintained at the DNA replication forks by means of the four-subunit GINS complex (; ). GINS was first described in yeast as a result of genetic analyses aimed at the discovery of proteins that interact with DNA polymerase B possible subunit 11 (Dpb11) (). The complex is comprised of four conserved proteins—Sld5, Psf1 (Partner of Sld5), Psf2 and Psf3—each distantly related to each other and with no known folding motifs. Three of them were discovered independently by a functional proteomics approach on the basis of induced proteolysis (). The GINS complex is essential for initiation of DNA replication and the normal progression of the replisome (; ; ). Previous electron microscopy images of rotary-shadowed GINS complex suggested ring-like or C-shaped structures (). In budding yeast, chromatin immunoprecipitation studies have shown that GINS is recruited at the paused replication fork together with Mediator of replication checkpoint 1 (Mrc1), Topoisomerase interacting factor 1 (Tof1), polymerases α and ɛ, Cdc45 and the MCM2–7 complex (). GINS, Cdc45 and the MCM2–7 could form the core of a large complex known as the ‘replisome progression complex' (). So far, the physiological role and the biochemical features of the GINS complex are poorly understood. A suggestion comes from the recent purification of a stable high-molecular-weight complex formed by Cdc45/MCM2–7/GINS (CMG) from that shows helicase activity (). Conversely, the interaction of the human GINS complex with DNA polymerase α-primase seems to stimulate its activity (). Here, we have characterized the three-dimensional structure and DNA binding of a recombinant human GINS complex. By using single-particle electron microscopy and three-dimensional reconstruction, we have obtained a medium-resolution volume of the human GINS complex showing its horseshoe shape. The arrangement of the subunits in the structure was shown using a combination of mass spectrometry of the intact complex and subcomplexes generated in solution or gas phases, and monoclonal antibody mapping using electron microscopy. The DNA-binding preferences of GINS have been also studied. The three-dimensional structure, in conjunction with DNA-binding experiments, suggests the possible role of GINS in the CMG helicase complex. The open reading frames of the Sld5, Psf1, Psf2 and Psf3 proteins were cloned in a T7 promoter polycistronic vector. The recombinant protein complex was isolated in three steps by using affinity, anion exchange and gel filtration chromatography (see the Methods and online). SDS–polyacrylamide gel electrophoresis of purified recombinant human GINS complex () showed four bands identified as its subunits by mass spectrometry (data not shown). Analytical ultracentrifugation ( online) and nano-flow mass spectrometry of the intact complex () showed that human GINS is a heterotetramer with 1:1:1:1 stoichiometry. The molecular mass of the intact complex, measured by mass spectrometry, showed a mass of 98,373±12.7 Da, which is in close agreement with the theoretical value (98,122.0 Da) of the complex, with a Tobacco etch virus (TEV) cleavage site and a His-tag. To address the binding of the purified human GINS complex to DNA, different DNA probes resembling several replicative structures were analysed by using electrophoretic mobility shift assays (EMSA; ). Human GINS showed a clear preference for the probes consisting exclusively of single-stranded DNA (ssDNA) or containing stretches of ssDNA (‘ssDNA', ‘3′ end', ‘5′ end' and ‘bubble') than for a probe consisting of only double-stranded DNA (dsDNA; ). Remarkably, a supershift was observed with the ‘bubble' probe. This could be caused by the loading of more than one human GINS complex on each ssDNA region. These results represent the first experimental evidence that human GINS can associate directly with DNA and indicate its role within the CMG helicase (see below). The human GINS complex was applied to carbon-coated grids and negatively stained with uranyl acetate. Despite the low molecular mass of human GINS complex for electron microscopy analysis, a clean distribution of single particles was observed (; for details, see the online). The refined volume of the human GINS complex at 33 Å resolution shows a horseshoe shape. The approximate molecule dimensions are 130 × 60 × 80 Å (). The complex shows a central hole of 30–35 Å in diameter, which is large enough to accommodate either dsDNA or ssDNA. The upper part of the three-dimensional volume is wide open, whereas the opposite side of the central hole is narrower. Hence, the central hole is arranged in a manner similar to a funnel with an upper diameter of approximately 70 Å and a bottom diameter of approximately 25 Å (; online), indicating the possibility of different functions for each side of the complex. Although the human GINS three-dimensional structure forms an open ring, the shape of the volume resembles the structure of proliferating cell nuclear antigen (PCNA)—an essential processivity factor for DNA polymerases ( online). The different human GINS subunits could not be identified in the electron microscopy three-dimensional structure owing to the limited resolution, therefore a combined approach of mass spectrometry and monoclonal Fab labelling was used to show the subunit organization. Mass spectrometry of the intact human GINS complex showed the heterotetrameric oligomerization state of the complex (). Interestingly, the Psf2 subunit readily dissociated on activation and tandem mass spectrometry (MS/MS), indicating that Psf2 has fewer intersubunit contacts and is likely to locate at one end of the horseshoe-shaped structure (, inset). Interactions between the subunits in the human GINS heterotetramer were determined by generating subcomplexes using in-solution perturbation and gas-phase dissociation of the resulting complexes (). After the addition of 42% methanol, two additional charge state series were observed (). The measured masses (47,758 and 70,895 Da) indicate that the two series correspond to the Psf2–Sld5 heterodimer and a Psf2–Sld5–Psf1 heterotrimer, respectively. As Psf2 is located at one end and it interacts with Sld5, the Psf1 subunit should be located on the opposite site of the Psf1/Sld5/Psf2 heterotrimer. Thus, a model of the subunit organization in the complex comprises a central core formed by Sld5 and Psf1, and Psf2 and Psf3 are located at the tips of the horseshoe (). This arrangement is in agreement with the network of interactions of the GINS subunits proposed in yeast using genetic and two-hybrid methods (). On the basis of the interactions observed by mass spectrometry and the restrictions imposed by the subunit organization inside the three-dimensional structure, our model of the human GINS architecture could be confirmed by localizing Psf2 within the complex. Thus, the human GINS complex was incubated with a monoclonal Fab fragment that recognizes Psf2, and the human GINS–Fab complex was purified ( online). To obtain the three-dimensional structure of the human GINS–Fab, the purified complex was negatively stained and analysed by using electron microscopy (). A total of 2,000 images were selected and processed similarly to the volume representing the human GINS complex alone (without Fab). The resultant three-dimensional volume () resembles the human GINS structure and shows an additional mass on one tip of the horseshoe-shaped human GINS structure, which corresponds to the size and shape of a Fab molecule (). This result confirms the localization of Psf2 at one end of the structure and, combined with the mass spectrometry data, supports the proposed model of the organization of human GINS subunits within the complex. A certain parallel could be drawn between the structures of human GINS and PCNA (). Indeed, it has been proposed recently that GINS binds to and enhances the activity of DNA polymerase α-primase (). However, we believe that the structural similarities are not sufficient to indicate that GINS, as PCNA, has the characteristics of a DNA processivity factor. First, the dimensions of PCNA (90 × 40 × 90 Å) are smaller than the human GINS complex (see previous section), according to the number of components and their molecular weight. Second, the PCNA structure is a closed ring, whereas human GINS is an open ring. Third, although the internal diameter of the central hole has similar dimensions of around 30–35 Å in both, the PCNA internal channel does not show an internal funnel-like shape similar to that observed in human GINS (; online). Furthermore, the EMSA assays () indicate that human GINS does not show preferential binding to dsDNA, which is the molecule bound by PCNA during DNA replication (). Finally, PCNA possibly does not change its overall conformation on DNA binding. This might not be the case for human GINS and a conformational change induced by DNA binding could occur. The structure suggests that DNA binding might promote a more compact complex to embrace the nucleic acid. An attractive idea is that the biochemical function of GINS resides within the recently described CMG complex consisting of Cdc45, GINS and the MCM2–7 hexamer. All the components of the CMG are present at DNA unwinding sites (; ), and a purified CMG complex from shows ATP-dependent helicase activity (). The association of the MCM2–7 hexamer with these two cofactors seems to stimulate DNA unwinding and strand displacement activities, which have been predicted and experimentally sought for the MCM2–7 hexamer for a long time (). The need for essential activators of the helicase activity represents a change in the model about the mode of action of eukaryotic replicative helicases and could help to explain the delay between the assembly of the MCM2–7 complexes on the chromatin during late telophase/early G1 and the initiation of DNA replication several hours later (). Previous models on the eukaryotic replicative helicase function, based on steric exclusion (; ) or rotary pumps (; ), were focused on the MCM2–7 complex as the unique assembly responsible for the unwinding and strand displacement activities. On the basis of the described association of GINS with the MCM2–7 complex and Cdc45 to form a molecular machine that unwinds DNA (), and on our observation that purified human GINS shows preferential binding for DNA structures containing ssDNA, it is tempting to speculate about the possible role of GINS after its association with the other components of the CMG complex. We foresee two main possibilities (). In both cases, the MCM2–7 complex would work as an engine to unwind the dsDNA coupled to ATP hydrolysis and GINS as a crucial structural element required for the successful separation of the two DNA strands. In the first model (), MCM2–7 pumps dsDNA through its inner channel by helical rotation, destabilizing the double helix. Hence, the GINS complex would function as a strand displacement blade, or ‘ploughshare' (), located where unwound DNA exits from the MCM hexamer, preventing re-annealing and providing room for the activity of the polymerases. In the second model (), GINS would be located in front of the MCM2–7 complex and would have a more active role in DNA unwinding. The main difference is that, in this case, only one strand of DNA goes through the MCM2–7 inner channel. This model would share more structural features with the recently proposed mode of action of the MCM4–6–7 helicase () and the viral E1 helicase (). The two hypothetical models in represent two alternatives of cooperation between a motor engine formed by MCM2–7 and a ‘strand displacement unit' provided by GINS, but other variations could also be envisioned. So far, no structural information on Cdc45 is available and its position between GINS and MCM2–7 is speculative. However, it is worth noting that an immunoprecipitation with anti-Cdc45 was the original method to isolate the CMG complex (). Full protocols are available in the online. The complementary DNAs of the human GINS subunits were cloned in a polycistronic vector and expressed in Rosetta (DE3) cells (Novagen, Madison, WI, USA). Transformed cells were grown in lactose broth medium supplemented with ampicillin (at 100 μg/ml) and chloramphenicol (at 34 μg/ml). The cells were induced with 0.5 mM isopropyl-β--thiogalactopyranoside (IPTG) overnight at 16°C. The recombinant human GINS complex was isolated using nickel affinity, anion exchange and gel filtration chromatographic steps. Fractions containing human GINS after the gel filtration were pooled, concentrated and stored at −80°C in small aliquots. Different DNA structures were obtained by hybridization of the P-labelled 60-mer oligonucleotide. Protein–DNA binding reactions were carried out by incubating recombinant GINS complex (1–10 pmol) with 150 fmol of each probe in buffer B (50 mM Tris–HCl (pH 8.0), 100 mM NaCl, 0.5 mM EDTA, 10% glycerol and 1 mM dithiothreitol) at 25°C for 30 min; protein was always added last. After incubation, the mixtures were resolved by in 5% polyacrylamide–TBE non-denaturing gel electrophoresis. Gels were dried and exposed to autoradiography. For negative staining, a few microlitres of purified human GINS complex and its anti-Psf2-Fab bound complex were diluted to an appoximate concentration of 0.1 and 0.2 mg/ml, respectively. Samples were applied to glow-discharged carbon-coated copper–rhodium grids, negatively stained with 2% uranyl acetate (w/v) and observed in a JEOL 1230 electron microscope at an accelerating voltage of 100 kV. The human GINS complex images were recorded under a low-dose condition at a nominal magnification of × 60,000, and images of a human GINS–Fab complex were taken at × 25,000. Good micrographs were digitized in a Dimage Scan Multi Pro scanner (Minolta, Osaka, Japan) at 2,400 d.p.i. and averaged to a final 3.56 Å/pixel at the specimen for human GINS complex and 4.2 Å/pixel at the specimen for human GINS–Fab complex. Mass spectra collected for the intact protein complexes were recorded on a QSTAR XL mass spectrometer (MDS Sciex, Concord, Canada) modified for high-mass detection (). The human GINS complex (1 μg/μl) was exchanged into 300 mM ammonium acetate (pH 7.5) by using microbiospin-6 columns (Bio-Rad Laboratories, Hercules, CA, USA), and 2 μl aliquots were introduced by gold-coated nanoflow capillaries prepared in-house. The conditions within the mass spectrometer were adjusted to preserve noncovalent interactions (). The mass spectrometer was operated at a capillary voltage of 1,200 V and a declustering potential of 40 V. An MS/MS spectrum of the intact human GINS complex was obtained by MS/MS of an isolation at 4,685 with collision energy of 100 V. The intact human GINS complex was disrupted through the stepwise addition of methanol up to 42% (v/v) and MS/MS of the resulting subcomplex was carried out at collision energy of 80 V. The structure of this complex has been deposited at the European Bioinformatics Institute, with the unique accession code EMD-1355. is available at online (). Following the submission of this paper, the crystallographic structure of a truncated mutant of the human GINS complex has been published (Kamada K : 388–396). It is interesting to note that although the intersubunit interactions are similar in both studies, the overall conformation described by Kamada is different from our structure.

Chylomicron remnants (CMR) carry lipids of dietary origin from the gut to the liver for processing and there is now a large and growing body of evidence indicating that these lipoproteins are strongly atherogenic. They have been shown to be taken up into the artery wall as efficiently as LDL ; remnant-like particles containing apolipoprotein E (apoE) have been isolated from human aortic intima and atherosclerotic plaque ; and delayed clearance of CMR from the circulation correlates with lesion development . Moreover, we and others have shown that CMR induce extensive lipid accumulation causing foam cell formation in human monocyte-derived macrophages (HMDM) and in human and murine monocyte/macrophages cell lines . Low density lipoprotein (LDL) plays a major role in atherogenesis and in foam cell generation, but oxidation of the lipoprotein particles, a process which can occur within the artery wall, is necessary before extensive lipid accumulation is induced . In striking contrast, CMR do not require prior oxidation to cause macrophages to form foam cells . However, our studies have demonstrated that incorporation of lipophilic antioxidants into the particles enhances, rather than inhibits, lipid uptake and accumulation in the cells , suggesting that the oxidative state of CMRs may play a role in their induction of foam cell formation, but in the opposite way to that of LDL. Oxidized CMR could occur either in the artery wall by the action of the cell-associated lipoxygenase and myeloperoxidase which are believed to oxidize LDL, or in the circulation, because dietary oxidised lipids, which are produced when fat is cooked at high temperatures, have been shown to be transported in these lipoproteins . Clearly, therefore, it is important for the understanding of the atherogenicity of CMR to establish how their oxidation influences their uptake and induction of foam cell formation and the pathways by which CMR are internalised by the cells. It has been demonstrated that CMR are taken up by the liver by apolipoprotein E (apoE)-dependent pathways mediated by the LDL receptor (LDLr) and the LDL receptor-like protein (LRP) . The exact mechanisms by which CMR are taken up by macrophages, however, are not yet definitively established, and nothing is known about the effects of oxidation of the particles on the routes by which they are internalised. The LDL receptor (LDLr) appears to play a part , but as it is down-regulated by the influx of cholesterol into cells, native LDL does not induce foam cell formation , and our studies have suggested that the delivery of cholesterol to macrophages by CMR has a similar effect. Thus, other mechanisms are also likely to be involved, and evidence from experiments using antibodies to the LDLr and animals lacking the LDLr supports this view . Candidates include the LRP , the apoB48 triacylglycerol-rich lipoprotein receptor (apoB48r) , an as yet unidentified 43 kDa protein described by Elsegood et al. and scavenger receptors such as scavenger receptor A (SR-A) and CD36 . Phagocytosis has also been suggested as a possible mechanism . Since the route of uptake of LDL by macrophages is profoundly changed after oxidation from the regulated LDLr to the unregulated scavenger receptors , it is important to establish whether the oxidative state also alters the mechanisms of uptake of CMR by the cells. The aim of this study is to investigate the effects of the oxidative state of CMR on their uptake by macrophages and on the accumulation of lipid within the cells, and to determine how oxidation affects the pathways involved in the internalisation of the particles. Chylomicron remnant-like particles (CRLPs) at three different levels of oxidation (CRLPs, oxidized CRLPs (oxCRLPs) and CRLPs containing the antioxidant probucol (pCRLPs)) and macrophages derived from the human monocyte cell line THP-1 were used as the experimental model, and the mechanisms of uptake were evaluated using specific inhibitors of the processes believed to be involved. The findings clearly demonstrate that oxidation of CRLPs reduces the rate of their uptake by THP-1 macrophages and decreases lipid accumulation in the cells, and further show that this is due to differential interaction with apoE dependent receptors. The lipid composition of CRLPs, oxCRLPs and pCRLPs is shown in . The concentration of TG and TC in CRLPs and pCRLPs was similar, but that of oxCRLPs was a little lower, reflecting the dilution of the preparations during the oxidation procedure. The mean TG:TC ratio, however, was between 5 and 6 for all 3 CRLP types. Previous work in our laboratory has shown that the phospholipid content of CRLPs, oxCRLPs and pCRLPs prepared by the methods used here are not significantly different . Analysis of the TBARS content of the particles showed that there were significant differences in their oxidative states, with values being higher in oxCRLPs as compared to CRLPs, while those for pCRLPs were lower (one way ANOVA, Bonferroni's test post hoc) (). The apoE content of CRLPs, oxCRLPs and pCRLPs was evaluated by SDS PAGE electrophoresis. The results showed that the particles contained apoE, but no apoE was detected in the top fraction of the 1.063–1.21 g/ml fraction of human plasma incubated and centrifuged in the absence of lipid particles (A), indicating that the apoprotein was bound to the CRLPs during the incubation. No apoCs or other apolipoproteins were detectable. There were no significant differences in the apoE content of the CRLPs, oxCRLPs or pCRLPs as assessed by optical density volume analysis (B). Incubation of THP-1 macrophages with CRLPs or oxCRLPs for 5,24 or 48 h caused a marked increase in the total lipid found in the cells (A) at all time points, but lipid accumulation was greater with CRLPs as compared to oxCRLPs ( < 0.001). This effect was mainly due to greater accumulation of TG in the presence of CRLPs ( < 0.001) (B), as the trend for an increase in TC in CRLP- as compared to oxCRLP-treated macrophages did not reach significance (C). The uptake of DiI-labelled CRLPs, oxCRLPs and pCRLPs by THP-1 macrophages was assessed by confocal microscopy and FACS analysis (). Examination of cells incubated with DiI-labelled particles for periods up to 24 h with the confocal microscope showed that the fluorescence associated with the cells increased with time in all cases, but that there was clearly more in pCRLP- and less in oxCRLP-treated as compared to CRLP-treated cells (A). Quantification of the cell-associated fluorescence and analysis by ANOVA repeated measures showed that the rate of uptake of CRLPs over 24 h was significantly higher than that of oxCRLPs ( < 0.01) and significantly lower than that of pCRLPs ( < 0.01) (B), and a similar result was obtained using FACS ( < 0.05, CRLPs vs. oxCRLPs or pCRLPs (C). To investigate whether apoE is necessary for uptake of CRLPs, THP-1 macrophages were incubated with DiI-labelled CRLPs or CRLPs without apoE for 2 h and the fluorescence associated with the cells was determined by FACS analysis. In the absence of apoE, the uptake of CRLPs was reduced by about 90% (fluorescence values: CRLPs, 98.6 ± 7.1; CRLPs without apoE, 8.4 ± 0.9,  = 3). The potential role of apoE dependent receptors in the uptake of CRLPs by the cells was studied using excess LDL and the LRP ligand, lactoferrin, to block entry via the LDLr and LRP, respectively. The effects of lactoferrin (2 mg/ml) on lipid accumulation in macrophages exposed to CRLPs or oxCRLPs as assessed by Oil red O staining are shown in . Lactoferrin had no effect on the lipid content of cells incubated in the absence of CRLPs, but reduced that in macrophages treated with either CRLPs or oxCRLPs ( < 0.05, both cases). When the effects of excess LDL (200 μg cholesterol/ml) and lactoferrin (2 mg/ml), on the fluorescence associated with the macrophages after incubation for 1, 4, 16 and 24 h with CRLPs, oxCRLPs or pCRLPs were assessed by confocal microscopy (A–C), no significant inhibition was observed in the presence of excess LDL, but lactoferrin caused a decrease of > 90% in experiments with CRLPs (A) and oxCRLPs (B) and > 80% with pCRLPs (C) ( < 0.0001, all cases). Addition of LDL and lactoferrin together completely abolished detectable uptake of all three types of particles (A–C). FACS analysis of macrophages treated with CRLPs, oxCRLPs or pCRLPs in the presence or absence of lactoferrin (2 mg/ml), LDL (300 μg cholesterol/ml) or LDL + lactoferrin for 2 h also showed that lactoferrin caused a marked decrease of about 75% in the uptake of all three types of CRLPs ( < 0.01), and in this case significant inhibition (− 30–40%) was observed in the presence of LDL ( < 0.05) (D). Moreover, in a further separate experiment in which the excess of LDL added was increased by lowering the concentration of CRLPs to 10 μg cholesterol/ml and raising that of LDL to 500 μg cholesterol/ml, the uptake of both CRLPs and oxCRLPs was also significantly inhibited (fluorescence values ( = 3): CRLPs, 94.8 ± 6.2; CRLPs + LDL, 58.7 ± 5.8 ( < 0.05); oxCRLPs 56.7 ± 1.5; oxCRLPs + LDL, 29.2 ± 7.3 ( < 0.05)). As observed in the experiments using confocal microscopy, adding both LDL and lactoferrin to the incubations caused a further decrease to similar minimal levels in the uptake of all three types of CRLPs. Assessment of the fluorescence associated with THP-1 macrophages after incubation with DiI-labelled acetylated LDL (acLDL) (10 μg cholesterol/ml) for 2 h showed that acLDL uptake, in contrast to that of CRLPs, was not significantly inhibited by excess LDL (500 μg cholesterol/ml) or lactoferrin (2 mg/ml) (). The effects of blocking the class A scavenger receptor SR-A, the class B receptors CD36 and SR-B1 or phagocytosis on the uptake of CRLPs by THP-1 macrophages was investigated using poly-I, a known ligand for SR-A, a blocking antibody to CD36, excess HDL, which binds to SR-B1, and cytochalasin D, which blocks the polymerization of actin microfilaments . Evaluation of the fluorescence associated with the cells after incubation with DiI-labelled CRLPs, oxCRLPs or pCRLPs for 1, 4,16 or 24 h by confocal microscopy (A–C) show no significant change in the presence of poly-I (5 μg/ml) or excess HDL (300 μg/ml). Significant decreases, however, were observed in the presence of anti-CD36 (1 μg/ml) (− 38% after 24 h) ( < 0.05) and cytochalasin D (10 μg/ml) (− 27% after 24 h) ( < 0.01) in experiments with CRLPs (A), although no change in the uptake of oxCRLPs or pCRLPs was detected with these treatments (B, C). FACS determinations after incubation of macrophages with CRLPs, oxCRLPs or pCRLPs for 2 h confirmed that poly-I and HDL did not inhibit the uptake of the particles (D). In contrast, however, the uptake of DiI-labelled acLDL was decreased by about 90% in the presence of poly-I (50 μg/ml) (). As found in the experiments using confocal microscopy, anti-CD36 and cytochalasin D significantly inhibited the uptake of CRLPs, but in this case effects of a similar magnitude were also seen with oxCRLPs and pCRLPs (After 24 h:anti-CD36; CRLPs, − 38%, oxCRLPs − 40%, pCRLPs, − 39%: cytochalasin D; CRLPs − 39%, oxCRLPs − 35%, pCRLPs − 38%), although the changes with oxCRLPs did not reach significance (D). In THP-1 macrophages exposed to CRLPs (30 μg cholesterol/ml) for 24 h, LRP mRNA levels were increased in CRLP- treated THP-1 cells (fold change, 2.7 ± 1.01,  = 3), while LDLr mRNA abundance was decreased (0.27 ± 0.11(range),  = 2). Since homogeneous CMR cannot be obtained easily from human blood without contamination with other lipoproteins of a similar density such as chylomicrons and very low density lipoprotein (VLDL), model CRLPs were used in this study. These particles are similar in size, density and lipid composition to physiological remnants , and also contained human apoE, thus they differ from physiological CMR only in lacking apoB48. Importantly, extensive previous studies in both humans and experimental animals have demonstrated that chylomicron- and chylomicron remnant-like particles without apoB48 are cleared from the blood and metabolised in a similar way to the corresponding physiological lipoproteins . In addition, CRLPs lacking apoB48, but containing apoE from the appropriate species, have been found to have effects which mimic those of physiological remnants in rat hepatocytes and pig endothelial cells , and our earlier work has shown that CRLPs cause extensive lipid accumulation in THP-1 macrophages and human monocyte derived macrophages (HMDM) which is comparable to that found in experiments with physiological CMR from rats and the murine macrophage cell line J774 . For the current investigation, the use of CRLPs facilitated the manipulation of the oxidative state of the particles by inclusion of the lipophilic antioxidant probucol or by exposure to oxidising conditions similar to those used extensively in studies with oxLDL . In previous work, we have shown that incorporation of probucol into CRLPs increases their uptake by THP-1 macrophages , and that this is due to protection of the particles from oxidation, since lycopene, a chemically unrelated antioxidant, has a similar effect . For the present study, therefore, we used pCRLPs in addition to CRLPs and oxCRLPs, so that 3 different oxidative states were tested. Assessment of TBARS in the particles used showed that the oxidative states of CRLPs, oxCRLPs or pCRLPs were significantly different, and there were no significant differences in their apoE content thus they provided a convenient and suitable model for the study. Earlier work in our laboratory and others has demonstrated that CMR cause the extensive lipid accumulation associated with foam cell formation in HMDM, THP-1 macrophages and murine macrophage cell lines . In experiments with THP-1 macrophages and HMDM, we found that CRLPs induced a greater increase in the intracellular total lipid content than oxLDL (at the equivalent cholesterol level), and that, as might be expected, this was mainly due to greater accumulation of TG, while TC levels were raised to a comparable extent with both types of lipoprotein . In the current study we compared the effects of oxidized and non-oxidized particles and, although both caused rises in cellular lipid content compared to untreated cells, the increase was clearly smaller with oxCRLPs as compared to CRLPs, mainly because of decreased accumulation of TG (). These results suggest that the uptake of CRLPs by the cells may be inhibited by oxidation, and this was confirmed in confocal microscopy and FACS studies (), which showed that the rate of uptake of CRLPs was decreased by oxidation and increased by incorporation of probucol into the particles to protect them from oxidation (). These results clearly demonstrate that oxidation of CMR inhibits their uptake by macrophages and attenuates foam cell formation. This is an important new finding, since this effect is strikingly different from that found with LDL, where oxidation is required to induce macrophages to form foam cells. CMR are known to be cleared from the circulation by the liver mainly by apoE-dependent pathways involving the LDLr and the LRP, and the present study shows that apoE also plays an important role in the entry of CRLPs into THP-1 macrophages, since a markedly reduced rate of uptake was observed when apoE was not present in the particles. These finding are in agreement with earlier work which has suggested that both the LDLr and the LRP are involved in the uptake of CMR by macrophages . Our results indicate that both receptors are able to mediate the uptake of CRLPs regardless of their oxidative state, although lactoferrin, a ligand for the LRP, caused a marked reduction in lipid accumulation and uptake after exposure of THP-1 macrophages to all three types of CRLPs (), while the maximum effect of excess LDL on uptake observed was more modest in all cases (). This lesser role of the LDLr may be due to its down-regulation on the influx of lipoprotein into cells. We have shown previously that the expression of mRNA for the LDLr in THP-1 macrophages is decreased by CRLPs while the expression of LRP mRNA is increased , and the present study, which shows a decrease of about 73% in LDLr mRNA and a 2.7fold rise in LRP mRNA after exposure of the macrophages to CRLPs, is in agreement with these findings. The differences in rates of uptake of oxCRLPs, CRLPs and pCRLPs were retained in the presence of inhibitors of both the LDLr and the LRP, although when both the LDLr and the LRP were blocked, uptake of oxCRLPs, CRLPs and pCRLPs was reduced to a similar low level (), suggesting that the difference in their rate of uptake is apoE-dependent. Thus, an important novel finding of this study is that oxidation of CRLPs, unlike LDL, does not change their major routes of uptake by macrophages, with apoE-dependent receptors being of major importance and the LRP playing the predominant role, regardless of the oxidative state of the particles. ApoE does not bind to the LDLr family in its lipid-free state, as interaction with lipid is necessary to induce a conformational change which promotes high affinity for the receptors . In addition, apoE has been shown to adopt different conformations when complexed to different lipids. Thus, changes in the lipid composition of lipoproteins such as an increased content of oxidized lipids or the presence of lipophilic compounds such as probucol may alter the conformation of the protein on their surface, and not all apoE molecules on a particular remnant particle may be able to act as ligands . The different rates of uptake of CRLPs oxCRLPs and pCRLPs by macrophages demonstrated here, therefore, could be explained by effects on interaction with the LDLr and the LRP caused by differences both in the conformation of apoE and in the number of apoE molecules able to bind to the receptors, even though the total amount of apoE associated with the particles is not changed. Although we cannot rule out the possibility that the faster internalization of pCRLPs is due a specific effect of probucol on the surface structure of the particles, the finding that CRLPs are taken up more slowly after oxidation and more rapidly when the antioxidant is present suggests the changes are more likely to be related to the amounts of oxidized lipid in the particles. The expression of scavenger receptors such as SR-A and the class B receptors CD36 and SR-B1 is a characteristic feature of macrophages, and SR-A and CD36 are know to play a part in the induction of foam cell formation by oxidized or chemically modified LDL . It is possible, therefore, that this type of receptor may also be involved the uptake of oxCMR by macrophages. In the present study, poly I, a ligand for SR-A, reduced the uptake of acLDL by THP-1 cells by about 90% as expected, but had no significant effect on the uptake of CRLPs, irrespective of their oxidative state. In addition, excess HDL, which binds to SR-B1, did not prevent the entry of CRLPs, oxCRLPs or pCRLPs into the cells. In contrast, in the presence of anti-CD36, the uptake of all three CRLP types was inhibited by about 35–40% so that the differences in their rates of uptake were retained (). Thus, SR-A and SR-B1 do not appear to play a significant part in the uptake of oxCRLPs by macrophages, although CD36 may have a role which is unaffected by the oxidative state of the particles. Mamo et al. have reported previously that the phagocytosis may be important in the uptake of CMR by rabbit alveolar macrophages, although in a later electron microscopy study with HMDM they found no evidence for entry via this route . In our experiments with THP-1 macrophages, however, cytochalasin D inhibited the uptake of all CRLPs types tested by 35–40%, suggesting that some phagocytosis of CRLPs of all oxidative states does occur in these cells (). It has been suggested that the apoB48r may be involved in the uptake of chylomicron remnants by macrophages , and Kawakami et al. have reported that it is responsible for the induction of macrophage foam cell formation by remnant lipoproteins from hyperlipidemic patients. Since apoB48 is an integral protein incorporated during the assembly of chylomicrons in intestinal cells, it is not possible to bind it to model CRLPs in a physiological way. The particles used in the present work, therefore, do not contain apoB48 and we were unable to study involvement of this receptor in the uptake of CMR by macrophages. However, antibodies to apoB48 have been found not to inhibit the uptake of chylomicron remnants by rat macrophages , and Elsegood et al. who were unable to detect binding of chylomicron remnants to a protein with a molecular weight corresponding to the apoB48 receptor in THP-1 macrophages have suggested that it may be specific for VLDL remnants rather than chylomicron remnants. The results of this study demonstrate that oxidative modification of CMR as compared to LDL has profoundly different effects on the uptake of the particles and the subsequent induction of lipid accumulation in macrophages. Instead of markedly enhancing foam cell formation, oxidation of CMR slows their uptake by macrophages and reduces the amount of lipid subsequently accumulated in the cells. This difference may be due to the different receptor mechanisms involved, since oxidation of LDL shifts the main route of uptake from the regulated LDLr to the unregulated scavenger receptors, while our experiments suggest that CMR are taken up mainly by the LRP with some contribution from the LDLr, with CD36 and phagocytosis playing only minor roles, irrespective of their oxidative state. These findings provide important new information about the way in which oxidation of CMR influences their induction of foam cell formation and the mechanisms involved, and has important implications for the role of dietary factors such as oxidized lipids and antioxidants which are transported in CMR in the promotion of atherosclerosis.

The process of extracting biochemical content from genome annotations and literature sources to computationally catalog and interconnect the metabolic pathways available to the cell (i.e., metabolic reconstruction) is well established and has been carried out for a growing number of organisms on the genome scale (). This network reconstruction process ultimately results in the generation of a ochemically, enomically and enetically (BiGG) structured database that can be further utilized for both mathematical computation and analysis of high-throughput data sets. Goals of such computation and data integration efforts are to gain a better understanding of the observable phenotypes and coordinated functions of the cell, as well as to apply developed models for biological discovery and engineering applications. For mathematical computation, a number of methods have been developed to characterize models built from a metabolic reconstruction (; ), and reconstructions are becoming increasingly important in understanding high-throughput experimental data (). Thus, a well-curated metabolic reconstruction has a variety of uses and is of common interest to those studying systems biology relating to cellular metabolism. The Gram-negative rod-shaped bacterium, , has been an ideal target for metabolic reconstruction, since it is arguably the most studied and best characterized microorganism in terms of its genome annotation, functional characterization and knowledge of growth behavior (; ). Reconstruction of the metabolic network of has been progressing since 1990 (reviewed in ). This network reconstruction has been through a series of expansions and refinements (; ; ; , ; ; ; ), with each iteration building on previous work while incorporating new knowledge. Applications utilizing the reconstruction have had implications in a number of fields (for a list of applications and references, see ). For metabolic engineering applications, modeling enables examination and simulation of metabolism as a whole, circumventing the possible shortcomings of methods that rely on manual assessment of a limited number of interactions and possibly fail to detect non-intuitive causal interactions (; ). For studies of bacterial evolution, a reconstruction serves as a highly curated database and model to examine and simulate evolutionary hypotheses (, ). Network analyses have been applied to genome-scale reconstructions of to identify sets of reactions or metabolites whose activity is interdependent. These studies have obvious implications in aiding therapeutic interventions along with other systemic analyses (; ). Additionally, for the prospective goal of biological discovery, genome-scale reconstructions drive discovery by identifying specific areas where knowledge is lacking, or disagreements with observations, and provide a framework for the integration of high-throughput data (; ). In this study, we expand and refine the reconstruction of the metabolic network in . The new additions include: an up to date accounting for open reading frames (ORFs) in that have metabolic annotations and an alignment of the content in EcoCyc (), leading to the inclusion of 1260 ORFs (an increase of 356 ORFs over the previous reconstruction ()), an improved breakdown of the biomass composition, the maintenance requirements for growth and sustenance and a sensitivity analysis on the parameters used in computational modeling and thermodynamic information about the chemical transformations accounted for in the reconstruction. The thermodynamic properties estimated for the model reactions and compounds were utilized to test the thermodynamic consistency of the reactions included in the reconstruction (). This expanded version of the metabolic network will allow for additional and more comprehensive computational and experimental studies of the systems properties of metabolism. We give several such examples that use the new network reconstruction. #text Metabolic reconstruction and subsequent mathematical computation has become a useful tool in the post-genomic era by aiding both biological computation and experimentation. In this work, we present, characterize and utilize the AF1260 metabolic reconstruction of K-12 MG1655. The reconstruction serves as both a BiGG database containing the current knowledge of metabolism, as well as a framework for mathematical analysis. Accordingly, the major contributions from this work are: an expansion in size, scope and detail of the metabolic network of , effectively exhausting the available literature, an enumeration and description of the parameters and methods needed to utilize the reconstruction as a predictive model; examples of simulation results compared with high-throughput experimental data are presented and the inclusion of thermodynamic information and a novel thermodynamic consistency analysis for chemical transformations accounted for in the reconstruction. AF1260 represents the largest metabolic reconstruction of any unicellular organism and accounts for 1260 ORFs (28%) in the current genome annotation (). Furthermore, 1161 of the included ORFs (92%) have experimentally-based functions, conferring a high degree of confidence in the corresponding interactions. Just as gene annotation and sequence databases are used to identify and characterize genes in newly sequenced genomes, AF1260 will similarly serve as a primary reference for future metabolic reconstructions. Because of its curation history and size, future reconstructions, especially those for closely related organisms, will draw directly from this content. This process will further be aided by the synchronization and mapping with the EcoCyc database. The next step in the expansion of the metabolic network will require further discovery of metabolic functions and computational methods are needed that can facilitate this process (). In addition to expanded content, significant advancements in reconstruction techniques and methods used to determine network capabilities were presented. Thermodynamic consistency analysis represents a novel way to flag or highlight highly improbable intracellular and transport reactions for further evaluation. This approach can be added to future metabolic reconstruction and modeling projects. It effectively constitutes a QC/QA test that should improve the utility and scope of modeling predictions. Additionally, the use of a core biomass BOF (BOF) has identified an improved strategy to probe gene essentiality for growth. Previous analyses examining gene essentiality have utilized a BOF, which is based on measurements from a specific growth environment and is also constant in the type and relative proportion of metabolites. A common problem that arises when using a BOF based on wild-type measurements is that potential false positives can be generated when conditionally essential metabolites are inappropriately included in the BOF (; ). The BOF presented here, with continual refinement guided by experimentation, should increase the accuracy and utility of computational predictions with respect to mutant phenotype predictions. The approach taken to evaluate reaction reversibility in AF1260 was to prevent the inclusion of reactions that were highly unlikely to be reversible. This approach was carried out by using the thermodynamic consistency analysis and subsequent analysis of reaction thermodynamic estimates. Due to the thermodynamic coupling of reactions operating simultaneously, reactions that are individually thermodynamically reversible under physiological conditions will not necessarily be reversible when operating in concert with the other reactions in the cell. In line with this, using reaction reversibility determined from the thermodynamic analysis of individual reactions alone with FBA will result in improper model behavior due to the operation of thermodynamically infeasible pathways and cycles. Only if thermodynamic constraints are used in conjunction with the mass balance constraints of FBA to prevent the operation of these thermodynamically infeasible pathways (for example, ) can the reaction reversibility determined for individual reactions be used. Therefore, utilization of the thermodynamic information presented to fully assign reversibility and irreversibility in modeling simulations automatically requires additional implementation of methods, which consider thermodynamics on the systems level. With the increasing use of network reconstruction and the constraint-based modeling approach, a need has emerged to clearly define and demonstrate the steps required to computationally utilize a reconstruction. By outlining these steps and examining the sensitivity of modeling parameters used in computations, we have both explicitly defined the protocol and revealed the impact of modeling parameters on predictions. A computational software package is also available to efficiently implement such metabolic modeling (). A sensitivity analysis of key strain-specific parameters, using an early version of the reconstructed network (), found that the P/O ratio significantly affects the GR and flux predictions, whereas varying the BOF had relatively little effect. However, our analysis shows a greater dependence on the maintenance parameters calculated for these conditions. This result is primarily due to our testing of a broader maintenance value range (±50% of the calculated values as opposed to 20% by ). This larger value was selected because it is approximately the amount of the GAM that is difficult to quantify (i.e., unknown maintenance that accounts for gradient maintenance, protein turnovers and so forth; ) and produced a range that can be justified by examining different growth data (results not shown). Future projects should take into account the impact of the influential parameters (i.e., P/O ratio, growth maintenance) when designing their computational studies. The culmination of the increased size and expanded coverage of the reconstruction, in combination with the improved reconstruction techniques, has broadened the scope and accuracy of computational predictions. Comparisons of AF1260 simulations with experimental data for gene-essentiality and growth phenotypes showed an overall increase of 4 and 16% over JR904 predictions, respectively. Specifically, AF1260 is markedly improved in analyzing and predicting a wider range of minimal media growth conditions (see ). It can also better predict and screen the essential genes needed for viability in (see ). The one area where it appears that the model's ability to match experimental data decreased was where ORFs were found to be experimentally essential but computationally non-essential. This area can be addressed through further expansion of the reconstruction's scope (e.g., by including the transcriptional and translation machinery in as well as transcriptional regulatory effects) and targeted experimentation (e.g., elucidating the entry step into the biosynthesis of biotin). Future directions for improvement of the metabolic reconstruction of remain. As previously stated, the scope of the reconstruction will continually increase. Dead-ends and lumped reaction in the reconstruction point to specific areas of metabolism that can be further characterized in this expansion effort. A computational approach to resolve these dead-ends that utilizes constraint-based methods can be used in this effort. Additionally, an area for further compartmentalization of metabolites in the reconstruction is for metabolites located in the lipid bilayers. For example, a lipid on the inner leaflet of the outer membrane is different than one on the outer leaflet of the inner membrane, but currently in the reconstruction, they are both located in the periplasm. Further advancements in modeling will also be achieved through acquisition of additional experimental gene essentiality studies under different minimal media conditions to better define the core metabolites needed for viability and improve overall computational accuracy. Advancements are also likely to arise from additional incorporation of reaction and system thermodynamics. In summary, AF1260 represents a significantly expanded and comprehensively verified reconstruction of the metabolic network with broadened and enhanced predictive capabilities. With the growing number of studies based on previous versions of this reconstruction appearing, this work will enable a wider spectrum of studies focused on both proximal (i.e., immediate) and distal (i.e., over time) causation in biology. As the field of systems biology expands to incorporate cellular interactions from multiple core functions (e.g., regulation, signaling, etc.) on the genome scale, AF1260 will serve as a key component for the study of by providing an extensive picture of cellular metabolism. The reconstruction process has also been previously outlined (; ). Here, we provide certain details specific to this work. Starting from the metabolic network for JR904 (), additional reactions were added to the network based on -specific biochemical characterization studies (see for a complete list of references) and other reactions were removed (see Results). This process was aided by comparing the content of JR904 with the EcoCyc database (see below). The genome annotation () was used as a citation source for biochemical characterization studies and a framework upon which translated metabolic proteins, and subsequently reactions, were assigned to form gene to protein to reaction (GPR) assignments. The SimPheny™ (Genomatica Inc., San Diego, CA) software platform was used to build the reconstruction. For each reaction entered into the reconstruction, the involved metabolites were characterized according to their chemical formula and charge determined using their p value for a pH of 7.2. Metabolite charge was determined using its p value(s). When the metabolite p was not available, charge was determined using the p of ionizable groups present in a metabolite (). All of the reactions entered into the network were designated as enzymatically catalyzed reactions or spontaneous reactions, were both elementally and charged balanced and are either reversible or irreversible. Reversibility was determined first from primary literature for each particular enzyme/reaction, if available (see for references). Additionally, general heuristic rules, like those applied by , were used to enter reversibility using knowledge about the physiological direction of a reaction in a pathway (sometimes including regulatory knowledge) and/or basic thermodynamic information (such as reactions hydrolyzing high-energy phosphate bonds are almost always irreversible). Furthermore, a thermodynamic analysis of reversibility was utilized to assign the directionality of some reactions (see above). The comparison between the content of the AF1260 and the EcoCyc () and MetaCyc () databases was performed in three phases. Initially, a list of metabolic ORFs contained in EcoCyc and not in JR904 (the previous reconstruction) was manually evaluated for inclusion in AF1260 in an effort to merge content. A total of 176 out of 308 ORFs from this list were included into AF1260 from manual analysis of this list or were included before this analysis from primary literature in a separate effort. Many of the inclusions in this phase were transporter encoding ORFs. A common type of ORF that was not included were those acting on nonspecific metabolites (e.g., nonspecific drugs), proteins or RNA molecules. The second phase of the comparison consisted of generating a complete mapping of the metabolites contained in AF1260 and EcoCyc or MetaCyc. This phase permitted the inclusion of compounds in each database that were missing from the other and identified possible errors in enzyme substrate specificity and metabolite structure. It also provided a future reference for linking of the metabolite content between the two resources. In an initial automated effort, mappings between metabolites in AF1260 and EcoCyc/MetaCyc were established computationally using textual matching between the official name in AF1260 to the common name and/or synonyms of metabolites in EcoCyc/MetaCyc, version 10.6. In addition, when available in both data sets, KEGG identifiers and CAS numbers were used to double-check matches or to make additional matches. After this computational step, 871 out of 1039 metabolites in AF1260 were mapped to EcoCyc/MetaCyc. The remaining metabolites were mapped manually and changes to the content of AF1260 made during this mapping process were facilitated by cross-referencing the ORFs that encoded for the proteins that acted on specific metabolites in AF1260 with their annotation in EcoCyc (see Results for findings and for the mapping). The final phase of the comparison was an automated mapping between reactions contained in AF1260 and EcoCyc/MetaCyc. This phase generated a list of high-confidence reactions that both AF1260 and EcoCyc contained, and provides a future reference for a full merging of the reaction content between the two resources. The automated reaction mapping was performed with software written specifically for this task, to accommodate frequently occurring types of differences between the models. The matcher parses the equations of every reaction in AF1260 and uses the previously described metabolite mappings to find the reaction object in EcoCyc/MetaCyc that contains the same set of metabolites as does . Numerous reactions in AF1260 contain protons in the equation that do not appear in EcoCyc, and the matcher can take into account this and other similar differences. The matcher also tries to find a generic reaction in EcoCyc that is specified in terms of compound classes, if the metabolite instances used in the equation in AF1260 did not yield a direct match. The biomass BOF was generated by defining all of the major and essential constituents that make up the cellular biomass content of . To determine these metabolites and their quantity, we used the dry weight composition data for an average B/r cell growing exponentially at 37°C under aerobic conditions in glucose minimal medium, with an approximate doubling time of 40 min having a dry cell weight of 2.8 × 10 grams () (; ). Each cellular biomass macromolecule (i.e., protein, RNA, DNA, etc.) was divided into its corresponding metabolic precursors present in the reconstruction (for example, -alanine, UTP or dTTP, respectively). Each of the precursor metabolites was assigned a value that it contributes to the total percentage of the macromolecule, except for the soluble pool metabolites (e.g., thiamine diphosphate). This process was followed so that if the overall quantities of macromolecules were changed, the corresponding precursor metabolite would be scaled appropriately (see Sensitivity analysis section). The quantity of soluble pool metabolites (approximately 2.9% of the total biomass) was taken from experimentally measured values or alternatively, it was estimated as a 0.1 mM intracellular concentration (see for a complete list of references). From this data, a linear biomass BOF was formulated based on the wild-type cell composition for and an ATP maintenance approximation to account for non-metabolic processes (see Supplementary information). Using FBA, the model was analyzed to determine if each BOF metabolite could be generated from the defined minimal medium under both aerobic and anaerobic conditions with -glucose, -ribose and glycerol as the carbon and energy source. Only metabolites identified as cofactors could not be generated from the glucose minimal medium (discussed in ). Using the BOF, gene essentiality and published data, a ‘core' BOF was formulated that was consistent with the minimal set of macromolecular molecules needed for cell viability. The 20 common amino acids, inorganic ions and nucleotide metabolites were all considered essential (). For the other BOF metabolites, each metabolite was evaluated individually to determine if the genes that were necessary to synthesize the metabolite from minimal media substrates (see ) were essential (; ). One macromolecule, glycogen, was not essential for cell viability because there were no essential ORFs encoding for enzymes in the synthesis or breakdown of glycogen. The essential metabolites were defined by identifying the end product from the closest essential reaction to the BOF metabolite () in the possible pathway(s) for biosynthesis. Molecules in this group, such as riboflavin, were determined to be essential, whereas the wild-type outer membrane K-12 LPS molecule was not found to be essential. However, a precursor of the common wild-type LPS molecule, KDO-Lipid A, was found to be essential for cell viability (). Alternatively, ‘core' metabolites were also determined from specific published studies. For example, thiamine diphosphate was found to be essential (), whereas phosphatidylglycerol was determined not to be essential (). where the number of ATP equivalents hydrolyzed is characterized in the GAM variable. The entire BOF is given in mathematical terms in . Aside from the BOF maintenance, an NGAM (mmol ATP DW h) value was used as an energy ‘drain' on the system during the linear programming calculations and accounts for non-growth cellular activities (). The NGAM was represented as a defined flux in the reaction flux vector, (see below and ). A stoichiometric matrix, ( × ), was constructed for AF1260, where is the number of metabolites and is the number of reactions. The corresponding entry in the stoichiometric matrix, , represents the stoichiometric coefficient for the participation of the th metabolite in the th reaction. FBA was then used to solve the linear programming problem under steady-state criteria () represented by the equation: where ( × 1) is a vector of reaction fluxes. Since the linear problem is normally an underdetermined system for genome-scale metabolic models, there exist multiple solutions for that satisfy . To find a particular solution for , the cellular objective of producing the maximal amount of biomass constituents, represented by the ratio of metabolites in the BOF, is optimized for in the linear system. Additionally, constraints that are imposed on the system are in the form of: where α and β are the lower and upper limits placed on each reaction flux, , respectively. For reversible reactions, −∞⩽⩽∞, and for irreversible reactions, 0⩽⩽∞. The constraints on the reactions that allow metabolite entry into the extracellular space were set to 0⩽⩽∞ if the metabolite was not present in the medium, meaning that the compounds could leave, but not enter the system. For the metabolites that were in the medium, the constraints were set to −∞⩽⩽∞ for all except the limiting substrate(s) (e.g., glucose and/or oxygen). The reaction flux through the BOF was constrained from 0⩽⩽∞. Linear programming calculations were performed using SimPheny™ (Genomatica, San Diego, CA) and the LINDO (Lindo Systems Inc., Chicago, IL) or TOMLAB (Tomlab Optimization Inc., San Diego, CA) solvers in MATLAB (The MathWorks Inc., Natick, MA) with the COBRA Toolbox (). When comparing the flux distribution in central metabolism to experimentally reported values (), all of the comparisons were performed using computational results when optimal growth is predicted using the BOF, the 152 regulated reactions under these conditions constrained to zero (see above), a split in the flux ratio between the two NADH dehydrogenases of 1:1, an NGAM value of 8.39 mmol ATP DW h, a GAM value of 59.81 mmol ATP DW and AF1260. An FVA on the optimal flux distribution yielded no flexibility in the central metabolism pathways examined in this study. From the study, data from growth in reactor conditions were used because the oxygen uptake and CO secretion rates were reported, and the flux values that were used were based off C-constrained flux balancing. The sensitivity analysis was performed under aerobic glucose-limiting minimal medium conditions. For each analysis, the parameter being examined was varied while the GUR was sequentially set between 0 and 10 mmol DW h for a series of simulations, with the maximum OUR set to 18.5 mmol DW h. This maximum uptake rate was chosen, since it closely matched the maximum uptake rate of oxygen observed (e.g., see ; ). All other modeling parameters were set to those determined in the application of AF1260 to predict cellular phenotypes section. The BOF was used in all simulations (except those stated otherwise), since the predicted GR and OUR were found to be insensitive to the use of either the BOF or BOF. To determine the carbon, nitrogen, phosphorus and sulfur sources that could support simulated growth, we screened all of the metabolites that could be exchanged with the environment (i.e., exchange reactions) in the AF1260 and JR904 models. The identified metabolites formed the potential substrate sets (). Through subsequent simulations, we set an arbitrary maximum flux of 20 mmol substrate DW h for each potential substrate tested (consistent with maximum observed substrate uptake rates ) and optimized for flux through the BOF using FBA and either AF1260 or JR904. An OUR of 18.5 mmol DW h, the BOF, a NGAM of 8.39 mmol ATP DW h, a GAM of 59.81 mmol ATP DW and no regulatory constrains were used during the growth condition analysis of AF1260 (for JR904, see ). During the analysis, the reactions CAT, SPODM and SPODMpp were constrained to zero to prevent generation of cellular energy equivalents through reactions involved in 's response to oxidative stress. If a positive flux could be generated through the BOF reaction (>0), then the substrate was considered a viable source. Experimental data used in the comparison were provided by Biolog () and both ‘weak' and ‘positive' readings from the biolog data were considered as a positive growth condition. To determine the effect of a gene deletion, the reaction(s) associated with each gene in AF1260 were individually deleted from and FBA was used to predict the mutation growth phenotype. The simulations were performed using glucose minimal medium conditions with a GUR of 10 mmol DW h, an OUR of 20 mmol DW h, the BOF, an NGAM of 8.39 mmol ATP DW h, a GAM of 59.81 mmol ATP DW and zero flux through the 152 reactions regulated under glucose aerobic conditions (see ). The flux through the BOF was optimized in the mutated network, ′, and a positive flux through the BOF (>0) was considered non-essential (). Experimental criteria for gene essentiality are described in detail in . All Δ′ and Δ′ calculated for the reconstruction using the group contribution method are based upon the standard condition of aqueous solution with pH equal to 7, temperature equal to 298.15 K, zero ionic strength and 1 M concentrations of all species except H, and water. In the cases where multiple charged forms of a molecule exist at pH 7 (i.e., ATP and HATP), the most abundant form is used. This is consistent with the form of the molecules used in the fitting of the group contribution energy values (MD Jankowski and V Hatzimanikatis, in preparation). The charges of the molecules and the proton balances for the reactions included in the reconstruction are based on a reference pH of 7.2. In order for the Δ′ values included with the reconstruction to match the reference pH of the reconstruction, all Δ′ calculated using the group contribution method (based on a reference pH of 7) were adjusted to a reference pH of 7.2 using the method described in . The adjusted Δ′ values were used in the calculation of Δ′ and for all other thermodynamic analysis performed on the reconstruction. The p values for the compounds in the reconstruction used in the transformation of Δ′ to a reference pH of 7.2 were estimated from the molecular structures of the compounds using the MarvinBeans software developed by ChemAxon. The Δ′ calculated for all reactions contained in the reconstruction is based on the reference state of 1 mM concentrations for all species except H, water, H and O. The reference concentrations for H and O are the saturation concentrations for these species in water at 1 atm and 298.15 K. All Δ′ values reported in this work also include the energy contribution of the transmembrane electrochemical potential and proton gradient for all reactions involving transport across the cytoplasmic membrane assuming a periplasmic pH of 7.7 and a cytoplasmic pH of 7.2. All Δ′ calculated for reactions in the AF1260 model are listed in . We also determined the direction of flux required in the reactions contained in AF1260 to achieve near optimal growth (90–100%) on each of 174 carbon sources using FVA () and the BOF. It is worthwhile to note that the same set of reactions can or cannot be utilized in FVA simulations when examining approximately 5–95% of the optimal flux value achievable for the BOF under glucose aerobic conditions (one exception is the cytochrome oxidase bo and oxygen transport reactions, which are needed for generating the necessary energy to achieve approximately 80% or greater of the BOF flux). During the FVA of conditions corresponding to glucose aerobic growth, the reactions CAT, SPODM and SPODMpp were constrained to zero to prevent generation of cellular energy equivalents through reactions involved in 's response to oxidative stress, and the reaction formate hydrogenlyase, which appears to be involved in regulating cytosolic pH (), was also constrained to zero to prevent the production of significant amounts of hydrogen gas that is not typically observed for most buffered experiments around pH 7. The results of the FVA indicated that some of the reactions in the reconstruction consistently operated in the reverse direction. During the calculation of Δ′ for these reactions, the forward direction of each reaction was redefined to be in the direction of flux required for near optimal growth to occur. Because of this adjustment, all negative Δ′ and Δ′ values reported (see ) indicate reactions that are thermodynamically feasible in the direction of flux while positive values indicate thermodynamically infeasible reactions. The range of possible values for the Δ′ of a reaction depends not only on Δ′ but also on the uncertainty in the estimated Δ′ (), the activities of the metabolites involved in the reaction and for transport reactions, the energy contribution of the electrochemical potential and proton gradient across the cytoplasmic membrane (Δ) (). Δ′ can deviate from Δ′ because the activity of a metabolite can deviate from the reference value of 1 mM. The maximum and minimum values for Δ′ were calculated using the following equations. where is the minimal metabolite activity assumed to be 0.00001 M, and is the maximum metabolite activity assumed to be 0.02 M. The physiological range of activities for the dissolved gasses H, O and CO is much lower than the range of activities for other metabolites involved in metabolism. For this reason all of the values for H, O and CO were set to 10 M, which is approximately equivalent to one molecule per cell, and the values for H, O and CO were set to the saturation concentrations for these gasses in water at 298.15 K and 1 atm, 0.000034, 0.000055 and 0.0014 M, respectively. The activity terms for H and HO were left out of and because these activities have already been lumped into the Δ′. The Δ′ ranges encompassed by Δ′ and Δ′ calculated for the reactions in the reconstruction were used to assign reversibility and directionality to the reactions based on the thermodynamic estimates. Reactions with exclusively negative Δ′ values were identified as thermodynamically irreversible in the forward direction, reactions with exclusively positive Δ′ values were identified as thermodynamically irreversible in the reverse direction and reactions with both positive and negative Δ′ values were identified as thermodynamically reversible. FVA was then utilized to determine the directions in which each of the reactions in the reconstruction operated during near optimal growth on 174 carbon sources. In this way, reactions for which the direction of operation indicated by FVA conflicted with the direction of thermodynamic feasibility indicated by the Δ′ ranges were identified.

The number of new molecules generated by the chemical and pharmaceutical industry has boomed in the last few years owing to the emergence of combinatorial chemistry along with the demand for novel industrial, agricultural and therapeutic products (). The number of natural or man-made organic compounds present in the biosphere is somewhere between 8 and 16 million molecular species, of which as many as 40 000 are predominant in our daily lives (). Microorganisms are key players in determining the environmental fate of novel compounds because they can be used as carbon and energy sources (). Microbial metabolism may not only cause the complete elimination of a given chemical compound but it can also generate chemical species that are as toxic or as persistent as the original ones. In the case of complete metabolism, microbial biodegradation can be exploited for waste treatment and used in directed bioremediation processes or (). Therefore, knowing whether a novel chemical compound is likely to be metabolised by microorganisms is crucial for assessing the environmental risks associated to its production, transportation, utilisation and disposal (; ). However, after 50 years of research on microbial biodegradation, detailed knowledge about biodegradative pathways is available for only about 900 chemical species (; ). New pesticides and pharmaceuticals are being produced at rates that cannot be matched by experimental attempts to determine the outcome when spilled or released into the environment. This makes essential to develop systems that can predict the fate of chemical compounds (; ) before experimentally assessing the capacity of the microbiota to degrade them. Although hydrophobicity, water solubility and the presence of xenophores (; ; ) have been invoked for assessing the biodegradability of given compounds, there are many examples in which the presence/absence of certain functional groups do not match the experimental results. As an alternative, we have approached the problem of predicting the environmental fate of chemical species from an experience-based perspective, using a (micro)biological logic rather than a purely (bio)chemical appraisal, for example, making the most out of available information about known microbial catabolic reactions on organic pollutants. To this end, we have exploited the wealth of knowledge on the genetic and biochemical basis of microbial metabolism available at the University of Minnesota Biodegradation and Biocatalysis Database (UMBBD; , ) and the Biodegrative Strain Database of the Michigan State University (BSD; ) to train a rule-based classification system () for detecting the association between certain chemical compound descriptors and environmental fates. Such descriptors are based on the deconstruction of chemical structures in atomic triads (also referred to as ). A machine learning system () was then used to identify explicit rules that associate compound vectors to environmental fates as inferred from the analysis of the metabolic network that represents the global biodegradative potential of microorganisms. Finally, a scheme to predict the fate of new chemical compounds, using the previously identified rules, was implemented as a server. The results obtained include the evaluation of the prediction capacity of the system and its application to several sets of compounds provided by the or obtained from the database —for most of which there are no data on their biological fate. Herbicides seem to be the group of functional molecules that have less favourable prospects of recycling through the global microbial biodegradation network. At the time of starting this work, the UMBBD contained information on 850 compounds and 903 reactions (, ). The first issue at stake was whether structural features of the target molecules could be significantly correlated to their known environmental fate. To this end, we resorted to describing each chemical structure as a whole of 152 descriptors that represented atomic triad frequencies, molecular weight (MW) and water solubility, the latter expressed both quantitatively and qualitatively. Such atomic triads (or ) included 149 groups of three consecutive, connected atoms that can be identified on the structure of a compound, taking into account the type of connecting chemical bonds. For example, the atomic triad C–C–H is different from C=C–H, whereas C=C–H is equal to H–C=C (). The choice of atomics triads instead of focusing on reactive groups or functional motives reflected the tradeoff between having significant structural information and the handling of a minimal number of attributes (see the Discussion section). Deconstruction of each compound in this way is achieved by first translating the SMILES () representation of each molecule, which is available from UMBBD, into other forms of chemical depiction that include explicit information regarding atom connectivity and chemical bond types. Then, the frequency in which the different atomic triads appear for each compound is recorded. MW is also available from UMBDD and compound solubility is, in some cases, accessible through links to the corresponding entry in ChemFinder (). The collection of atomic triad frequencies, the MW and the solubility were then assembled to generate molecular descriptors, henceforth referred to as ( and ). Through these criteria, nine compounds out of the 850 listed were not associated to any vector because they had less than three atoms or because their entries in UMBBD did not include SMILES strings. About 718 distinct vectors represented the remaining set of 841 compounds, indicating that the correspondence between compounds and vectors is not equipotent. A one-to-one relation between compounds and vectors existed for 625 compounds, whereas 93 vectors described the remaining 216 compounds. Many-to-one relations between compounds and vectors is explained by the fact that positional isomers in which functional groups have changed between equivalent positions may share the same pattern of atomic triad frequencies even if they do have different connectivity and different SMILES strings. That is the case for pyrogallol versus phloroglucinol, and also the case of 2-formil-1-indanone versus 1-formil-2-indanone. In addition, as stereoisomers have the same atomic connectivity and identical composition in terms of atomic triads, they are encoded by the same vectors. This kind of information was expected to enter some noise in the predictive system, although (as explained below) not as important as one could anticipate. Description of chemicals as of this sort ( and ) was used to feed the training algorithm for classification of the molecule according to its fate in the global network shaped by the global microbial metabolism (see below). Once each compound had been expressed in a vector form, the reactions in which the chemical is known to take part, as a substrate or as product, were retrieved from the database (, ; ). To categorise the environmental outcome of the complete list of 850 chemical species under study, we exploited all known metabolic reactions for organic chemicals (independent of their specific bacterial host) to delineate a global network of microbial catalysis (). Such an inclusive biodegradation network has been described before as an entity with topological properties that resemble single-cell metabolic transactions. Although a network of this kind includes interconnected pathways that may not stand alone in a single organism (; ; ; ; ; ), it does represent the known biodegradative potential of microbial communities at a global scale (). Such a biodegradation network (, ) was employed to pinpoint the channelling of every compound into one of three final destinations () as follows. The first was composed of 38 compound entries in UMBBD that were annotated as belonging to the . We extended this category of chemicals by including all molecules that participated in pathways through the network leading them to the central metabolism. In this way, a group of 533 chemical species were defined as (CMs). On the other hand, we labelled as , nonbiodegradable compounds those that do not participate as in any reaction documented in UMBBD, and thus can never reach the central metabolism or being biodegraded otherwise. After scrutiny of the global biodegradation network, 108 compounds of the database unequivocally fulfilled that criterion. In addition, two pairs of somewhat special compounds (arsenate/arsenite, benzyldisulphide/benzylmercaptane) that were linked by bidirectional reactions but had no other outgoing connections were also classified as nonbiodegradable. The operative list of recalcitrant compounds included, therefore, 112 compounds. Yet, as before, we extended the nonbiodegradable whole to those molecular species that were directly or indirectly connected to recalcitrant compounds as precursors of ultimately intractable chemicals (). The extended set of such molecules included 353 specimens, which were operatively tagged as (NB). This set of compounds did overlap by 112 compounds with the previously defined set of CM compounds. This indicated that many chemicals can either be degraded upon being channelled into the central metabolism or accumulated in the environment if diverted into nonproductive reactions. The nonredundant set that contained all CM and NB compounds included 774 chemical species. The 76 remaining molecules were not connected to either central metabolism or nonbiodegradable compounds. Instead, they belong to various pathways that go straight into carbon dioxide and water, without converting into any of the typical intermediates of the central metabolism. Although they are of course biodegradable, the lack of connections to the central metabolism rules out their classification as CMs. On the other hand, they cannot be classified as NB compounds either, as CO is not a bona fide recalcitrant, terminal molecule: it can be captured back to metabolism by a formylmethanofuran dehydrogenase reaction or (in practice) by many other CO-fixing microbial processes. We thus established a separate, extended type of compounds, which were directly or indirectly positioned in pathways leading to CO. This group, which includes 329 molecular specimens, was termed as (CDs). One further extension of this criterion was to take CO and central metabolism as the same final fate, and group all compounds connected to them. The resulting set thus comprises central metabolism and carbon dioxide path compounds (CMCDs) and includes 634 chemicals. CMCDs correspond to what can be considered intuitively as the set of . In summary, as shown in , each compound can be ascribed to each of three environmental fates (CM, NB and CDs), in which the sum of CMs and CDs forms the operative biodegradable (CMCD) category. The four types of biodegradative fates (CM, NB, CD and CMCD, ) did overlap to a significant extent. To refine further the sorting of the chemicals and to generate better classifiers for the compounds, we established four separate, binary categorisation schemes that would label out each chemical as belonging to each of the groups or to the cognate negated classes. Accordingly, we defined the following four classification classes: (i) CM or No CM, (ii) NB or No NB, (iii) CD or No CD and (iv) CMCD or No CMCD. Obviously, the most important categorisation for our purposes is the last (CMCD no CMCD), as it reflects either eventual recalcitrance or amenability to biological recycling. Yet, the other classifiers do hold a considerable practical value as well (see the Discussion section). A second consideration was related to the structural similarity between the different types of compounds. One could suspect that chemicals belonging to the same group (CM, NB, CD, CMCD, or the corresponding negated classes) might share some structural features, especially if they are part of the same metabolic pathway. To examine rigorously this issue, chemical compound similarity was estimated for each pair of compounds using their atomic triad frequencies for calculating a modified version of the Tanimoto association coefficient τ (). This coefficient reflects the ratio between the number of atomic triads that two compounds have in common and the number of atomic triads that they do not have in common, and can be used as a measure of the distance between compounds, in respect to their chemical similarity. The distribution of such distances for the whole of compound pairs (), indicated that the collection of chemicals was quite diverse. Although 90% of the pairs had τ values <50, only 1% of the pairs had τ figures ⩾80. When the distribution of distances was calculated for pairs of compounds that belong to the same classes of environmental fates, we found that all groups were equally diverse (data not shown). The degree of clusterisation, however, varied among the different groups, as measured by their average clustering coefficient (; ). Having categorised the compounds according to the four binary classification schemes mentioned above, and having defined compound vectors that describe their composition, topology, MW and solubility, the machine learning algorithm c4.5 () was used to generate classifiers in the form of sets of rules. This program uses an inductive decision tree process that generates classification schemes matching the attributes of the training examples to given classes. These schemes can later be used for assigning new (unseen) examples to such classes (). Each classification involves a tree structure, which can be also be expressed as a set of rules, in which internal nodes represent a test condition (formulated in terms of the attributes), whereas the terminal (leaf) nodes represent classes. Such an approach was preferred over other available learning methods, for example, neural networks, because (i) the machine learning of choice can handle missing values (such as water solubility, unknown for some compounds), and (ii) the rules created by c4.5 can be easily interpreted by the user. In addition, such rules produce generalised models that become instrumental to make predictions on molecules not previously visited by the learning machine. Rules take the form of propositions with two sides: the left-hand side contains a conjunction of attribute-based tests, and the right-hand side is a class. For example, the rule in states that a compound with more than 19 triads of the type C–C–C, more than one triad of the type O–C–C and three or less triads of the type O–C–C, should belong to the NB class, with a support confidence of 94%. Each of the four final classifiers was composed of a set of 16–23 rules, and each rule was composed, in average, of 3.3 attribute-based tests (standard deviation 2.07, range 1–12). To gain some insight on the relationship between chemical structure (i.e., the frequency of triplets) and environmental fate, the rules were reanalysed to assess the weight of each of the attributes. Out of such 152 traits (149 frequencies of atomic triads, MW, quantitative solubility and qualitative solubility), only 52 were included as part of the propositional rules of all classifiers. These attributes are listed in , together with a graphical depiction of the frequencies in which each of them appears in the rules. For example, attribute-based tests referring to the frequency of atomic triad O–C=O come out in about 45% of the rules that conform the classifier for the scheme CD or No CD. Also, although MW and solubility are taken into account by the classifier NB or No NB, they are useless for the classifier CM No CM, (). This reflects that not all attributes have the same importance for each of the environmental fates. To assess the predictive capacity of the system, we followed a fivefold cross-validation strategy. For this, the data set was divided into five blocks; four of them were used as a training set, to generate the classifiers (rules), and the remaining block was used as a test set. This allowed measuring the ability of the classifiers for predicting the environmental fate of chemicals not included in the training set. The process was repeated five times, changing the block that was used as a test set. The accuracy of the system (i.e., the percentage of compounds correctly classified as belonging to any of the NB, CM, CD, CMCD classes, or their negation) was averaged for each classification scheme. The resulting averaged accuracies ranged from 73 to 87%, for the different classification schemes (). A more detailed picture of the predictive capacity of the system was obtained by calculating its sensitivity and specificity for the prediction of specific classes. The values for sensitivity (fraction of compounds correctly classified as belonging to a specific class, relative to the total number of cases of that particular class) and specificity (fraction of compounds correctly classified as belonging to a specific class, relative to the total number of predictions for that class) ranged from 50 to 93%, and from 66 to 85%, respectively, in the different classification schemes (). In general, the classification scheme CM or No CM is the one for which the best predictive performance was achieved. This could be explained in part by the fact that CM is the group with the highest average clustering coefficient, , making it easier for the c4.5 algorithm to generate rules that represent the more similar compounds of the group (). Consistently with this explanation, we have observed that the relationship between the average clustering coefficient and the sensitivity of the predictions for any given class can be adjusted to a regression line with a correlation coefficient =0.53 (). To authenticate the significance of the figures generated above and uncover possible biases of the dataset on the predictive value of the classifiers, we compared the performance of our c4.5-based system with one employing random predictors. Possible biases due to overrepresentation of classes were corrected by having such predictors assigning compounds randomly to one of the two fates for each classification scheme, with a probability that was proportional to the population of each of the classes. shows the average accuracy obtained for the different classes (versus their negated classes) for the real and the randomised dataset. It can be seen that the real dataset produces a considerably higher accuracy for all the classes, which is more pronounced in the CM and NB groups. Details of the analyses of the randomised dataset (sensitivity or specificity) are shown in the . To assess the statistical significance of the differences between the c4.5-based system and the random predictors, the dataset was subject to a This analysis compares the performance of two methods based on the number of cases that one of them provides a correct response, whereas the other fails and vice versa. A P(N) is thereby obtained which can be interpreted as the probability for the null–hypothesis, that is that the observed differences are happening by chance. These probabilities are shown in . The result clearly demonstrates the superiority of the c4.5-based predictors as compared to their equivalent random counterparts with values of P(N) in the order 10–10. To compare the efficacy of our predictive system with experimental data, we made use of an additional set of compounds that had not been included as part of the training set as they were not available at the time of its setup. To this end, we took the 147 compounds entered in the UMBBD along with fresh information on its biodegradability from 17 November 2003 (when the predictive system was first set) to 27 November 2006. As before, the structures of these new compounds were translated into SMILES formats, used to generate compound vectors and fed as inputs for each of the four classifiers established previously. The complete set of biochemical reactions involved in their biodegradation (as of 27 November 2006) was collected as well, thereby allowing a refinement of the global biodegradation network () and the assignment of the corresponding compares the actual classification of each of the compounds to the predictions emitted by the system. The accuracy values for the four classification schemes ranged from 40 to 69%, lower than those of the fivefold cross-validation tests reported above with the original dataset (73–87%). However, the sensitivity values for the major classes ranged from 72 to 91%, which are comparable to those of the cross-validation experiment (78–93%). The best accuracy value was that for the classification scheme CMCD or no CMCD (69%). Consistently with this, the prediction of compounds that belong to the CMCD class achieved the highest sensitivity value, 91% of the compounds actually belonging to this category being classified as such. The least sensitivity was associated to the prediction of compounds belonging to No CM. Only 16% of these compounds were predicted as such. This is not unexpected, as this set of compounds is very heterogeneous, holding the lowest clustering coefficient (). To put the prediction system into operation as a user-friendly resource, it was implemented as a public server called (BDPServer, ). The input for the BDPServer () requires the expression of the formula of the chemical under study in SMILES format, although an integrated Java applet allows the user to draw the chemical structures directly, instead of typing SMILES strings. Quantitative and qualitative solubility information can also be entered (), but it is optional because the learning machine c4.5 can deal with missing values. The prediction engine within the BDPServer (a Perl script called zPredict, ) uses OpenBabel for adding hydrogen bonds and translating SMILES strings into BS and Alchemy formatted reports. The connectivity between atoms is extracted from such reports, which includes the type of chemical bonds between atom pairs. zPredict then calculates the composition of each compound in terms of atomic triads. OpenBabel is also used to calculate the MW of compounds. A vector that conveys the compound descriptors is then generated. These include atomic triad frequencies, solubility (if provided by the user), and MW. Predictions are generated by the module of the c4.5 package, using the classification models mentioned above. The output of the system consists of four independent predictions, associated to confidence factors that are calculated by If a user-provided SMILES string happens to match another string in the database, the server returns the classification of the compound, and predictions are not shown unless the user chooses to force the forecast. An example of how the BDPServer works and its predictive ability operates is shown in the exercise summarised in the . In this case, we set out to classify a collection of compounds that belong to well-characterised microbial pathways for biodegradation of toluene—which were part of the training data. Such pathways, which include a total of 42 different compounds, were taken from the MetaRouter server () by querying the database for all reactions connecting toluene to the central bacterial metabolism. SMILES strings of each of the 42 chemicals were submitted to the BDPServer and the predictions compared to the actual classification of the compounds, according to the four binary schemes defined previously. The number of compounds that belong to each of the classes, and the absolute number of successful BDPServer predictions, within each of the classification schemes, are shown in . All compounds were correctly categorised for the classification schemes CM or No CM and CMCD or No CMCD. Suboptimal—but still significant—results were obtained for the classes CD or No CD, and NB or No NB in which 37 out of 42 and 40 out of 42 compounds, respectively, were classified correctly. Given the fact that the pathways for degradation of toluene were included in the training of the system, it is likely that these figures overestimate the capacity of the system. Yet, they represent the type of result and error margin that one would expect from the analysis of compounds that participate in full metabolic routes. An additional exercise was designed to test the ability of the system for recognising compounds that are clearly linked to the central metabolism. To this end, the KEGG database was used to generate a collection of 733 molecules that fulfilled the following conditions: (i) they were components of metabolic pathways; (ii) they were not part of metabolic pathways involving xenobiotic or recalcitrant compounds; (iii) they were composed of more than three atoms; and (iv) they had an associated description in MOL format that could be converted into SMILES format. The BDPServer was then used to generate biodegradability predictions for such compound set. As the result of this, the BDPServer predicted that 662 KEGG compounds (90.31%) could be assigned to the CMCD class (central metabolism plus carbon dioxide sinks), that is, the class that defines the group of bona fide biodegradable chemicals. This figure is consistent with the prediction that 597 (81.44%) out of the 733 KEGG compounds belong to the class of chemicals not connected to recalcitrant compounds (No NB class) and that 501 molecules (68.34%) could be directed to central metabolism (CM). As a control, random predictions were generated as above, that is, assigning compounds arbitrarily to one of the two fates for each classification scheme, proportionally to the population of each of the classes in the original training set. The accuracies for these random predictors are 76.8% for CMCD, 58.66 for No NB and 60.84 for CM. These differences in performance are statistically supported by the associated values of P(N): 1.52 × 10 (CMCD), 9.45 × 10 (No NB) and 3.3 × 10 (CD). With the tools described above in hand, and after having evaluated the reliability of the system in different sets of chemicals with a known biodegradative fate, we set out to produce global predictions for compounds found in lists that are subject to regulations through the Such lists include (i) 3365 dangerous substances incorporated to directive 67/548 of the European Commission, which regulates the classification, packaging and labelling of hazardous chemicals, last updated in April 2004 (the so-called Annex-I); (ii) 2747 High Production Volume Chemicals (HPVCs, defined by directive 793/93 as molecules that are produced or imported in quantities exceeding 1000 tons per year); and (iii) 7829 Low Production Volume Chemicals (LPVCs, between 10 and 1000 tons per year). As each of the three catalogs contain many substances of poorly defined composition (such as petroleum derivatives) that cannot be analysed by our system, we filtered the lists with the SMILECAS database, to obtain refined inventories of defined compounds with associated SMILES strings. The curated lists contained 1766, 1653 and 5645 compounds, respectively. The overlap between those from Annex-I with HPVCs and LPVCs included 595 compounds in one case and 366 chemicals in the other. Upon blind testing of such molecules with the BDPServer (), about 5% of these compounds were automatically rejected because they had less than three atoms or because their SMILES entry was not correctly interpreted by OpenBabel. Yet, the system predicted that, for any of the three lists, ∼60% of the compounds would be connected to central metabolism (CM), ∼20% would be linked to carbon dioxide (CD) and ∼70% would be connected to either central metabolism or carbon dioxide (CMCD). Therefore, more than two thirds of the compounds could be in principle biodegraded. However, about 47% of the compounds of any list were either recalcitrant or could evolve into nonbiodegradable compounds (NB). The highest percentage of NBs (55%) was found in the subset of compounds that are part of the curated Annex-1 but not of HPVCs (data not shown). In a subsequent step, we analysed sets of chemical species of the PubChem database that were explicitly labelled as (1707 compounds), , , , and . As indicated in , our system exposed that the percentage of CM (∼54%) or CMCD (∼70%) compounds within these lists were roughly similar to those of the species listed by the . However, the percentage of compounds connected to CO (CD, ∼34%) and to nonbiodegradable end products (NBs, ∼59%) was significantly higher. The highest percentage of predicted difficult-to-degrade compounds was observed in the collection of herbicides included in PubChem (NB ∼74%). The much feared flame retardants () incorporated in the study turned out to be in principle amenable to microbial degradation leading to central metabolism or carbon dioxide (CMCD, 100%), although four of them were also classified as precursors of eventually nonbiodegradable compounds (NB, ∼66%). The detailed predictions for all the sets of compounds mentioned in this section are available on-line at . The growing production of new chemicals make the early diagnoses of their environmental fate and their microbial metabolism necessary (; ). Previous attempts to predict biodegradability (for instance, UMBBDpredict) have focused on the identification of specific metabolic pathways that a compound might follow on the basis of the presence of predefined functional groups (; ). The user must choose one of the possible transformations to generate, iteratively, a virtual degradation route. Such a manual, iterative approach lengthens the procedure if a large number of chemicals are being analysed. In addition, strategies that focus on functional groups have the limitation of being restricted to predefined structures that have been manually collected. Other schemes such as Meteor and Catabol evaluate only the pathways that are most likely to occur, instead of predicting all possible pathways. In these cases, transformation rule libraries have to be constructed manually from the literature and generalised through chemical criteria (; ). As an alternative, we have tackled the problem from an experience-based perspective, using a computational machine learning approach trained with all known microbial catabolic reactions on organic pollutants (; ). This approach allows the combination of continuous and discrete attributes (descriptors), permits dealing with missing values, and generates classification rules in human-readable forms endowed with biological meaning. This is an important difference with other machine learning techniques (such as neural networks), which generate classifications with opaque rules. The application of such predictive system to lists of chemicals released into the environment thus represents an early tool for tentatively classifying the compounds as biodegradable or recalcitrant. The pivotal feature of our predictive system is the vectorial representation of chemical compounds as sets of 152 descriptors that express atomic composition and topology in terms of atomic triad frequencies (), plus MW and water solubility. This approach is related to some QSAR-type systems that use, as chemical descriptors, all possible subfragments of connected atoms that can be obtained from a compound (; ; ; ; ; ). In our hands, deconstruction of any given chemical as an assembly of atomic triads was a far superior descriptor of the molecules, biodegradation-wise, than any other representation tried. This codification seems to represent an optimal tradeoff between significant structural/chemical properties and the processing of a minimal number of attributes by the machine learning sytem employed. Indeed, because the reactivity properties of any given atom generally depends on its neighbours up to a distance of 2, triplets do hold a considerable information on the Chemistry of the molecule while keeping low the number of descriptors (possible atomic triads) per molecule. But is such a correlation casual or does it embody a biological meaning? We argue that the frequency of atomic triad presets the susceptibility of the compounds to the global biodegradation network. In fact, the outcome of the approaches presented in this paper suggest that enzymatic activities of catabolic pathways coevolve to target discrete molecular motifs which can be shared by many chemicals, rather than adapting to deal with one specific molecule, with obvious consequences for the evolution of the substrate recognition sites of the enzyme pool (). It is thus reasonable that confrontation of a diverse microbial community with a mix of chemical compounds (i.e., the most frequent environmental pollution scenario) results in the encounter of a multispecies biodegradation network with a landscape of chemotopes—rather than dealings of single type of bacteria with an unique chemical species. The consequences of such a situation for surveying the degradation gene scenery through experimental and computational means deserve further research. The strategy sketched in has two major incentives. First, it is fully automated and, therefore, it can be quickly applied to massive lists of compounds, as we have shown above. Second, it is not restricted to known functional groups and, therefore, it may provide hints about the environmental fate of compounds that contain undocumented structures, allowing an early prediction of their environmental fate before releasing them into the environment. Its simplicity makes it suitable as a screening predictive tool and provides and early rationale for putting interventions into practice and setting priority procedures. These applications will probably be intensified by the growingly restrictive European Union Regulatory Framework for Chemicals (REACH; ec.europa.eu/enterprise/reach) and other international rules, for example, the Pollution Prevention Framework (). In the meantime, our analysis () indicates that hundreds (if not thousands) of the compounds which are produced in large quantities by the chemical industry may not have a chance of ever being biologically degraded—at least as understood with our current level of knowledge of the microbial metabolism. In this respect, although our prognostic system says nothing on the possible kinetics of degradation of specific compounds, we expect predictive approaches of the sort presented in this paper to inform decisions about acceptability of the release of current and future chemicals into the environment. UMBBD () regularly compiles information on experimentally characterised biodegradative reactions. MetaRouter () is a system mainly based on UMBBD (, ) designed to maintain heterogeneous sets of data related to biodegradation and bioremediation. ChemFinder () is a database that contains a variety of information about all types of chemical compounds. The lists of compounds known as Annex-I, HPVC and LPVC were kindly provided by Rémi Allanou, of the (). The three lists included compound names and Chemical Abstract Service (CAS) Registration Numbers. The SMILECAS database contains SMILES strings for more than 100 000 compounds that are referred to by their names and CAS Registration Number, and was kindly provided by Bill Meylan, from Syracuse Research Corporation (). SMILES strings () are linear text representations of the atomic structure of molecules. Atoms are represented with the standard nomenclature and specific signs are used to express different types of chemical bonds and to denote branching, cycles, and other molecular features (). Although SMILES strings unambiguously represent the structure and connectivity of any given compound, each chemical may have several alternative SMILES strings. PubChem Compound () is a database maintained by the NCBI that contains information about more than five million unique chemical structures, including their SMILES strings. OpenBabel is a program and library designed to interconvert file formats used in molecular modelling and computational chemistry (). c4.5 is a machine-learning algorithm for the construction of decision trees and rule-based classifiers (). JME () is a Java applet that generates SMILES strings from drawings of chemical compounds produced with a graphical interface, and was kindly provided by Peter Ertl from Novartis AG. BioLayout JAVA is a program designed for the visualisation of biological networks (). All data preparation and manipulation was carried out by means of scripts written in Perl language. Each of the 903 reactions described in UMBBD as in November 17 2003, was deconstructed into all possible pairs of compounds that sustain a substrate–product relation. For example, for the reaction A → B+C, the following pairs of connected compounds would be generated: A → B, A → C. By assembling a single, nonredundant list of compound pairs, a directed graph representing the global biodegradation network was defined in which nodes correspond to compounds, and edges to reactions, as described previously (). Compositional and topological information about each chemical was expressed as series of atomic triad frequencies. Triads were preferred over other possibilities (pairs, tetrads, etc.) because the two nearest neighbours of any given atom in a molecule determine intrinsic reactivity the most. To deconstruct given compounds into such atomic triad series, the SMILES strings associated with each chemical was processed with the OpenBabel sofware (see above), which adds hydrogen atoms not explicitly represented in SMILES strings, and translates the results into BS and Alchem formatted reports. Atom names and connectivity information were extracted from such BS reports in the form of , whereas bond types were extracted from Alchemy reports. The frequencies of atomic triads were then calculated from the information on their connectivity. With such criteria, 149 different atomic triads were identified and categorised for each of the 850 chemical compounds included in the 17 November 2003 update of UMBBD (), and their absolute frequencies were determined for every target molecule. Water solubility figures were obtained from the ChemFinder database. Because solubility can be expressed either qualitatively or quantitatively, both types of records were mined from the database and converted into the reciprocal form according to a scale of solubility. Apart of the numerical data on solubility, an operative qualitative scale was set by examining the distribution of solubility values and classes found in the collection of compounds as indicated in . Alas, information on solubility was available for only 214 of the studied compounds. For the rest, the solubility values were left as , which is a circumstance that can be handled by c4.5 (). A modified version of the Tanimoto association coefficient (τ) was used for expressing the degree of chemical similarity between compounds in a fashion that was coherent with their description as series of atomic triads. This coefficient is particularly well suited for dealing with molecular representations consisting of strings of binary descriptors that may indicate, for example, whether predefined substructures are present or absent in a compound (). In our case, the Tanimoto association coefficient was calculated with nonbinary data (i.e., atomic triad frequencies) by means of the following formula: where and are the number of atomic triads in two compounds, and is the number of those that they have in common. The Tanimoto coefficient ranges between 0 and 100, and its value can be interpreted as the degree of identity between compounds relative to their atomic triad composition. Once chemical similarity for each pair of compounds had been defined, we studied the distribution of chemical distances for the whole set of compound pairs () and for pairs of compounds that belong to specific classes of environmental fate. We also examined to what extent the whole set of compounds and the environmental fate classes of chemicals involved clusters of similar molecules. To this end, we calculated the clustering coefficient () for each compound () with the rule: where is the number of compounds that are connected to , and is the number of connections between the compounds that are linked to (). To define whether two compounds were connected or not, we considered two different thresholds for the Tanimoto coefficient: τ⩾50% and ⩾80%. The lower and upper limits of are 0 and 1, respectively: compounds that are not connected to any other molecule are considered to have a clustering coefficient of 0, whereas those that belong to clusters in which many of the members are linked have clustering coefficients closer to 1. Finally, we calculated the average clustering coefficient for the whole set of compounds and for the specific environmental classes of compounds (). The average clustering coefficient of a given class represents the cliquishness of that set of compounds (). The machine learning algorithm c4.5 () was employed to generate rule-based classifiers that associate the properties of chemical compounds with one of the two predictable fates for each of the four independent binary classification schemes defined upon the analyses of the biodegradation network (see the Results section). To assess the predictive capacity of the system, we followed a fivefold cross-validation strategy, in which the data set was divided into five blocks; four of them were used as a training set, to generate the classifiers (rules). The remaining block was used as a test set, for measuring the ability of the classifiers in predicting the environmental fate of chemicals not included in the training set. The process was repeated five times, changing the block that was used as a test set, in such a way that all compounds were part of both sets, at least once. Only those compounds with an associated vector were taken into account (841 out of the original set of 850 compounds). As a basic measure of the predictive capacity of the classifiers, we averaged the robustness of the predictions for the five iterations of the cross-validation experiment. In this context, accuracy was defined as the percentage of correctly classified cases, relative to the total number of them taking together the majority and the minority classes. On the contrary, sensitivity is the fraction of compounds correctly classified as belonging to a specific class, relative to the total number of cases of that particular class. Specificity is the fraction of compounds correctly classified as belonging to a specific class, relative to the total number of predictions for that particular class. Therefore, the last two features (sensitivity and specificity) were independently calculated for the majority and the minority classes. This was made because the generalisation process carried out by c4.5 produces classification models in which one of the classes is defined as the one. This class is usually the most frequent in the training set, as it happens with the four classifiers generated by this work. When the models are used to classify new cases, those that are not covered by any rule are assigned to the default class. Therefore, by calculating the sensitivity and specificity of the predictions for each distinct class, it is possible to generate a more detailed estimation of the predictive performance of the system, than that represented by accuracy only.

Dynamic transcriptional regulatory networks underlie most complex cellular responses. Understanding the structure and coordinated behavior of these networks is fundamental to systems biology. Yeast has proven to be an excellent model for understanding eukaryotic transcriptional regulatory networks. Genome-wide chromatin localization and expression data reveal a general hierarchical four-tiered network structure (), and superimposed on this structure are compact units of recurring patterns in network architecture (). Each of these network motifs confers specific properties to the system including temporal control, coordinate expression with other genes, speed or stability of responses, sensitivity to continuous or transient stimulus and noise suppression (; ). Multi-input motifs involve regulation of a group of targets by multiple factors and have been attributed to a response being required for multiple growth conditions, or the involvement of the target genes in multiple metabolic pathways. Here, different stimuli activate different transcriptional regulators leading to the activation of both common and distinct classes of genes (). The effect of concomitant control of a multi-input motif, that is multiple factors regulating the same targets at the same time, has not yet been investigated on a global scale. However, it has been suggested that because there is so much overlap between the targets of individual regulators, this is likely an important mechanism to confer specificity to transcriptional responses (). The transcriptional response to fatty acid exposure serves as an excellent model for studying the concomitant control of multi-input motifs. Yeast cells respond to oleic acid exposure by inducing genes responsible for fatty acid metabolism, but, as with other external stimuli, they additionally coordinate other related processes (e.g. glucose metabolism, stress response, etc.). The transcriptional response controlling fatty acid metabolism has been characterized and is outlined in . For many genes related to this process, heterodimers of Oaf1p and Pip2p bind to upstream oleate response elements (OREs) and activate transcription in the presence of oleate (). The specificity of the response is controlled at two levels: Oaf1p is activated directly by oleate (), and the expression of is activated by Oaf1p/Pip2p heterodimers, as it has an upstream ORE. A third transcriptional activator, Adr1p, which is also involved in regulating genes involved in the metabolism of other carbon sources (; ), directly activates () and some other ORE-containing targets involved in fatty acid metabolism (feedforward regulation). Adr1p is necessary for full ORE-mediated activation and may also be necessary for derepression of these genes in the absence of glucose (; ; ). The extent of the influence of these factors on the transcriptome, and how the response is propagated to coordinate other cellular processes have not been systematically explored. Among the numerous coordinated cellular processes are two different stress responses (; ). One of these is an acute and immediate activation of an oxidative stress response, which is proposed to involve Yap1p and to be a response to fatty acid-induced uncoupling of the respiratory chain (). The other is a downregulation of general stress response genes, which appears to be related to metabolic reprograming in response to the environmental change and mediated, in part, by the exit of Msn2p and Msn4p from the nucleus (). Here, complementary high-throughput experimental techniques and various data integration strategies, including a novel network topology-based clustering method, were used to characterize a dynamic transcriptional regulatory network controlling fatty acid metabolism. Genome-wide condition-specific chromatin localization data were generated and used to construct physical interaction networks in the presence and absence of oleate. For each network, targets were clustered based on their network topology and the control of each cluster was inferred from statistical analysis of its size, and expression and functional properties of its members. This approach, combined with targeted experimental validation of the network, demonstrates that in this context, Oaf3p (heretofore uncharacterized) acts as a negative transcriptional regulator implicated in multiple cellular responses, and reveals a dynamic multi-input network structure in which Oaf1p acts as both a negative regulator of the general stress response and a positive regulator of the fatty acid metabolism response. The behavior of Oaf1p in this network suggests a mechanism by which the same regulator can control and synchronize related biological processes through involvement in different multi-input network motifs. We characterized a dynamic regulatory network involving four fatty acid-responsive transcriptional regulators and their primary targets to gain insight into the effects of their combinatorial control. The network was seeded with Oaf1p, Pip2p and Adr1p, the three activators known to conditionally cooperate to activate expression of genes involved in fatty acid metabolism (), as well as Ykr064p, which is renamed Oaf3p and which has been implicated as a transcription factor (; ), but has not been characterized (see below). Oaf3p was included because preliminary data suggested that it plays a role in regulating fatty acid-responsive genes: was first identified as one of 224 genes that displayed expression profiles similar to genes implicated previously in fatty acid metabolism or peroxisome biogenesis () (see also ), and it is a predicted Zn-Cys transcriptional regulator (; ). Oaf3p is also localized to the nucleus (), and a BLAST search of the yeast proteome revealed that Oaf3p is most similar to Pip2p and Oaf1p, with -values of 3.6 × 10 and 5.1 × 10, respectively (using WU-BLAST 2.0 with default parameters; released May 10, 2005; Gish, W (1996–2004) ) (). The targets of Oaf1p, Pip2p, Oaf3p and Adr1p were determined after growth in the presence of 0.1% glucose and 5 h after a switch to medium containing the fatty acid oleate as the sole carbon source. The 5 h time point was chosen to maximize the detection of targets of each factor based on the average expression profile of peroxisome-related genes during oleate induction (). At 5 h, upregulation of these genes is robust, but not yet maximal, suggesting that oleate-induced transcriptional regulation is active and not declining. The targets of each factor were identified by genome-wide chromatin localization analysis of myc-tagged versions of each factor (). Each strain was analyzed by chromatin immunoprecipitation (ChIP) followed by two-color DNA microarray analysis comparing DNA in the ChIP fraction to that in a whole-cell extract (WCE) on yeast intergenic microarrays (see Materials and methods). Data for three biological replicates of each experiment were merged and normalized and differential enrichment ratios were calculated. Next, VERA and SAM analysis tools were used to identify intergenic regions significantly enriched in the IP fractions (, ). This was carried out by using an estimated error model to improve the accuracy of the differential enrichment ratio and to assign a λ value representing the likelihood of differential enrichment for each intergenic region. For each experiment, a λ value threshold was chosen to yield an estimated false discovery rate (FDR) of 0.001 (see Materials and methods). For each condition, intergenic regions enriched in the ChIP fraction that were above the threshold were selected as targets for each factor (). Physical interaction networks representing chromatin localization data for all four factors were generated for each growth condition using Cytoscape software (). Networks of factor binding on oleate and low glucose are shown in (left panels). Each network consists of the four regulators (labeled nodes) connected by directed edges to the intergenic regions to which they bind (unlabeled nodes). After exposure to oleate, the number of targets in the network increased from 221 to 571. Specifically, the number of targets for Oaf1p, Pip2p, Adr1p and Oaf3p increased from 128, 52, 53 and 26 to 394, 212, 137 and 261, respectively. There was also an increase in network connectivity, as reflected by the >3-fold increase in connectivity score (targets with multiple edges/total number of targets) (). These data suggest the existence of oleate-specific functions for the four factors involving multi-input motifs. The large-scale ChIP networks were analyzed for multi-input motifs appearing to represent functionally relevant trends as described below. For each condition-specific network, targets were grouped into clusters based on their network topology (). For each cluster, the significance of the cluster size was determined by calculating the probability of having the observed size or greater (using the cumulative distribution function (CDF) with error correction) with the null hypothesis that each binding event of the four factors is independent (see Materials and methods). The results are displayed in (right panels) where network topology clusters are listed along the -axis and significance of overrepresentation for each is represented by a bar. For the glucose-specific network, only three topology clusters had significance scores greater than 2 (CDF -values <0.01) (colored clusters) and only one (AO; bound by Adr1p and Oaf1p) had a score greater than 25 (-value <1 × 10), reflecting very high confidence in cooperation of Adr1p and Oaf1p in the glucose network. In the oleate network, seven clusters had significance scores greater than 2 and three (AOY, OPY and AOPY; bound by combinations of Adr1p (A), Oaf1p (O), Pip2p (P) and Oaf3p (Y)) had scores greater than 25, suggesting more cooperation among the factors in the presence of oleate than in glucose. The conditional overrepresentation of these network motifs suggests oleate-specific regulation by at least three different multi-input motif instances (represented by AOY, OPY and AOPY clusters) with potentially different regulatory mechanisms and outputs. The topology clusters were also analyzed using an FDR of 0.01 for the large-scale ChIP data. This analysis yielded similar results, but increased the significance scores for highly connected, high-scoring network clusters (AOY, AOPY and OPY), in part, because lowering the stringency reduces false negatives in the network. False negatives have the effect of reducing the population of highly connected clusters and erroneously increasing the occupancy of various less-connected clusters. However, the lower stringency networks are larger (801 and 570 targets for oleate and glucose networks, respectively) and presumably contain an increased number of false positives. Therefore, to maximize the accuracy of the networks and to increase statistical power of subsequent analyses, ‘combined threshold' networks were generated, for which membership in the network was determined using the high stringency threshold (FDR ∼0.001) and topology of the network was determined using the FDR threshold of ∼0.01. Intergenic regions and their topologies for these networks are listed in . These networks were used for all subsequent analyses. To facilitate the integration of gene expression data with the network, each intergenic region was translated into target genes with adjacent start sites using the genome annotations generated by . Because of errors introduced by this conversion and technical limitations inherent to the genome-wide ChIP analysis (; ), rather than focusing on the characterization of individual targets in network clusters, we identified significantly overrepresented gene properties in each network topology cluster. This was performed using the CDF with error correction (see Materials and methods) to determine the probability of finding the observed (or greater) overlap between genes with a given property and genes in a topology cluster. The null hypothesis was that the given property is independent of membership in a topology cluster. This method of analysis not only provides a measure of confidence in the results, but also broadens the analysis to global trends and therefore is likely to reveal an insight into system level network regulatory function. The first gene properties measured were the results of a time-course transcriptome profiling study (), which was conducted under conditions similar to those of the ChIP experiments performed here (i.e. carbon source was switched from low glucose to oleate). For this analysis, several measurements were obtained within minutes after the switch to oleate, a time when transcription appears to be changing most rapidly; thus they classified responses that might otherwise be indistinguishable. This study identified five distinct expression profile clusters of oleate-responsive genes, and integration of gene ontology (GO) and DNA motif data revealed that the expression clusters represented different biological processes including oxidative stress response, general stress response and peroxisome biology (). To analyze the relationship between network structure and oleate-specific expression, we searched for significant overrepresentation of genes of each expression cluster in each network topology cluster. The analysis revealed significant enrichment of two of the five expression profile clusters in the networks (-value <0.01 for at least one topology cluster). Each of these clusters (peroxisome-related genes and general stress response genes) appears to be regulated by multiple factors in an oleate-specific manner. In the presence of low glucose, genes of the peroxisome-related expression cluster were overrepresented in the cluster targeted by Pip2p with low confidence, but were significantly enriched in the cluster targeted by Oaf1p, Pip2p and Oaf3p (OPY) or all four factors (AOPY) in the presence of oleate with higher confidence (). In contrast, the general stress response expression cluster was most significantly enriched in topology clusters targeted by Oaf1p (and/or Adr1p) (O/AO) in the presence of glucose, but by Adr1p, Oaf1p and Oaf3p in the presence of oleate (AOY) (). As mentioned above, the expression profile clusters were named based on correlations found between expression profiles and either GO annotations or the presence of transcription factor binding motifs. Therefore, network enrichments of relevant DNA binding motifs and GO Slim terms, which are a high-level view of GO terms (), were also analyzed (see Materials and methods). OREs () and Msn2p/Msn4p targets () had similar profiles to the corresponding expression clusters, peroxisome-related and general stress response, respectively. In addition, the distribution of OREs in the glucose network further suggests that some of these elements can be bound by a combination of Adr1p, Oaf1p and Pip2p in the absence of oleate. The GO Slim terms analyzed also had similar network distributions to the corresponding gene expression clusters (data not shown). For this analysis, genes annotated with ‘peroxisome' (component slim term) and ‘response to stress' (process slim term) were significantly enriched in oleate network clusters OPY (-value of 1.9 × 10) and AOY (-value of 8.8 × 10), respectively. Together these data suggest the dynamic regulation of two biological functions in the network, the oleate-induced upregulation of peroxisomes and downregulation of general stress response, which appears to involve oleate-specific targeting of the OPY and AOY topology clusters, respectively. To determine the transcriptional response of each topology cluster, the transcriptional activity of the oleate network was measured. First, microarray analysis was used to compare poly A RNA isolated from yeast grown in the presence of 0.1% glucose to that of yeast grown for an additional 5 h in the presence of oleate (see Materials and methods). Replicate experiments were merged and processed as described previously (), and VERA and SAM analysis tools () were used to generate an error model that was used to generate a λ value reflecting the likelihood of differential expression for each gene. Log expression ratios and λ values for each experiment are provided in . Next, genes with significant differential expression in response to oleate were identified (see Materials and methods). A total of 79 and 137 genes significantly increased and decreased in expression in response to oleate, respectively. The abundance of each class of genes in each topology cluster in the oleate network was statistically analyzed. This was performed using CDF with error correction as described above except that the null hypothesis was that the environmental change has no effect on genes in the topology cluster (i.e. the frequency of genes up- and downregulated by oleate in the cluster is equal to the estimated FDR in the expression data) (see Materials and methods). The results are shown in . The significance of enrichment of genes up- and downregulated after a 5-h oleate induction is represented by red (above the -axis) and green bars (below the -axis), respectively. The OPY (and OP, AOPY and O) topology clusters were enriched for genes that increased in expression in response to oleate, whereas the AOY (and OY) clusters were enriched for those that decreased in response to oleate. These data are consistent with the previous analysis () and a similar network topology analysis of different microarray expression data () comparing glycerol-grown cells to those grown in oleate for 6 h (data not shown). To study the influence of each factor on the network, microarray analysis was used to determine the transcriptome profiles of each deletion strain (Δ, Δ, Δ and Δ) compared to that of wild type after a 5-h induction in oleate. The deletion experiments identified 194, 104, 175 and 76 genes with significant differential expression, respectively. Genes significantly up- and downregulated were identified and data were overlaid onto the oleate network and statistically analyzed as described above. The topology clusters with the highest significance scores () were the same as those identified in the analysis of overrepresented clusters with data from oleate-induced wild-type cells ( and ). Clusters OPY (and AOPY and OP) were enriched for genes whose expression was reduced by deletion of , or . These data reflect the role of these three factors in the upregulation of peroxisome-related genes in the presence of fatty acids (), and the role of Adr1p in directly upregulating (), which is also supported by the Adr1p– interaction identified here (). There is also evidence of regulation of the AOY (and OY) clusters by Oaf1p and Adr1p. Adr1p appears to have a positive influence on genes of the AOY cluster, whereas Oaf1p appears to negatively regulate genes in this cluster (and in the OY cluster). This role in negative regulation appears to be independent of Pip2p because deletion of had no significant influence on the expression of these clusters. The negative regulatory activity of Oaf1p is not likely controlled by absence of dimerization with Pip2p as it has previously been shown that overexpression of can activate transcription from an ORE-containing promoter in a deletion strain (). Instead, regulation of this activity appears to involve a DNA-binding context because unlike clusters positively regulated by Oaf1p, those under negative regulation (AOY and OY in ) are not significantly enriched for OREs (). Interestingly, these data suggest that a third cluster, genes targeted by only Oaf1p in the oleate network, might have biological significance. This cluster is enriched for genes that increase in expression in response to oleate (panel A) and those that are positively regulated by Adr1p (panel D). We do not know the biological function of this enrichment, but it may be due in part to the presence of genes in the O cluster that bind both Oaf1p and Pip2p that are false negatives for Pip2p binding. This is consistent with the weak enrichment of OREs in this cluster (). To determine if the influences of the regulators detected here contribute to the oleate-specific expression patterns that are enriched in the network (), overlap between the three data sets (oleate topology clusters, expression profile clusters and discrete transcription factor deletion data) was analyzed. Topology clusters AOY and AOPY/OPY were reduced to include only the relevant expression cluster genes (14 general stress cluster genes and 10 peroxisome cluster genes, respectively) and then reanalyzed for significant enrichment of genes affected by transcription factor deletions. The intersection of the AOY cluster and the general stress response expression cluster was enriched for genes that are negatively influenced by Oaf1p, whereas the intersection of the OPY cluster and the peroxisome-related expression cluster was enriched for genes that are positively influenced by Oaf1p, Pip2p and Adr1p (all -values ⩽1 × 10), suggesting that the factors indeed contribute to oleate-specific expression patterns associated with these clusters. The data suggest that the negative regulatory activity of Oaf1p is independent of Pip2p. While the Pip2p-dependant role of Oaf1p is well characterized, the Pip2-independent role is not. Therefore, we measured the effects of transcription factor deletions on representatives of the AOY cluster in the oleate network. and , which are both negatively regulated by Oaf1p and not regulated by Pip2p (), were used as reporters. The levels of Dip5p and Ato3p GFP chimeras were determined after exposure of wild-type and isogenic mutant strains (Δ, Δ, Δ or Δ) to oleate (). Deletion of resulted in increased levels of both chimeras, supporting the conclusion that Oaf1p negatively influences the expression of these genes. In contrast, deletion of had no detectable effect on either protein, supporting the conclusion that Oaf1p does not cooperate with Pip2p in this context. Interestingly, deletion of resulted in increased levels of Dip5p-GFP (1.4-fold), suggesting that it is a negative regulator of . Consistent with this, the deletion of resulted in 1.5-fold increased expression of by microarray analysis (), but the λ value reflecting the likelihood of differential expression was 29, which fell below our threshold of 36.23. These data suggest that the influence of deletion on network gene expression, as detected by microarrays, is modest (see also ). To investigate further the role of Oaf3p, we analyzed the deletion array data by combining wild-type time-course expression, deletion strain expression and large-scale ChIP data sets. We first identified a set of 65 genes that were differentially expressed in the oleate time-course expression study (), and significantly upregulated at 5 h on oleate in the deletion strain as compared to wild type. Next, we determined the frequency of Oaf3p binding (on oleate) to the 53 of these genes that were represented on the intergenic microarray, and compared this frequency to the average frequency of Oaf3p binding among all oleate-responsive genes (see Materials and methods). Using this approach, we determined that Oaf3p binding is enriched among genes transcriptionally responsive to oleate and negatively regulated by on oleate (-value=0.0023). These data support a role for Oaf3p as a negative regulator of expression of its target genes in response to oleate. We therefore looked for further validation of Oaf3p influence on network targets by looking at the protein levels of its targets under various conditions. The effect of deleting on the levels of Cta1p (encoded by an AOPY group gene) was analyzed by immunoblotting (). A wild-type strain with a TAP-tagged version of Cta1p was compared to isogenic strains deleted for either or . In the wild-type strain, levels of Cta1-TAP were not detected in cells grown in raffinose-containing media (non-inducing condition), but levels increased after growth in the presence of oleate or antimycin, a second peroxisome-inducing condition (). Deletion of resulted in little or no induction as expected from the previously published data (), but deletion of resulted in oleate expression levels that were higher than those in the wild-type strain. The protein levels of other Oaf3p targets were also analyzed in an overexpression strain. Various GFP-tagged strains were transformed with pYEX-OAF3 (a plasmid with under the control of the copper-inducible promoter of ), induced for 20 h in medium containing copper and oleate, and analyzed by fluorescence-activated cell sorting (FACS). The results are shown in . In most cases tested, overexpression of (red lines) resulted in decreased protein levels of the corresponding target gene (panels 1–4; 1.5- to 2.4-fold decrease), but in some cases, no detectable effect on protein levels was evident (e.g. Faa1-GFP). This is not unexpected, as not all members of the topology groups responded identically to other perturbations. As overexpression of had a negative effect on the transcription of , a gene involved in fatty acid metabolism, an overexpression strain was also analyzed for its ability to metabolize the fatty acid myristic acid (). An equal number of cells of a wild-type strain transformed with either the empty plasmid or the overexpression plasmid was induced with copper and then grown on turbid fatty acid medium. Overexpression of resulted in a reduction in the size of the halo around the cell patches, indicating a reduced ability to metabolize fatty acids (). Taken together, these data suggest a role for Oaf3p as a negative regulator of transcription in the presence of oleate. To explore the potential for temporal coordination of the network, representative GFP-tagged strains were analyzed by FACS after growth in the absence of oleate (YPBG) and after 5 and 20 h inductions in the presence of oleate (YPBO) (). Consistent with the gene expression data (), levels of proteins corresponding to the AOY cluster decreased in the presence of oleate, whereas those corresponding to the OPY and AOPY clusters increased in response to oleate. Interestingly, although the decreases in expression were clear after a 5-h induction, increases in expression were prominent after 20 h in oleate. These data appear to reflect the fact that the transcriptional response of general stress response genes precedes that of the peroxisome-related genes as identified previously by microarray analysis () (). This suggests temporal coordination of the two biological processes and provides insight into the function of the network as described in Discussion. A dynamic transcriptional regulatory network generated from genome localization and transcriptome profiling data was characterized using a novel topology-based clustering approach. The analysis used simple, widely available tools that can be applied to the characterization of other regulatory networks. Results from this analysis were integrated with literature data to infer a dynamic model of network function (). This was carried out by first generating a model for the 5 h time point. For this condition, oleate-specific interactions (edges) between regulators and targets were inferred from ChIP data (; and ), negative (red) and positive (green) influences were inferred from these data along with expression and protein abundance data of wild-type, deletion and overexpression strains (, and and ). The states of these network connections at other time points (before and after 5 h) are predictions made from these data along with time-course microarray data of the factors (; ) and network targets (; ; ) and from protein levels of the target genes at different time points (). Other interactions were added from data in the literature as indicated. The network is described in detail below. Immediately after the carbon source switch (immediate response; top panel), the presence of oleate is recognized directly by constitutively present Oaf1p (), which binds upstream of genes of the AOY, general stress response cluster. Genes of this cluster are known to be acutely and transiently activated when respiration is turned off (), and may be involved in reprogramming cellular metabolism in response to changes in respiratory state (). Here, these genes are transiently repressed by Oaf1p (and likely other factors). This Oaf1p-mediated repression appears to coincide with an efflux of Msn2p and Msn4p from the nucleus, which also contributes to maintaining these genes in an off state before subsequent metabolic reprogramming (see below) (). The carbon source switch signals the rapid and transient increase in expression levels of (immediate response; top panel; ). As transcripts accumulate and are translated, Adr1p levels begin to rise, which positively and directly influences the expression of (), genes in the AOY, and AOPY clusters and likely other pathways controlling utilization of other carbon sources (; ) (early response; second panel). The third panel shows the response approximately 5 h after the carbon source switch. Here, the activation of by Adr1p leads to accumulation of Pip2p, which (as a heterodimer with Oaf1p) transmits the oleate signal to genes directly involved in fatty acid metabolism and positively influences genes of the OPY and AOPY clusters. As the levels of mRNA are declining at this time (), the direct influence of Adr1p on its targets is shown as a dashed line representing weak influence in this panel. At this time, Oaf1p/Pip2p dimers also feedback positively on the expression of . This self-regulation, in combination with the Adr1p-mediated activation of and peroxisome-related genes of the AOPY cluster, constitutes coupled feedback and feedforward circuitry to activate peroxisome-related genes. The dual influence of Oaf1p on OPY and AOY clusters appears to mediate temporally synchronized regulation: the negative influence of Oaf1p on general stress response genes immediately follows the carbon source switch because levels of Pip2p (and thus Oaf1p/Pip2p heterodimers) in the initial condition are low. Time-delayed accumulation of Pip2p resulting from feedforward and feedback regulation (; ) subsequently leads to dramatic upregulation of peroxisome-related genes by Oaf1p/Pip2p heterodimers. The late response (lower panel) shows that the influence on the peroxisome-related cluster remains strong as the negative influence on the general stress response subsides. This difference in duration reflects steady-state gene expression data () (), and is likely caused by the rising Pip2p level, which increasingly draws more Oaf1p molecules from AOY to OPY and AOPY targets through heterodimerization. As discussed earlier, Oaf3p appears to be a weak negative regulator in the network (). Its influences in the 5 h and late response panels reflect expression levels, which are reduced immediately after a shift to oleate and gradually increase over time (). The specific function of Oaf3p in the network is not yet known. We and others have shown that transcriptional repression can ensure a more homogeneous expression level of a target gene across a cell population, in the context of network structure such as negative feedback regulation (; ; ; ). Kinetic modeling of the response, as performed to understand negative regulation in the galactose utilization pathway (), will help to elucidate its role. Modeling can be facilitated by refining the qualitative network model by investigating other potential influences such as other transcriptional regulators not yet considered, as well as mRNA and protein degradation, which have been implicated previously in regulating carbon source utilization (). The topology-based clustering and analysis methods outlined here identified a transcriptional regulatory network involving the participation of a bi-functional regulator in multiple multi-input network motifs to control and synchronize related biological processes. The data suggest that, beyond providing specificity in promoter binding, heterodimerization of transcription factors can contribute temporal control of tightly coordinated biological responses. Haploid deletion strains are isogenic to either BY4742 or BY4739, and are from the commercially available deletion set (Invitrogen, Carlsbad, CA). Haploid strains with myc-tagged genes were made by genomically tagging target genes with the sequence encoding 13 copies of the c-myc epitope from pFA6-13MYC () by homologous recombination into BY4742 using a previously described PCR-based procedure (). Strains with no apparent growth defects and appropriately sized fusion proteins were used. Haploid strains tagged with (S65T) GFP or TAP tag are isogenic to BY4741 and are from the commercially available GFP-clone collection (Invitrogen, Carlsbad, CA) or TAP fusion collection (Open Biosystems, Hunstville, AL), respectively. Strains containing both gene deletions and gene tags were made by mating, sporulation and tetrad dissection. For all experiments, control strains were isogenic to test strains. Unless otherwise stated, strains were grown in YPD (1% yeast extract, 2% peptone, 2% glucose); SCIM (1.7 g yeast nitrogen base without amino acids and ammonium sulfate (YNB−aa−as)/l, 0.5% yeast extract, 0.5% peptone, 0.79 g complete supplement mixture/l, 5 g ammonium sulfate/l) containing either 0.1% glucose or 0.5% Tween 40 (w/v) and 0.15% (w/v) oleate, or both; or YPB (0.3% yeast extract, 0.5% potassium phosphate (pH 6.0), 0.5% peptone) with either 3% glycerol (YPBG) or 0.5% Tween 40 (w/v) and 0.15% (w/v) oleate (YPBO). For each transcriptional regulator, genome-wide chromatin localization experiments were performed to comprehensively identify DNA-binding locations by ChIP followed by microarray analysis as developed previously (; ). YPD-grown cultures of strains containing myc-tagged regulators were grown in SCIM medium containing 0.1% glucose for 16 h to a density of ∼8 × 10 cells/ml, and then transferred to SCIM containing 0.1% glucose or 0.15% oleate and 0.5% Tween 40, and grown for an additional 1.75 or 5 h, respectively. Proteins were crosslinked to their cognate DNA binding sites (and each other) with formaldehyde. Cells were disrupted and chromatin sheared into fragments by glass bead lysis followed by sonication. myc-tagged factors were collected by ChIP with magnetic beads. Crosslinking was reversed in fractions of the ChIP and WCE, linkers were annealed to the DNA ends and DNA was amplified and labeled by PCR in the presence of Cy5-dUTP and Cy3-dUTP, respectively. DNA in ChIP and WCE samples was compared by hybridizing both samples together to yeast intergenic DNA microarrays. For each experiment (i.e. ChIP of one factor under one growth condition), there were three biological replicates, each hybridized to different microarrays. As each array contains four replicates of each intergenic region, the total number of replicates for each condition is 12. Data were processed by a previously published method () using the SBEAMS microarray database software (). Processing included normalizing scan intensities, subtracting background, merging data from replicate spots and generating an error model. A likelihood statistic, λ, was computed for each intergenic region to determine whether its abundance was significantly enriched in either the ChIP fraction or the WCE fraction. Two threshold λ values were chosen that yielded an approximate FDR of 0.01 and 0.001, corresponding to 64 and 6 false positives per 6438 intergenic regions, respectively. This was determined by utilizing the fact that differential enrichment for ChIP microarray data is one directional (i.e. DNA targets tend to be enriched in the ChIP fraction only) as opposed to expression array data, for which genes can go up or down in expression. For each experiment, the λ value was identified, above which there were 64 or 6 targets enriched in the WCE fraction instead of the ChIP fraction (false positives). Intergenic regions with λ values above this threshold (minus the known false positives enriched in the WCE fraction) were chosen as target intergenic regions. Genes with start sites adjacent to these target intergenic regions were chosen as target genes. For each condition, physical interaction data for each of the four factors were combined and graphically displayed using Cytoscape network visualization software version 2.2 (). Targets were grouped based on their network topology. For every topology cluster, the CDF () was used to calculate a -value equal to , the probability of the cluster size being equal to or greater than the observed size by chance (using R 2.3.0), with the null hypothesis that the four factors have independent sets of targets. is the observed cluster size, is the number of trials (the number of intergenic regions on the array in this case) and is the probability that a given target intergenic region will have this particular network topology assuming the null hypothesis. The calculation of is shown in , in which represents an estimate of the population size and is the number of intergenics on the array and is the number of targets of a given transcription factor. The subscript denotes the subset of the four transcription factors analyzed that target the cluster, whereas represents those that do not. An example calculation shows the probability of an intergenic region being in cluster AO (i.e. targeted by Oaf1p and Adr1p but not by Pip2p or Oaf3p) (). To reduce error due to multiple statistical comparisons of a single data set, the α-level was adjusted to 15 representing each of the clusters analyzed (Bonferroni correction). In the graphs in , bars reflecting statistical significance are shown only if the cluster contained more than two members. Time-course expression profiles and expression profile clusters for genes in response to oleate exposure were obtained from previously published microarray analyses (; ). GO Slim terms were downloaded from the Genome Database website (). For binding motif enrichment studies, Fuzznuc, the nucleic acid pattern search algorithm component of EMBOSS software (), was used to identify intergenic regions containing one or more ORE(s) conforming to the consensus CGGNTN(A/G)NCCG as defined previously (). Msn2p/Msn4p targets were defined as intergenic regions that interacted with one or both of the factors in genome localization data () as determined by (<0.005 and conserved in three species). The experimental conditions of this data set include various stress-inducing environments. To facilitate the integration of gene attribute and gene expression data with the network, each intergenic region in the combined threshold networks () was translated into target genes with adjacent start sites using the genome annotations generated by . Topologies of gene targets are given in . Genes assigned to more than one network topology, due to the fact that multiple intergenics regions can be assigned to one gene, have one entry with the multiple topologies merged together. For each test, the CDF formula and Microsoft Excel software were used to calculate a -value, equal to the probability that by chance, the enrichment of the attribute in the topology cluster is greater than or equal to that observed, with the null hypothesis that annotation with the gene attribute and membership in the topology cluster are independent. The equation used is similar to except that is the observed number of genes with the attribute in the topology cluster, is the total number of genes in the topology cluster and is replaced by , the probability of a gene with the attribute being in the cluster assuming the null hypothesis. To reduce error due to multiple statistical comparisons of a single data set, the α-level was adjusted to 15 representing each of the clusters analyzed (Bonferroni correction). In all graphs, bars reflecting statistical significance are shown only if the cluster contained more than two members with the attribute. For analysis of binding motifs, intergenic region networks were used instead of gene networks. For the comparison of mRNA levels in each of four deletion strains (Δ, Δ, Δ or Δ) to those in wild-type cells and for the wild-type versus wild-type control experiment, all strains were grown in YPD overnight and then transferred to SCIM with 0.1% glucose and without oleate and Tween 40, and grown for 16 h to ∼8 × 10 cells/ml, oleate and Tween 40 were added to the medium and cells were grown for an additional 5 h and then harvested. For the comparison of glucose- to oleate-grown cells, the experiment was carried out the same way except that the reference culture was harvested before the 5-h oleate induction. For all experiments, poly A+ RNA was extracted and cDNA was synthesized with incorporated Cy3 or Cy5 fluorescent dyes, and equimolar amounts of each label were mixed and hybridized to yeast ORF oligonucleotide microarrays as described previously (). There were two biological replicates for each experiment, and for each replicate both label orientations were analyzed on arrays containing four replicate spots of each gene, resulting in a total of 16 replicate spots per gene. Spotfinding was performed using Analyzer DG software (Molecularware, Irvine, CA) and data analysis was carried out as described previously () except that the λ likelihood value threshold of 36.23 was chosen which resulted in an FDR of 0.01 as determined from a wild type versus wild-type control experiment. Genes having λ values above the threshold were annotated as up- or downregulated as a result of the deletion, and genes with λ values below the threshold were annotated as unchanged. The statistical analysis of the representation of genes up- and downregulated in each network topology cluster was performed as described above for other gene attributes except the null hypothesis was that the environmental change has no effect on genes in the topology cluster. For example, for the analysis of upregulated genes, -value (the probability of a gene being upregulated and in the cluster assuming the null hypothesis) is 0.005, or half of the estimated FDR of differential expression. In cases where the actual rate of upregulated genes was lower than the estimated FDR, the actual rate was used. First, we determined the significance of enrichment of genes negatively regulated by in a subset of genes transcriptionally responsive to oleate. Time-course microarray data measuring the transcriptional response to oleate were taken from . The λ values for all microarray expression ratios for this and the deletion experiment were normalized by dividing by the mean un-normalized λ value for the wild-type versus wild-type control experiment described above. -values were computed from the normalized λ values using the right-tailed CDF of the χ distribution with one degree of freedom (). For each gene, a vector of -values was obtained from all the ratios of the time-course experiment. From this vector, a -value threshold of 0.0012 was determined to correspond to an FDR of 0.005 (). Using this -value threshold, 3871 genes were determined to be significantly differentially regulated relative to the glycerol condition, in at least one replicate-combined wild-type time-course measurement. These genes were then analyzed in the Δ versus wild-type experiment. A total of 2240 genes (of the 3871) were found to have a positive expression log-ratio Δ/wild type. From the vector of -values for these genes, an FDR of 0.1 was found to correspond to a -value threshold of 0.0057. A total of 65 genes were found to be significantly differentially expressed in Δ relative to wild type, based on this significance threshold. The binding of Oaf3p to the 53 (out of 65) -repressed genes represented on the ChIP array was then analyzed. The enrichment of binding of Oaf3p (in oleate) to the 5′ flanking intergenic regions of the 53 genes was computed, relative to the background set of all 3387 oleate-responsive genes on the ChIP array (resulting in a more conservative enrichment estimate than if all intergenic regions represented on the ChIP array were used as the background set). Using the CDF of the hypergeometric distribution, the enrichment -value was found to be 0.0023. The plasmid pYEX-OAF3 was made by amplifying the open reading frame of from genomic DNA by PCR with primers containing flanking H1 and 1 sites, and ligating into the corresponding sites of pYEX-BX (Clontech Laboratories, Mountain View, CA). Strains analyzed by FACS were grown overnight in CM−ura−leu (0.77 g CSM minus uracil and leucine (BIO 101 (Carlsbad, CA))/l, 1.7 g YNB−aa−as/l, 5 g ammonium sulfate/l) containing 3% glycerol and 0.5 mM CuSO, transferred to CM−ura−leu containing 0.5% Tween 40, 0.15% oleate and 0.5 mM CuSO and grown for 23 h. Clear zone assays were performed as described previously () except that BY4742 cells containing either pYEX-OAF3 or pYEX-BX were grown in CM−ura containing 2% glucose overnight and induced for 2 h by adding 0.5 mM CuSO to the medium. Cells were washed with water and ∼10 000 cells of each strain were spotted onto solid minimal myristate medium (1.7 g YNB−aa−as/l, 5 g ammonium sulfate/l, 0.5% potassium phosphate buffer, pH 6.0, 0.77 g CSM minus uracil/l, 2% agar, 0.5% Tween 40, 0.125% myristic acid) and grown for 3 days. Fluorescence intensity of individual cells was measured using an FACS Caliber flow cytometer (BD Biosciences, San Jose, CA). Data analysis was performed using WinMDI 2.8 (available from ), a forward scatter gate of >52 units, event normalization and smoothing of 25 units.

Substrates of the ubiquitin pathway are covalently modified by the attachment of a small protein called ubiquitin and as a result are targeted for degradation or other cellular fates (; ; ). Ubiquitination involves the sequential action of three enzymes: E1 (ubiquitin-activating enzyme), E2 (ubiquitin-conjugating enzyme) and E3 (ubiquitin-protein ligase) (). The E3 enzyme, which is responsible for the specificity of the reaction, associates with substrates (; ; ), and defects in this interaction have been implicated in numerous diseases (; ; ; ). A significant fraction of the proteome is regulated by the ubiquitin pathway and eukaryotic genomes express hundreds of E3 ligases to coordinate the ubiquitination of cellular proteins (; ). Currently, most E3 enzymes have not been linked to any specific substrate and any platform that would allow for the systematic discovery of enzymatic E3 substrates would be tremendously useful for advancing our understanding of the ubiquitin pathway. Rsp5 is a yeast E3 ubiquitin ligase that belongs to the Nedd4 family (). It contains a C2 domain, a catalytic HECT domain and three WW domains that can bind substrates directly by recognizing a (L/P)PxY sequence (PY motif) (, ; ; ; ). Ubiquitination of proteins by the Nedd4 E3 family has been implicated in numerous cellular functions, including endocytosis, sorting and trafficking (; ; ). For example, Nedd4 (or Nedd4-2), the human Rsp5 homologue, ubiquitinates the epithelial sodium channel (ENaC) to regulate its endocytosis, and mutations that inhibit the Nedd4-2:ENaC interaction cause Liddle syndrome, a hereditary hypertension (; ; ). Similarly, Rsp5 was demonstrated to regulate endocytosis and sorting of several yeast plasma membrane proteins (; ). Moreover, Rsp5 has been implicated in the regulation of several other cellular functions, including mitochondrial inheritance, drug resistance, intracellular pH, fatty acid biosynthesis and transcriptional control (see below). Despite the biological importance of the Nedd4/Rsp5 family of E3 ligases, only a few substrates have been identified to date for this ubiquitin ligase family. Thus, our goal was to globally identify Rsp5 substrates in the yeast proteome. For that, we chose to use protein microarray technology as our experimental platform. The arrays used in this study contain thousands of purified proteins (most of the proteome) immobilized at a high spatial density on standard sized slides and can be readily used to probe the yeast proteome using traditional biochemical approaches (; ; ; ; ; ). To date, few studies have assayed enzymatic activities using this technology. In the current study, we have successfully used yeast () protein microarrays to assay the enzymatic (ubiquitination) activity and binding of Rsp5 to its substrates, and we have identified previously reported and novel ubiquitinated substrates and interacting partners of this E3 ligase. Our results also demonstrate how this approach can yield informative data regarding the binding mechanisms and substrate specificity of an E3 enzyme. For this study, ubiquitinated Rsp5 substrates were identified using commercially available yeast protein microarrays (Invitrogen ProtoArray Yeast Proteome Microarray). These protein microarrays are based on technology described previously () and contain more than 4000 GST- and 6 × HIS-tagged yeast proteins from spotted in duplicate on nitrocellulose slides (ProtoArray Yeast Proteome Microarray nc v1.1). Before assaying for ubiquitinated proteins on the protein microarray, we developed conditions in which Rsp5 could ubiquitinate one of its known substrates, the C-terminal domain of Rpb1 (CTD) (). The ubiquitination of CTD was dependent on the budding yeast E1 enzyme, an E2 enzyme (Ubc4), ubiquitin, Rsp5 and ATP, and was visualized by Western blotting. This control reaction was used to optimize conditions for Rsp5-dependent ubiquitination on nitrocellulose-coated glass slides. In these experiments, the CTD and other proteins were robotically spotted onto slides and incubated with a reaction mixture containing Rsp5 and FITC-labeled ubiquitin. The proteome array was then assayed for Rsp5-dependent ubiquitination using the optimized conditions (). Following the reaction, protein microarray slides were washed, scanned and proteins modified by ubiquitin were identified by quantifying the intensity of the FITC signal produced compared with the background (). Although detection of protein–protein interactions on microarrays is generally highly reproducible (; ) (), we repeated the Rsp5 ubiquitination reaction on two separate protein microarray slides to increase the quality of our data set. Based on selection criteria for identifying positive hits described in Materials and methods, we generated a data set of 150 Rsp5 substrates (henceforth referred to as the ‘relaxed Rsp5 substrate set'). From this set of substrates, we selected a ‘high-confidence' data set, which comprises the 40 proteins that produce the strongest signal. These proteins were considered for further study ( and ). #text xref italic #text We used an established ubiquitination assay to confirm that the proteins identified as Rsp5 substrates on the protein microarray are modified by this E3. Traditional approaches for monitoring ubiquitination involve subjecting specific purified proteins to ubiquitination by an E3 and using a Western blot approach to visualize ubiquitination. Fifteen proteins from the Rsp5 high-confidence substrate set, and six proteins that were not identified as substrates of Rsp5, were purified from yeast using glutathione affinity purification, incubated in ubiquitination reactions containing Rsp5 and the above described E1 and E2 (Ubc4), and assayed for ubiquitination using anti-ubiquitin antibodies and Western blots. All of the proteins whose ubiquitination was detected on the protein microarray were verified to be ubiquitinated by Western blot analysis (; ). Most of the proteins were efficiently polyubiquitinated or ubiquitinated on multiple lysines. In contrast, the six proteins tested whose ubiquitination was not detected on the protein microarray did not appear to be ubiquitinated after Western blot analysis (), confirming that the enzymatic activity detected is specific and that the data generated by the protein microarray approach are consistent with established methods of detecting ubiquitination. To further validate our data, we tested for ubiquitination of several putative substrates (known or suspected to be involved in sorting/endocytosis), by comparing ubiquitination of these proteins expressed in (WT) or mutant yeast cells. is a temperature-sensitive mutant that reduces Rsp5 expression upon temperature shift to 37°C (an -null mutant is lethal). As shown in , Lsb1 and Sna3 (both known interactors or substrates of Rsp5; ; ; ), as well as Sna4, were ubiquitinated by Rsp5. Although the function of Sna4 is unknown, it is a vacuolar resident protein, much like Sna3, and we thus anticipate that it too utilizes interactions with Rsp5 for vacuolar targeting. Our preliminary data also revealed ubiquitination of other substrates by Rsp5 (e.g. Yip5, Rcr1 and Rcr2—data not shown). To directly test Rsp5 substrate binding using the protein microarrays, and to compare these data to the ubiquitination data sets above, we screened the protein microarrays for proteins that bind Rsp5. Purified Rsp5 was labeled with Alexa 647 and incubated with the protein microarray in two separate experiments. After washing and scanning the slides, the data were analyzed and a data set of 155 Rsp5 binding proteins was generated ( and ). A sequence search revealed that the Rsp5 binding set was significantly enriched for proteins containing PY motifs (<0.01—exact randomization test). Ten proteins in the Rsp5 binding set have previously been identified in Rsp5 pathways. Twelve proteins in the high-confidence Rsp5 substrate set and 52 proteins in the relaxed Rsp5 substrate set were also present in the Rsp5 interaction set. Conversely, 34% of the proteins in the Rsp5 interaction set were ubiquitinated by Rsp5. Eleven of the 12 proteins that both bound to and were ubiquitinated by Rsp5 contain PY motifs. We examined the amino-acid sequences of Rsp5 substrates to determine whether additional amino-acids residues in the PY motif ((L/P)PxY) may contribute to substrate specificity. A total of 38 PY motifs were present in 29 proteins of the high-confidence subset. Considering the third (x) position in the motif, the most frequent motifs were PPY (ten), PPY (five) and PPY (five). Comparisons with sets of randomly selected proteins containing PY motifs showed that Ser and Ala (but not Pro) were both significantly overrepresented at the third position within our experimentally determined Rsp5 substrates (<0.001; randomized exact test) (). To further confirm these findings, we performed a modified phage display screen to explore substrate specificity of each of the three WW domains of Rsp5. All peptides identified through this screen (over 300) were found to contain a PY motif. Consistent with data presented above, Ser and Ala were both found to be preferred at the third position (). More interestingly, the most common amino-acid residue associated with the third position was Pro, suggesting an important biological role for this residue. In this report, we demonstrate that protein microarrays can be used to identify, on a global scale, ubiquitinated substrates and binding partners of a yeast E3 ubiquitin ligase, Rsp5. A combination of techniques was used to validate the protein microarray data and contributes to our understanding of Rsp5 substrate interaction mechanisms. The high-confidence Rsp5 substrate set contains 12 proteins previously reported in Rsp5 pathways. Six of these (Ygr068c, Aly2, Lsb1, Ylr392c, Dia1 and Rim4) were reported in other HTP screens (; ; ; ), while the remaining six (Rod1, Rog3, Rvs167, Bul1, Sna3 and Ack1) were validated as substrates using a combination of genetic and biochemical approaches (; ; ; ). Most of the high-confidence Rsp5 substrates contained at least one PY motif, usually PPxY (). However, a few substrates did not (e.g. Sgt1, Cue5, Sip5). Sgt1, Cue1 and Sip5 are known to be involved in the ubiquitin pathway. The precise role of Sgt1 is not clear, but the association of this protein with Rsp5 is interesting, since it has been implicated as an activator of SCF E3 enzymes (; ). Cue1 has a ubiquitin binding motif and its affinity for ubiquitin may facilitate its monoubiquitination (). Alternatively, it is possible that it might have bound FITC-ubiquitin non-covalently; however, this is unlikely because its ubiquitination was also detected on a Western blot. Sip5 may not be an Rsp5 substrate, since it has a RING/U-box domain and likely produced a positive signal in the screen because it used the ubiquitin machinery present during the reaction for autoubiquitination. In addition to the known Rsp5 substrates described above, the relaxed Rsp5 substrate set contains six other proteins that were previously identified as Rsp5 substrates or implicated in Rsp5 pathways. These include, Rpb7 (), Tef2 (), Ubi4, Uba1 (), Rpl40B () and Rpl40A (; ). The identification of 18 proteins known to participate in Rsp5 pathways or to be ubiquitinated directly by Rsp5 suggests that the protein microarray experimental approach is a valid tool for the discovery of ubiquitinated E3 substrates, and that this approach is capable of discovering physiological substrates of Rsp5. Not all known Rsp5 substrates were identified in our screen. First, some of the known substrates were not printed on the array (e.g. Mga2, Rpb1, Hpr1, Bsd2 and Pma1). Second, our approach is likely to have missed Rsp5 substrates that do not bind Rsp5 directly or require cofactors for their interaction. This is a plausible explanation because some of the known substrates that were not identified on the array (Gap1, Fur4, Rfa1, Zrt1 and Tat2) do not have PY motifs, do not bind Rsp5 directly, and may require adaptor proteins (e.g. Bul1 and Bul2; ) to bind Rsp5. Third, it is impossible to tell how the purification and printing process used to make the array may have affected the accessibility of some substrates to bind Rsp5. In the high-confidence data set, 20 novel Rsp5 substrates were identified. Consistent with the well-established role of Rsp5 in ubiquitinating proteins at the plasma membrane or Golgi and affecting their sorting to vesicles, endosomes and to the vacuole, there are nine proteins in the high-confidence Rsp5 substrate list (Ymr171c, Rcr1, Rcr2, Sna3, Sna4, Yip5, Ydl012C and Alg6) that localize to either the plasma membrane, the vacuole, or have otherwise been implicated in the secretory pathway. Further characterization of these substrates is necessary in order to better understand their role in Rsp5-dependent cellular pathways. Twelve proteins in the high-confidence Rsp5 substrate set, and 52 proteins in the relaxed Rsp5 substrate set, were also present in the Rsp5 interaction set. Of the 12 proteins that bound Rsp5 in the high-confidence substrate set, seven had been previously described in Rsp5 pathways, and four are novel substrates. A recent study used the same protein microarray to find binding substrates for individual WW domains of Rsp5 (). Their network of interactions identified 124 interactions, of which eight had previously been reported as Rsp5 substrates (compared with 10 known substrates in our Rsp5 interaction data set). Fourteen proteins in this Hesselberth data set overlapped with our high-confidence Rsp5 substrate set and 58 overlapped with the relaxed Rsp5 ubiquitinated substrate set. Fifty-eight proteins from the Hesselberth data set are also present in the Rsp5 interaction set. The number of known substrates identified by probing the protein microarray for enzymatic (ubiquitination) substrates as opposed to binding partners of Rsp5 suggests that this experimental approach results in a much higher quality data set. One explanation for this is that the binding affinity for WW domains and PY motifs is relatively weak (low to mid micromolar range) (, ) and transient Rsp5 substrate interactions may therefore be missed when probing the array for interactions. If these transient interactions result in the enzymatic transfer of ubiquitin, however, it will likely be detected because ubiquitin becomes covalently, and therefore permanently bound to its substrate. Furthermore, polyubiquitination (or multi-monoubiquitination) of substrates on the proteome microarray may result in the amplification of fluorescent signal and thus a higher sensitivity may be achieved. Finally, in addition to being more sensitive, probing for ubiquitination is a more direct assay of Rsp5's biological role and therefore more likely to yield information that is more physiologically relevant. As expected, the high-confidence substrate set was significantly enriched for proteins containing (L/P)PxY motifs. Statistical analysis of proteins identified from binding and ubiquitination protein microarrays and the protein screens from phage display experiments, reveal a preponderance of Pro, Ala and Ser residues in the third (x) position of the PY motif. These findings are in accordance with a survey of previously reported Rsp5 substrates containing PY motifs; Ala and Ser are found in the PY motifs of six proteins (Sna3, Bul1, Bul2, Rod1, Rog3 and Rvs167) and absent in two (Spt23 and Mga2; ). Furthermore, it was shown that the WW domains from three distinct proteins (KIAA0082, Ras GTPase-activating-like protein IQGAP1 and Transcription Factor CA150) have a binding preference to motifs with Pro and Ala at the third position (). Finally, an earlier phage display study, which characterized the binding preferences of WW domains from Rsp5 and other proteins, had also demonstrated that Ala and Ser residues are favored within the PY motif of the first domain, whereas Pro was found to be most abundant in the second and third domains (). The mechanism of how Rsp5 recognizes PY motifs in substrates is not well understood. By highlighting the preference for a limited number of residues within the (L/P)PxY motif, these analyses may provide additional insights into the mechanism of substrate recognition by Rsp5 and ultimately a better general understanding of WW domain–substrate interactions. The availability of complete genomes for an additional 17 species of related fungi (Ascomycetes) was exploited to explore the evolution of the Rsp5 substrates identified in . Orthologous sets of proteins for each Rsp5 substrate were generated and used to determine the presence of conserved (L/P)PxY motifs ( and ). Conservation of the PY motif was observed in the majority of orthologues, particularly within the closely related species. Comparisons of Ka/Ks ratios for the orthologues identified across other fungal species found that the nucleotide sequence underlying the (L/P)PxY motif is under stronger purifying selection (and hence more highly conserved) than the neighboring residues (Ka/Ks <0.1; )), reflecting their functional importance. In addition, this was associated with the conservation of Ser, Pro and Ala at the third position (position ‘x' in the (L/P)PxY motif) in all but two orthologous sets (), in agreement with the data shown in and . These data are consistent with the hypothesis that the sequence (L/P)P(S/P/A)Y is important to maintain the interaction between the substrate and the WW domain of the Rsp5 orthologue in each species. To help assess the physiological relevance of the ubiquitination and binding assays, we integrated the data from this current study with recently generated physical and genetic interaction datasets, to derive an Rsp5 interaction network () (; ; ; ). Consistent with a role in ubiquitination, both positive (alleviating) and negative (aggravating) genetic interactions were observed with the deubiquitinating enzymes Ubp13 and Ubp3/Bre5, suggesting that these enzymes may share similar substrates. Further inspection of the Rsp5 map reveals that a large number of physical and genetic interactions place Rsp5 into pathways responsible for the regulation of chromatin function and transcription. Consistent with this, Rsp5 was originally isolated in a genetic screen for suppressors of mutations in Spt3 (), a subunit of the SAGA (Spt/Ada/Gcn5/acetyltransferase) complex (; ), which modulates the transcription activity of RNA polymerase II (RNAPII). Although it is known that Rsp5 ubiquitinates the large subunit of RNAPII (Rpb1) under conditions of UV irradiation, its exact role in transcriptional regulation is not well understood. From the interaction network it is clear that Rsp5 is physically and genetically linked to several complexes involved in chromatin remodeling and/or transcription regulation, including the histone deacetylase Rpd3C(L) complex, the histone Snf1 kinase complex, SAGA, SWI/SNF and Mediator. In addition to transcriptional regulation, Rsp5 seems to be functionally linked to several other processes including rRNA metabolism and mRNA splicing. For example, Dis3 has previously been shown to be the key regulator of the exosome (responsible for degradation of snoRNAs, mRNAs and rRNAs) (). Here, we have shown that Rsp5 is both capable of binding and ubiquitinating Dis3, suggesting that one mode of its (and hence the exosomes) regulation may occur through the ubiquitin pathway. Finally, two previously uncharacterized proteins, Yjl084c and Ykr021w, were not only found to bind to and are substrates of Rsp5 but also were recently shown to be physically associated with Rsp5 . (; ). Interestingly, these are two out of the four budding yeast proteins that contain an arrestin N domain, which in metazoans is linked to inactivation of G protein-coupled receptors and cross-talk with other signaling pathways (). These relatively specific connections suggest this region may be mediating the binding and/or activity of Rsp5. It is expected that the interaction network presented in will facilitate the generation of many more testable hypotheses of Rsp5 function. In summary, by applying the ubiquitination assay to the protein microarray platform, we have developed a sensitive assay that can be used to discover numerous substrates simultaneously, covering the whole proteome. Our approach should support the screening of other E3 systems both in humans and in yeast. Furthermore, once proteins on the array are ubiquitinated by a particular E3, deubiquitinating enzymes could be screened for specific substrates. In such experiments, loss of signal would indicate that a particular protein has been deubiquitinated. Yeast E2 gene was expressed in strain BL21 (DE3) from pET15b plasmids as described previously (). Transformed cells were grown at 37°C to an absorbance of 590 of 0.6 in 2 l of Luria broth and expression was induced by addition of 1 mM isopropyl-β-1-thio--galactopyranoside (IPTG). After 12 h of induction at 16°C, the cells were harvested and lysed by sonication in binding buffer (20 mM HEPES, pH 8.0, 500 mM NaCl, 10% glycerol, 10 μM ZnCl, 0.5 mM tris(2-chloroethyl) phosphate (TCEP)) and protease inhibitor tablets (one tablet per 50 ml of buffer, Roche Applied Science) containing 5 mM imidazole. Lysates were clarified by centrifugation at 100 000 for 1 h at 4°C and His-tagged proteins were purified from the clarified lysate on a 2-ml nickel–nitrilotriacetic acid superflow agarose column (Qiagen). Bound proteins were washed with binding buffer supplemented with 30 mM imidazole and eluted with binding buffer supplemented with 500 mM imidazole. The GST-Rsp5 expression plasmid (pGEX-6P2-RSP5) () was used to express GST-Rsp5 in using the same method as described for the E2 enzyme, except that imidazole was omitted from the binding buffer. The recombinant proteins were purified from the cell lysate on a column containing 3 ml of glutathione–Sepharose resin (Amersham Biosciences), washed once with 50 ml of binding buffer, followed by a wash with 25 ml of PreScission cleavage buffer (PCB: 50 mM Tris–HCl, pH 7.0, 150 mM NaCl, 1 mM TCEP, 10% glycerol). Rsp5 was proteolytically cleaved from the GST moiety by incubating the resin for 4 h with 1 ml of PCB containing 40 U of PreScission protease (Amersham Biosciences). Rsp5 was purified as described above, except that 50 mM Tris–HCl was replaced with 50 mM HEPES in the PCB. Purified Rsp5 (1 mg/ml) was labeled with AlexaFluor 647 using the Microscale Protein Labeling kit (Molecular Probes) according to manufacturers' instructions. A 50 μg weight of Rsp5 was used for the reaction. Final concentration of Alexa647-Rsp5 was 0.1 mg/ml in a volume of 100 μl. The GST-CTD expression vector (pET21a-GST-TEV-CTD) was constructed by Dr N Fong and generously provided by Dr D Bentley. GST-CTD was expressed in and purified using the same method as described for the GST-Rsp5, except that proteins were eluted with binding buffer containing 15 mM glutathione. The collection of yeast strains expressing GST proteins was a generous gift from Dr M Snyder. Culture of the yeast strains and expression of recombinant GST proteins were carried out as described previously (). Proteins were purified from 50 ml of growth media using 100 μl of glutathione–Sepharose resin and eluted in 100 μl of binding buffer containing 15 mM glutathione (Amersham Biosciences). The final yield of purified proteins varied from 10 to 200 μg. All purified proteins were resolved on 12% SDS–polyacrylamide gels and visualized by Coomassie Blue staining (Sigma B-7920). Immunoblotting was performed using a mouse monoclonal anti-GST antibody (B-14, Santa Cruz Biotechnology). Purified proteins were frozen in ethanol and dry ice and stored at −80°C. Reactions contained 3 μl of 5 × assay buffer (250 mM HEPES, pH 7.4, 25 mM MgOAc, 2.5 mM TCEP, 500 mM NaCl and 50% glycerol), 1 μg of ubiquitin (b-Ub), 0.16 μg of yeast E1, 3.8 μg of Ubc4 E2, 1.2 μg of Rsp5 E3, 8 pmol of GST-tagged substrate and 3.3 mM ATP (Sigma). E1 and b-Ub were purchased from Boston Biochem. Water was added to each reaction to bring the final volume of all reactions to 15 μl. ATP was either omitted or added last in order to minimize autocatalytic ubiquitination reactions by the ubiquitination enzymes. Reactions were allowed to proceed for 4 h at room temperature and stopped by boiling in 5 μl of sample buffer. Detection of a shift to high MW, indicative of ubiquitination, was performed by immunoblotting with anti-GST antibodies. The and the corresponding wild-type () strains expressing the desired HIS-tagged proteins were grown to log phase in Ura− synthetic drop-out media containing 2% raffinose. The temperature was then changed to 37°C; the expression of the proteins was induced by the addition of galactose to 2% and growth was continued at the restrictive temperature for 2 h. The cells were then lysed and the proteins were purified from the cells using anti-HA antibodies, and immunoblotted with anti-ubiquitin antibodies (Covance). For the purpose of developing the ubiquitination assay using protein microarrays, protein samples were spotted onto ArrayIt™ or eight-pad FAST™ nitrocellulose slides at various concentrations using a Piezorray™ (Perkin Elmer) platform. Following printing, the slides were stored at −20°C. Invitrogen ProtoArray Yeast Proteome Microarray was purchased for subsequent experiments. Slides were removed from the freezer, briefly allowed to thaw, rinsed with 0.5%. PBST and then blocked for an hour in 5% skim milk prepared with 0.5% PBST. Following blocking, the slide was washed three times for 5 min each with 0.5% PBST. After washing, 0.6 ml of a ubiquitination reaction mixture (50 mM HEPES, 5 mM MgCl·6HO, 0.5 mM TCEP, 10% glycerol, 4 μg of FITC-labeled ubiquitin (Boston Biochem) 0.64 μg of E1, 15.2 μg of E2 (Ubc4), 4.8 μg of E3 Rsp5) was gently pipetted onto the surface of the slide after 2 μl of 100 mM ATP was added to start the ubiquitination reaction. The reaction mixture on the slide was kept humid using wet filter paper and was allowed to proceed for 3 h. Following the reaction, the slide was briefly rinsed with 0.5% PBST followed by three 10-min washes with 0.5% PBST. The slide was dried by centrifuging for 4 min at 1000 and visualized by fluorescent laser scanning at 10 μm resolution using a 488 nm laser on a ProScan Array HT™ scanner (Perkin Elmer). Printed slides containing fluorescent proteins and dyes were kept in the dark for the duration of the experiment. Slides were removed from the freezer, briefly allowed to thaw, rinsed with PBS, followed by a rinse in 0.1%. PBST, and then blocked for one hour in 5% skim milk made with 0.1% PBST for 2 h. Probe solution (2.5 μl of Alexa647-Rsp5 (0.1 mg/ml) diluted in 0.75 ml of reaction buffer (50 mM HEPES, 5 mM MgCl·6HO, 0.5 mM TCEP, 10% glycerol) was carefully pipetted over the entire area of the slide, kept humid with a wet filter paper and the allowed to incubate for 1.5 h. The slide was washed 3 × in 0.1% PBST for 10 min and dried by centrifugation (4 min at 200 ). The slide was then scanned at 10 μm resolution using a 633 nm laser on a ProScan Array HT™ (Perkin Elmer) scanner. Printed slides containing fluorescent proteins and dyes were kept in the dark for the duration of the experiment. Data from the two ubiquitination assay slides were analyzed using ProScan Array HT™ (Perkin Elmer) software. Spots on which 50% of the pixels produced signal greater than two standard deviations above the background were identified as ‘hits'. These proteins were eliminated from the Rsp5 substrate list unless both the duplicated spots met this criterion on both of the assayed slides (i.e. all four spots had to meet the criteria). Once the Rsp5 substrate list was generated, the spots were ranked according to their signal intensity calculated as (signal intensity=mean signal on the spot−background)/concentration of the protein on the spot). Once this list was generated, all the proteins with a signal intensity >2 (40 proteins) were chosen as ‘high-confidence' Rsp5 substrates. Data analysis to generate the Rsp5 interaction list was performed using the same quantitative parameters. The protein complements for the 17 yeast species were obtained from the following online databases: , . , . , . , . , . and . — genome database (); , , , —Genolevures () (); , , , , —the fungal genome initiative at the Broad Institute (); —National Center for Biotechnology Information (); The Wellcome Trust Sanger Institute (). Orthology was determined using a modified implementation of the widely used reciprocal best BLAST hits approach (). In brief, BLASTP () comparisons of protein complements were performed for every pair of species. For each query protein, the top 10 hits (ranked on the basis of sequence similarity) for a particular species were extracted. Each hit was then subjected to a more sensitive Smith–Waterman alignment against the original query protein () and hits ranked according to their bit scores. Those proteins that were found to be the best reciprocal hits (in terms of the bit scores) were then defined as ‘putative orthologues'. For species in which no such ‘putative orthologue' could be determined, proteins, which displayed the highest bit score to the query protein in the query protein, were manually inspected and compared with ‘putative orthologues' identified in other species. These proteins were defined as ‘highest scoring homologues'. Multiple sequence alignments were generated for groups of orthologue proteins using the program—MUSCLE (). The nucleotide sequences were mapped onto this alignment using a script written in-house and the resultant alignments used for the calculating the Ka/Ks values using the codeml program available through the PAML software suite (). A library of random dodecapeptides fused to the N-terminus of the M13 gene-8 major coat protein was constructed and cycled through rounds of binding selections with the bacterially expressed WW domain immobilized on 96-well Maxisorp immunoplates (NUNC), as described previously (; ). Phage were propagated in XL1-blue (Stratagene) in medium supplemented with M13-KO7 helper phage (New England Biolabs) to facilitate phage production and 10 μM IPTG to induce expression of the library. After four rounds of selection, individual phage were isolated and analyzed in a phage ELISA. Phages that bound to the WW domain were subjected to DNA sequence analysis. Unique binding sequences were aligned to derive a specificity profile.

Nuclear transport serves as a key regulatory step in signal transduction, cell cycle progression, and mRNA processing (; ; ). Access to the nucleus is provided by nuclear pore complexes that allow passive diffusion by small molecules, but restrict translocation by molecules larger than ∼40 kDa (; ). Entry or exit of large molecules is usually mediated by soluble receptors such as the karyopherins, which are associated with the nucleoporin proteins that line the pores. Importins bind cargo in the cytoplasm and release it in the nucleus, whereas exportins reverse this process (). Transport is driven in both directions by the high concentration of the GTPase Ran bound to GTP in the nucleus. Importins can only bind cargo in the absence of RanGTP. The association of RanGTP with an importin induces cargo release (). The RanGTP gradient across the NPC is maintained by the restriction of RCC1 (a guanine nucleotide exchange factor, RanGEF) to the nucleus and of the GTPase-activating protein, RanGAP (), to the cytoplasm and NPC. With each transport cycle, one RanGTP is exported and one RanGDP is returned to the cytoplasm by the transport receptor NTF2 (; ). The majority of known import cargoes contain a nuclear localization signal (NLS) (; ). This ‘bar-code' can be a monopartite stretch of seven basic amino acids or a longer, bipartite sequence. The NLS is recognized by an adapter protein, importin-α (Impα), which binds to the karyopherin importin-β (Impβ). In the nucleus, RanGTP binding to Impβ disassociates the complex. RanGTP–Impβ then translocates back to the cytoplasm, where RanGAP (assisted by RanBP1) hydrolyzes RanGTP, releasing Impβ for another round of transport (). Impα in the nucleus is exported by a specific exportin called CAS, which also promotes release of cargo from the adapter (; ; ). Some import cargoes bind Impβ directly. Without the need to export Impα from the nucleus, this import pathway uses only one GTP cycle rather than two. Because the Impα pathway utilizes more energy and more protein production, strong selective pressure must have driven its evolution in the cell. One possibility is the flexibility that adapter proteins provide to allow specific populations of cargoes to be imported at different times or different cellular states. At least five isoforms of Impα exist in mammalian cells, and some isoforms show specificity for particular protein cargoes. However, budding yeast expresses only a single Impα. Another possibility is that, as Impα-mediated transport is coupled to the hydrolysis of two GTP molecules, it might drive a higher nuclear/cytoplasmic cargo gradient than direct Impβ-mediated import. We now demonstrate, using a combined /experimental approach, that contrary to expectations, Impα-mediated transport is actually less efficient than direct import by Impβ. Direct import is faster, and can drive a higher nuclear/cytoplasmic cargo gradient. In addition, we show that a bipartite NLS can accumulate in the nucleus to a higher concentration than a monopartite NLS, as predicted by our computer model. However, an sensitivity analysis shows that Impα provides a greater dynamic range of control over import than Impβ. To test this prediction , we use a combination of recombinant protein co-injection and siRNA knockdown. To investigate cargo gradients in both types of import, we developed a 3-compartment transport model (). Details of the model are in Materials and methods, and the complete schematic for the cargo import, Ran transport, and Karyopherin transport modules can be found in the by . Addition of either type of cargo to the cytoplasm was simulated by instantaneously stepping its concentration from 0 to 4 μM and measuring nuclear accumulation over 1800 s. Unexpectedly, cargo imported directly by Impβ had a greater initial rate and a higher steady-state nuclear accumulation than cargo imported via the adapter, Impα (). This difference results from the greater reaction rate for a bimolecular interaction, faster cycling time of Impβ between the nucleus and the cytoplasm, and the slightly higher permeability for the Impβ–cargo complex through the NPC, as compared to the Impα/β–cargo complex. To evaluate steady-state accumulation of Impα/Impβ cargo in our experimental system, we built GST-NES-GFP-NLS, which contains both an import and an export signal. The export signal is an NES from protein kinase inhibitor (PKI) that is recognized by CRM1 (). To compare Impα adapter-mediated import with direct import, we prepared a second cargo protein, GST-NES-GFP-IBB, which contains an IBB motif (). The IBB domain is a 41 amino acid arginine-rich fragment from Impα that is representative of a class of NLSs that bind directly to Impβ (). To describe these shuttling cargoes in the model, we added pathways for export through CRM1 and cofactor RanBP3 (). We injected each cargo into the cytoplasm of HeLa cells and recorded the nuclear/cytoplasmic ratio (N/C ratio) after 30 min (Materials and methods). The direct Impβ (IBB) cargo achieved a significantly larger nuclear/cytoplasmic gradient than the Impβ/Impα (NLS) cargo (). To compare initial rates of import, we used a GST-GFP-IBB cargo. After microinjecting either GST-GFP-IBB or GST-GFP-NLS into the cytoplasm of HeLa cells, initial import rates were recorded as described previously (). Import of GST-GFP-IBB was significantly faster than that of GST-GFP-NLS (). Next, we sought to determine if the results for the monopartite NLS were generalizable to a bipartite NLS. Bipartite NLSs, like that found in nuclear CAP-binding protein subunit p80 (CBP80), are known to bind to Impα with a greater affinity than monoparite NLSs (; ). Simulation of bipartite NLS cargo import predicted a similar initial rate, but increased N/C ratio as compared to monopartite NLS cargo (). To test this prediction, we measured initial import rates and steady-state accumulation for GST-GFP-CBP80 and GST-NES-GFP-CBP80 in HeLa cells. GST-GFP-CBP80 shows a similar initial rate to the monopartite cargo and GST-NES-GFP-CBP80 a much higher N/C ratio at steady state, in agreement with the model (). If adapter proteins do not allow the production of a greater cargo gradient, what other advantage might they offer? Previously, we used sensitivity analysis to explore the coupling of reactant concentrations to cargo import rates (). Impα had the largest dynamic range of control over initial rate, although Ran and NTF2 also functioned as limiting reactants in the system. Surprisingly, Impβ, CAS, and the guanine exchange factor RCC1 inhibited import at higher levels of concentrations. As steady-state cargo concentrations are likely to have greater cellular consequences than initial rates, we performed a sensitivity analysis to correlate reactant concentrations with steady-state cargo accumulation. Initial reactant concentrations were varied individually from 0.0001- to 10 times that of their original values. After allowing the system to reach steady state, cytoplasmic injection of a shuttling cargo was simulated by increasing concentration of the cargo instantaneously from 0 to 4 μM. After a return to steady state, the ratio of nuclear/cytoplasmic concentration was calculated (). Steady-state nuclear/cytoplasmic cargo ratio followed the same general trends as the initial import rates (); Impα, Ran, and NTF2 act as limiting reactants, whereas high concentrations of RCC1 and Impβ reduce the N/C ratio. The inhibitory effect of Impβ has two primary sources. First, excess Impβ can travel through the NPC without cargo and bind with RanGTP in the nucleus before returning to the cytoplasm. This process, called ‘futile cycling', depletes the RanGTP gradient. Second, Impβ must react with RanGTP on the nuclear side of the NPC to be released from a binding site on nucleoporins in the nuclear basket. Without sufficient RanGTP, Impβ, and cargo complexes arrest within the nuclear pore, blocking traffic in both directions. Knockdown of Impβ reduced nuclear accumulation in agreement with the sensitivity analysis (). Knockdown of Impα1 shows that changing Impα1 levels exert larger corresponding changes in nuclear cargo accumulation (), indicating a larger dynamic range of control. Co-injected recombinant Impα1 strongly upregulated the nuclear accumulation of the shuttling cargo (), whereas co-injected Impβ suppressed nuclear accumulation of the cargo, as predicted. Titration of co-injected Impα1 and Impβ levels shows how increases in Impα1 concentration have a greater effect on the range of cargo accumulation (). Although our current model includes competition for nucleoporins, it does not include a three-dimensional spatial representation of the NPC that could capture the blocking effect of Impβ in the confined space of the nuclear basket. This difference may explain the somewhat greater inhibitory effect of excess Impβ seen . Using a combined modeling/experimental approach, we have shown that adapter-mediated import offers no advantage in driving a cargo gradient, despite using twice as much energy as that of direct import. However, Impα shows a large dynamic range of control over cargo accumulation. In contrast, Impβ shows an ‘inverted U' response curve in which either increasing or decreasing Impβ inhibits import. This behavior results from the high-affinity binding of Impβ both to RanGTP and to the nuclear basket. Excess of Impβ can cause futile cycling and deplete the RanGTP gradient. Diminished nuclear RanGTP can cause accumulation of Impβ at the nuclear basket to the point at which it begins to occlude the pore, severely restricting transport of cargo complexes. Adapter proteins may, therefore, have evolved as a means to more flexibly control cargo gradients under different cellular conditions. For example, recent work has shown how the modulation of nuclear transport rates maintains the fidelity of wave propagation in cell cycle progression (). Flexibility in biological systems describes an organism's ability to adjust to changing environments. This control comes at the expense of efficiency and requires an additional expenditure of energy. But the advantage that comes with this flexibility is increased robustness, the ability of a system to remain stable in the face of external perturbations (; ). This trade-off between regulation and efficiency is fundamental to all control systems and is likely to be widespread in cell biology. Another group has recently looked at the dynamics of cargo import in yeast cells (). They found a simple linear relationship between initial cytoplasmic concentration and initial import rate similar to that previously found in mammalian cells. This relationship remained constant even up to 100 μM initial concentration of cargo, demonstrating the remarkable capacity of the NPC to handle large amounts of cargo transport. Import rates were found to be 0.07–1.2 cargo molecules/NPC/μM, comparable to that found previously in our study of import in mammalian cells. also found that the initial rate of an NLS-GFP cargo increased with higher Karyopherin concentrations up to about 15 μM, at which point import rates began to show saturation kinetics. Although the contrast this result with our finding that Impβ inhibits import at higher concentrations, Kap95p is most closely related to Impβ in mammalian cells. Kap95p shows a high affinity to the nucleoporins in the nuclear basket in a similar way to Impβ, so a test of Kap95p abundance in yeast cells would be a more definitive comparison. We suggest that the high-affinity binding of Impβ and Kap95p to the nuclear side of the NPC make these Karyopherins especially sensitive to limitations in the Ran gradient as well as competition for binding sites that can cause the inhibition of transport we have observed at high concentrations of these receptors. Import rates were successfully fit using a simple model of cargo-binding kinetics, passive import, and passive leak kinetics. interpreted these findings as evidence that nuclear transport follows a simple ‘pump-leak' model in which import rates are largely determined by the number of Karyopherin–Cargo complexes that form and the rate at which cargo passively diffuses back to the cytoplasm. This model is consistent with our findings that the Karyopherin Impα acts as a limiting reactant for import and the bipartite NLS (which binds Impα more tightly) leads to greater nuclear accumulation of cargo. Both these factors would act to increase the amount of effective Karyopherin–cargo complexes . also use their model to predict that the Ran gradient is not limiting for import. However, we have shown previously () that both Ran and NTF2 are limiting for import in whole HeLa cells. Our results do agree with in showing that the capacity of the NPC is enormous and not likely to be limiting for cargo import rates. Recent work from another group has shown that the rate of cargo transport through the NPC could be modulated ∼10-fold by Impβ in permeabilized mammalian cells (). However, permeabilized cells have an NPC depleted of native Karyopherins and would not show competitive inhibition seen in whole cells. This is reflected by experimental evidence that import rates in whole cells are more than an order of magnitude less than those of permeabilized cells (). Therefore, any quantitative result from permeabilized cells should be interpreted with caution. As total Karyopherin concentration in whole cells has been estimated to be 15 μM (), Impβ might be expected to start inhibiting transport above these levels in permeabilized cells. Our nuclear transport model is the first to include a detailed reconstruction both of nuclear import through Impα/Impβ and export through CRM1. It will allow future work to explore how the regulation of nuclear import couples to signal transduction pathways such as PKA/CREB and Jak-Stat in which shuttling factors may increase responsiveness of the system to changes in the state of the receptor at the cell surface. Our original computer model for nucleo-cytoplasmic transport considered only the receptor-mediated import of cargo. In addition, it ignored the complexities of translocation through the NPC as a complex, and described nucleo-cytoplasmic movements by single permeability constant (). Experimental work has shown that this simple linear model can successfully describe the translocation process to a first approximation, and has the advantage that the permeability constant can be derived experimentally. However, more complex behaviors, such as competition between transport receptors at the NPC, cannot be represented in this way. To create a more realistic model of the translocation process, we added a third compartment that represents the entire volume of all nuclear pores embedded in the nuclear envelope. Compartmental models assume that all species diffuse completely within compartments. Therefore, nucleoporins were modeled as freely diffusing but effectively trapped within the nuclear pore. Although most nucleoporins do not freely diffuse , this approximation should be reasonably accurate, as all kinetic measurements of nucleoporins have been measured from dilute solutions. Nucleoporins show an increasing gradient of affinity for Impβ, with interior nucleoporins showing an affinity of approximately 100 nM (). We chose this value as an average representational value for all nucleoporins in the NPC compartment of the model. The complete model was simulated using Jarnac, a Biochemical simulation package for Windows (). The reactions were converted internally by Jarnac to a series of coupled ODEs. Jarnac was then used to solve the ODEs based on a set of initial values. Cellular concentrations for Impα, CRM1, and RanBP3 were measured experimentally as described below. Rate constants for CRM1/RanBP3 export were taken from the literature wherever possible (). The entire model will be available from the Biomodels database () in SBML format. The GST-GFP-NLS construct has been described previously (). The IBB domain (Impα1 residues 1–88) was first subcloned from GFP-IBB into GST-GFP to produced GST-GFP-IBB. GST-GFP-CPB80 has been described previously (). The PKI NES (LALKLAGLDI) was then subcloned into GST-GFP-NLS, GST-GFP-IBB, and GST-GFP-CBP80 to produce the shuttling constructs GST-NES-GFP-NLS and GST-NES-GFP-IBB, and GST-NES-GFP-CBP80. Concentrations of recombinant Ran, Impα, CRM1, and RanBP3 were first quantified by comparison to BSA standards on SDS–PAGE stained with Coomassie blue. Known concentrations of Ran and each transport receptor from 0.5 to 0.05 μg were then run on SDS–PAGE, together with 20 μl HeLa cell lysate. After blotting for both Ran and each transport receptor individually, concentration of proteins in the HeLa cell lysate was estimated by comparison to the recombinant standards. Estimated cellular concentration for transport receptor proteins (Impα 1 μM; CRM1 0.3 μM; RanBP3 0.05 μM) was then calculated based on the previously published concentration of Ran in HeLa cells (6 μM). HeLa cells were transfected with Impβ siRNA, Impα siRNA, or the control siRNA (Dharmacon random 21-mer) using SiPort (Ambion) (Impα), or Oligofectamine (Invitrogen) (Impβ) according to the manufacturers protocol. After 72 h, the media was changed to physiological saline and cells were cytoplasmically injected with the shuttling GST-NES-GFP-NLS protein. Media was then changed to DMEM (5% FCS, 5% CS, 1% PS) and the cells were incubated for 30 min at 37°C. Following incubation, cells were fixed and immunostained as described previously () for Impβ (ABS) or Impα (Transduction Laboratories). Microinjection and microscopy were performed as described previously (). Nuclear to cytoplasmic ratio was defined by the ratio of mean nuclear pixel intensity to mean cytoplasmic pixel intensity 30 min after injection.

The problem of prognosis differs from that of diagnosis in two important ways. First, the goal of diagnosis is to assign patients to discrete categories (affected or unaffected), whereas the goal of prognosis is to provide a probability that a given outcome will occur. Second, for many diseases that have been characterized by molecular markers, clinical parameters (such as age or race) are not relevant to their diagnosis, but are often of substantial prognostic value. Since recent studies have shown that diagnosis can be enhanced by multivariate approaches (; ; ; ), we set out to develop a strategy that addresses the challenges unique to prognosis. Our strategy incorporates information from clinical variables as well as molecular markers, is not biased by assumptions about the relationships between variables and outcome, and can be implemented in the clinic without introducing new and expensive technology. As proof-of-concept, we developed a prognostic test for patients initiating kidney dialysis. In the United States alone, end-stage renal disease (ESRD) affects ∼100 000 individuals per year and there are at present ∼400 000 individuals undergoing chronic hemodialysis (). Of patients with ESRD, ∼10% die within the first 3–4 months of initiating treatment and there is currently no way to predict early mortality. In general, patients with renal failure have excess inflammation, and inflammation has been implicated in cardiovascular events and infection—the two leading causes of death among ESRD patients (; ; ). As such, many dialysis-related studies have focused on cytokines as potential prognostic markers (; ; ; ; ; ; ). To date, no single marker has been discovered that accurately predicts outcome, and cytokine levels are not used routinely in the clinical management of dialysis patients. We set out to develop a model that predicts which patients are most at risk of dying within the first 15 weeks of initiating treatment. Critical decisions may be aided by such a model, including setting priorities for renal transplantation, changing the frequency or dose of dialysis, and identifying a subset of patients at whom clinical trials could be directed. To address this problem, we turned to Accelerated Mortality on Renal Replacement (ArMORR), a prospective study of ESRD patients that initiate dialysis at any one of >1000 dialysis centers in 34 US states (). ArMORR contains detailed demographic and clinical data, as well as serum samples, for all participants. For this study, we selected 208 consecutive patients who died within 15 weeks of initiating dialysis to serve as cases, and 260 consecutive patients who survived for at least 15 weeks to serve as controls. Serum samples were collected within 14 days of initiating dialysis. To identify putative prognostic markers, we searched the literature for cytokines or other blood proteins whose levels correlate with kidney disease. We also included proteins associated with hypertension or diabetes, the two leading causes of ESRD (). From this initial list of proteins, we chose 14 that are present in the serum of dialysis patients and for which matched pairs of antibodies, as well as purified antigens, are commercially available: angiogenin (Ang), EGF, Fet-A, ICAM, interleukin-12 (IL-12), IL-1α, IL-8, MIP-1β, RANTES, TNF-β, TNFR2, TNFR1, vascular cell adhesion molecule-1 (VCAM-1), and VEGF (). To facilitate rapid and accurate measurement of all 14 markers in all 468 patient samples, we developed a high-throughput, multiplexed assay that mimics a sandwich immunoassay, but in a microarray format. Capture antibodies were arrayed at high spatial density in each well of 96-well microtiter plates (), and serum samples were applied to each array. Captured cytokines were detected using a cocktail of biotinylated antibodies, which were then visualized with a fluorescent conjugate of streptavidin. By using a very bright fluorophore (PBXL-3), we were able to achieve exquisite sensitivity without requiring enzyme-mediated signal amplification: most cytokines could be detected at a concentration of ∼1 pg/ml. This greatly facilitated the rapid processing of hundreds of arrays. In addition, multiplexing did not compromise the assay; biotinylated detection antibodies did not cross-react with capture antibodies and capture antibodies did not cross-react with non-cognate antigens when tested individually. The absolute concentration of each cytokine in each sample was determined by relating the fluorescence intensity of microarray spots to a standard curve, generated for each cytokine in a multiplexed fashion using one column of each microtiter plate (). This strategy minimized both plate-to-plate and day-to-day variation, since a separate standard curve was generated on each assay plate. For redundancy, each array contained five replicate spots of the capture antibodies and every sample was analyzed on two arrays. Overall, the average coefficient of variation was 6.6% for replicate spots within an array and 11% for replicate samples on separate arrays. Using these microarrays, cytokine levels were measured in all 468 patient samples ( and ). A cursory inspection of the data showed that for all 14 cytokines, their distribution in the population of patients who died closely matched their distribution in the population of patients who survived (). This is consistent with previous studies showing that no single marker is predictive of early mortality. Although it is possible that prognostic information is embedded in correlations between pairs of biomarkers, including cross-terms in any analysis would increase the number of variables from 14 to 182, and thus substantially increase the false discovery rate. We therefore focused our efforts on the 14 first-order terms, which are also more readily interpretable. We found that standard data-mining methods (), including hierarchical clustering, -means clustering, nearest-neighbor methods, and principal components analysis, all failed to distinguish those who died from those who survived. These methods rely on metrics that quantify the ‘distance' between patient profiles and hence require arbitrary rescaling of variables. More important variables are not weighted appropriately, and hence these methods are weakened by noise and outliers. Decision trees and adaptive boosting with decision stumps () also failed to segregate those who survived from those who died. While these methods do not require rescaling of variables, they work by converting continuous variables into binary data and so discard much of the information embedded in the quantitative dataset. More importantly, all of these methods are best suited to classifying samples, but our goal was to develop a continuous predictor of early mortality. We therefore turned to regression methods as a way to extract the relationships between variables and outcome. where is the value of the -th variable (e.g., age or IL-12 concentration) and and are constants. It is important to note that the probabilities in are calculated with respect to the patients in our nested case–control study and not with respect to the general population. We intentionally over-sampled patients who died (stratified sampling); we correct for this difference later based on an early mortality rate of 10%. Since clinical data are routinely collected on each patient, we started by building an additive model using these data alone. We focused on 11 clinical parameters previously shown to be associated with dialysis-related mortality (, ; USRDS, 2006): gender, age, race, body mass index, diastolic blood pressure, underlying disease, method of vascular access, serum albumin level, serum calcium level, serum phosphate level, and blood hemoglobin content. To avoid over-fitting and to construct a model that incorporates only as many variables as are necessary, we adopted the following strategy. If is the number of variables in the model, we started with =1 and, in an incremental manner, performed an exhaustive search for the best -variable model. We continued to increment until no -variable model could be found in which all of the parameters were significant (<0.05 for each ). Based on this criterion, the best model was obtained using four clinical parameters: age, diastolic blood pressure, serum albumin, and method of vascular access (arm or neck). We then repeated this procedure using the serum cytokine levels measured on our microarrays. In this case, we found that the best model was obtained using three cytokines: angiogenin (Ang), interleukin-12 (IL-12), and vascular cell adhesion molecule-1 (VCAM-1). Although linear models are easy to implement, there is no reason why risk should vary linearly with any clinical or molecular variable. Indeed, there is no reason why any parametric function should describe these relationships. To capture non-linearities, we refined our efforts by building generalized additive models (), in which the log-odds of death is given by the following equation: where () is a spline, composed of piecewise cubic polynomials, with the requirement that two connected polynomials have the same slope where they meet. Since any curve can be approximated by a spline, generalized additive models are not constrained by investigator bias. Since they are non-parametric, however, there is no straightforward way to calculate a -value for each variable. We therefore relied on our previous variable selection and used non-parametric methods to refine the models. To avoid over-fitting, we constrained the nominal degrees of freedom of each spline to 2. The two degrees of freedom were not concentrated at any part of the spline, but were instead spread evenly across the spline. In addition, since minimizing the sum-of-squared error tends to skew the model to outliers, we took a maximum likelihood approach. As anticipated, the generalized models picked up fine features in the relationship between death risk and each variable, providing further clinical insight. We found that the death risk increases abruptly when age increases above ∼60 years, when diastolic blood pressure drops below ∼80 mmHg, and when serum albumin levels drop below ∼3.5 g/dl (). These inflection points, which cannot be identified using linear models, provide therapeutic goals for clinicians striving to optimally manage diastolic blood pressure or serum albumin levels. The same non-parametric method applied to cytokines shows that the slopes of the splines vary as cytokine levels change (). Interestingly, we found that high levels of IL-12 and Ang are associated with low risk of early mortality. IL-12 is primarily produced by peripheral blood mononuclear cells such as macrophages () and enhances the cytotoxic activity of NK cells and the activation of T cells. The serum level of IL-12 is therefore an indicator of immune capability, which is often impaired in patients with renal failure. Similarly, Ang, although originally implicated in tumor angiogenesis, has been shown to be protective against bacterial and fungal pathogens () and appears in circulation during the acute phase response to infection (). Ang also protects against neutrophil degranulation, a side effect of dialysis (). Unlike IL-12 and Ang, increased levels of VCAM-1 were found to be associated with increased risk of death. VCAM-1 is normally absent from resting endothelium. Uremia (excessive urea in the blood stream) induces an increase in the expression of adhesion molecules on vascular endothelial cells and shedding of these molecules into the circulation (). In addition, VCAM-1 is involved in atherosclerosis (). Since cardiovascular events are the most common causes of death among dialysis patients, it is possible that antagonizing VCAM-1 will have beneficial therapeutic effects. Interestingly, the three molecular markers are produced by and act on different cell populations. This may explain why a simple additive model is sufficient to capture their associations with early mortality. Cytokines acting on the same cell often exhibit synergistic or antagonistic effects (), but IL-12, Ang, and VCAM-1 are, to a first approximation, independent. As a first step toward a unified model, we prepared a scatter plot in which the two models are presented jointly, with the clinical predictor on the horizontal axis and the cytokine predictor on the vertical axis. These predictors, which provide the probability of early mortality within the sample population, were obtained by first calculating the log-odds of death by adding each variable's contribution, as well as the appropriate constant term, ( and ). Log-odds was then converted to a probability by taking the inverse logit according to the following equation: Using only seven parameters, the combined model is able to separate patient outcomes effectively. While there are outliers in any human population, the centroids of the two patient populations are well separated (). Since the goal of our approach is to provide a continuous predictor of outcome, we estimated probability densities for death () and survival () using kernel methods. Kernel methods amount to convolving discrete data with a Gaussian window to obtain continuous estimates for densities. In other words, the density estimate at each location is a weighted average of all the discrete samples, with the weight of each sample decreasing with increase in distance between the sample and that location. To ensure that our density estimations are not biased by sample size, we generated 100 bootstrap data sets (sampling with replacement) and performed kernel density estimation on each data set. The final density estimate is the average of all 100 bootstrap density estimates. This procedure is often referred to as ‘bagging'. Based on and , and adjusting for our over-sampling of patients who died, we went on to compute predictors that give the overall risk of death among new dialysis patients in the general population, based on an orthogonal combination of clinical and cytokine data (). Numerical values for our model are provided as . Although a continuous predictor is more appropriate for prognosis than a binary classifier, there are situations in which it is useful to classify patients based on their expected outcome. For example, high-risk patients can be selected for clinical trials aimed at altering their outcome. To classify patients, a simple decision boundary can be applied to our model: patients with a risk of early mortality above the boundary are projected to die, whereas those below the boundary are projected to survive. Clearly these projections will sometimes be incorrect, especially for patients who are close to the boundary, but model-based selection should prove more accurate than random selection. In order to assess the accuracy of our model with respect to binary classification, we performed five-fold cross-validation, each time using 80% of the data for model fitting and 20% of the data as a naïve sample for model testing. The five runs gave near-identical results, indicating that our approach is robust (). At a decision boundary equal to the overall risk of early mortality (0.10), our model classifies patients in the general population with 73±2.5% (s.e.m.) sensitivity, 76±1.8% specificity, and a positive predictive value (PPV) of 25±1.4% (for definitions, see legend to ). When it is desirable to favor specificity over sensitivity, a decision boundary of 0.20 enables patients to be selected with reasonable sensitivity (39±4.3%) and high specificity (94±1.1%), yielding a PPV of 43±6.4% (). This strategy can be used to enrich high-risk cases in a clinical trial by 4.3-fold relative to a trial run without patient selection, thereby substantially decreasing expenditures. This is particularly relevant given that several recent trials designed to improve survival among dialysis patients were negative (; ; ). In addition to defining non-linear relationships between variables and outcome, our method also highlights non-linearity in the relationship between clinical variables and molecular markers () and suggests a simple strategy for patient management (). As highlighted by the combined model, serum cytokine levels are most useful among patients that are identified as being at risk based on their clinical variables. If the clinical predictor is low (left side of ), little additional information is gained by measuring the patient's cytokine levels (). If the clinical predictor is high (right side of ), serum cytokine levels markedly improve risk assessment (). Thus, we find that cytokine levels are informative, but only in a subset of patients. This may explain why reports aiming to identify prognostic markers without taking into account clinical variables are either conflicting or find that cytokine levels have marginal prognostic value (; ; ). Our combined model highlights potentially important interactions between clinical variables and cytokine levels that are readily interpretable. At a broad level, patients at risk of early mortality based on advanced age or vascular access through the neck are more susceptible to additional insults, such as excess inflammation (IL-12), infection (Ang), or cardiovascular compromise (VCAM-1). We can also speculate on more specific, synergistic interactions between the clinical and cytokine variables. Low serum albumin levels have been strongly linked to impaired immune function, bacteremia, and sepsis in hemodialysis patients (; ). In the context of impaired immunity, low levels of IL-12 and Ang could exacerbate a predisposition to infection. Similarly, low diastolic blood pressure is thought to reflect an underlying impairment of cardiac reserve (). Impaired cardiac reserve, superimposed on endothelial dysfunction and accelerated atherosclerosis (high VCAM-1), would render patients especially vulnerable to cardiac-related mortality. Thus, specific cytokine alterations are particularly important among patients otherwise predisposed to related adverse outcomes. This finding converges with the current trend toward personalized medicine: just as certain drugs are only effective in specific subsets of patients (), so too prognostic tests based on molecular markers may be most informative following patient selection (). Since commercial assays already exist for Ang, IL-12, and VCAM-1, our model can be implemented without introducing new and expensive technology into clinical laboratories. ArMORR is a nationally representative prospective cohort study of US patients who initiate chronic hemodialysis at any one of >1000 dialysis centers operated by Fresenius Medical Care, North America. Clinical data are collected prospectively and entered into a central database uniformly by practitioners at the point of care. Likewise, all patient blood samples are uniformly shipped to and processed by Spectra East (Rockland, NJ), a GCP-accredited central laboratory. Antibodies were spotted onto aldehyde-displaying glass substrates using a piezoelectric microarrayer. Ninety-six identical microarrays were fabricated in a 12 × 8 pattern on the glass substrate, and the glass was subsequently attached to the bottom of a bottomless 96-well microtiter plate using an intervening silicone gasket. Serum samples were diluted 1:3 with HBS (10 mM HEPES, 10 mM NaCl, 0.004% NaN, pH 7.4) supplemented with 1% bovine serum albumin (w/v) and applied to the arrays. To generate eight-point standard curves for each cytokine, recombinant cytokines were mixed, diluted in HBS supplemented with 25% fetal bovine serum, and applied to the wells in column 12 of the microtiter plate. Captured cytokines were detected with a cocktail of biotinylated detection antibodies, followed by a PBXL3 conjugate of streptavidin. Replicate spots from duplicate wells were averaged and related back to the appropriate standard curve to obtain the concentration for each antigen in each sample. For non-cytokine variables, missing data and outliers (∣−2∣>2(3−1)) were mostly due to uncorrectable errors in clerkship, and were replaced with 2 (1, 2, and 3 represent the first, second, and third quartiles, respectively). For cytokines, undetectable levels were replaced with 1/10 of the lowest non-zero value measured, while outliers (∣−2∣>5(3−1)) were replaced with 2±5(3−1) to minimize their influence on model fitting. Outliers and missing data represented less than 2.5% of the data. Let (1⩽⩽, 1⩽⩽) be the value of the -th variable of the -th patient, and be the outcome of the -th patient: =0 for survival and =1 for death. Linear additive models were fit using the ‘glmfit' function of MatLab (The MathWorks Inc.). Variable selection was based on the -values of 's. where is a constant, and () is a smoothing spline. To ensure uniqueness of the fitting, was constrained to be zero for each . To avoid over-fitting, the nominal degree of freedom for () was constrained to be two for each (the trace of the smoother matrix was set to three by adjusting the smoothing parameter). Minimizing deviance (−2•(log-likelihood)), the solution was attained by backfitting (). To mimic Bayesian posterior calculations, a collection of =100 bootstrap samples was generated from the original data set and fit to the models. This resulted in a collection of fits, from which we computed the means and variances of the constants, 's, and the splines, ()'s. Let () and () be the values of the non-cytokine and cytokine predictors for the -th patient, respectively. and the kernel was chosen to be Gaussian. Variance of the density estimates was minimized by bagging (averaging estimates over a collection of 100 bootstrap samples). We denote the ‘bagged' estimates as and . The probability of death as a function of and was computed using the following equation: where δ=0.1 is the overall death rate of hemodialysis patients by the 15th week of therapy. Further details for all methods are provided as . JRC and TK are co-first authors. TK developed and performed protein microarray experiments. JRC carried out data analysis and model building. RT orchestrated the ArMORR sample collection and database. GM directed and supervised the study. JRC and GM wrote the manuscript, with contributions from TK and RT. All four authors contributed to interpretation of the results.

Disulfide bonds have long been recognized as structural elements stabilizing proteins in harsh extracellular environments. More recently, an additional concept has emerged: some disulfide bonds operate as dynamic scaffolds capable of regulated rearrangement into a variety of functional forms (). Consistent with this notion, various cell surface processes have long been known to depend on catalyzed thiol-disulfide exchange including cell adhesion (), uptake of bacterial toxins () and viral fusion with the host membrane (). Moreover, a variety of cell surface signaling receptors appear to exist in more than one thiol-disulfide configuration, for example CD28 (). However, in most cases, neither the catalyst driving thiol-disulfide exchange nor the functional differences between the redox forms have been elucidated. A number of thiol-disulfide oxidoreductases are known to be secreted and to act on the cell surface. One of these redox catalysts is protein disulfide isomerase (PDI), a member of the thioredoxin (Trx) superfamily. Cell surface-PDI has been found to act on transmembrane and surface-associated proteins, including the envelope protein of HIV-1, to cause its fusogenic conformation () and integrins, to mediate platelet adhesion (). Another thiol-disulfide oxidoreductase associated with extracellular functions is Trx1. Best known for its intracellular roles, Trx1 reduces transiently formed disulfide bonds of cytosolic and nuclear target proteins and thereby participates in a multitude of fundamental processes, ranging from oxidant scavenging and DNA synthesis to regulation of apoptosis and cell proliferation (). In addition, Trx1 is released to the extracellular environment by a variety of normal and neoplastic cells (). Human Trx1 was first purified as a cytokine-like factor from supernatants of virally transformed lymphocytes and initially named adult T-cell leukemia-derived factor (), Tac-inducing factor (), B-cell stimulatory factor or ‘B cell IL-1' (). Extracellular Trx1 is present in the circulation of healthy subjects and its levels increase under inflammatory conditions, including viral infection (). Circulatory Trx1 acts as a chemoattractant for monocytes, neutrophils and lymphocytes (), and inhibits neutrophil migration into inflammatory sites both and (). More recently, Trx1 was found to be secreted by dendritic cells upon cognate T-cell recognition and to contribute to subsequent T-cell activation (). At present, the mechanism(s) and pathway(s) by which extracellular Trx1 influences cellular behavior remain unknown. As many of its reported extracellular activities depend on a functional active site, it appears likely that Trx1 catalyzes thiol-disulfide exchange in one or more cell surface target proteins through its enzymatic activity. However, thiol-disulfide exchange reactions, even if highly specific, are too transient to be detected by conventional techniques. To date, only a single cell surface receptor, CD4, a member of the immunoglobulin superfamily, has been shown to be susceptible to Trx1 redox activity (). Other cell surface proteins targeted by the enzymatic activity of Trx1 await identification. In this study, we address the question as to which cell surface receptors expressed on lymphocytes specifically interact with extracellular Trx1 by way of disulfide bond exchange. Using a kinetic trapping technique that enables the detection and isolation of otherwise short-lived reaction intermediates on the surface of intact cells, we identify and validate the tumor necrosis factor receptor superfamily member CD30 (TNFRSF8) as the principal target molecule for Trx1 on infected and transformed lymphocytes. The cell surface activity of Trx1 is highly selective, discriminating between different members of the TNFR superfamily. Trx1-mediated thiol-disulfide exchange leads to a structural change in the CD30 ectodomain that can be detected with conformation-sensitive antibodies. We demonstrate that disulfide exchange between Trx1 and CD30 interferes with binding of the CD30 ligand (CD30L) to its cognate receptor and that Trx1 affects CD30-dependent changes in lymphocyte effector function. As CD30 is implicated in both stimulatory and apoptotic signaling, our findings suggest that Trx1 interacts with CD30 to modulate lymphocyte behavior and survival under conditions of infection and inflammation. #text Accumulating evidence indicates that the reduction and rearrangement of disulfide bonds constitutes a mechanism controlling protein function on the cell surface (). The idea that disulfide bonds can act as dynamical redox switches, specifically operated by secreted redox catalysts, represents a novel concept in signal transduction (). However, technical difficulties in detecting and analyzing individual disulfide rearrangements on the cell surface have made progress slow. Trx1 is recognized as one of the most important regulators of cellular and organismal redox homeostasis (). In particular, intracellular Trx1 counteracts oxidative stress, promotes cell growth and inhibits apoptosis. Under conditions of oxidative stress, Trx1 is released by cells and accumulates at sites of inflammation (). Numerous studies have reported that secretory Trx1 influences effector functions and proliferation of lymphocytes (). However, proteins and pathways coupling extracellular Trx1 redox activity to defined cellular responses have remained unknown. In this study, we addressed the question regarding which lymphocyte surface receptors are targeted and regulated by the redox activity of extracellular Trx1. For this purpose, we made use of a mechanism-based kinetic trapping approach to capture mixed disulfide intermediates formed between exogenous Trx1 and its target proteins on the cell surface of living cells. Activity-based techniques offer the opportunity to identify interactions too short-lived to be detectable by conventional methods. To our knowledge, this is the first reported application of kinetic trapping to identify novel target proteins of mammalian Trx1 and the first application of this technique to the surface of intact cells. We demonstrate that Trx1 interacts with intra- and extracellular target proteins in a highly selective manner, guided by specific protein–protein recognition rather than random encounters with disulfide bonds. Applying the approach to the surface of cell lines representative of the lymphoid lineage, we observed that Trx1 basically targets a single cell surface protein, subsequently identified as TNF receptor superfamily member 8, also known as CD30. The pronounced preference of Trx1 for one particular target protein might seem surprising, but could be due to the fact that we assessed Trx1 reactivity of proteins as they are embedded in their natural microenvironment, namely the intact surface of the active plasma membrane of living cells. It is conceivable that protein disulfide exchange interactions are limited and controlled by their native context and location. To scrutinize the specificity of the observed interaction, we asked if the preference of Trx1 for CD30 might be caused by an unusual density of disulfide bonds within CD30 and/or exceptional cell surface expression levels. Although the CD30 ectodomain harbors a significant number of predicted disulfide bridges within CRDs, it does not appear to be unusual in terms of disulfide bond composition/density when compared to other members of the superfamily. When tested experimentally, Trx1 failed to interact with other CRD-containing proteins, including the EGFR featuring a total of 25 ectodomain disulfide bonds. The preference for CD30 could not be explained by exceptional surface expression levels either. While Hodgkin's disease cell lines typically express high levels of CD30, other cell lines including LCL-721.220 or CCRF-CEM show at least 20- to 100-fold lower expression as determined by flow cytometry, yet the same selective targeting was observed. Conversely, ectopic overexpression of several related TNFR superfamily members in HeLa cells did not lead to their interaction with Trx1, yet CD30 strongly interacted on the same cells under the same conditions. Consistent with these findings, recent experiments demonstrate that Trx1 targets a particular site within the CD30 ectodomain (Y Balmer and TP Dick, unpublished data). To facilitate identification of low-abundance cell surface proteins, trapping experiments were typically performed using Trx1 concentrations of 1–3 μM. However, when disulfide exchange was subsequently tested at lower concentrations, Trx1(CSAAA) concentrations in the low nanomolar range (4–40 nM) were found to give rise to the formation of proportional amounts of Trx1-CD30 mixed disulfide intermediates (), thus demonstrating that the observed interaction is compatible with the expected physiological concentration range of secretory Trx1 (see below). Wild-type Trx1 is known to act as a multiple-turnover catalyst if a suitable reducing system and electron source is provided for its regeneration. In agreement with these considerations, we observed that sustained reduction of CD30 in cell culture requires a Trx1 regenerating system. Using flow cytometry to monitor conformational changes in the CD30 ectodomain, CD30 was found to respond to Trx1 concentrations in the nanomolar range, starting at around 100 nM (, lower panel). However, the minimal Trx1 concentration required for sustained CD30 reduction might be substantially lower in specific environments, which are efficient in delivering reducing equivalents and preventing oxidative inactivation of Trx1. At present, it is not clear how extracellular Trx1 is regenerated . Despite the overall oxidizing character of the extracellular compartment, reductive processes are known to take place on the cell surface. On the one hand, there is long-standing evidence for the existence of transplasma membrane redox systems delivering electrons to the cell surface (). On the other hand, Trx1 may be regenerated by co-secreted reductants, as Trx1 secretion in DC-T co-culture is accompanied by the release of reduced cysteine and the creation of a reducing microenvironment between interacting cells (). In addition, TrxR was found to be secreted by activated monocytes and might be part of an extracellular Trx1 reducing system (). The concentration of Trx1 in human plasma is in the low nanomolar range (1–5 nM), and is found to be elevated several-fold under inflammatory conditions (). However, plasma Trx1 is oxidized and appears to represent systemic dilution of Trx1 previously released within tissues. Accordingly, local tissue concentrations of secretory Trx1, for example, within activated lymph nodes, are expected to be markedly higher than in plasma. Overall Trx1 concentrations in mammalian tissues can be as high as 20 μM (). Certain Trx1-secreting cell types, including macrophages and dendritic cells, distinctly upregulate expression of Trx1 upon activation (). studies of Trx1 secretion suggest that a substantial fraction of intracellular Trx1 can be released within a few hours (). Although direct measurements of extracellular Trx1 within tissues are not available, physiologically relevant extracellular Trx1 concentrations may well reach into the upper nanomolar, if not lower micromolar range. We found that Trx1-mediated disulfide reduction changes the conformation and functional properties of the CD30 ectodomain. In the reduced state, CD30 lost its ability to interact with its cognate ligand CD30L or agonistic antibodies. The presence of catalytically active Trx1 impeded CD30-dependent signaling in the YT lymphoma cell line, as demonstrated by its effect on CD25 and FasL expression, as well as its influence on cytotoxicity against Fas-expressing target cells. The physiological role of the CD30-CD30L system has remained unclear. studies focusing on CD30 lymphoid malignancies showed that triggering of CD30 signaling can induce either proliferation, activation, growth arrest or apoptosis, depending on cell type and stimulatory conditions (). , cell surface expression of CD30 appears to be tightly regulated and restricted to B and T lymphocytes undergoing activation in lymphoid tissues. It has been proposed that CD30 provides proliferation and/or survival signals during lymphocyte responses (). Under inflammatory conditions, CD30 expression is markedly induced. activation of CD30 can be monitored by the release of sCD30, shed from the plasma membrane upon CD30L binding (). Similar to serum Trx1, serum sCD30 is increased in infection, autoimmunity and allergy, for example systemic lupus erythematosus, rheumatoid arthritis and atopic dermatitis (). Both Trx1 and CD30 appear to play a role in the regulation of the antiviral inflammatory response. Both Trx1 secretion and CD30 expression have been associated with virally transformed lymphocytes. Elevated levels of sCD30 occur during viral infection. Likewise, viral infection leads to elevated Trx1 plasma levels and several studies indicate that secreted Trx1 modulates the antiviral inflammatory process (, ). In this study, we have identified an enzyme–substrate relationship between Trx1 and CD30, a receptor of activated lymphocytes involved in the regulation of inflammation. As lymphocytes migrate between different microenvironments, it is conceivable that Trx1 catalyzes disulfide exchange dynamically, activating or inactivating the CD30 pathway in response to the redox environment. The interaction between Trx1 and CD30 might represent a regulatory link between oxidative stress and lymphocyte function. Understanding of this relationship awaits the generation of suitable experimental tools. BL-41, CCRF-CEM, HDLM-2, Jurkat, RMA and U937 cells were cultured in RPMI 1640 (Gibco) supplemented with 10% heat-inactivated fetal bovine serum, 2 mM -glutamine, 100 U/ml penicillin and 100 μg/ml streptomycin (Gibco). LCL-721.220 and YT cells were cultured in IMDM (Gibco), HeLa and A431 cells in DMEM (Gibco) with the same supplements. Depending on the type of experiment, recombinant trapping mutant was applied to cytosolic preparations, human serum ultrafiltrate or intact cells. A detailed description of the different substrate trapping protocols is provided as . For reduction of cell surface CD30, 2.5 × 10 cells were incubated with 5 μM Trx1 together with 200 μM DTT or 100 nM human Trx reductase (TrxR)/500 μM NADPH for 30 min at 4°C. To monitor reduction of cell surface CD30, cells were stained with anti-human CD30 monoclonal antibody MAB229 (R&D Systems), anti-human CD30 monoclonal antibody Ki-1 (Santa Cruz) or anti-human CD30 monoclonal antibody Ber-H2 (DakoCytomation) followed by incubation with R-PE-conjugated goat F(ab') anti-mouse Ig's (Biosource). For control, cells were stained with R-PE-conjugated anti-human CD28 monoclonal antibody (BD Pharmingen). Cells were analyzed using a FACSCalibur (Becton Dickinson) and CellQuest software. A total of 2.5 × 10 cells were incubated with 5 μM Trx1 (SBP-CCCCC) and 200 μM DTT for 30 min at 37°C, washed three times and incubated with 250 ng/ml recombinant CD30L-His (R&D Systems) for 10 min at RT. After washing, cells were stained with anti-polyHis monoclonal antibody (Sigma) followed by incubation with R-PE-conjugated goat F(ab') anti-mouse Ig's (Biosource). HeLa cells were seeded on coverslips and transfected with expression constructs using CaCl precipitation. After 2 days, transfected cells were fixed with 3% formaldehyde and 2% sucrose in PBS for 7 min at RT. Fixed cells were washed three times with PBS and incubated with different Trx1 constructs or recombinant CD30L (R&D Systems). Proteins were visualized using appropriate primary antibodies (Anti-CD30L polyclonal antibody (R&D Systems), anti-CD30 monoclonal antibodies Ki-1 (Santa Cruz) or Ber-H2 (DakoCytomation), anti-CD95 monoclonal antibody (a kind gift from Dr P Krammer), anti-Trx1 polyclonal antibody (M Preuss and TP Dick, unpublished) followed by FITC-conjugated anti-goat IgG, FITC-conjugated anti-rabbit IgG or TRITC-conjugated anti-mouse IgG and analyzed with a Nikon C1Si confocal microscope.

DNA looping has been proposed as a fundamental mechanism for action at a distance in the control of gene expression and DNA recombination [reviewed in ()]. It is of interest to understand whether the intrinsic physical properties of DNA are consistent with DNA looping constrained only through protein binding at the ends of the loop, or if looping also requires the action of proteins that enhance the apparent flexibility of the intervening DNA. There is active debate over the best model for describing local DNA stiffness, including recent controversial results on the probability of large bends in short pieces of DNA (). However, it is clear from experiments that DNA strongly resists bending and twisting over distances shorter than its ∼150 bp persistence length. For example, the efficiency of recombination using the invertasome system falls dramatically as the length of DNA looped between recombination sites is reduced below one persistence length (). That this effect is due to bending stiffness is confirmed by the observation that restriction fragments shorter than ∼200 bp are poor substrates for cyclization by DNA ligase (). Both reactions involve contact between distant DNA sites, and their rates are both increased dramatically when sequence-non-specific architectural proteins (bacterial HU or eukaryotic HMGB proteins, respectively) are present. Architectural binding proteins are sequence-specific or non-specific proteins whose main functions appear to be reshaping DNA and/or changing its apparent stiffness (). Despite considerable study, it remains unclear whether these proteins function to increase apparent flexibility by creating static bends at various locations or by inducing flexible DNA kinking. The lifetimes of many such complexes on unperturbed DNA are unknown. Both prokaryotes and eukaryotes must dramatically compact their genomic DNA while maintaining access of the genetic material to replication, transcription and repair machinery. If it were rigid, the 4.5 × 10 bp chromosome would form a circle 1.5 mm in circumference. The intrinsic flexibility of DNA should allow spontaneous collapse only to a volume of ∼200 µm (i.e. ∼400 times larger than the nucleoid). Thus, significant further DNA compaction must be achieved (). Proteins can facilitate this additional compaction using binding-free energy available from protein/DNA interactions. Genome compaction in eukaryotes is achieved by the spooling of DNA onto histone octamers and packing of the resulting nucleosomes into 30 nm and higher-order fibers. Packaged prokaryotic DNA lacks defined structures analogous to nucleosomes. Rather, six key nucleoid proteins assemble with the bacterial chromosome (). These proteins include (in decreasing order of abundance during exponential growth): factor for inversion stimulation (Fis), heat-unstable nucleoid protein (HU), integration host factor (IHF), histone-like, nucleoid-structuring protein (H-NS), suppressor of td mutant phenotype A (StpA) and DNA-binding protein from starved cells (Dps). The present work focuses on HU, IHF and H-NS. HU is a heterodimer of 90-residue (9.2 kDa) HU-1 and HU-2 subunits encoded by the and genes, respectively (,). HU binds DNA without sequence specificity, but has been reported to bind with greater avidity to pre-bent or kinked DNA (), consistent with expectations for a DNA-bending protein. HU accumulates to ∼25 000 dimers per cell during exponential growth, and it has been proposed that the protein functions in many processes including DNA replication, transcriptional repression and recombination (). HU facilitates formation of a DNA repression loop in the operon, where the protein reportedly binds a specific sequence in the loop (). Analysis of co-crystals of HU and DNA show sharp and variable DNA bending around the protein (). The integration host factor IHF, named for its role in phage λ integration, binds specific asymmetric DNA sites (WATCAANNNNTTR; W, A or T; N, any base; R, A or G) as a heterodimer (,), bending DNA by 160°. IHF and HU monomers share ∼30% sequence identity and are related in both structure and function (,,). The IHFα (99 aa, 11.2 kDa) and IHFβ (94 aa, 10.7 kDa) subunits are ∼30% identical to each other (and to HU monomers) and are encoded by the and genes, respectively. Unlike HU, which can form either homodimers or heterodimers, IHF is isolated from cells only as a heterodimer, and IHFβ homodimers can only be produced at high subunit concentration . Such IHFβ homodimers bind DNA with 100-fold lower affinity than heterodimers, and IHFα homodimers are even less stable; thus, the removal of only the IHFβ subunit in our experiments should eliminate all DNA-binding activity (). Cells contain ∼10 000 IHF dimers during exponential growth (). Gene array profiling indicates that the expression of at least 100 genes is altered upon deletion of IHF, with 46 of the genes containing putative IHF-binding sites (). H-NS homodimers (137 aa, 15.4 kDa), the product of the gene, also accumulate to ∼10 000 dimers per cell (,,). The protein is believed to oligomerize into higher-order complexes via a coiled-coil domain (,,). The protein C-terminus binds DNA, appearing to prefer A/T-rich or curved DNA. H-NS acts as a transcriptional repressor or silencer, with an H-NS-deficient strain showing increased expression of >100 genes (). H-NS protein may be involved in DNA condensation, and overexpression is lethal (,). The H-NS-like Sfh protein carried by R27 plasmids of () appears to act as a general repressor of plasmid transcription, as the plasmid is better tolerated by the host in the presence as opposed to the absence of Sfh. This result suggests a general role of H-NS-like proteins in gene repression. We have developed an experimental system for measuring DNA flexibility in living cells and have used it to determine whether proteins play important roles in enhancing the apparent flexibility of DNA. The system, shown schematically in , is based on classic studies of DNA looping in repression of the lactose operon (). The reporter construct is a simplified operon, with a reporter gene placed downstream from the moderately strong UV5 promoter. The operon is modified so as to increase sensitivity to DNA looping by using a weak proximal O operator to mediate repression and a strong distal O operator upstream. The Lac repressor binds strongly to O, and it can form a repression loop by simultaneously binding at O. The stability of this loop is related to the energetic costs of bending and twisting the intervening DNA, which are, in turn, dependent on the distance of separation and helical phasing of the operators. The reporter is introduced in single copy on an F′ episome. Host strains with or without the genes encoding architectural DNA-binding proteins are used; they all express wild-type levels of the wild-type bidentate Lac repressor tetramer (). The development of this system and its application to flexibility induced by the rat HMG-B protein have been described in detail (). Measurement of the degree of promoter repression as a function of operator separation provides information about the longitudinal and torsional bending properties of DNA . Control experiments using IPTG (a stable allolactose analog) reveal the behavior of the system when the affinity of repressor for DNA has been dramatically reduced. Previous characterization of this system for loop lengths of 63–91 bp revealed that (i) DNA twisting rather than bending is the major obstacle to DNA looping; (ii) weak repression loops are still detected in the presence of saturating concentrations of IPTG; (iii) deletion of the architectural protein HU dramatically destabilizes repression loops; (iv) replacement of HU with heterologous mammalian HMGB proteins can partially rescue DNA looping and (v) the effect of HU loss on DNA looping does not result from changes in DNA supercoiling (). The key conclusion of these experiments was that the sequence-non-specific HU protein is required to stabilize small repression loops . These data as well as the original length-dependence work of Müller-Hill () have been analyzed recently by others, using several different formalisms (). In one analysis () it was concluded that the bending properties of DNA in cells lacking HU match those measured , but other work has used Lac operon looping to support the existence of surprisingly easily bent DNA (). An earlier rod mechanics model of the loop suggested that the repressor conformation changes depending on the supercoiling environment (), in agreement with earlier experiments (). The additional repression peaks caused by weak loops formed by the presence of IPTG-bound repression [conclusion (ii) above] have been identified in the Müller-Hill data set (,). Clearly, more experimental work is needed to clarify the loop geometry, the dependence of DNA flexibility on architectural proteins, and the effects of supercoiling. The present study explores how the loss of HU and other nucleoid proteins affects DNA looping. We have confirmed and extended our earlier results. The most surprising new finding is that H-NS acts to destabilize rather than stabilize small loops. Changes in either total or unrestrained superhelicity do not appear to explain the effects of architectural proteins on DNA looping. strains bearing gene disruptions are described in . The and genes were disrupted in parental strain FW102 () as described (,). Disruption of the gene (encoding the IHFβ subunit) was accomplished by gene-targeted recombination with a kanamycin selectable marker (complementary sequence in bold) amplified with primer pair LJM-2485 5′-ATCATGCAGCACAGCAGCGCTATGCTAGAC and LJM-2486 5′-AGCACGACAGTGCTCTCTCGTCAGTGAGTA by published methods (). Disruption of the gene similarly involved recombination with a selectable marker (complementary sequence in bold) amplified with primer pair LJM-2477 5′-TAGCTCTATACTACACACACATATAGTG and LJM-2478 5′-ATATCGCGCTGCGATAGCAGTGCATCTAC. In each case, the integrated selectable marker was removed in a second step involving recombination between FRT sites as described (). Genotypes of all deletion strains and the presence of looping assay episomes were confirmed by diagnostic PCR amplification following conjugation and selection. DNA looping constructs were based on plasmid pJ992, created by modifications of pFW11-null () as previously described (). Constructs contained a strong distal O operator and a weak proximal O operator. The O operator normally present within the coding region was destroyed by site-directed mutagenesis (). A construct with a proximal O but lacking upstream O was used as a normalization control. Test promoters did not contain CAP-binding sites. looping constructs were placed on the single copy F128 episome by homologous recombination between the constructed plasmids and bacterial episome. F128 carries the gene producing wild type levels of repressor. Bacterial conjugation and selections were carried out as previously described (). After mating and selection, correct recombinants were confirmed by PCR amplification. All chemicals were obtained from Sigma (St Louis, MO, USA). expression was measured by a liquid β-galactosidase colorimetric enzyme assay as described (). The assay was modified as follows to increase efficiency: cultures were grown in 1.1 ml LB/antibiotic in 96-well boxes (2 ml capacity per well) with shaking (250 rpm) at 37°C. Subcultures (1.1 ml of media) were then inoculated with 30 μl of overnight culture in the presence or absence of 2 mM IPTG. Subcultures were grown with shaking at 37°C until OD reached ∼0.3. For samples with low β-galactosidase activity, 800 µl of bacterial culture was assayed after centrifugation and resuspension in 1 ml Z-buffer (60 mM NaHPO, 40 mM NaHPO, pH 7.0, 10 mM KCl, 1 mM MgSO, 50 mM β-mercaptoethanol). For samples with high levels of β-galactosidase activity, 100 µl of bacterial culture was diluted with 900 µl of Z-buffer before analysis. Cells were lysed by addition of 50 µl chloroform and 25 µl 0.1% SDS, followed by repeated pipetting (10–12 times) with a 12 channel pipettor. Samples were equilibrated at 30°C for 5 min, followed by the addition of 200 µl of 4 mg/ml O-nitrophenylpyranogalactoside (ONPG) in Z-buffer. Incubation at 30°C continued with accurate timing until OD reached ∼0.5. Reactions were stopped with 500 µl 1 M NaCO and the reaction time was recorded. Cell debris was pelleted by centrifugation of the 96-well box for 10 min at 4000 × . Three hundred and fifty microliters of cleared samples were transferred to 96-well plates. Sample OD readings were measured on a Molecular Devices SpectraMax 340 microtiter plate reader. β-galactosidase activity () was calculated according to: Assays were performed with a total of six colonies from each independent strain repeated on two different days. The enhancement of repression due to specific DNA looping is expressed in terms of the normalized expression parameter  ′, according to: Note that Equation is corrected from the original report (). A previously described statistical weights/DNA mechanics model () was used for simultaneous fitting of experimental ′ and data to expressions for the distribution of possible states of the O operator under repressed and induced conditions. The experimentally derived fraction of O that is bound by repressor () is modeled as a function of DNA spacer length () with five adjustable parameters: the optimal operator spacing in bp (), the DNA helical repeat (), the apparent torsional modulus of the DNA loop (), the equilibrium constant for specific O–O loop formation when operators are perfectly phased () and an equilibrium constant for non-specific looping (). The term describes all forms of O-dependent enhanced binding to O other than the specific loop; for example, it could include looping between O and a pseudooperator overlapping O, or enhanced O binding via sliding or hopping from O. The total linking number deficit of plasmid DNA isolated from cells (500 ng) was assayed by separating topoisomers on multiple agarose gels at different concentrations of chloroquine (,). Electrophoresis was performed at 2 V/cm for ∼18 h in 1X TAE buffer (40 mM Tris–acetate and 1 mM EDTA) through 0.8% agarose gels containing 0–10 μg/ml chloroquine. Gels were then stained with 0.5 μg/ml ethidium bromide until bands were barely visible, nicked for 30 s with long-wave UV irradiation, and then stained for an additional 30 min prior to photography. This procedure assures that the ethidium bromide binds equivalently to the different topoisomers. Trimethylpsoralen (TMP; Sigma) photo-crosslinking of plasmid DNA followed by Southern analysis was performed with modifications of a published method (). Briefly, bacterial strains containing plasmid pJ992 () were subcultured and grown to mid-log phase, A ∼ 0.6, typically in 5 ml of LB medium containing antibiotics. Cultures were pelleted by centrifugation at 3000 × for 10 min, washed in M9 salts () and resuspended in M9 salts at 1/10 of the original volume, all at 4°C. TMP treatment and the initial steps of DNA harvesting were performed in the dark at 4°C unless otherwise indicated. An ethanol-saturated solution of TMP (5 µl) was added to 500 µl cell culture in a 6-well plate and allowed to equilibrate for 5 min. Samples were irradiated for various times using long-wavelength UV light (∼366 nm) delivered from a hand-held (Mineralight) lamp at an intensity of 0.6 mW/cm. To compensate for TMP autoinactivation, additional TMP was added every 10 min (). Cells were then lysed and DNA harvested as previously described (). DNA was digested with I to yield an ∼950 bp restriction fragment from plasmid pJ992. A sample (5 µg) of the total resulting DNA was denatured in 100 mM NaOH by treatment for 2 min at 55°C followed by acid neutralization, and electrophoresed on a 1.2% agarose gel at 3 V/cm (). DNA was transferred to a Nytran membrane (Whatman) using an alkaline transfer system following the manufacturer's recommended protocol. After UV cross-linking to the membrane (Stratalinker apparatus), the membrane was hybridized with a radioactive probe specific for the DNA fragment of interest. The hybridization signal was quantitated using a Molecular Dynamics Storm PhosphorImager. Plasmids pJ1345 (original name pPHB94) and pJ1346 (original name pPHB95) are derived from previously described constructs () and were the generous gifts of P. Heisig. Plasmid pJ1345 contains the luciferase gene under control of the promoter cluster, while plasmid pJ1346 places the same reporter under the control of the promoter (). For comparisons of unrestrained supercoiling strain in different genetic backgrounds, the promoter cluster and promoter were assayed in two additional contexts. Plasmids pJ1454 and pJ1456 place the promoters on pJ992, a smaller pACYC184-based plasmid for comparison (,). Primers 5′-CGACGATCTATCGTACTCTGATG and 5′-GCTCGCTGCAGCGTGAGATGCAG) were used to amplify the promoter and reporter regions from plasmid pJ1345 (∼2300 bp product) and pJ1346 (∼2000 bp product), installing flanking I and I restriction sites. PCR products were cloned between the I and I restriction sites of pJ992 to create plasmid pJ1454 (luciferase gene under the control of the promoter) and plasmid pJ1456 (luciferase gene under the control of the promoter). The promoter-reporters from pJ1454 and pJ1456 were also moved onto the single copy F128 episome by homologous recombination (). Luciferase assays were performed using the Promega Luciferase assay kit with modifications to accommodate bacterial cells. Briefly, subcultures (5 ml) in LB media containing kanamycin were inoculated with saturated overnight culture (125 µl). Subcultures were grown at 37°C, with agitation, until A reached ∼0.6. A sample of the culture (90 µl) was combined with buffer (10 µl: 1 M KHPO pH 7.8, 20 mM EDTA). Samples were mixed and frozen at −80°C for 30 min. To each thawed sample, cell culture lysis reagent (200 µl; Promega, Madison, WI) and fresh lysozyme mix (100 µl of solution containing 5 mg/ml lysozyme and 5 mg/ml BSA) was added. Samples were incubated at 25°C for 10 min. Accurately measured samples (1–5 µl) were added to luminometer tubes followed by the addition of luciferase assay reagent (100 µl; Promega) and luciferase activity was measured on a Turner 20/20 luminometer. The normalized unrestrained supercoiling ratio () is given by the : reporter expression ratio, after normalization for cell density: We first collected expression data for the full set of operator spacings using recombinant F′ episomes carried in our standard laboratory strain of (FW102, labeled ‘WT’). Data are shown in A. The upper panel displays the conventional repression ratio (), while the lower panel shows normalized expression (′) both in the absence and presence of 2 mM IPTG as inducer. Data points were analyzed by a statistical weights/DNA mechanics model () with simultaneous fitting to and ′ data. The full data set is given in Supplementary Tables S1 and S2, and curve fit parameter estimates are provided in . Since our previous work, we introduced assay modifications that allow more rapid data collection and therefore more simultaneous measurements directly comparing different strains and conditions. The changes in assay timing, however, resulted in slightly higher measurements of gene expression for the most tightly repressed WT strains. Even so, the fit parameters in for WT cells are quite similar to the previous report (). The value for is slightly smaller reflecting less-efficient peak repression and the value of is slightly larger to compensate for the largely unchanged level of out-of-phase repression. All data reported here are new and were collected using the same methodology. The repression ratio data for WT cells are characterized by periodic oscillation, with peaks (maximal repression) corresponding to alignment of operators on the same DNA face separated by integral multiples of the helical repeat of the DNA. As previously noted (), the oscillation pattern contains secondary peaks, and it also appears to be damped as DNA length increases. The origin of this unexpected complexity is revealed by plotting the ′ data from both induced and uninduced WT cells (A, lower panel). It is clear that under both conditions reporter expression oscillates with the DNA helical repeat, but that the oscillations are not precisely in phase. This is evidence for weak looping by Lac repressor even under conditions of induction at saturating IPTG concentrations. Optimal operator spacing differs by ∼0.5 bp and the helical repeat differs by ∼0.9 bp when loops are fit under induced versus uninduced conditions (). The complex peak structure of the repression ratio curve reflects dephasing (due to the helical repeat difference) between the uninduced and induced ′ data rather than a spacing-dependent change in DNA looping energy. The apparent overall damping in the repression ratio is also due to dephasing, as well as to broadening of the torsional oscillations as the length increases. It does, however, appear that points at longer DNA length tend to fall slightly below the fit curves, suggesting a weak effect of longitudinal flexibility. This lack of a strong dependence of optimal loop probability on DNA length for such small DNA segments suggests enhanced apparent DNA longitudinal flexibility : if the loop-free energy were controlled by the DNA elastic bending energy, we would have observed much more dramatic length dependence. The low fit value of the apparent DNA torsional modulus (; ) suggests a similar increase in apparent torsional flexibility. Analysis of the ΔHU strain (B) substantiates previous observations that loss of HU substantially disables looping (). The observed ∼3-fold effects of genetic background on promoter strength are factored out of both repression ratio (upper panel) and  ′ data (lower panel), so reflects specific effects on DNA loop stability. Loss of HU protein causes a global ∼3-fold promoter derepression (B), reflected in a ∼3-fold decrease in estimated equilibrium constants for specific and non-specific loop formation ( and , ). Notably, induced expression shows almost no residual operator phase dependence (B, lower panel, filled circles), suggesting that the putative loop formed by IPTG-bound repressor requires HU for stability. The optimal operator spacing for uninduced DNA looping is ∼2 bp shorter in the absence of HU, suggesting a change in the optimum repression loop geometry, a reorganization of the DNA domain, or a change in the local supercoiling status in the absence of this nucleoid protein. Destabilization of DNA looping in the absence of HU could imply a direct, sequence-non-specific but possibly structure-specific binding of HU to looped DNA, as is seen for the related Gal repressor (). An alternative, though not mutually exclusive, possibility is that unrestrained negative superhelical strain in this region of the F′ element stabilizes DNA looping and loss of HU reduces this strain. Looping assays in cells lacking the sequence-specific nucleoid protein IHF show only a modest looping disability relative to WT cells (C). Residual repression under inducing conditions is marginally decreased relative to WT. The phase-dependence of repression in the absence of IPTG is slightly reduced (reflected in a smaller ), and the equilibrium constant for the most stable loop is ∼2-fold reduced in the absence of IHF (). IHF protein displays both sequence-specific and non-specific binding (), and the experimental DNA looping constructs do not contain specific IHF recognition sequences. These results suggest that non-specific IHF binding contributes only slightly to apparent DNA flexibility under these conditions. Cells that do not produce H-NS protein were also tested. Quite surprisingly, ΔH-NS cells showed substantially DNA looping compared to WT (D). Repression ratios were ∼2-fold improved for optimally aligned operators, while remaining unchanged for operators on opposite helical faces of DNA. These effects are also clear in the  ′ data (D, lower panel) where troughs (maximal repression in the absence of IPTG; open circles) are deepened in the absence of H-NS, while peaks (promoter leakiness with out-of-phase operators) are unchanged. Upon H-NS deletion, stabilization of the optimally aligned loop is most apparent, reflected in the model fitting parameters by a ∼1.7-fold increase in the looping equilibrium constant . For out-of-phase operators the corresponding stabilization is not observed due to the disappearance of significant nonspecific looping (a decrease in from ∼20 to ∼0). We caution that the parameter is often not precisely determined by the data, so our interpretations do not rest on it. Finally, upon H-NS deletion remains essentially the same, suggesting that H-NS does not change the twist flexibility of DNA significantly: in the context of the weak induced loop the protein may increase twist flexibility very slightly. These results suggest that the presence of H-NS primarily acts to destabilize DNA looping through its effect on DNA bending, not twisting. This loop destabilization could be due to decreasing the longitudinal flexibility of DNA, but we do not observe a progressive change in the peak and trough heights with length, as might be expected in this case. H-NS could also constrain the DNA in a conformation that tends to prevent looping, as discussed later. Overall, the changes in the and parameters suggest that H-NS constrains the DNA in some fashion that allows (or perhaps even enhances) non-specific contacts between DNA sites but tends to destabilize the specific O–O loop. Perturbations in the normal complement of nucleoid proteins could influence the stabilities of repression loops by direct and/or indirect mechanisms. For example, direct loop stabilization could occur if architectural proteins bind within the DNA loop so as to introduce at least transient favorable bends or sites of flexibility. Indirect effects on DNA looping could result from local changes in superhelical strain or superhelix geometry caused by the absence of a nucleoid protein. To differentiate among explanations invoking superhelical writhe, twist strain and protein binding, it is important to have measurements of supercoiling (due to DNA wrapping on protein surfaces), and supercoiling, which creates actual mechanical twisting strain (). The total DNA linking number deficit is the sum of these two components. Restrained supercoiling could be important in creating a DNA geometry favorable or unfavorable for looping, whereas unrestrained supercoiling has the potential to drive the compaction of naked DNA, facilitating formation of the repression loop. Enhancement of DNA looping by supercoiling has been reported () and (). To determine whether direct or indirect mechanisms better explain the effects of nucleoid proteins on DNA looping, we independently measured both total and unrestrained negative supercoiling in WT, ΔHU, ΔIHF and ΔH-NS strains using several methods. Total supercoiling was monitored by assessing plasmid topoisomer distributions after extraction from cells, using electrophoretic separation of topoisomers in agarose gels containing different concentrations of the intercalating agent chloroquine. The concentration of chloroquine required to alter DNA twist sufficiently to eliminate negative supercoils (thereby maximally reducing plasmid electrophoretic mobility) is directly related to the initial negative superhelical density of the extracted plasmid. and show the results when plasmids from WT, ΔHU, ΔIHF and ΔH-NS strains were electrophoresed in the presence of 1, 2, 3, 5, 7 or 9 µg/ml chloroquine. Plasmid topoisomer mobility decreased over this chloroquine concentration range, demonstrating that all plasmid populations were initially negatively supercoiled. Importantly, however, the degree of total negative supercoiling depended on the complement of nucleoid proteins. Inspection of the electrophoresis series showed that total negative supercoiling was indistinguishable in WT and ΔH-NS strains. In contrast, total negative supercoiling was strongly reduced in ΔHU cells but strongly increased in ΔIHF relative to WT (this is most clear , 7 µg/ml chloroquine). The result for ΔHU cells confirms our previous observations (). As discussed earlier, the chloroquine titration method assays total supercoiling, reflecting the sum of restrained and unrestrained supercoiling of plasmids in the cell. Because unrestrained supercoiling generates local twisting strain, while restrained supercoiling does not, it was important to determine if levels of unrestrained supercoiling follow the same trends as those seen in . One biochemical approach to the measurement of unrestrained supercoiling was developed for eukaryotic cells. It is based on the enhanced binding and photocrosslinking of intercalating psoralen derivatives to DNA under unrestrained superhelical strain (). We adapted this approach to monitor changes in superhelical strain in prokaryotes. The assay measures changes in the rate of trimethylpsoralen (TMP) crosslinking of plasmid pJ992. The extent of psoralen cross-linking was measured by plasmid isolation, restriction digestion, denaturation, electrophoresis and Southern blotting to measure the extent of crosslink-enabled rapid dsDNA renaturation. Sample autoradiograms are shown in B, and the results of image quantitation are given in C and . These data show that levels of unrestrained negative supercoiling in the WT and ΔHU backgrounds are similar, while negative supercoiling in ΔH-NS and ΔIHF strains was increased relative to WT (). These results generally track with measured levels of total supercoiling except for the ΔH-NS strain. We found the TMP crosslinking assay to be technically challenging and only moderately reproducible for the present purpose. We therefore performed a more direct bioassay that has recently been reported for the measurement of unrestrained supercoiling in bacteria (). This assay exploits the fact that the promoters driving expression of DNA gyrase (p) and topoisomerase I (p) are oppositely responsive to local levels of unrestrained negative supercoiling. The promoter is repressed by high unrestrained negative supercoiling and activated when unrestrained negative supercoiling is low. In contrast, the promoter cluster driving is activated by high unrestrained negative supercoiling and repressed when unrestrained negative supercoiling is low. This intuitively reasonable relationship supports bacterial supercoiling homeostasis. By placing reporter genes downstream from these promoters () in separate plasmid constructs or in the bacterial assay episome itself, the ratio of : reporter activities can be used to monitor unrestrained negative supercoiling . This assay approach was applied to WT, ΔHU, ΔIHF and ΔH-NS strains with reporter promoters either carried on high or low-copy number plasmids of different sizes or integrated into the large F′ episome, providing the results shown in . As seen for total supercoiling, the extent of unrestrained negative supercoiling was found to be very similar for WT and ΔH-NS strains, independent of the DNA construct carrying the reporters. This result was completely reproducible, though it does not agree with a previous report that loss of H-NS caused a modest increase in negative supercoiling (). In contrast, the diagnostic ratio for ΔHU cells depended on the location of the reporter genes. For plasmids, the ratios were significantly less than the WT ratio, but the value was more variable and somewhat increased versus WT when assayed in the F′ episome (). The basis for context-dependent changes in the unrestrained negative supercoiling in the absence of HU protein is unknown. The data in also show that unrestrained negative supercoiling was consistently and significantly higher in ΔIHF cells, with the diagnostic ratio 2- to 4-fold higher that observed for WT cells: for ΔIHF cells the results of all supercoiling assays are concordant. Three important conclusions can be drawn from these supercoiling assays. First, the promoter activity bioassays of unrestrained supercoiling in the test strains parallel the results of total supercoiling assays. In general it appears that a fairly constant fraction of total supercoiling is due to unrestrained supercoiling . Second, unrestrained supercoiling estimates from supercoiling-dependent promoters generally do not depend on whether the promoter-reporters are present on plasmids or episomes. Third, and most importantly here, the effects of eliminating different nucleoid proteins on repression loop stability be explained simply by invoking perturbations in local unrestrained supercoiling. Comparison of the data in and reveal between repression loop stability and levels of unrestrained negative supercoiling in the different genetic backgrounds tested. In particular, ΔH-NS cells showed strongly enhanced DNA looping but had a level of negative DNA supercoiling that was similar to WT cells (). The ΔIHF cells showed strongly increased unrestrained negative supercoiling but little change in DNA looping (). The ΔHU cells have only slightly decreased negative supercoiling but show substantial changes in loop stability and character. These results suggest that the observed changes in looping are not caused by indirect supercoiling effects. This conclusion also supports our previous observation that DNA looping was not reduced when total negative supercoiling in was reduced using the gyrase inhibitor Norfloxacin (). The data in indicate that loss of the H-NS protein enhances the formation of the experimental DNA loop in living . The data in and show that this effect does not involve changes in DNA supercoiling. How might the loss of H-NS facilitate DNA looping? Single molecule experiments have shown that H-NS, like other DNA-binding proteins (,), can either endow flexibility (at low binding densities) or stiffness (at high binding densities) (,). If H-NS concentrations were sufficient to drive high protein occupation of the repression loop, DNA stiffening might antagonize looping, but this seems unlikely, as exponentially growing cells are thought to contain only about one H-NS dimer per 1400 bp of DNA (,). Another model to explain H-NS inhibition of DNA looping is based on the observation that H-NS dimers have two DNA-binding domains and can bridge between DNA duplexes (,). Such cross-linking could inhibit DNA looping by the mechanism shown in . Looping requires the shortening of a segment of DNA and would be inhibited if DNA segments are cross-linked by H-NS proteins and not free to slide past one another in the crowded nucleoid. We have adapted classic work on the operon () to develop a system for studying DNA flexibility. Regulatory elements from the operon have been organized into a series of episomal constructs in which reporter gene repression is highly dependent on the stability of a small DNA loop between operators. We previously used this system to show that DNA twist inflexibility limits looping , while optimal loop stabilities are independent of DNA length over the range 63–91 bp (). We also demonstrated weak DNA looping by Lac repressor even when saturated by IPTG, and showed evidence that loss of the nucleoid protein HU destabilized repression loops. The present experiments confirm and extend these conclusions. We demonstrate that the simple statistical weights/DNA mechanics model used here can capture the observed experimental variations and suggest underlying physical interpretations. It has not been necessary to include explicit consideration of DNA bending persistence length changes or alternative loop shapes, but the lack of a DNA length effect and the calculated values of the torsional modulus confirm that the apparent effective flexibility of DNA is much greater than would be expected from experiments. The structure and composition of the bacterial nucleoid is ill-defined, and much remains to be learned regarding the roles of major nucleoid proteins. In one proposed classification (), nucleoid proteins are characterized as DNA bridging factors (H-NS, SMC, Lrp) or DNA bending factors (HU, IHF, Fis). Our study used DNA loop stabilization as a probe of the effects of HU, IHF and H-NS proteins on properties of the nucleoid. We conclude that loss of the sequence-non-specific DNA bending factor HU strongly destabilizes a repression loop, while loss of the sequence-specific bending factor IHF has much less effect. In contrast, loss of the DNA bridging factor H-NS from the nucleoid DNA looping. We show that the effects of these gene disruptions on DNA looping do not generally correlate with their effects on negative DNA supercoiling in the host. We therefore interpret the DNA looping phenotypes of HU, IHF and H-NS deletion strains as evidence for direct protein–DNA interactions that alter the apparent physical properties of the looped DNA. We cannot rule out that DNA looping effects observed upon removing nucleoid proteins could have more complex origins. For example, mutant strains may express altered levels of the remaining nucleoid proteins and/or other factors related to apparent DNA flexibility (). The different effects on DNA looping of IHF versus HU loss may be due to the lower abundance of IHF or to higher non-sequence-dependent DNA binding by HU in the DNA loop. Future studies will investigate whether IHF and HU participate in the experimental repression loop. The strong enhancement of DNA looping upon deletion of H-NS is surprising, and the mechanism remains unknown. It remains possible that the loss of H-NS might simply reduce competition for important HU sites near the DNA loop, but we suggest that loop stabilization is due to inhibition of DNA slithering. Comparisons among other bending and bridging proteins should resolve this issue. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Non-long terminal repeat (non-LTR) retrotransposons, also called long interspersed nuclear elements (LINEs), are the most abundant family among mobile elements. LINEs have been identified in all major groups of eukaryotes, with the exception of the bdelloid rotifers (). In human, up to 21% of the genome is comprised of LINEs (), which are implicated to be involved in the gene evolution and genome reconstruction (). LINEs are divided into two subtypes. The first has only one open reading frame (ORF) that encodes a type-2 restriction-enzyme-like endonuclease (EN), whereas the other has two ORFs including an apurine/apyrimidine endonuclease (APE)-like EN at the N-terminus of the second ORF (ORF2) (). ORF2 of the latter members of the LINE family, which include L1 (,), TRAS1 (), SART1 (), Tx1L () and R1 (), typically encodes an EN domain at the N-terminus, a reverse transcriptase (RT) domain in the middle region and a zinc-finger-like domain at the C-terminus. The first ORF (ORF1) has no sequence similarity to other known sequences, although some ORF1 proteins have been reported with RNA-binding () activities. The retrotransposition process of LINEs is known as target-primed reverse transcription (TPRT). In the TPRT model, the EN domain first recognizes and cleaves the target DNA and the RT domain immediately proceeds with reverse transcription using its own mRNA as the template (). Although most LINEs insert randomly throughout the genome, some elements show high-target sequence specificity. For example, TRAS1 (,) and SART1 () insert into the telomeric repeats of , Tx1L () inserts into the Tx1D sequence of and R1 inserts into 28S rDNA sequences of many insect species (,). Of these target-specific LINEs, TRAS1, SART1, and R1 belong to the R1 clade. Most retrotransposons in this group are target specific (). Previous studies have demonstrated that the EN itself has sequence specificity (,,,). Furthermore, a study of EN swapping between TRAS1 and SART1 showed that the target specificity of EN determines its insertion site (). Structural studies of TRAS1 () and human L1 () have revealed that the LINE EN has a folding pattern similar to that of apurinic/apyrimidinic (AP) EN, but has an extra β-hairpin at the DNA-binding surface. A mutagenesis study of the TRAS1 EN indicated that the β-hairpin region is necessary for sequence recognition (). In spite of these reports, molecular basis for the sequence specificity of EN encoded in LINEs are still unclear. R1Bm, the R1 retrotransposon species of , inserts into a 14-bp region of 28S rDNA. Biochemical analysis has shown that R1Bm EN cleaves the sequence ACA!GTG (! = cleavage site) on the bottom strand followed by the sequence ACT!ATC on the top strand (), although the detailed mechanism remains unknown. In this study, to elucidate how R1Bm EN achieves such complicated sequence recognition, we performed a substituted oligonucleotide cleavage study for the target DNA sequence and determined the crystal structure of R1Bm EN at 2.0 Å resolution. Furthermore, mutagenesis studies of the DNA-binding surface of R1Bm EN based on the crystal structure revealed some amino acids involved in the sequence specificity. The R1Bm EN domain was amplified by PCR with Turbo DNA polymerase (Stratagene) using primers R1EN-NdeI-s (5′-AAAAACATATGGATATTAGGCCCCGACTTCG-3′) and R1EN + 19aa-XhoI-a (5′-AA AAACTCGAGTTACGGCTCGCCCGGCCTCAAATCGC-3′) with R1WT-pAcGHLTB as a template (). The first primer contains an NdeI site. The second primer includes a stop codon followed by an XhoI site. The PCR product was digested with NdeI and XhoI, and subcloned between NdeI and XhoI sites of the pET16b expression vector (Novagen). The resulting plasmid, named pR1EN, includes the first 717 bp of ORF2, corresponding to the first 239 amino acids, and the sequence encoding 10-His tag at the N terminus of R1Bm EN. All point mutations were generated by a QuickChange site-directed mutagenesis Kit (Stratagene) according to manufacturer's instructions. The sequences of the primers used for the introduction of these mutations are available on request. The mutation of each plasmid was confirmed by DNA sequencing. The pR1EN and its mutant derivatives were transformed into BL21(DE3)/pLysS strain. The transformants were cultured at 37°C in 50 ml of Luria broth until the optical density at 600 nm reached ∼0.8. Isopropyl- β--thiogalactopyranoside (IPTG) was then added to a final concentration of 1 mM, followed by further incubation at 25°C overnight. Cells were pelleted by centrifugation and frozen in liquid nitrogen. Purification of R1EN was conducted following the protocol from Qiagen (catalog number 30210). Cell pellets were thawed at 4°C for 10 min, suspended in 0.6 ml of sonication buffer [50 mM sodium phosphate (pH 6.0), 0.5 M NaCl, 100 mM imidazole, 20 mM 2-mercaptoethanol, 2% triton X-100] and sonicated for 1 min on ice. The cell extracts were clarified by centrifugation at 20 000 × for 10 min and filtration through a 0.45 μm membrane (Millipore). Because the total volume of cell extracts increased when the mass of cell pellets are taken into account, the imidazole concentration was readjusted to 100 mM, the supernatant was mixed with 30 μl of pre-equilibrated nickel NTA agarose (Qiagen) at 4°C for 2 h. The resin was washed three times with 1 ml of sonication buffer, three times with 1 ml of washing buffer [50 mM sodium phosphate (pH 6.0), 1 M NaCl, 100 mM imidazole] and once with 1 ml of washing buffer containing 0.35 NaCl. Finally, the protein was eluted with 0.25 ml of elution buffer [50 mM sodium phosphate (pH 6.0), 0.35 M NaCl, 0.3 M imidazole]. The eluted protein was ultrafiltered and concentrated in storage buffer [50 mM sodium phosphate (pH 7.0), 0.35 M NaCl, 10% glycerol, 10 mM 2-mercaptoethanol] with Microcon YM-10 (Millipore). The purified protein, the concentration of which was determined by SDS–PAGE, was diluted with storage buffer at 0.5 mg/ml and stored at −80°C. The concentration of R1Bm EN was determined by comparing the intensity of the band with an analytical curve obtained from Coomassie blue stained SDS–PAGE gel with that of known amounts of bovine serum albumin. For crystallization, BL21(DE3)/pLysS cells carrying pR1EN were cultured in LB medium at 37°C. When OD had reached to 0.6, IPTG was added to the medium at final concentration as 0.75 mM, and cells were further incubated for 6 h at 26°C. Cells were collected by centrifugation and stored at −80°C. Stored cells were thawed and sonicated in lysis buffer [40 mM Tris–Cl (pH 7.5), 0.5 M NaCl, 50 mM imidazole, 0.1% triton X-100 and 0.1 mM PMSF], centrifuged and the supernatants were subjected to Nickel-trapped HiTrap Chelating column (GE healthcare) and eluted by elution buffer [40 mM Tris, 0.5 M NaCl, 0.35 M imidazole, (pH7.5)]. The proteins were precipitated by adding ammonium sulfate and collected by centrifugation, and the pellet was dissolved with 10 ml of digestion buffer [50 mM Tris–Cl (pH 7.5), 0.3 M NaCl, 1 mM DTT, 2 mM CaCl], followed by Factor Xa (New England Biolab) digestion for overnight at 10°C. The sample solution was loaded on HiTrap SP column (GE healthcare) followed by Superdex 200 (GE healthcare), resulting the highly purified R1Bm EN. The protein was concentrated in sample buffer (40 mM HEPES-Na pH 7.2, 200 mM NaCl, 1 mM DTT, 1 mM EDTA) with 10 kDa Microcon (Millipore) and protein solution was prepared to 6 mg/ml. In that case, the concentration of R1Bm EN was determined by absorption of 280 nm spectra with extinction coefficients of 27 310 M cm. Crystals of the R1Bm EN were obtained by hanging drop vapor diffusion method at 283 K. Two microliters of protein solution were mixed with 1.5 μl of reservoir solution containing 2.4 M sodium acetate (pH 6.9), 10 mM ammonium sulfate, and 1–2% Jeffamine M-600 reagent. The hexagonal rod-shaped crystals, whose size of 0.2 × 0.2 × 0.5 mm, were gown within 2–3 weeks. Prior to the data collection, crystals were transferred into cryoprotectant solutions containing 9% (v/v) ethylene glycol and 9% (v/v) PEG-200 in reservoir solution and flash frozen in cryo nitrogen stream. Dataset was collected at the Photon Factory (Tsukuba, Japan) on beamline BL-6A. Data processing and reduction were carried out with the programs MOSFLM () and SCALA (). Phases were calculated by molecular replacement method with the MolRep () program using TRAS1-EN structure (PDB entry: 1WDU) as a search model. Further model was built with program O () and structure refinement calculations including simulated annealing and B-factor refinement were done with CNS (). Summary of the crystallographic statistics are shown in . The P-labeled substrates containing the R1Bm target site were prepared exactly as previously described (). The 40-bp top or bottom strand oligonucleotides were radio-labeled, annealed with the complementary non-labeled oligonucleotides and gel-purified. The cold substrates were also prepared by annealing of non-labeled top and bottom strand oligonucleotides. The mixture of labeled and non-labeled substrates was used for the cleavage reaction. The reaction mixture containing 50 mM PIPES-NaOH at pH 6.0, 17.5 mM NaCl, 1 mM MgCl, 200 ng of purified proteins and 1 or 3 pmol of substrate DNA (the molar ratio of protein:DNA is 15:2 or 15:6) in a total volume of 10 μl was incubated at 25°C for 60 min. The reaction was stopped by the addition of 10 μl of denaturing solution (95% formamide, 50 mM EDTA, 0.01% bromophenol blue). The reaction product was denatured for 3 min at 95°C, immediately chilled on ice, and separated on a 30% polyacrylamide denaturing gel. The cleavage efficiency was quantified with BAS 5000 imaging analyzer system (Fujifilm). In the experiment of E, oligonucleotide cleavage assays were performed with the reaction mixture containing 10 mM MgCl and 0.1 pmol of substrate DNA, and incubation time was increased from 1 to 5 h. Radio-labeled oligonucleotides with the same sequence as the expected cleavage products were used for the size markers: 5′-ACGAGATTCCCACTGTCCCTATCTACT-3′ and 5′-GGTTTCGCTAGATAGTAGATAGGGACA-3′ as for top and bottom strand cleavage, respectively. To determine cleavage sites precisely, sequencing markers were used in some experiments. Sequencing markers were generated by primer extension with TaKaRa Taq Cycle Sequencing Kit (Takara) using the manufacturer's suggested protocol except for the PCR condition. The reaction mixture was denatured at 94°C for 20 s, followed by 25 cycles of 94°C for 20 s, 30°C for 20 s and 72°C for 1 min, and then another 15 cycles of 94°C for 20 s and 72°C 20 s. To generate the sequencing template, 247 bp of 28S rDNA sequence containing the R1Bm target site was amplified by PCR with Turbo DNA polymerase (Stratagene) with the primers 28SrDNA-XhoI-s (5′-AAAAACTCGAGGCGCGGGTAAACGGCGGG-3′) and 28SrDNA-BamHI-a (5′-AAAAAGGATCCCGCGAAACGATCTCCC-3′) using pBmR161 () as template. The PCR product was purified with GenElute PCR Clean-up Kit (Sigma) prior to use. The sequencing primers used for the bottom and the top strands are 5′-GGTTTCGCTAGAT-3′ and 5′-ACGAGATTCCCAC-3′, respectively. p B shows that the 28S rDNA sequence was cleaved by R1Bm EN not only at the target site, but also at other sites. Because the signal patterns for the non-specific products were not random, we focused on them and tried to determine the sequence tendency recognized and digested by R1Bm EN. The cleavage activity was estimated from the density of each band, and they were aligned by cleavage site to allow comparison (). In all those in the high-cleavage group (>50% that of the wild-type sequence), which included X, Y, 7, 4, 5 (), and B8* (), adenine is located at the −3 position from the cleavage site. In five of the six highly cleaved sequences, thymine is located at the +2 position. Furthermore, guanine or adenine is located at the +1 position and thymine or adenine at the −1 position in all sequences, and five of the six sequences have cytosine or guanine at the −2 position. Taken together, these results suggest that the consensus sequence for the high-cleavage group is 5′-A(G/C)(A/T)!(G/A)T-3′ (! = cleavage site). The weakly cleaved sequences (<50% that of the wild-type sequence), which included 1, 2, 3, 6 and B5

Cockayne syndrome (CS) is a rare autosomal recessive genetic disorder, classified as a segmental premature-aging syndrome (). The clinical features of this disease include poor growth (‘cachectic dwarfism’), neurological abnormalities and cutaneous photosensitivity. However, in contrast to xeroderma pigmentosum (XP) patients—who also exhibit increased sensitivity to ultraviolet (UV) irradiation—individuals with CS do not display elevated cancer risk. CS is divided into two complementation groups: CSA (mutation in CKN1) and CSB (mutation in ERCC6). Of the patients suffering from CS, ∼80% have mutations in the gene (). The CSB protein is composed of 1493 amino acids, and based on sequence homology, has been placed into the SWI2/SNF2 family of proteins that harbor seven helicase-like ATPase motifs (,). Although no helicase activity has been ascribed to CSB (,), the protein possesses a DNA-dependent ATPase activity (). Moreover, since purified CSB (i) promotes alterations in the DNA conformation upon binding to the double-helix and (ii) alters the arrangement of nucleosome complexes (at the expense of ATP hydrolysis), the protein has been suggested to function as a chromatin remodeling factor (). This function appears dependent on the ability of the protein to wrap and unwrap DNA molecules (). More recently, CSB was found to possess homologous DNA strand pairing activity (). Numerous studies indicate that CSB participates in transcription-coupled nucleotide excision repair (TC-NER), as well as in global genome DNA repair and general transcription (,). In particular, CSB mutant cells exhibit hypersensitivity to a number of DNA-damaging agents, including UV light (), 4-nitroquinoline-1-oxide (4-NQO) (), and -acetoxy-2-acetylaminofluorene (NA-AAF) (,). In the case of UV, the hypersensitivity has been attributed to a specific defect in the ability of CSB mutant cells to repair DNA damage within actively transcribed genes (,). A link of CSB to TC-NER is supported by the observations that (i) CSB-defective cells are unable to recover RNA synthesis following DNA-damaging agent treatment () and (ii) CSB binds RNA polymerase II when arrested at a template lesion and promotes recruitment of TFIIH, a factor involved in transcription and NER (). Results also indicate that CSB plays a more general role in DNA repair, promoting changes in the chromatin structure to facilitate damage processing, particularly within active genes (), and assists RNA polymerase I- or II-directed transcription (,). Accumulating evidence suggests a role for CSB in base excision repair (BER) (). BER is responsible for correcting most spontaneous, oxidative, or alkylation forms of DNA base or sugar damage. The observation that CSB cells, at least certain cell types, display hypersensitivity to agents that generate reactive oxygen species (ROS), such as IR, paraquat and hydrogen peroxide, supports a role for the encoded protein in the repair of oxidative lesions (). Moreover, biochemical assays using extracts from mutant cells indicate that CSB is responsible for promoting incision at 8-oxo-dG, a frequent oxidative base lesion and a marker of oxidative damage (,). In fact, global genome as well as mitochondrial DNA repair of 8-oxo-dG requires a functional CSB gene product (). CSB mutant cells also exhibit a defect in the global repair of 8-hydroxyadenine, another oxidative base modification (). Work from Flohr . () suggests that the efficient repair of oxidative base lesions through poly(ADP-ribose) polymerase 1 (PARP-1), a key DNA damage response protein, is likewise dependent on CSB. In addition, Thorslund . () identified a physical and functional interaction between CSB and PARP-1, and reported that poly(ADP)-ribosylation of CSB inhibits its DNA-dependent ATPase activity. Spivak . () have shown that reactivation of a plasmid containing a thymine glycol is defective in CSB mutant cells, further supporting a role for this protein in oxidative DNA damage repair. Although evidence exists for a function of CSB specifically in TC-BER, such a role has recently become unclear (). It has been postulated that a defect in gene-specific and/or global genome repair of endogenous DNA damage gives rise to the premature-aging symptoms and increased neurological deficiencies associated with CS patients (). In mammalian cells, APE1 is the major apurinic/apyrimidinic (AP) endonuclease, operating to initiate repair of mutagenic and cytotoxic AP sites by incising the DNA backbone immediately adjacent to the lesion (). AP sites are formed by spontaneous base loss (at a frequency of ∼10 000 events per mammalian genome per day), as well as by increased base release due to chemical modification (e.g. alkylation or oxidation) or through the action of DNA repair glycosylases that excise specific base damages. Following APE1 incision, subsequent participants of the BER pathway remove the 5′-terminal abasic fragment, replace the damaged nucleotide and seal the final nick to restore genome integrity. We describe herein data indicating a role for CSB in the repair of BER substrates/intermediates, and a specific functional interaction of CSB with the predominant AP site repair enzyme, APE1. The DiscoverLight Protein Array Kit (Pierce, Rockford, IL, USA) was used to screen for interactions between purified CSB and select proteins. As described elsewhere (,), various protein amounts were spotted on the membranes, which were then dried and blocked in PBS-T (1 × PBS, 0.1% Tween 20) containing 5% bovine serum albumin (BSA) for 1 h at room temperature. The membranes were then incubated with CSB (10 ng/ml) for 1 h at room temperature. A duplicate membrane was incubated with blocking buffer alone as a control for possible cross-reactivity with the CSB antibody. After washing with PBS-T, the membranes were probed with monoclonal mouse anti-CSB antibody (1 : 2000, kindly provided by Dr Jean Marc-Egly, Institut de Génétique et de Biologie Moléculaire et Cellulaire) overnight. After washing, the membranes were incubated with secondary antibody (1 : 10 000, anti-mouse IgG-horseradish peroxidase (HRP); Vector, Burlingame, CA, USA) for 1 h at room temperature, washed again and then developed with the SuperSignal West Pico chemiluminescent kit (Pierce). Full-length, recombinant CSB and APE1 proteins were expressed and purified as previously detailed (,). Enzyme-linked immunosorbent assays (ELISAs) were performed essentially as described (). Briefly, in the direct ELISAs, either BSA or APE1 (30 fmol) was coated onto a 96-well plate and then incubated with increasing amounts of CSB for 2 h at 37°C. After several washes, bound CSB was detected with anti-CSB antibody, followed by HRP-conjugated secondary antibody. For the indirect ELISA, either CSB (Santa Cruz Biotechnology, Santa Cruz, CA, USA) or APE1 (Trevigen, Gaithersburg, MD, USA) polyclonal antibody was coated overnight at 4°C on a 96-well plate, and then incubated with either CSB (6 fmol = 1 ng) or APE1 (30 fmol = 1 ng) protein, respectively. After incubation with increasing amounts of the partner protein (i.e. APE1 or CSB) for 2 h at 37°C, bound protein was detected with either CSB or APE1 (Trevigen) monoclonal antibodies, followed by the secondary antibody steps described above. Where indicated, ethidium bromide (Invitrogen, Carlsbad, CA, USA) or DNase I (Sigma–Aldrich, St Louis, MO, USA) was added at 10 µg/ml or 5 µg/ml, respectively. The protein interaction was detected using OPD substrate (Sigma–Aldrich). Reactions were terminated after 3 min incubation with 3 N HSO and plates were read at 490 nm using a Bio-Rad Benchmark Plus microplate spectrophotometer and Microplate Manager 5.2 software (Bio-Rad Laboratories, Hercules, CA, USA). ECFP-CSB-CS1AN cells, which were created as described (), were lysed in radioimmunoprecipitation assay (RIPA) buffer [50 mM Tris–HCl (pH 7.4), 1% NP-40, 0.25% Na-deoxycholate, 150 mM NaCl, 1 mM EDTA, 1 mM phenylmethylsulfonyl fluoride, 1 mM NaVO, 1 mM NaF, 10 U/ml DNase I and 1 tablet Complete protease inhibitor cocktail (Roche, Basel, Switzerland) per 50 ml] by sonication. After cell disruption, the suspension was centrifuged at 14 000  for 15 min. The supernatant represented the whole cell extract, and the protein concentration was determined using the Bio-Rad Protein Assay (Bio-Rad Laboratories). For immunoprecipitation, ECFP–CSB whole cell extracts were pre-cleared with Protein G-Agarose beads (Invitrogen). The pre-cleared extracts (4 mg each) were then immunoprecipitated with either negative control rabbit IgG antibody (Santa Cruz Biotechnology), living colors full-length A.v. (i.e. anti-ECFP; 1 : 100) polyclonal rabbit antibody (BD biosciences, San Jose, CA, USA), or mouse monoclonal APE1 antibody (Novus, Littleton, CO; 1 : 50) for overnight at 4°C. Samples were next incubated with Protein G-Agarose beads (30 μl) at 4°C for 1 h, followed by multiple washes. Bound proteins were eluted by boiling in SDS sample buffer and were analyzed by SDS-PAGE and western blotting using mouse anti-CSB (1 : 1000; Dr Egly), or mouse anti-APE1 (1 : 1000; Novus) antibodies, followed by chemiluminescent analysis (Pierce). The AP site-containing (i.e. the tetrahydrofuran, F, analog) 42-mer oligonucleotides used were: CCGCTGAATTGCACCCTCGACTAGGTCGATGATCCTAAGCA (42F-11), TGCTTAGGATCATCGACCTAGGTCGAGGGTGCAATTCAGCGG (42F comp) and TGCTTAGGATCATCGAGGATCGAGCTCGGTGCAATTCAGCGG (42F bubble comp). Following the generation of normal (42F-11:42 comp) or bubble (42F-11:42F bubble comp) P-end labeled duplex substrates, incision assays were performed at 37°C for 10 min with APE1 (10 or 30 fmol) and analyzed on denaturing polyacrylamide gels as previously described (). Reactions consisted of 50 mM HEPES-KOH, pH 7.4, 50 mM KCl, 5% glycerol, 10 mM MgCl, 100 µg/ml BSA, 0.05% Triton X-100 and 200 fmol of DNA. Where indicated, increasing amounts of CSB were added. endonuclease IV and truncated Δ29APE1 were purified as described (). Whole cell extract total AP endonuclease activity was determined for CSB-V and CSB-WT cells according to previously published methods (). Immunodepleted CSB supernatant fractions were obtained according to the procedure of (). In brief, 15 µg of recombinant HA/His-tagged CSB was mixed with protein A magnetic beads (New England Biolabs, Ipswich, MA, USA) that had been pre-incubated with rabbit polyclonal anti-HA antibody (Santa Cruz Biotechnology). After an overnight incubation, the supernatant was separated from the beads by centrifugation, and kept as the immunodepleted fraction (ID CSB). Beads were washed and CSB was eluted with 0.5 mg/ml of HA peptide in buffer A () containing 0.5 M KCl to generate the immunoprecipitated (or immunopurified) CSB protein (IP CSB). The number of AP sites in isolated chromosomal DNA was determined using an aldehyde reactive probe reagent (′-aminooxymethyl-carbonyl-hydrazino--biotin) and the DNA damage quantification kit of Dojindo Molecular Technology (Gaithersburg, MD, USA) as described (). CS1AN.S3.G2 SV40-transformed human skin fibroblast cell lines (,)—stably transfected with either the mammalian expression vector pcDNA3.1 (CSB-V), pcDNA3.1 containing the wild-type human gene (CSB-WT), or pcDNA3.1 containing the E646Q ATPase domain II mutant (CSB-E646Q) ()—were cultured in Minimal Essential Medium plus 15% fetal bovine serum, antibiotics and 400 µg/ml of geneticin (Invitrogen). To investigate methyl methanesulfonate (MMS) and 5-hydroxymethyl-2′-deoxyuridine (HmdU) sensitivity, colony forming assays were performed essentially as described (). Briefly, 5 × 10 cells were plated onto a 60-mm dish, incubated for 16 h, and then treated with different concentrations of MMS for 1 h. For HmdU treatment, cells were plated in complete medium at a density of 5 × 10 cells per 35-mm dish. After overnight incubation, HmdU was added, and the cells were incubated for an additional 24 h to permit incorporation. After agent treatment, cells were re-plated in complete media at a density of 1 × 10 per 60-mm dish. Colonies were stained and counted after 10 days of growth. CSB-V and CSB-WT cells were grown in the presence of [C]thymidine (0.02 µCi/ml, Amersham or GE healthcare) for 3 days to label chromosomal DNA. Cells were then washed with PBS prior to a 1 h treatment with 1 mM MMS. Fresh media was subsequently added to allow cells to recover prior to measuring RNA synthesis at the indicated times. To measure transcription, cells were pulse-labeled with 5 µCi/ml [H]uridine for 1 h at 37°C, washed twice with PBS, and lysed in 10 mM Tris (pH 8.0), 150 mM NaCl, 1 mM EDTA buffer containing 0.5% SDS and 100 µg/ml proteinase K for 1 h at 37°C. Trichloroacetic acid (TCA) (20%) was added to the cell lysate and the samples were spotted onto glass fiber discs (Whatman, Maidstone, UK). The filters were sequentially washed in 5% TCA, 70% ethanol and acetone. The TCA-precipitable radioactivity was scintillation-counted. Evidence suggests a role for CSB in facilitating global genome, and possibly transcription-coupled BER (). As a means of elucidating the molecular involvement of CSB in BER, we examined for physical interactions of CSB with proteins operating in this and related DNA repair pathways. Specifically, initial studies employed a dot blot technique, where select proteins were (i) spotted and fixed to a capture membrane, (ii) incubated with purified CSB in solution and (iii) probed for binding of CSB using CSB-specific antibodies. Potential interactors identified using this technique were the strand break sensor protein PARP-1 (data not shown), the structure-specific endonuclease FEN1, the tyrosine kinase c-Abl and the major human abasic endonuclease APE1 (Supplementary A: See online supplementary material for a color version of this figure). No significant interaction was seen with the tumor suppressor p53, the replication processivity factor PCNA, the end-joining binding complex Ku70/80, the Werner helicase, the protein defective in Nijmegen Breakage Syndrome NBS1, the telomere repeat binding protein TRF1, the recombination protein Rad51 and the single-stranded DNA binding protein RPA (Supplemental A and data not shown). Detailed studies describing the interaction of CSB with PARP-1 have been reported elsewhere (), and experiments with other putative binding partners (e.g. c-Abl and FEN1) are ongoing. To interrogate the CSB–APE1 interaction further, we performed both direct and indirect ELISAs. Direct ELISAs, where APE1 was bound to the microtiter dish surface, supported that CSB physically interacts with APE1, as revealed by a concentration-dependent increase in the CSB protein signal that is not observed with the negative control protein BSA (Supplementary B: See online Supplementary Data for a color version of this figure).Reciprocal experiments, however, where CSB was fixed to the surface, did not reveal an interaction (data not shown). Thus, as an alternative method, an indirect ELISA approach, where either an APE1- (A) or CSB-specific (B) antibody was bound to the dish first and then incubated with the appropriate purified protein, was employed. In these studies, a reciprocal interaction was observed (albeit weaker when using APE1 as the binding probe; B), and the increased signal in both cases was generally unaffected by the presence of ethidium bromide (EtBr) or DNase I (A and B). The APE1 and CSB antibodies also did not cross-react non-specifically in the indirect ELISA experiments. These findings indicate that the APE1–CSB binding is the product of a direct, physical association and is independent of DNA. Analysis of the indirect ELISA experiments revealed an apparent binding affinity (i.e. the concentration at which approximately half maximal binding was observed) for CSB and APE1 of ≥4.5 (when APE1 was coated) to ≥20 nM (when CSB was coated). We next examined whether APE1 and CSB existed in a common protein complex in human cell extracts using immunoprecipitation techniques. CSB mutant human cells (i.e. CS1AN.S3.G2) transfected with the pECFP–CSB expression vector were employed [ECFP-CSB-CS1AN; ()], largely because antibodies directed against human CSB worked inconsistently in our hands in immunoprecipitation experiments. As shown in C, immunoprecipitation with antibodies specific for either ECFP or APE1 pulled down both CSB and APE1 simultaneously from the ECFP-CSB-complemented CS1AN extracts. In contrast, neither the IgG antibody alone (negative control) nor the anti-ECFP antibody used against extracts containing only the ECFP fusion portion (vector control) precipitated either repair protein (C). Semi-quantitative densitometry analysis of the western blot signals indicated that, depending on the experiment, between roughly 5% and 50% of either APE1 or CSB was co-immunoprecipitated with the other protein. The reason for the variability is presently unknown, but may reflect differences in cell culture conditions, cell cycle status or experimental handling. As APE1 and CSB were found to physically interact and co-immunoprecipitate from human cell extracts (Supplementary and ), we next explored whether this association had a functional effect on either APE1 or CSB enzymatic activities. Using a 42F normal (i.e. fully base-paired) duplex or a 42F 11-nt bubble substrate [where F is an abasic site analog ()], we measured the effect of CSB on APE1 AP site incision activity. As shown in A, APE1 cleavage of normal double-stranded AP–DNA was stimulated significantly in a CSB concentration-dependent manner, with roughly a 2-fold activation being observed at a molar ratio of APE1 to CSB of 1 : 3, and a 4-fold activation being detected at a 1 : 10 APE1 to CSB ratio. As expected, CSB itself exhibited no AP endonuclease activity. Moreover, no activation was observed with the CSB storage buffer (A), suggesting that the stimulation was dependent on the CSB protein itself. With the 11-nt AP site-containing bubble substrate, which mimics a DNA transcription intermediate (), CSB protein was again found to stimulate APE1 endonuclease activity in a concentration-dependent manner (B). Significantly, CSB activation at either an equal molar ratio (2-fold) or at a higher ratio (up to 6-fold) was more profound with the bubble substrate than with the fully paired AP duplex (up to 4-fold). This observation suggests some DNA specificity in the activation, and is consistent with the higher affinity of CSB for transcription bubble structures than for normal double-stranded DNA ()—an observation that supports that the stimulatory effect on APE1 is CSB-dependent. As above, CSB did not itself display AP endonuclease activity on the bubble substrate. Notably, addition of ATP did not affect the CSB stimulation of APE1 (C). Last, APE1 had no effect on CSB ATPase activity in the presence of a 90-mer (35 nt)-bubble substrate (data not shown) or the 42F()-bubble substrate (Supplementary : See online Supplementary Data for a color version of this figure). To determine whether the activation of APE1 was indeed dependent on the CSB protein, we performed incision assays using immunodepleted and immunoprecipitated CSB protein fractions (see Materials and Methods section). As revealed by silver staining, immunodepletion of CSB with anti-HA antibodies reduced CSB protein levels in the recovered supernatant by ∼90% (data not shown). Thus, as expected, and supportive of the stimulation being CSB-dependent, the immunodepleted supernatant (ID CSB) had a ∼10-fold reduced ability to stimulate APE1 activity; that is, roughly equal stimulation was observed at a 10-fold higher amount of the immunodepleted CSB extract, relative to the purified CSB protein or the immunoprecipitated (i.e. immunopurified, IP CSB) CSB protein (A). As another means of evaluating the specificity of the CSB stimulation, we performed abasic site incision assays with purified bacterial AP endonuclease IV. Endonuclease IV is an protein with no sequence or structural homology to the human APE1 protein, although both enzymes exhibit many of the same biochemical repair functions (). As shown in B, CSB had no effect on endonuclease IV incision activity of the 11-nt bubble substrate, even when present at a molar ratio of 100:1 (i.e. at 10 fmol CSB); some inhibition was observed at the higher molar ratios. This result indicates that the AP endonuclease activation is specific to the human protein combination (i.e. CSB and APE1). As an aside, the findings here indicate that the ability to incise at AP sites in transcription bubble configurations is a general property of abasic endonucleases, both (unrelated) prokaryotic and eukaryotic. We next examined whether CSB could stimulate the endonuclease activity of a truncated APE1 protein lacking the first 29 amino acid residues (i.e. Δ29APE1). As shown in C, purified CSB did not activate Δ29APE1 incision activity of the 11 nt AP site-containing 42-mer bubble substrate, even at a molar ratio of 20 : 1. This result indicates that CSB physically interacts with the unique, unstructured N-terminal (REF-1) domain of APE1 () and that the protein contact is critical for the observed activation. Employing a previously established electrophoretic mobility shift assay, which omits the divalent cation Mg essential for APE1 catalysis (), CSB was found not to obviously promote APE1 binding to fully paired or bubble-containing 42-mer AP–DNAs (Supplementary : See online Supplementary Data for a color version of this figure). This finding argues that the endonuclease stimulation is mediated through a direct physical association with the APE1 protein (perhaps by promoting a favorable conformational change in APE1 through the N-terminal interaction) and/or the introduction of a topological alteration in the DNA substrate that permits a more efficient incision reaction. Interestingly, APE1 induces a kink in DNA to facilitate both recognition and strand cleavage of AP sites (). Finally, some apparent binding of DNA was observed by CSB, particularly at the higher protein concentrations (Supplementary ). As a means of evaluating the function of CSB in AP site repair, we measured the steady-state level of abasic lesions in total chromosomal DNA isolated from both CSB-V and CSB-WT fibroblasts using an established aldehyde reactive probe method (). These studies found that both human cell lines maintain similar AP site levels (A). Consistent with this, both CSB-WT and CSB-V whole cell extracts exhibited comparable total AP site incision capacities on fully paired 34-mer duplex substrates (B and C). In addition, as with the whole cell extracts, we did not detect any difference in incision efficiency of normal duplex or bubble substrate AP site-containing DNAs using nuclear extracts prepared from CSB-WT or CSB-V cells (data not shown). These results suggest that CSB does not obviously modulate global genome repair of abasic lesions. However, the data does not exclude a more specialized role for the CSB–APE1 interaction in the repair of a subset of APE1 substrates, perhaps in regions of complex DNA structure. Significantly, addition of recombinant CSB protein to whole cell extracts prepared from CSB-V cells resulted in a concentration-dependent stimulation of AP site incision efficiency (D), supportive of the idea that situations of high-localized concentrations of APE1 and CSB may indeed foster AP site repair. We note that semi-quantitative western blot analysis using CSB-WT whole cell extracts indicated an overall ratio of APE1:CSB of ≥100:1 (data not shown). As an additional means of evaluating the cellular contribution of CSB to the repair of BER substrates/intermediates, we determined the sensitivity of the SV40-transformed CSB mutant fibroblast cell line CS1AN.S3.G2 complemented with either a control vector (CSB-V) or a recombinant plasmid expressing wild-type human CSB (CSB-WT) () to the monofunctional alkylating agent MMS and the thymidine analog HmdU. For MMS, the most critical biological lesion is presumed to be the -methylation base products, which frequently give rise to AP sites via enhanced hydrolysis of the -glycosylic bond or DNA glycosylase-mediated base release (). HmdU is an oxidative base product formed by attack of intracellular ROS as well as by way of deamination of HmdC, another oxidative base lesion. Repair of HmdU in chromosomal DNA proceeds through a SMUG1-initiated BER response (). Prior studies have established that exposure to exogenous HmdU in the culture medium induces cell killing in a manner that is dependent on the formation of BER intermediates in DNA, and not the base itself (). As shown in , CSB-V cells exhibit an ∼3-fold increase in sensitivity to both MMS (panel A, 0.48 mM) and HmdU (panel B, 5.4 μM), as determined by the LD doses relative to the CSB-WT control (MMS, 1.3 mM; HmdU, 17.1 μM). These findings provide novel evidence for a direct role of CSB in the repair of BER substrates/intermediates, possibly AP sites or single-strand breaks (SSBs), which are common to the two agents. As seen previously (), we did not detect an increased sensitivity of the CSB-V cells (studied here) to hydrogen peroxide (data not shown). As ATP was not critical to the APE1 activation (C), we determined the contribution of the ATPase function of CSB to MMS resistance. Our studies reveal that CSB-deficient CS1AN cells expressing an ATPase domain II mutant CSB protein (i.e. CSB-E646Q) display intermediate sensitivity to MMS challenges (A). This finding indicates both a structural and an enzymatic role for CSB in the MMS response. We report herein that CSB and the major human AP endonuclease, APE1, directly interact, as well as exist in a common protein complex in human cells (Supplementary and ). As seen upon UV exposure (), the composition of the CSB protein complex likely changes depending on the cellular environment, perhaps explaining the variability seen in the co-IP studies reported within (see Results section). Investigations to further analyze the nature of the CSB complex during different cell cycle stages or upon challenges with disparate DNA-damaging agents (e.g. oxidizing or alkylating compounds) are warranted. We also found that CSB stimulates APE1 AP site incision activity on abasic site-containing DNA substrates—in an ATP-independent, species-specific manner (no activation of endonuclease IV was observed by CSB)—likely via a direct interaction with the N-terminal ‘REF-1 region’ of APE1. This stimulation was more pronounced on a transcription-mimic bubble structure in comparison with a fully base-paired, ‘classical’ abasic site BER substrate (). For instance, we detected a 2-fold enhanced incision activity of APE1 on the bubble duplex at a 1 : 1 CSB:APE1 molar ratio [keep in mind that CSB likely operates as a functional dimer ()], whereas no stimulation was observed at this ratio on the fully paired double-stranded AP-DNA. Thus, in addition to coordination with DNA glycosylases in endogenous base damage repair and PARP-1 in DNA damage responses (see Introduction section), the interaction with APE1 suggests a more general function of CSB in modulating BER processes. Despite the strong biochemical stimulation noted above, we found no clear evidence for a role of CSB in the global genome repair of endogenous, natural (hydrolytic) abasic sites (). However, in light of the more pronounced activation seen with the bubble substrate, we favor a model whereby the CSB–APE1 interaction is most critical to regions of the genome where complex DNA structures are formed, such as during transcription or replication, or at sites of recombination (R-DNA or telomeres), where the local, relative concentrations of these proteins may also be higher. Moreover, a more specialized role for this interaction in the repair of a specific subset of APE1-targeted damages could exist. For instance, CSB may promote excision of oxidized AP sites or certain base modifications (e.g. 5,6-dihydro-2′-deoxyuridine, 5,6-dihydrothymidine, 5-hydroxy-2′-deoxyuridine, α-2′-deoxyadenosine and α-thymidine adducts), which have all been shown to be substrates for APE1 repair activity (,). We note that we did not see any obvious activation of the APE1 3′ to 5′ exonuclease activity on a 3′-recessed primer-template DNA duplex (unpublished data), supporting a specificity to the CSB–APE1 interaction. While the idea of TCR of BER substrates is controversial, it is possible that CSB facilitates APE1 repair within actively transcribed genes. Interestingly, our preliminary data indicate a delayed RNA synthesis recovery in CSB mutant cells following MMS exposure (Supplementary : See online Supplementary Data for a color version of this figure), suggesting that certain BER substrates/intermediates may indeed be handled in a gene-specific or potentially TCR-specific manner. Prior reports in fact indicate that hydrolytic or oxidized abasic lesions create a pause or arrest site for elongating RNA polymerases (), and a stalled RNA polymerase is critical for the initiation of TCR and the recruitment of CSB (,). SSBs are also blocks to RNA polymerase progression (), whereas alkylation base damage does not appear to be corrected in a TCR mechanism (). Innovative methodologies to accurately and specifically evaluate genome (region)-specific repair and TCR of BER lesions are needed. To our knowledge, we provide the first evidence that CSB mutant cells are defective in the repair of MMS-induced DNA damage, as well as DNA products generated after HmdU incorporation. Since both of these agents induce cell killing through the production of BER substrates/intermediates, the colony survival assays herein () provide compelling evidence that CSB mutant cells are defective in the efficient or productive removal of cytotoxic BER products, likely AP sites and/or DNA SSBs, which are common to these two agents. Consistent with this interpretation, prior studies have documented a clear and reproducible hypersensitivity of cells defective in a central BER participant to MMS and HmdU challenges. In particular, APE1, POLβ, XRCC1 and PARP1 mutant cells exhibit extreme sensitivity to both MMS () and HmdU [() and unpublished data]. Thus, combined with the reduced base damage removal and the increased sensitivity to oxidizing agents displayed by CSB mutant cells (see Introduction section), evidence is mounting that supports a role for CSB in facilitating a BER-related response. The lack of an effect of ATP on the CSB-dependent activation of APE1, and the observation that the ATPase mutant protein (E646Q) only partially corrects the MMS sensitivity of CSB mutant cells (A), suggests that this SNF2/SWI2 family member can function in BER in a chromatin remodeling-independent manner. This conclusion is in line with previous findings that uncovered normal hydrogen peroxide and γ irradiation sensitivity, as well as 8-oxo-dG incision activity, for CS1AN.S3.G2 cells transfected with the E646Q mutant (,). These results imply that CSB modulates BER primarily through protein complex associations (direct or indirect), or via a more subtle (ATP-independent) modification of the DNA structure. Notably, a functional ATPase activity, while necessary for chromatin remodeling, is not required for the induction of topological change in naked DNA (). Thus, a story is emerging that suggests a distinct role for CSB in facilitating BER in comparison with NER, where the ATPase function of the protein (and energy) is more critical to the repair of UV-induced DNA damage (,). With that said, the partial complementation of E646Q (A) does indicate a role for the ATPase function (and presumably the remodeling activity) of CSB in the MMS response as well. Further studies to more thoroughly define the precise biochemical contributions of CSB to the different DNA damage-response pathways (i.e. the contribution of protein–protein interactions, the ATP-dependent remodeling activity, or the ability to manipulate naked DNA) are a priority. In summary, we report a physical and functional interaction between CSB and APE1, where the CSB-dependent stimulatory effect on APE1 incision activity, particularly on a bubble structure, is quite pronounced relative to other APE1 interactions (). Moreover, we demonstrate that CSB mutant cells exhibit a profound hypersensitivity to DNA-damaging agents that create BER-type substrates/intermediates, namely MMS and HmdU, and display an impaired RNA synthesis recovery following MMS exposure. These findings provide further evidence for a role of CSB in the processing of BER-specific DNA substrates, potentially in regions of the genome that take on complex structures. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Eukaryotic RNA polymerase II (RNA Pol II) transcripts are matured through a highly coordinated program of processing steps prior to export from the nucleus to the cytoplasm where they are translated into protein by the ribosome (). These pre-mRNA processing events include 5′ end modification by 7-methyl-guanosine cap addition and binding of the nuclear cap binding complex, intron removal by the spliceosome, 3′ end cleavage and poly-adenosine tail addition, transcript-specific modifications such as adenosine deamination and binding of specific protein factors to regulate and promote mature mRNA export from the nucleus. Coordination of these events involves interaction between the machineries involved in each process. For example, the RNA Pol II transcription complex communicates and interacts extensively with the 5′ end capping, pre-mRNA splicing and 3′ end processing machineries (). While native pre-mRNAs contain multiple, often extremely large introns, pre-mRNA splicing reactions are carried out using synthetic pre-mRNA fragments containing a single, efficiently spliced intron of a size compatible with acrylamide gel electrophoresis analysis. Although the core pre-mRNA processing machinery will likely be very similar between different transcripts as well as for the multiple introns contained within a single transcript, the bulk pre-mRNA processing machinery purified from its native context is likely to contain a more comprehensive sample of the polypeptides required for or participating in the splicing of pre-mRNA in vertebrate cells. Several groups have purified and characterized spliceosomes formed on model vertebrate pre-mRNAs () and shown that they contain a remarkably large number of associated polypeptides. Nevertheless, as these synthetic precursors have generally been modified from their natural state by internal deletions within the intron and truncations of the exons, the pattern of associated proteins is inevitably less complex than on the full-length, generally multi-intronic precursors that exist . In addition, the spliceosomes purified from reactions were assembled on pre-mRNAs derived from either the adenovirus major late or β-globin loci. Thus, it is likely that there exist a number of factors that are required for or participate in pre-mRNA processing , yet are not present in previously purified splicing complexes because they are specific to one or more of the thousands of other pre-mRNAs present in metazoan cells. Finally, the pathway by which pre-mRNA processing complexes are assembled using salt-extracted nuclear fractions most likely bypasses many interactions relevant to this process . Thus spliceosomes purified following assembly are expected to contain additional components that reflect the native pathway, but are not required to effect model intron removal . Additionally, factors that assist in inter-spliceosome interactions in multi-intron substrates will be absent from mono-spliceosome purifications, and should be present in -purified complexes. Consistent with this view, other investigators have shown that endogenous pre-mRNA is processed in extremely large ribonucleoprotein particles, called supraspliceosomes () or polyspliceosomes (). Biochemical and structural analyses of these complexes have demonstrated the presence of RNA Pol II transcripts (,) and the pre-mRNA splicing machinery components (,) as well as functional interactions that mirror those in active splicing complexes assembled (). The higher order particles formed partly reflect the presence of multiple introns, an average of eight per pre-mRNA () with some transcripts possessing as many as 147 introns [Nebulin ()], that need to be faithfully removed prior to nuclear export. In , we present a schematic model of the pre-mRNA processing pathway that encompasses the concept of the supra/polyspliceosome. Whether the individual ‘spliceosome’ moieties are formed via stepwise snRNP assembly on individual introns () or via pre-formed penta-snRNPs () in vertebrates is still a matter of considerable debate, although recent chromatin immunoprecipitation experiments in human cells provide support for the penta-snRNP model (). With the goal of expanding our understanding of pre-mRNA splicing as it occurs in intact cells, we have purified the endogenous pre-mRNA processing machines from HeLa cells and from chicken DT40 pre-B cells () on a preparative scale and have defined their RNA and polypeptide compositions. We have chosen the chicken DT40 system to compare with the HeLa system for a number of reasons. First, we have shown that working with this rapidly growing cell type, which possesses high rates of homologous recombination, allows for downstream experimental flexibility in epitope-tagging of other genes (). The evolutionary distance between human and chicken will also allow us to assess the evolutionary conservation of the machinery as well as validating novel co-purifying factors. We show that these pre-mRNA processing complexes contain spliced and unspliced mRNAs, all five spliceosomal snRNAs and polypeptides involved in all aspects of pre-mRNA processing from transcription to nuclear export. Although our strategy may not be sensitive enough to identify very low abundance pre-mRNA-specific factors, it has allowed us to probe more deeply into the general pre-mRNA processing machinery present in vertebrate cells. Indeed, when combined with the data from -assembled spliceosome characterization and known splicing factors not detected in any complexes previously purified, we show there are at least 305 polypeptides involved in or present during the processing of nuclear pre-mRNA. Ten liters of HeLa cells (purchased from the National Cell Culture Center) were processed essentially as described (). Briefly, cells were washed in PBS (137 mM NaCl, 2.7 mM KCl, 10 mM NaHPO, 2 mM KHPO) and disrupted by mechanical breakage in a glass dounce (20 strokes, pestle ‘B’) in a hypotonic solution (30 mM Tris–Cl pH 7.5, 10 mM KCl, 5 mM MgCl, 10 mM 2-mercaptoethanol) at 4°C. Nuclei were pelleted at 1000 × at 4°C for 5 min through the hypotonic buffer containing 25% glycerol. The nuclei were washed three times in the hypotonic buffer containing 0.5% Triton X-100 and once with detergent-free hypotonic buffer. Nuclei were re-suspended in a low-salt buffer (LS+; 10 mM Tris–Cl pH 7.5, 100 mM KCl, 2 mM MgCl, 10 mM 2-mercaptoethanol, 0.15 mM spermine, 0.05 mM spermidine) and sonicated twice for 20 s at the maximum microtip setting. The resulting nuclear debris was pelleted at 14 000 × for 10 s, and the supernatant was layered onto a 15–45% glycerol gradient (11ml Beckman SW41) made isotonic to LS- buffer (LS buffer without polyamines) and sedimented at 40 000 × for 90 min. Fractions (420 μl) were collected from the top. Protein and nucleic acid were separated by phenol/chloroform extraction and precipitation with acetone () (protein) or ethanol (nucleic acid). Fractions corresponding to the supraspliceosomes were pooled from six velocity gradients run in parallel fashion, diluted to ∼8% glycerol with LS- buffer and incubated with 20 mg Y12 antibody which had been covalently attached to 1 g CnBr-sepharose (GE Biosciences) according to the manufacturer's instructions. After incubation with rotation for 2 h at 4°C, the sepharose matrix was washed with 200 ml LS- buffer by gravity flow in a column and supraspliceosome material was eluted with 0.2 M glycine. Protein and nucleic acids were separated by phenol/chloroform extraction as described above. Six liters of SmD3-TAP DT40 cells () were grown in Dulbecco's modified Eagle media supplemented with 5% chicken serum and 2.5% Fetalplex (Gemini Bio-Products) to a density of 7.5 × 10 cells/ml for TAP purification. Cells were harvested by centrifugation (1000 x for 5 min), washed twice with ice-cold PBS, allowed to swell in 10 ml of TM buffer (10 mM Tris–Cl pH 7.5, 3 mM MgCl) with 0.2 mM PMSF, 1 µg/ml leupeptin, and 1 µg/ml pepstatin for 10 min ice, and lysed with 25 strokes of a Dounce homogenizer at 4°C. The nuclei were pelleted and washed twice with 10 ml of TM buffer containing 0.1% NP40, re-suspended in 5 ml of low salt buffer (30 mM Tris–Cl, 125 mM KCl, 5 mM MgCl, 0.5% Triton-X100), and sonicated at the maximum output, twice for 20 s on ice with 1 min in ice between sonications. The sonicated mixture was centrifuged at 14 000 x for 1 min and the supernatant was used for TAP purification. TAP-tagged protein material for SmD3-TAP DT40 cells was affinity purified by the TAP procedure (). The TEV eluate was layered onto glycerol gradients and fractionated as described above for the human supraspliceosomes. Polyclonal antisera directed against the carboxyl-terminal 15 amino acids of KIAA0332 and NP_035897 (NCBI accession numbers) were produced by Genemed Synthesis and the IgG fraction was partially purified by ammonium sulfate precipitation at 50% saturation. Antiserum or non-immune serum was incubated for 1 h at 4°C with the sample(s) of interest prior to addition of 50 µl Protein-A agarose beads. This mixture was incubated one further hour with rotation at 4°C prior to washing with 4 × 15 ml IPP150. Proteins and nucleic acids were released from the matrix by incubation in IPP150 at 100°C for 5 min. The supernatant was collected and phenol extracted as described above to harvest, separate and precipitate the nucleic acids and proteins. Nucleic acids were transferred to Brightstar membranes (Ambion) and hybridized with snRNA probes consisting of antisense chicken snRNAs transcribed with αP-GTP using T7 RNA polymerase (U5) or SP6 RNA polymerase (U1, U2, U4, U6 snRNAs) from plasmids containing cDNA versions of the chicken snRNAs. Polypeptides were resolved in 10% polyacrylamide gels (), transferred to nitrocellulose membranes (Biorad) and blotted with antiserum as described in the text. The secondary antibody used was horseradish peroxidase-conjugated goat anti-rabbit IgG (Rockland) and the signal was detected by enhanced chemiluminescence (Perkin Elmer). Pooled supraspliceosome protein fractions were separated by polyacrylamide gel electrophoresis and stained with Coomassie Blue G-250 (). Discrete gel slices were dissected from the top of the gel lane to the bottom and all regions were subjected to trypsin digestion. Mass spectrometry and database searching was performed as previously described (). We discovered that in gently sonicated nuclei treated with low salt (), the majority of the snRNA, as judged by visual inspection of ethidium bromide stained gels (A), was engaged in very large (>80S) ribonucleoprotein complexes that closely resemble supraspliceosomes in sedimentation values and other properties (,,). These particles may also be related to the polyspliceosomes described in salt-extracted nuclei, which sedimented as complexes slightly smaller than our supra/polyspliceosome, likely reflecting salt-induced factor loss during nuclear extraction (). For purification of the human supraspliceosomes, we gently sonicated nuclei in a buffered low-salt solution and purified the material in the ∼200S region (A and B) as previously described (). This material was immunopurified using the antibody Y12 on a solid matrix. Remarkably, this treatment nearly quantitatively retained the detectable material from this region of the glycerol gradient as judged by coomassie gel staining, even after extensive washing, indicating that the majority of the nuclear contents of this size are Sm-antigen-containing complexes. The lack of polyribosomes in the rapidly sedimenting material (as judged by the absence of 5S and 5.8S rRNAs and ribosomal proteins by mass spectrometry) indicates that the nuclei we prepare are not contaminated with cytoplasm. The material purified on a preparative scale was separated by SDS-PAGE gels of two compositions to provide optimal resolution of the large number of polypeptides present (C). To demonstrate the specificity of these purifications, gradient-separated supraspliceosomes were subjected to affinity chromatography using identical beads and identical washing and elution conditions, but lacking the Y12 antibody. In Figure S1, we show that from the mock purification, there is no detectable coomassie-stained material in the resulting protein gel (panel A) and no snRNAs present (panel B). Using our recently developed CLEP tagging procedure (), we tagged the SmD3 polypeptide in chicken DT40 cells by introducing a TAP tag () at the native genomic locus. For each experiment, 6 l of SmD3-TAP-DT40 cells were harvested and processed as described for purification of supraspliceosomes from HeLa cells. Affinity chromatography was performed according to the TAP procedure () and the TEV eluate was sedimented through a glycerol gradient. The material corresponding to the supraspliceosomes was isolated; proteins and nucleic acids are shown in A and B, respectively. In Figure S1, we show the proteins (panel C) and RNA (panel D) resulting from an identical affinity purification procedure performed using extracts from untagged DT40 cells. The absence of proteins, beyond the contaminating TEV protease, and the absence of snRNAs indicates that the purification is specific and the proteins identified by mass spectrometry are likely to be supraspliceosome components. Additional confidence is provided in that there is size-selection as well as one or two steps of affinity chromatography. RNAs corresponding to the supraspliceosome fractions from human and chicken cells are shown in B and B, respectively. Identities of the RNAs were confirmed by northern blotting (data not shown). The presence of all five spliceosomal snRNAs in this material indicated that it contained a mixture of pre-mRNA splicing complexes in varying stages of assembly and activity, as both U1 and U4 snRNAs have been shown to be released from the spliceosome before the first catalytic step of the splicing reaction (). Alternatively, the presence of both U1 and U4 may reflect functional differences between our preparations and the spliceosomes assembled ; for example, it is possible that the U1 and U4 snRNAs do not completely dissociate in conjunction with catalytic activation , but are only destabilized and maintained locally. The presence of all five snRNAs in roughly equivalent amounts also lends experimental evidence to the participation of the penta-snRNP in these functional complexes (). Polyacrylamide gel lanes from the entire human and chicken supraspliceosome fraction were dissected and the material analyzed by tandem mass spectrometry (). Though a wide variety of polypeptides were identified, it is notable that we detected very little background contamination of factors known to be unrelated to gene expression. In , we categorize the identified polypeptides according to function. Remarkably, we detected 222 distinct polypeptides in the chicken supraspliceosomes, and 177 distinct polypeptides in the HeLa supraspliceosomes. These numbers are significantly higher than the number of polypeptides detected in any one of the three previously published spliceosome purifications (). By mass spectrometry, we identified nearly all of the known pre-mRNA splicing snRNP-associated polypeptides (). We were initially surprised by the apparent absence in our preparations of a subset of snRNP-associated proteins found in most or all of the previously purified spliceosomes. However, upon closer inspection, we observed that for each polypeptide not represented in our mass spectrometry results, the inability to be detected correlated with the presence of an abundant hnRNP protein of similar molecular weight. The coverage of the major spliceosomal snRNP proteins was more complete for the chicken supraspliceosomes. Indeed, the CLEP tagging and purification procedure was sensitive enough to detect the presence of two minor AT–AC spliceosome components, the U11/U12–65K and U11/U12–48K polypeptides () in the chicken fractions, whereas no AT–AC specific splicing components were detected in the human complexes. The ability to detect all of the Sm proteins, but not all of the LSM proteins may reflect the difficulty in detecting all of these proteins in splicing complexes as shown previously (,), the 5-fold abundance differences of the two classes of proteins or perhaps due to the LSM proteins leaving the spliceosome during the process of pre-mRNA splicing (). Previous spliceosome purifications did not yield polypeptides known to be involved in snRNP biogenesis. We note in that there are three of these present in the chicken supraspliceosomes (SIP1, SMNrp30 and Coilin). These factors, which are involved in the assembly of snRNPs, are contained in Cajal bodies (CBs), nuclear organelles enriched for pre-mRNA splicing factors (). We hypothesize that the CLEP tagging procedure may allow purification of a subset of snRNPs in the process of being re-targeted to the CBs. We did not detect this class of polypeptides in the HeLa supraspliceosomes. Other factors present in the chicken supraspliceosome but not those from HeLa cells were cyclophillins and chromatin remodeling proteins, which may be related to procedural differences in the purification methods. In the second half of , we compile a list of the 59 SAPs identified in one or more of the spliceosome purifications. The -assembled complexes contained pre-mRNA-interacting factors such as U2AF (,), PTB (,), and the cap binding complex proteins (), which were present in some but not all of the -assembled spliceosomes. We detected the majority of PRP19-complex (NTC) related components as well, including homologues of Prp19p (,), Syf1p (), Syf2p (), Syf3/Clf1p (,), Isy1p (), SKIP/Prp45p (,) and CDC5/Cef1p (,), which were also detected , and BCAS2/SPF27, which was only detected in supraspliceosomes assembled . Interestingly, our preparations included a number of polypeptides that are snRNP-associated in yeast but have not been identified in purified metazoan snRNPs or spliceosomes. Among these factors are putative orthologues of yeast Prp38p (), Prp39p (), Prp40p (), Aar2p () and Luc7p (). Other polypeptides exclusively contained in the purified supraspliceosomes include, NONO/p54nrb, PNN/Pinin, CWC22, which have been previously implicated in splicing (). Conversely, 13 polypeptides were exclusively associated with the purified spliceosomes, possibly reflecting specificity for the pre-mRNAs upon which these were assembled. Alternatively, the differences in composition may result from differences in the procedures. Most striking, however, is the absence of SF1/BBP from our supraspliceosome preparations. This may be due to the fact that SF1/BBP interacts very early with the pre-mRNA, remains associated for a short time and is replaced by U2 snRNP at the branchpoint sequence (). We identified a large number of DExH/D proteins in the preparations of supraspliceosome assembled . In addition to the RNA helicase-like polypeptides known to function in pre-mRNA splicing, such as DDX15/Prp43p (), DHX8/Prp22p (), DDX46/Prp5p (), DDX5/p68 and DDX17/p72 (), UAP56/Sub2p (), DDX16/Prp2p (,) and the snRNP-associated helicases U5-200K/Brr2p () and U5-100K/Prp28p (), we noted a number not previously implicated in pre-mRNA splicing and absent from the purified spliceosomes [ (‘Gg’ and ‘Hs’ to ‘N’, ‘R’ and ‘Z’)]. These include 13 additional polypeptides with sequence motifs indicative of DEAD, DEAH or Ski2p-like helicase family members, most of which were absent from spliceosomes assembled . Although we do not as yet have evidence that these proteins function in pre-mRNA splicing, the complete absence of DNA helicases in our preparations indicates a specificity (i.e. a specificity for RNA processing complexes and against chromatin components), which minimally suggests that they function in some aspect of RNA Pol II transcript processing. Polypeptides termed hnRNPs are highly abundant nuclear proteins known to interact with hnRNA. We detected virtually all of the known hnRNP proteins in both human and chicken cells, as well as other hnRNP-like proteins annotated in genome databases (). We note that in the spliceosomes purified from extracts, some hnRNPs were identified; however, perhaps owing to specific binding of some hnRNPs to the bulk hnRNA and not the single transcript used in the spliceosome assembly reactions, a greater number of these polypeptides were detected in the supraspliceosomes. Many splicing factors are rich in arginine and serine residues including long stretches of alternating dipeptides termed SR domains. These factors function at multiple steps in the pre-mRNA splicing pathway (), and were constituents of the purified splicing complexes. We detected 10 different SR family members in our purified supraspliceosomes from both chicken and human cells (), though not a complete set. One possible conclusion is that, due to the means by which these complexes were purified and analyzed, these SR proteins represent the major SR proteins functioning in these cells, and that those not detected in our preparations function in the splicing of a smaller subset of pre-mRNAs. In addition to the known snRNP-associated USA–CYP (,), we detected five additional potential proline isomerases co-purifying with spliceosomes from chicken, but interestingly, not from HeLa cells. Several of those from the chicken purification were also present in spliceosomes assembled (). As it is likely that these proteins also function in HeLa cells, their absence may represent operational differences in the ways in which the chicken and human cells were handled and the ways in which the complexes were purified and analyzed. There were 32 polypeptides identified among all of the splicing complex purifications possessing sequence homology to polypeptides believed to interact with RNA by virtue of containing RNA Recognition Motifs (RRM), double-stranded RNA binding domains (dsRBD) or other motifs implicated in RNA binding. Some were identified previously, such as the ELAV/Hu protein that binds AU-rich elements in both cytoplasmic () and nuclear RNAs (), the U2AF-related PUF60 protein (), the dsRBD-motif-containing NFAT45 and NFAT90 and RNA adenosine deaminases; these were previously shown to exist in large nuclear complexes () and believed to function in RNA Pol II transcript metabolism. Only a single predicted RNA binding protein was found in the -assembled splicing complexes but not in either the chicken or HeLa supraspliceosomes, while 22 were exclusively found in supraspliceosomes, but not in the -formed complexes. We believe this is most likely due to the use of a single pre-mRNA , while a broader spectrum of RNA binding proteins will be associated with bulk pre-mRNA. In we present a graphical representation of the 16 presumptive RNA binding proteins novel to our study and highlight the sequence motifs contained in each. Cap-binding proteins (CBC80 and CBC20) are present in both spliceosomes assembled and supraspliceosomes assembled (). In , we report the presence in our preparations of many 3′ end processing factors (CSPF, CSTF and poly-A binding proteins), a comprehensive set of proteins shown to be involved in mRNA export including the TREX complex (THO1/HPR1, THO2, THO3/TEX1, UAP56 and ALY) (), export factors such as TAP (,), GLE1 (), GLE2 () and GLC7 (), and exon junction complex constituents including Y14 and Magoh. We also note that a single component of the nonsense-mediated decay (NMD) pathway (UPF1) () was identified in the chicken material. As NMD is likely to be active only in a very small subset of pre-mRNA processing complexes (), we were surprised to observe even a single polypeptide implicated in this process. Recent data from several laboratories suggest a functional interaction between the structural proteins of the nuclear matrix and the gene expression machinery (). Consistent with this model, we detected a number of nuclear matrix proteins in our endogenously formed pre-mRNA splicing complex preparations including actin, spectrin, matrin3, numatrin, lamin B and a matrix associated protein MAP1. Although we cannot confirm the functional relevance of these associations, we note that a few structural proteins also co-purified with -assembled spliceosomes. We also note that a number of hnRNPs and other known splicing factors such as Prp19p () were initially termed nuclear matrix-associated proteins, indicating an intimate relationship between the pre-mRNA processing machinery and the nuclear matrix. Indeed, it is an attractive hypothesis that pre-mRNA and mRNA are trafficked to the nuclear pore via the nuclear matrix. A substantial number of nucleoporins (NUPs) are present in the purified supraspliceosome complexes from human cells but not in spliceosomes assembled , which may perhaps be due to our use of sonication to release complexes from the purified nuclei versus salt extraction for preparation of splicing extracts. In the chicken supraspliceosomes, we detected a smaller set of NUPs, which may be due to their release by the detergent NP40 present during purification of these complexes. NP40 was absent during the purification of the human complexes, which likely maintained the integrity of hydrophobic interactions believed to stabilize the interaction of export complexes with NUPs. A recent report from Muchardt and colleagues () has demonstrated that the SWI/SNF component Brahma/SMARCA4 (Brm) associates with the splicing apparatus and its presence favors the inclusion of alternatively spliced exons. The Prp4 kinase, which is present in both the human and chicken supraspliceosomes, has been reported to phosphorylate both Brm and the splicing factor U5–102K/hPrp6 () providing further evidence that it functions in Pol II transcript maturation. In the purified chicken supraspliceosomes, Brahma/SMARCA4, and a number of other SWI/SNF-related polypeptides were identified (), all with high degrees of confidence given the depth of the peptide identification. As our mass spectrometry data neither included structural proteins of chromatin- such as histones, nor the DNA replication machinery or other DNA binding proteins, Brahma and other polypeptides with chromatin-related functions must specifically associated with the pre-mRNA processing complexes. Our mass spectrometry peptide data revealed several novel and intriguing polypeptides in the supraspliceosome complexes (). We found chicken homologs of the yeast splicing factors Prp38p, Prp39p and Aar2p, previously unannotated in purified pre-mRNA splicing complexes. The other polypeptides of interest in the endogenous splicing complexes include the 5′ to 3′ exonuclease XRN2/Rat1p (), which has been implicated in linking transcription termination with polyadenylation. XRN2/Rat1p was found in both the human and chicken preparations, as were two uncharacterized AAA ATPases, ATAD3A and ATAD3B. The identities of 25 additional novel polypeptides are reported in . To demonstrate the authenticity and functional relevance of a novel polypeptide that co-purified with endogenous spliceosomes, we generated antiserum against ZFR () and used it to specifically immunopurify ZFR-associated components. In A, we show that the ZFR polypeptide is present in very high molecular weight complexes that co-migrate with supraspliceosomal material. In B, we show that the anti-ZFR antiserum, but not the pre-immune serum or the Protein-A beads, immunoprecipitates the U1, U2, U4, U5 and U6 snRNAs. As a positive control, we showed that antiserum directed against the known spliceosomal protein SR140, prepared and analyzed under identical conditions, also immunoprecipitated all of the snRNAs. We also tested for the presence of another pre-mRNA splicing factor, hPrp43 (DHX15), in the material immunopurified with anti-ZFR; C shows that the specific antiserum, but not the pre-immune serum or the Protein-A beads, immunoprecipitates hPrp43p/DDX15. This demonstrates that the novel spliceosome-associated factor ZFR is indeed associated with spliceosomal snRNAs and other spliceosomal proteins. In this work, we report the composition of the endogenous pre-mRNA processing machinery from human and chicken cells and provide a comparison between native supraspliceosome complexes and spliceosomes assembled on a model single-intron substrate from salt-extracted nuclear material. These supraspliceosomes have recently been shown to be functional in add-back experiments using micrococcal nuclease-treated extracts for splicing by Sperling and colleagues (), further enhancing the functional relevance of our findings. In addition to confirming the set of factors known to interact with Pol II transcripts during splicing, we discovered an extensive array of novel factors by purifying supraspliceosomes from two types of vertebrate cells. Many of these have been implicated in pre-mRNA maturation including a subset of the SWI/SNF chromatin remodeling complex proteins, recently shown to influence alternative splicing patterns and to co-purify with pre-mRNA splicing factors. The novel polypeptides discovered in the endogenous complexes will provide a rich source of new proteins to investigate, ultimately enhancing our understanding of this incredibly complex macromolecular machine. In we present a comparison of the polypeptides present in the supraspliceosomes purified in this work with those purified in previous spliceosome preparations. The core machinery (snRNPs, SAPs, SR proteins, etc.) is well represented in the material derived from all of the purification schemes. However, a great many other factors are present in all of the preparations as well, highlighting the amazing number of proteins required to remove even the single intron used in the experiments. The methods used for the purification of all of the complexes represented in were operationally distinct and the -assembled spliceosomes contained a larger number of proteins than did the spliceosome preparations formed . What is perhaps most remarkable about our results is the fact that, despite the operationally distinct purification strategies, the basal pre-mRNA processing machinery required to effect the removal of a single intron is not vastly different than that purified from complex mixtures of all of the pre-mRNAs in a vertebrate or human nucleus. The major differences in composition between the previous purifications and the one described herein involve (i) polypeptides predicted by sequence homology to interact with the pre-mRNA (ii) the depth of coverage for polypeptides involved in export and 3′ end processing and (iii) polypeptides that may require that the pre-mRNA in these complexes follow the path of RNA Pol II transcription and nuclear trafficking, such as the SWI/SNF complexes, structural proteins and NUPs. To complete the catalog of polypeptides that participate in the nuclear pre-mRNA processing pathway, we have compiled a list of factors know to function in processing of RNA Pol II transcripts, but not present in any of the five spliceosome preparations listed in . In , we outline this relatively short list of 14 factors. does not include factors implicated in yeast splicing, but for which no identifiable human or vertebrate homologue in the genomic databases. By adding together all of the polypeptides listed in , we arrive at an estimate of at least 305 for the number of proteins that co-purify with endogenous pre-mRNA splicing complexes. This is by far the largest cataloguing of factors potentially required for or participating in this process to date. Two possible classes of polypeptides may exist that are not detected in our preparations. First are those that are underrepresented because they may interact with only a small number of pre-mRNAs, such as intron- or exon-specific binding proteins. Other classes of proteins which may participate in pre-mRNA splicing but as absent from our analyses might include tissue-type developmental stage-specific factors which would not be present in our bulk supraspliceosome preparations due to the use of only two cell types. To date, such factors have generally been elucidated via genetic or molecular strategies focused on individual pre-mRNAs. However, with the introduction of the CLEP tagging technology to more cell types (), it may be possible to rapidly enumerate factors that function in regulated or alternative splicing. Although we cannot completely eliminate the possibility that there may be contaminants present in our preparations, the basic strategy we adopted is validated by the presence proteins such as Brahma that were not detected in the previously characterized -assembled spliceosomes yet have clearly been implicated in splicing through other means. Our analyses in aggregate greatly expand our knowledge of the protein factors that function both in basal and regulated splicing in vertebrate cells. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Viral vectors are efficient DNA delivery systems as they possess natural mechanisms to enter cells, to escape from endosomes and to transport their DNA into the nucleus. However, they also display important disadvantages, such as immunogenic response and safety risks when administered to patients. Non-viral carriers lack these disadvantages, but poor transfection efficiencies currently limit the usefulness of these vectors for gene therapy applications. The low gene transfer capacity of non-viral vectors is mainly due to their inability to translocate the therapeutic DNA into the cell nucleus. Indeed, it has been shown that microinjection of plasmid DNA (pDNA) in the cytoplasm of non-dividing cells resulted in <1% gene expression, while a massive gene expression occurred when the pDNA was injected in the nucleus (). In dividing cells the nuclear envelope disassembles on a regular base, which offers an opportunity for DNA to enter the nucleus (). However, the DNA that is waiting in the cytoplasm for the next cell division is sensitive to degradation by nucleases. Therefore, methods that can enhance the nuclear uptake of DNA into nuclei of both non-dividing and dividing cells are urgently needed in non-viral-based gene therapy. Nucleocytoplasmic transport proceeds through the nuclear pore complexes (NPCs) which form channels in the nuclear envelope with a diameter of ∼40 nm (,). Vertebrate NPCs have a mass of ∼125 MDa and contain 30–50 different proteins, which are called nucleoporins. Small molecules with a molecular weight up to 30 kDa can passively diffuse through the NPC. In contrast, the translocation of larger macromolecules into the nucleus occurs via an active mechanism involving nuclear transport receptors. The majority of the nuclear transport pathways are mediated by receptors of the importin family. Proteins or other cargo molecules that carry a classical nuclear localization sequence (NLS) are recognized by importin-α, which subsequently forms a complex through its importin-β-binding domain with importin-β (). NLSs can be highly diverse in nature and range from the short bipartite classical NLS to extended protein domains, as is the case for histones or ribosomal proteins (). To promote the nuclear import of DNA, NLS peptides, NLS-containing proteins and even importin-β () have been attached to DNA via several strategies: electrostatic () or covalent () binding, via protein–DNA interaction (,), via PNA clamps (,) and coupled to polymers (), lipids () or recombinant lambda phage (). Nevertheless, all these attempts to improve the transport of DNA to the nucleus through the use of NLSs or importin-β have achieved only limited success. It has recently been shown that the nuclear uptake of macromolecules can be enhanced significantly by addition of the amphipathic molecule -cyclohexane-1,2-diol (TCHD) (). The mechanism by which TCHD causes nuclear localization of macromolecules can be explained based on the inner channel properties of the nuclear pores. It is believed that these nuclear pores are filled with a hydrophobic phase through which importins, but not inert hydrophilic substrates, can dissolve. The addition of TCHD causes a temporary, non-selective gating of the pore and allows passage of molecules which would otherwise be rejected from passage. In other words, a non-selective gating of the nuclear pore channel by TCHD renders the actual translocation through the pore channel independent of nuclear transport receptors. This can be explained by the fact that TCHD causes disruption of the hydrophobic interactions between the hydrophobic phenylalanine–glycine repeats of the nucleoporines, which consequently causes collapsing of the permeability barrier of the NPCs. Importantly, the effect of TCHD is reversible and does not cause damage of the nuclear pores (). In this paper we studied whether a non-selective gating of nuclear pores by amphipathic molecules like TCHD could also be used as an alternative method to facilitate nuclear entry of plasmid DNA. Therefore, we examined the effect of TCHD (a) on the nuclear import of macromolecules and pDNA and (b) on the transfection efficiency of naked pDNA and non-viral nanoparticles, such as poly- and lipoplexes. In summary, we found that TCHD was able to make the nuclear membrane permeable for both high molecular weight dextrans and pDNA at non-toxic concentrations. Furthermore, TCHD enhanced the transfection efficiency of both naked pDNA and DOTAP:DOPE-based lipoplexes, but had no effect on the linear PEI-based polyplexes. Dulbecco's modified Eagle's medium (DMEM), -glutamine (-Gln), heat-inactivated fetal bovine serum (FBS), phosphate-buffered saline (PBS) and penicillin/streptomycin (P/S) were supplied by GibcoBRL (Merelbeke, Belgium). The secreted alkaline phosphatase (SEAP) expression plasmid was a kind gift from Prof. Dr Tavernier (Ghent University, Belgium) and 22 kDa linear polyethyleneimine (PEI) from Prof. Dr Wagner (University of Munich, Germany). The pGL3-control plasmid and luciferase assay kit were obtained from Promega (Leiden, The Netherlands). One hundred fifty-eight kilodaltons of tetramethylrhodamine isothiocyanate-labeled dextran (TRITC-dextran) and TCHD were purchased from Sigma Aldrich. 1,2-dioleoyl-3-trimethylammonium-propane (DOTAP) chloride salt, 1,2-dilinoleoyl--glycero-3-phosphoethanolamine (DOPE) and 1,2-distearoyl--glycero-3-phosphoethanolamine--[Amino(Polyethylene Glycol)2000] (DSPE-PEG) were obtained from Avanti Polar Lipids (Alabaster, AL, USA). A549 (lung carcinoma cells; ATCC number CCL-185) and Vero (African green monkey cells; ATCC number CCL-81) cells were cultured in DMEM containing 2 mM -Gln, 10% heat-inactivated FBS, 100 U/ml P/S and grown at 37°C in a humidified atmosphere containing 5% CO. Plasmid DNA was covalently labeled with Cy5 using the IT kit of Mirus Corporation (Madison, WI, USA) according to the manufacturer's recommendations. Briefly, LabelIT reagent, containing Cy5, and 100 µg DNA were mixed in 1 ml Hepes buffer (20 mM Hepes, pH 7.4) at Cy5:DNA ratio (w:w) of 0.5:1 and incubated at 37°C for 1 h. Subsequently, the labeled pDNA was separated from unattached label by precipitation in the presence of ethanol and 0.5 M NaCl and reconstituted in 20 mM Hepes buffer (pH 7.4). Microinjection studies were conducted using a Femtojet® microinjector and an Injectman® NI 2 micromanipulator (Eppendorf, Hamburg, Germany). Vero cells were chosen for these microinjection experiments as they have a well-defined nucleus and large cytoplasm. Vero cells (2.5 × 10 cells/cm) were plated onto sterile glass bottom culture dishes (MatTek Corporation, MA, USA) and allowed to adhere for 1 day. The cells were then washed with PBS and transferred into 2 ml serum-free medium supplemented with 20 mM Hepes (pH 7.4) to improve the buffering capacity of the medium during microinjection. Five microliters 158 kDa TRITC-dextran (1 mg/ml) or Cy5-labeled pDNA (1 mg/ml) was back-loaded into Femptotip II microinjection needles and cells were injected using an injection pressure of 100 psi, a backpressure of 30–50 psi and injection duration of 0.5 s. Where mentioned, the medium was replaced after microinjection by TCHD-containing serum-free medium, supplemented with 20 mM Hepes (pH 7.4). For the -scan analysis of the fluorescence after cytoplasmic microinjection of the Cy5-labeled pDNA (Cy5-pDNA), the confocal volume (∼1 fl) of a BioRad MRC 1024 CLSM (Hemel Hempstadt, UK) equipped with the ConfoCor 2 fluorescence correlation spectroscopy (FCS) setup (LSM510 ConfoCor 2, Zeiss, Göttingen, Germany) was positioned in a randomly selected site in the nucleus. The fluorescence, along the -axis at this selected XY site and perpendicular to the cell surface, was recorded with the avalanche photodiodes of the ConfoCor 2 system before and every 10 min after addition of 1% (w/v) TCHD dissolved in serum-free medium supplemented with 20 mM Hepes (pH 7.4). Polyplexes consisting of 22 kDa linear PEI were prepared as described by Fayazpour . (). Briefly, polyplexes were prepared in 20 mM Hepes pH 7.4 by adding the PEI all at once to the pDNA at a N/P ratio of 10. Subsequently, the mixture was vortexed for 10 s and the polyplexes were allowed to equilibrate for 30 min at room temperature prior to use. The final pDNA concentration in the polyplexes was 0.126 µg/µl. Liposomes composed of DOTAP:DOPE:DSPE-PEG (molar ratio 5:5:0.2) were prepared as described previously (). Briefly, appropriate amounts of lipids were dissolved in chloroform and mixed. The chloroform was subsequently removed by rotary evaporation at 37°C followed by flushing the obtained lipid film with nitrogen during 30 min at room temperature. The dried lipids were then hydrated by adding Hepes buffer till a final lipid concentration of 10.2 mM. After mixing in the presence of glass beads, liposome formation was allowed overnight at 4°C. Thereafter, the formed liposomes were extruded 11 times through two stacked 100 nm polycarbonate membrane filters (Whatman, Brentfort, UK) at room temperature using an Avanti Mini-Extruder (Avanti Polar Lipids). The extruded liposomes were subsequently mixed with pDNA in a ±charge ratio of 4 and incubated at room temperature for 30 min prior to use. The final pDNA concentration in the lipoplex dispersion was 0.126 µg/µl. The average particle size and zeta potential of the liposomes, lipoplexes and polyplexes were measured by photon correlation spectroscopy (PCS) (Autosizer 4700, Malvern, Worcestershire, UK) and particle electrophoresis (Zetasizer 2000, Malvern), respectively. The liposome, lipoplex and polyplex dispersions were diluted 40-fold in Hepes buffer before the particle size and zeta potential were measured. The average (±standard error) of the liposomes and lipoplexes was 118 ± 1 and 242 ± 6 nm, respectively and their average zeta potential equaled 26 ± 4 and 14 ± 1 mV, respectively. The diameter and zeta potential of the linear PEI polyplexes were 165 ± 4 nm en 33 ± 2 mV. The influence of TCHD on the cell viability was determined using the CellTiter-Glo® Assay (Promega) according to the manufacturer's instructions. Briefly, 2.5 × 10 cells/cm were seeded in a 96-well plate and allowed to adhere. After 24 h, cells were washed with PBS and incubated with serum-free medium containing increasing amounts of TCHD. After 1 h, the TCHD was removed and replaced by culture medium. After 48 h, the plate was incubated at room temperature for 30 min and 100 µl CellTiter-Glo® reagent was added to each well. After shaking the plate for 10 and 2 min incubation at room temperature, the luminescence was measured on a GloMax™ 96 luminometer with 1 s integration time. Cells were seeded into 24-well plates at 2.5 × 10 cells/cm and allowed to attach overnight. Subsequently, the culture medium was removed, and after two washing steps with serum-free medium, 0.4 g pDNA, polyplexes or lipoplexes (both containing 0.4 µg pDNA) were added to each well. After 2 h the pDNA or non-viral nanoparticles were removed from the cells and the cells were post-incubated for 1 h with serum-free medium containing increasing amounts of TCHD. Subsequently, this medium was replaced by culture medium and the cells were further incubated at 37°C. After 48 h both the SEAP (or luciferase) activity, as well as the total cellular protein concentration were measured. To determine the SEAP activity, 100 µl of the culture medium above the cells was taken and incubated at 65°C for 30 min. Subsequently, 100 µl dilution buffer (0.1 M glycine, 1 mM MgCl, 0.1 mM ZnCl, pH 10.4) and 15 µl 4-methylumbelliferyl phosphate (4-MUP, 5.1 µg/µl in distilled water) was added. The obtained mixtures were then incubated at 37°C and the fluorescence was measured on a Wallac Victor2 fluorescence plate reader (Perkin Elmer-Cetus Life Sciences, Boston, MA) using an excitation and emission wavelength of 360 and 449 nm, respectively. The experimental data in this report are expressed as mean ± standard deviation (SD). One way ANOVA was used to determine whether data groups differed significantly from each other. A -value <0.05 was considered statistically significant. It has been demonstrated that TCHD enhances the rate of nuclear entry of the maltose binding protein (). However, this protein has a rather low molecular weight (40 kDa) and is consequently not totally excluded from the nucleus. Therefore, we studied whether TCHD could induce nuclear entry of higher molecular weight compounds like 158 kDa dextrans and especially pDNA, since nuclear transport of therapeutic genes forms an important bottle neck in non-viral gene delivery. In a first approach we microinjected 158 kDa TRITC-dextrans in the cytoplasm of Vero cells and followed their nuclear influx in the absence and presence of TCHD by confocal laser scanning microscopy (CLSM). In the absence of TCHD, no TRITC-dextran could be detected in the nucleus, not even after 1 h of incubation (A and B). This is as expected, since it is well-known that molecules larger than ∼70 kDa cannot move passively through the NPC network (). When TRITC-dextran microinjected cells were incubated with 1% (w/v) TCHD-containing medium, a rapid nuclear localization of the TRITC-dextrans was detected (C till G). Indeed, as soon as 10 s after addition of TCHD to the cells, TRITC-dextran was already detected in the nucleus. After 10 min the TRITC-dextran fluorescence was homogeneously distributed throughout the cell. These data clearly demonstrate that TCHD opens the NPCs what results in nuclear passage of macromolecules that otherwise are excluded from the nucleus. We also co-injected TCHD (2% w/v) and TRITC-dextran (158 kDa) in the cytosol, but under these conditions we could not observe nuclear localization (data not shown). One likely explanation is that TCHD, which is an amphiphilic compound and contains a polar ethylene glycol moiety and an apolar butylene moiety, can rapidly cross cell membranes and thus becomes rapidly diluted in the surrounding medium (). Next, we tested whether TCHD can also facilitate the nuclear uptake of pDNA, since pDNAs are much larger (2–10 MDa) than 158 kDa TRITC-dextran and have dimensions in the range of the inner diameter of the channels formed by the NPCs. When Cy5-pDNA was microinjected in the cytoplasm of Vero cells, a fluorescent spot was visible at the injection site (A, position 1). After 1 h incubation with TCHD, we could not observe accumulation of the pDNA inside the nucleus by CLSM (data not shown). Importantly, during that time the fluorescent microinjection spot became more diffuse but remained visible, indicating a restricted mobility of pDNAs in the cytoplasm. This is in agreement with the observations by Lukacs . () who showed that the diffusion of pDNA in the cytoplasm may be an important rate-limiting barrier in gene delivery utilizing non-viral vectors. Hence, after 1 h only a small fraction of the microinjected pDNA will have reached the nuclear membrane. Additionally, as stated earlier the inner diameter of the NPC is in the size range of pDNA. Therefore, even in the presence of TCHD the number of pDNA molecules that enter the nucleus is probably low and beyond the detection limit of the CLSM. To monitor nuclear pDNA influx with higher sensitivity, we used a CLSM equipped with a fluorescence correlation spectroscopy (FCS) set up, which can detect as few as ∼1 fluorescent molecule in a femtoliter range confocal volume (). We performed time-dependent -scans, perpendicular to the slide surface and through position 2 (A) before and after microinjection of Cy5-pDNA in the cytoplasm at position 1 (A), hereby crossing first a part of the cytoplasm beneath the nucleus, then the nucleus and finally the cytoplasm above the nucleus (B). The black squares in C show that after cytoplasmic microinjection of the Cy5-pDNA, a fluorescence signal could be detected in the cytoplasm, but not in the nucleus. The difference in fluorescence intensity detected below and above the nucleus is most likely due to a difference in distance from the cytosolic injection site. Subsequently, 1% (w/v) TCHD was added to the cells and z-scans through position 2 were performed every 10 min. The -scan after 60 min, represented by gray circles in C, shows a clear elevated Cy5-pDNA fluorescence signal in the nucleus compared to the fluorescence profile before addition of TCHD. Subsequently, we studied the pDNA influx in the nucleus in the presence and absence of TCHD. Therefore, the confocal volume of the FCS set up was parked in the middle of the nucleus (at -value −0.048 mm). A gradual increase of Cy5-pDNA was observed when TCHD was added to the cells (D; black circles). In contrast, when no TCHD was present, no increase in fluorescence could be detected in the nucleus (D; gray squares). This demonstrates that the time-dependent increase in fluorescence after addition of TCHD is not a result of passive diffusion of small degradation products of the Cy5-pDNA into the nucleus. It has been demonstrated that the effect of 7% (w/v) TCHD on the NPC permeability is reversible and that it does not cause denaturation or leakage of nucleoporins out of the NPCs (). Nevertheless, the reversible non-selective opening of NPCs may result into an unwanted leakage of cellular biomacromolecules from the cytoplasm into the nucleus or vice versa and hence interfere with essential cellular processes. Therefore, to ascertain that TCHD does not cause cytotoxic effects, we determined the cell viability 48 h after exposure to TCHD by the CellTiter-Glo® assay (Promega). This assay assesses the cytotoxicity by quantifying the intracellular ATP levels, which is a sensitive marker of cell viability as within minutes after a loss of membrane integrity, cells lose the ability to synthesize ATP and endogenous ATPases destroy any remaining ATP. shows that incubation of Vero and A549 cells during 1 h with 1% (w/v) TCHD slightly reduced the viability of these cells. These results demonstrate that the TCHD does not cause drastic cytotoxic effects under the conditions of the nuclear uptake experiments above, i.e. incubation of the cells with 1% (w/v) TCHD for 1 h. Furthermore, we noticed a cell-dependent TCHD sensitivity. Indeed, TCHD at concentrations above 1% (w/v) significantly decreases the viability of Vero cells, whereas almost no cytotoxic effects are observed in A549 cells incubated with up to 3% (w/v) TCHD. The effect of TCHD on the transfection efficiency of naked pDNA was evaluated by incubating Vero and A549 cells with naked pDNA for 2 h. Subsequently, pDNA that was not incorporated into the cells was removed by washing, and the cells were exposed to increasing percentages (w/v) of TCHD for 1 h (). The incubation with TCHD clearly increased the gene expression. This increase reached a maximal value at a TCHD percentage of 0.5 and 1.5% in Vero (A) and A549 cells (B), respectively. At these optimal concentrations, a 3- and 66-fold increase in gene expression was observed in Vero and A549 cells, respectively. At higher percentages no further increase and even a drop in gene expression was observed. Most likely this indicates that TCHD at these concentrations affects cellular processes that are not detected by the MTT assay. Indeed, it has been shown that the sensitivity of the MTT assay depends on the mechanism causing cytotoxicity (). Between 0 and 1.5% (w/v) TCHD, a gradual increase in gene expression was observed in A549 cells (B), which may indicate that the extent of NPC opening by TCHD is concentration dependent. Whether at 1.5% (w/v) TCHD a maximal opening of the NPCs is reached is not certain, since above this concentration also cytotoxic effects can play a role. This also explains the lower effects of TCHD in Vero cells. Indeed, in these cells the optimal concentration of TCHD to increase gene transfer is 0.5%. Based on the results in A549 cells, we can deduce that at such low TCHD concentration the opening of the NPC has not reached its maximum. To exclude that the higher gene expression was due to a perforation of the cell membrane by TCHD, which would allow an increased internalization of pDNA through the plasma membrane, we tested whether TCHD causes cytoplasmic entry of 158 kDa TRITC-dextrans. shows that TRITC-dextrans did not enter the cells in the absence nor in the presence of TCHD (1% w/v). Consequently, we can state that after 1 h incubation TCHD does not cause membrane perforation. In a next step, we analyzed the effect of TCHD on the transfection efficiency of linear PEI-based polyplexes, a quite efficient non-viral pDNA carrier that induces endosomal release via the proton sponge mechanism (). Surprisingly, none of the incubation protocols with TCHD resulted in a significant increase in gene expression mediated by linear PEI-based polyplexes even not when the Vero and A549 cells were post-incubated with 2% (w/v) TCHD (Supplementary and B). These data may indicate that, in agreement with Grosse . (), linear PEI-based polyplexes are mainly released from the endosomes as intact complexes. Because of the latter and taking into account the diameter of the linear PEI-based polyplexes (∼165 nm), TCHD is not expected to be able to enhance the nuclear transport and hence transfection efficacy of such linear PEI-based polyplexes. The endosomal escape mechanism of DOTAP:DOPE-based lipoplexes is based on a different mechanism and results in the release of uncomplexed pDNA in the cytosol (). This implies that, when the free pDNA reaches the nuclear membrane, TCHD should be able to induce its nuclear translocation, similar to the naked pDNA transfections. Therefore, we analyzed the effect of TCHD on the transfection efficiency of DOTAP:DOPE-based lipoplexes. The same incubation protocols and cells were used as in the experiments above. Incubating the cells with TCHD for 1 h immediately after incubation with the lipoplexes, indeed caused an increase in transfection efficiency in both Vero and A549 cells (A and B). A maximal increase was observed when the Vero and A549 cells were incubated with 0.5 and 1.5% (w/v) TCHD, respectively. This increase is lower than observed with naked pDNA transfection, which can be attributed to cell division. Compared to naked pDNA transfection, transfection with lipoplexes introduces a higher amount of pDNA in the cytoplasm, which can translocate to the nucleus with higher probability either by entry through the pores or during cell division. Hence, the few extra copies that enter the nucleus after treating the cells with TCHD will not tremendously increase gene expression. In contrast, the amount of pDNA that reaches the cytoplasm and subsequently the nucleus is extremely low in case of naked pDNA transfection. Therefore, if TCHD can cause nuclear translocation of only one pDNA molecule, this effect is much more spectacular. italic #text S u p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

DNA microarrays are versatile tools that allow the massively parallel study of gene expression. Microarray applications rely on the specificity of target-probe binding, when cRNAs form stable duplexes with complimentary surface-attached DNA probes. The availability of the Affymetrix spike-in study () has led to a significant effort in exploring the relationship between concentration and microarray signal (). Visual examination of the signal response to concentration in the spike-in data set reveals a nonlinear relationship between transcript abundance and signal; see . A simple derivation of the microarray hybridization isotherm is possible by equating the rates of hybridization and denaturation at the surface, and leads to the equation known as the Langmuir adsorption isotherm: Equation () is based on a simple physical ‘on/off ’ model that likely deviates from the complex process of microarray hybridization—herein we refer to this as the simple Langmuir model. It has been shown that the Langmuir adsorption isotherm captures the shape of GeneChip® hybridization (,,). According to the simple Langmuir model, signal response to concentration should saturate at the same level for all probes. However, as shown in , it is apparent that amongst different probes, there is up to a 30-fold difference in saturation levels and variation in these levels was found to correlate with probe sequences (,). Fitting the signal-versus-concentration curves into the Langmuir isotherm equation revealed that the equilibrium constant, , does not exhibit strong dependence on the probe sequence as predicted by the simple Langmuir model (,). On the contrary, although the nature of the observed saturation range was not addressed experimentally, the saturation levels themselves were successfully predicted by fitting the log saturation intensities to the sum of position-dependent nearest-neighbor stacking energy parameters (,). Several hypotheses have been proposed to explain the discrepancies between the simple adsorption model and the observed signal behavior. Alternative non-Langmuir models claim that saturation intensity is, in fact, constant and discrepancies in predicted parameters arise from fitting the wrong model. For example, Vainrub and Pettitt (,) developed a mean field model of the electrostatic effects in oligonucleotide microarrays that describe a non-Langmuir binding isotherm, suggesting that there are substantial differences in hybridization thermodynamics between DNA free in solution and surface-tethered DNA. According to this model, the binding of negatively charged RNA increases the total charge on the microarray surface and significantly affects probe–target-binding kinetics, thus leading to a modified binding isotherm. Another explanation, introduced by Peterson . (), is based on the concept of probe–target non-homogeneity, i.e. probes and targets have different lengths, hence binding occurs at varying rates. The Sips model provides a generalization of the Langmuir isotherm in which the single binding energy utilized in the Langmuir version is replaced by a distribution of binding energies. Multi-step reaction models explain the discrepancies between the simple adsorption model and the observed signal behavior by suggesting that probe–target-binding is a slow, complex multi-step reaction with multiple intermediate states (). This can lead to a dependence shape similar to the Langmuir isotherm, while saturation would then occur at an intensity that is different from the maximum occupancy state (,,). We hypothesize that the difference in saturation intensity can be explained by the post-hybridization wash. It is generally assumed that the primary purpose of the wash step is to mechanically remove the remains of unbound or loosely bound RNA from the chip surface. We predict that in a typical microarray experiment, a significant portion of non-specific RNA competes with specific RNA for binding sites due to the high complexity of the background, and thus a significant fraction of duplexes are not completely zipped after a standard round of hybridization. Those duplexes are removed during the washing step at a rate dependent on the probe–target duplex binding strength. The suggested model can be tested by introducing several slight modifications to the existing hybridization protocol. Briefly, the chip is scanned prior to the stringent wash cycle, and scanning is repeated once the wash cycle is complete. If the proposed wash model is correct, we expect to observe a modest intensity variation prior to the wash and a significant sequence-dependent intensity decline following the wash cycle. In this study, we alter the standard Affymetrix fluidics protocol in order to investigate the effect of the wash cycle, target concentration and hybridization time on the final hybridization signal. Scanning the chips prior to the stringent wash cycle allows for quantification of the effect of the wash on both specific and non-specific signal components. By varying the length of the hybridization cycle, the issue of attaining probe–target binding equilibrium can be addressed. Selective labeling of spiked clones and a complex biological background mixture allow the impact of specific and non-specific binding on the total signal to be monitored. In order to elucidate the various factors that influence the kinetics and thermodynamics of target hybridization to surface immobilized probes in a microarray setting, we introduced numerous alterations to the standard Affymetrix protocol for cRNA sample preparation, fragmentation, hybridization, washing, staining and scanning. To examine the contributions of both the specific and non-specific signal components, we devised an approach to selectively label the targets and complex background while minimizing other differences in cRNA preparation. We exploited the fact that the pOTB7 vector, in which each of the cDNAs employed in this study had been directionally cloned, has a T7 promoter sequence located 3′ to the cDNA coding region. This facilitated the use of either Affymetrix's GeneChip® IVT Labeling Kit or Ambion's MEGAscript® T7 Kit to generate antisense cRNAs that differ from one another chiefly in the presence or absence of a biotin label on the transcript. To study the effects of the wash cycle, we modified the standard fluidics script so as to obtain two measures of hybridization intensities, one prior to and one subsequent to the stringent wash. We further probed the issue of saturation intensity by varying the concentration of the spike-in targets. Following sample preparation, as described in the Materials and Methods section, the fragmented target RNA cocktail was hybridized and further processed according to either standard protocol or an altered washing protocol; see . To assess the effect of signal changes due to the altered protocol, an identical set of control mixtures was hybridized using the standard and altered hybridization, washing and staining protocols. The scatter plot of intensities for the same probes on two chips shown in reveals no significant differences between the post-stringent wash signal corresponding to the altered fluidics script and the signal obtained using the standard fluidics script. Thus, these changes in the fluidics script do not introduce any artifacts and do not significantly affect the signal. Hence, we can employ the altered fluidics script to study the effect of the washing cycle on microarray signals. In order to relate the wash effect (pre/post wash intensity ratio) to the probe sequence composition, a notion of affinity must be introduced. This measure is unknown and could either be measured by direct experimentation or predicted based on probe sequence composition. There are no data available for the measured affinities on the Affymetrix chips, however Zhang . [] suggested a method using public data sets to predict them based on sequence composition. According to the simple Langmuir model [Equation ()] and under the assumption that probes do not attain saturation at 512 pM, we can use the pre-wash intensity as a measure of affinity. To assure reproducibility, three independent sets of clones were labeled and spiked into unlabeled complex background; see for more details. In particular, consideration was given to the amount of non-specific background, the concentration of the spike-in clones and the total amount of RNA hybridized. The effect of the stringent wash is shown in . A shows the log observed intensities before and after the wash for PM and MM signals. It is clear that the intensities of specific probes signal are altered by the wash cycle and that the post-wash intensity is lower than the corresponding pre-wash intensity. B clearly demonstrates that the wash effect depends on probe affinities and reveals the dependence of the change in intensity resulting from the stringent wash on the pre-wash signal. An extremely pronounced wash effect is observed in which probes with pre-wash intensities ranging from 3000 to 22,000 (on the natural scale) are impacted by a factor of 2–16 as a result of the stringent wash. In order to further investigate the mechanism underlying the wash process, an additional round of stringent washing and scanning was performed in this experiment to validate the washing model suggested by Held . (). In the paper (), the authors propose that the drop 55 in intensity during the stringent wash step could be described by the exponential decay law. According to the suggested model, the ratio of signal before and after the second wash should then be the same as in the first round of stringent wash. From , it is apparent that the second wash does not affect the intensity to the same extent as the first wash; this behavior contradicts theoretical models proposed in the past (). Surprisingly, MM probes that are only partially bound by design are also not affected by the second wash cycle. This suggests that the population of probe–target duplexes remaining after the first stringent wash is significantly different from the probe–target duplex population that exists prior to the wash. We believe that those probe–target duplexes that survive the stringent wash are fully bound, whereas prior to the stringent wash the probe–target duplex population contains a significant fraction of partially bound complexes. Selective biotin-labeling of spiked clones and non-specific complex background RNA enabled us to explore the properties of specifically and non-specifically formed duplexes. presents a comparison of the signal behavior of specifically and non-specifically bound targets. A shows a boxplot of non-specifically bound probe intensities before and after the stringent wash. Prior to wash, the majority of non-specific probes demonstrate a noticeable signal of 7–9 on the log scale (or 130–500 on natural scale), while after the stringent wash, the interquantile range shifts to the 5–6 range on the log scale (or 30–60 on natural scale), corresponding to nearly ‘optical background’ level. B offers a superimposed scatterplot of specifically and non-specifically bound probe intensities before and after the stringent wash. Examination of this plot reveals that a small fraction of non-specific probes generates a high response, comparable to the intensity of the specific probes, in both pre- and post-wash conditions. Similarly, a fraction of probes representing specifically bound targets demonstrate a low signal, comparable to non-specifically bound or non-responding probe intensities. We explain these observations by significant cross-hybridization in the first case and presence of non-responding probes and/or probes with no complimentary match in target sequences in the second case. Overlaying before-versus-after wash intensities for specific and non-specific target signals in , we observe that the washing properties appear to be uniform across both specifically and non-specifically bound probes. This allows us to conclude that the hybridization/washing mechanism for both specific and non-specific targets is the same and the relationship between pre- and post-wash intensities is universal for all duplexes. According to the simple Langmuir theory, all probes in the pre-wash state are expected to saturate at the same level. However, reveals that probe intensities differ even in the pre-wash state. As mentioned earlier, due to differences in the hybridization kinetics of individual probes, a concentration of 512 pM may not be sufficient to saturate certain probes. To confirm this hypothesis, additional hybridization experiments were performed in which four clones were spiked-in and hybridized to two chips at a concentration of 1 and 10 nM, respectively, each in the absence of complex background; see . illustrates the effect of clone concentration on the hybridization signal. A compares pre-wash PM and MM log intensities between chips with 10 and 1 nM clone concentrations. It shows that for high-intensity probes, an increase in concentration does not affect the intensity, i.e. these probes are saturated. However, an increase in concentration results in a nearly proportional increase in the responsiveness of low intensity probes. Thus, the observed difference in pre-wash intensities can primarily be explained by the inability of some probes to saturate at a concentration of 512 pM. The microarray scanner used in our experiments has been well calibrated and the maximum observed intensity was below scanner saturation limits at all times. B compares the observed log difference in wash effects for 10 nM and 1024 pM plotted against the log pre-wash intensity at 1024 pM. This comparison reveals that the wash effect at the higher concentration is significantly lower than the effect observed at the lower concentration (Wilcoxon signed-rank test, < 0.001) and that PM and MM probes behave similarly in this context. To study the effect of hybridization time on duplex stability, the standard hybridization time (16 h) was extended by a factor of 2.5 (40 h); see . In order to control for possible experimental variation, the hybridization mixture was prepared as described in the Materials and Methods section, split into two parts and hybridized in parallel on two chips. After 16 h, the first chip was washed, stained and scanned. Following an additional 24 h period, the second chip was washed, stained and scanned. Examination of reveals that the washing effect was reduced following 40 h of hybridization for probes that were not saturated. Probes that were already near saturation levels were not altered by the wash step, while the overall intensity of the specifically bound probes for 40 h is less than that for 16 h. There is a significant difference observed in washing ratios for 16 h versus 40 h of hybridization (Wilcoxon signed-rank test, < 0.001) This study was designed to objectively validate and determine the cause of the variation in saturation levels observed for Affymetrix GeneChip® arrays. Whereas some earlier studies show a quantitative relationship between the saturated intensity signal [] and the probe sequence composition, experimental validation has not been previously described. Here we study the hybridization and washing steps of the standard Affymetrix protocol to provide a qualitative description of the microarray hybridization mechanism. This is achieved by altering the standard protocols for hybridization, washing and staining. In particular, by modifying the standard protocol we study the effect of the stringent wash, hybridization time, target concentration and the presence of complex background on the microarray hybridization signal. We establish that the major factor contributing to the discrepancy between the signal predicted by the simple Langmuir model and the intensity observed in public Affymetrix spike-in experiments is the stringent wash cycle, which results in decreased signal in accordance with probe affinity. By increasing the duration of hybridization we observe that low-affinity probes do not reach full equilibrium over the standard 16 h cycle and this partially explains the variation in saturation levels prior to the stringent wash step. By increasing the clone concentration we gain insight into the mechanism of probe–target binding kinetics. We observe a diminished washing effect at higher spike-in concentrations, an effect that cannot be dismissed as the result of normal variability. Consecutive stringent wash experiments, in which chips are subjected to two rounds of stringent washing and scanning, demonstrate the heterogeneous nature of the formed duplexes. We observe that 60–95% of specifically bound targets are washed off during the first stringent washing cycle, while a second stringent washing cycle only mildly alters the signal. In light of this observation, we can hypothesize why the observed energy of duplex formation for microarrays appears to be smaller than the one observed in solution: only a limited fraction of duplexes become fully zipped during hybridization. This would also explain the discrepancies between theoretically calculated equilibrium constants and those observed in experimental data. Our findings explain the mysterious difference in saturation intensities and provide insight into the theoretical and statistical modeling of Affymetrix microarray hybridization signals. Investigators should consider the nature of the hybridization mechanism, the heterogeneity of probe–target interactions, and the effect of the stringent wash when modeling hybridization signals. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

The discovery of naturally occurring RNA enzymes (ribozymes) (,) has postulated so-called the RNA world where RNAs could have served both genetic and catalytic roles (). This notion has motivated us to not only search more ribozymes from nature () but also artificially generate ribozymes with a wide variety of catalytic functions. Representative examples include phosphoryl transfer (,), acylation (), alkylation (,), Diels-Alder (), aldol () and redox () reactions. Among them, selection of ribozymes that utilize nucleotide triphosphates (NTPs) has been of great interest since NTPs are playing diverse but essential roles in biological systems. ATP, for example, is one of the building blocks of RNA and at the same time serves an energy or a phosphate source as well as in cofactors. Because of their potential biological significance, tremendous efforts have been made to evolve NTP-utilizing ribozymes from an RNA pool of random sequences. Indeed, it has been thus far reported that ribozymes catalyze three types of chemistry with nucleotide substrates. The first example is ribozymes that use ATP as a phosphate source. Polynucleotide kinase ribozymes (,) are capable of catalyzing 5′- or 2′-phosphoryl transfer reaction. The second example is a 5′-capping ribozyme (), which utilizes GDP (as well as GMP or GTP) to form a 5′–5′ phosphoanhydride bond that is analogous to the 5′-cap structure of mRNA. The third example is those that can elongate their own or other oligonucleotide chain by NTPs (,). Earlier ribozymes of this kind were isolated as those capable of ligating an oligonucleotide to its own 5′-end. One of these ribozymes, called Class I ligase ribozyme, was further evolved using various ingenious strategies and successfully turned into an RNA polymerase ribozyme (). Significantly, this ribozyme is able to add multiple nucleotides to the 3′-end of a substrate RNA according to the external template sequence ; thus, its 5′→3′ polymerization function is analogous to the naturally occurring counterparts. Most of earlier demonstrations described above have dealt with chemistry where the selection strategy could be rather easily devised; i.e. a selection method allowed ones to fish out desired ribozymes with only one possible function or with a dominant function competing with other potential reactions to occur. However, this does not mean that the ribozyme chemistry using NTPs or other nucleotide phosphates is limited to these reactions. For example, phosphoryl transfer to the 3′-hydroxyl or 5′-triphosphate (5′-ppp), and the 5′-nucleotidyl transfer (3′→5′ extension) could be catalyzed by RNA using ATP (or other NTPs) as a substrate. In principle, RNA sequences that catalyze one of these reactions can be selected by a conventional strategy that yielded kinase ribozymes using ATP-γS as a substrate (,,,); yet, no ribozyme with such functions has been reported. Therefore, a challenge is how ribozymes that catalyze unreported reactions are isolated and identified from a random RNA pool competing with those catalyzing other reactions reported previously. Accordingly, the selection strategy should be the most critical factor toward the success. To isolate ribozymes capable of catalyzing the above reactions, we developed a strategy involving three layers of selection constraints to enrich 5′-ppp-dependent catalytic species in a random RNA pool. Executing this selection indeed yielded a novel RNA sequence that catalyzes one of the above reactions: 3′→5′ nucleotide extension. We report here selection and characterization of this ribozyme that shows intriguing promiscuous recognition to purine nucleotides. The pool DNA (5′-GGATCGTCAT TGCATTGAGA-N70-GGTGGTATCC CCAAGGGGTA-3′; constant region for the PCR and random sequences are shown) was amplified using the 5′ primer containing the T7 promoter (5′-GGTAACACGC ATATGGGGATCGT CAGTGCATTG AGA-3′; T7 promoter sequences are underlined) and the 3′ primer (5′-TACCCCTTGG GGATACCACC-3′). Four copies of each sequence in the initial pool DNA (complexity of about 1.6 × 10) were transcribed and purified on 10% denaturing PAGE. From the second round of selection onward, the pool RNA was body-labeled with [α-P]GTP (ICN) during the transcription. Conditions for the transcription are described below. Reactions were carried out in 50 mM EPPS buffer (pH 7.5), 100 mM KCl, 100 mM MgCl and 10 mM ATP-γS. The RNA pool (4 μM for the first round and 1 μM thereafter) was placed in EPPS and KCl and was denatured at 95°C and then brought to room temperature. MgCl was added and let RNAs refold for 5 min. Reactions were initiated by adding ATP-γS (Sigma) and let proceed for 24 h at room temperature. Reaction volumes were 4.5 ml for the first round and 45 μl thereafter. RNAs were precipitated from the reaction once by isopropanol and once again by ethanol. 5′-ppp RNAs were prepared by transcribing PCR template with 2 mM each of ATP, CTP, UTP and GTP. Total 5–10 μCi of [γ-P]GTP or [α-P]GTP was added to the transcription reaction to prepare end-labeled or body-labeled RNAs, respectively. 5′-ppp RNA was treated with alkaline phosphatase (Promega) to prepare 5′-hydroxyl RNA (5′-OH RNA) according to the manufacture's manual. The substrate strand for the reaction was prepared by the action of 10–23 DNA enzyme [3′-AGCAACATCGATCGG-5′, priming sequences to the RNA substrate are underlined, ()] that was raised to the RNA transcript 5′-pppA↓AAAAG-3′ (priming sequences to the DNA enzyme are underlined and the site of cleavage is marked with an arrow). All RNAs were purified on appropriate PAGE. Usually the aimed transcript was heavily contaminated with + 1 transcript (up to 50–60% in the transcription of M4) and less significantly with + 2 or − 1 transcripts. Since 3′-heterogeneity did not affect the activity of the ribozyme these transcripts were not extensively purified unless otherwise noted. Reaction conditions were optimized to 100 mM EPPS (or 100 mM HEPES), 72 mM MgCl, 8 mM MnCl, 100 mM KCl, 10 mM ATP-γS (for other substrates, see figure legends), pH 7.5 and 1 μM RNA (20 μM RNA were used for the reaction). All reactions were carried out at room temperature for 24 h. Thiophosphate-containing RNAs were separated from the rest using either the biphasic [(-acryloylamino)phenyl]mercury polyacrylamide gel electrophoresis (APM–PAGE) () or the SAv gel shift assay with post-biotinylation with PEO-iodoacetyl biotin. Radiolabeled RNAs on the gel were visualized in a phosphoimager (Bio-Rad). Alternatively, cold RNAs were stained with SYBR Green II to be detected in the fluorescence scanner (Kodak). Software Quantity One (Bio-Rad) was used to quantify the RNAs in the gel. However, it should be addressed that this thiophosphate-containing RNA quantification was ‘relative to control reaction’ not ‘absolute quantification of the reaction’, for two reasons; () substrates ATP-γS and GTP-γS are contaminated with significant amount of ADP or GDP respectively (probably from the hydrolysis), () thiophosphorylated RNAs are not stable in terms of their terminal thiophosphate-phosphate diesters (only 50–70% of GTP-γS primed transcript that was purified from APM–PAGE survived after 24 h in the reaction buffer, data not shown). Reaction of cold M4 with [α-P]-ATP was quenched by precipitation with isopropanol. Residual [α-P]-ATP was removed from RNA by repeated precipitation with ethanol. Resulting RNA was subjected to the dephosphorylation reaction with 0.1 U/μl shrimp alkaline phosphatase (Roche) according to the manufacturer's manual. Samples were taken after 0, 10, 30 and 60 min of incubation at 37°C and were analyzed on 22% PAGE. [α-P]-ATP was treated with alkaline phosphatase to be used as a phosphate control. To make sure the phosphate release is the consequence of the enzyme action, the same set of reaction was carried out without alkaline phosphatase. Cold M4 was allowed to react with [γ-P]-ATP and was loaded on the Bio-Spin 30 column (Bio-Rad) to remove unreacted [γ-P]-ATP. Resulting RNA was digested with RNase T2 (Sigma) in 50 mM ammonium acetate at 42°C for 2.5 h. 5′-end-radiolabeled M4 was digested with RNase T2 for being used as charge standards in the TLC. The digests were spotted on DEAE-cellulose TLC plate (J.T. Baker). TLC was run at step gradient of ammonium acetate (from 0.2 to 0.25 M) in 9 M urea and 1 mM EDTA, after a brief pre-run in water. The thiophosphate-containing RNA was purified from the reaction of M4 on the APM–PAGE. Either this product RNA or starting substrate strand RNA in 0.5 μl water was mixed with 0.5 μl of 0.5 M diammonium hydrogen citrate and 1 μl saturated solution of 3-hydroxypicolinic acid (TCI) in acetonitrile:water:ethanol (50:45:5, v/v). Samples were spotted on the target before mass values were taken by MALDI-TOF (Bruker). Synthetic RNA standards (6 and 14 nt long) were used to calibrate the mass externally. In order to enrich unreported catalytic species from the RNA pool, the selection strategy was built upon three layers of constraints below (). The first constraint was to use ATP-γS as a substrate. This reagent had been used in the selection of kinase ribozymes where active RNAs were tagged with a thiophosphate transferred from ATP-γS to its own 5′-OH or internal 2′-OH (,). In theory, this reagent should tag any active species that transfer thiophosphate to any sites including the 3′-OH and 5′-ppp (if 5′-ppp-RNA were used) groups, although such ribozymes were not reported previously. The second constraint was to use a pool of 5′-ppp-RNA as a substrate for self-modification. The triphosphate at the 5′-end of RNA can offer a unique reaction center not only as an electrophile, but also as a nucleophile. At the same time, this should eliminate the appearance of a previously reported catalyst, 5′-OH kinase (), in the active populations. The third constraint was aimed at distinguishing 5′-ppp-independent species from 5′-ppp-dependent ones in the enriched pool. The positive selection was performed the same as in earlier rounds (positive selection, ). On the other hand, the counterselection involved removal of 5′-triphosphate on the enriched RNAs to expose the 5′-OH group, and then active species were selected (counterselection, ). We carried out these selections at the round 14 in parallel, giving pool 14-I and 14-II, respectively (). Since pool 14-I would contain both 5′-ppp-dependent and independent species, whereas pool 14-II would not contain 5′-ppp-dependent species, the comparison of the sequences found in these pools should allow us to identify the 5′-ppp-dependent species existing in the pool 14-I. In the sequence alignment of clones from pools 14-I and 14-II, we identified two unique sequences, C21 and C06, in the pool 14-I (A). The remaining sequences were found multiple times in both pools and classified to class I–IV. These unique clones and representative sequences of each class were then verified for the 5′-ppp dependence (B). The clones represented from classes I to IV were active independent from the presence of 5′-ppp group as expected. The activity of C21 was also independent from the presence of 5′-ppp group. Moreover, it was confirmed that periodate oxidation of the 3′-terminus of these clones did not diminish the activity, suggesting that all these clones likely self-thiophosphorylated onto the internal 2′-OH groups (data not shown). However, removal of 5′-ppp on C06 wiped out the activity, indicating that the 5′-ppp group is essential for activity (B). Thus, C06 was chosen for further studies. A potential secondary structure of C06 was predicted by RNAstructure v4.2 () to have four stems (termed P1–P4), three loops (L1–L3) and one junction loop as shown in A. Based on this structure, a series of mutations/deletions/substitutions were performed on P2, P3 and P4 domains (A). Disruption of the P3 stem by making misparings between A74–C76 and G90–U92 was detrimental to the catalytic activity, while its compensatory pairing gave a mutant M1 with 2.7-fold higher activity than C06. Therefore, the scaffold of M1 was used for further mutation studies toward miniaturization. Deletion of P4 domain in M1 gave only a minor decrease in activity (M2), ruling out a potential modification within this domain such as a hydroxy group at the 3′-end. On the other hand, substitution of L3 with UUCG tetraloop virtually eradicated the activity. Moreover, single and double deletions of the tandem guanosines at the 5′-overhang also were detrimental to the activity. These results indicate that both motifs in L3 and the tandem Gs play critical roles in the formation of catalytic core of this ribozyme. Interestingly, the deletion/substitution of P2 domain with UUCG tetraloop reduced the activity, but yet this miniaturized mutant (M3) had an activity 1.5-fold higher than that of the parental ribozyme C06. We therefore used M3 for further miniaturization. The P1 stem of M3 was further shortened to M4 consisting of a 6 bp stem, resulting in minor reduction in the activity; M4 exhibited a 0.7-fold activity of the wildtype C06 ribozyme (B). Substitution of the junction loop G68–A72 between P1 and P3 stems with UUCG tetraloop was detrimental to the activity, indicating the indispensability of the sequence and length of this region (B). On the other hand, double substitutions of P1 and P3 stems to unrelated stem sequences afforded M5 with slight increase in the activity presumably due to a better folding (C). Taken together, the parental C06 ribozyme was miniaturized to 45-nt M4 ribozyme consisting of the 5′-tandem guanosines and two stems that were linked via an indispensable 5-nt junction (B). Thus, M4 ribozyme was used to determine chemistry occurring in its active site. To characterize the M4-catalyzed reaction, we checked the 5′-terminal phosphate, the charge and the mass differences after the reaction. Prior to the series of experiments below to determine the product, we found that M4 ribozyme could be radiolabeled upon treatment with [α-P] or [γ-P]-ATP (data not shown). Moreover, because the reaction took place in the 5′-ppp-dependent manner, the radiolabeling occurred most likely at the 5′-terminus. We therefore took an advantage of this labeling method to characterize the product. Reaction product of M4 with [α-P]-ATP resulted in radiolabeled RNA as describe above. The resulting RNA was subjected to alkaline phosphatase-mediated dephosphorylation reaction. If M4 catalyzes the capping reaction, the resulting structure that contains a phosphoanhydride bond should be inert toward this enzyme action. However, the transferred [P]-radiolabel on M4 was readily removed as a phosphate in an alkaline phosphatase-dependent manner, enabling us to rule out the capping reaction from the possible chemistry (A). For a charge analysis, we used the RNase T2-mediated digestion of the product RNA. Treatment of M4 with [γ-P]-ATP followed by RNase T2-digestion of the product yielded a mixture of a [P]-labeled nucleotide and other non-labeled mononucleotides 3′-phosphate. We then analyzed this mixture on DEAE-cellulose TLC. Since the migration of nucleotides on DEAE-cellulose TLC is mainly determined by the number of negative charges (slower migration indicates more negative charges), this analysis allowed us to estimate the number of phosphates on the radiolabeled substance. As a sample for the comparison with the above product, we carried out [γ-P]-GTP–primed transcription of M4 generating 5′-p*pp-M4 (p* indicates [P]-labeled phosphate) and its RNase T2-digestion yielding p*ppGp (6 negative charges; −6). We found that the reaction product migrated slower than the control p*ppGp, indicating that it has more than 6 negative charges (B). If an ATP transfer reaction at the 5′-terminus took place and the ATP-adduct contained a 3′–5′ phosphate linkage, RNase T2-digestion should yield p*ppAp (−6). The mobility of p*ppAp should be the same or as close as p*ppGp, and thus the above observation ruled out this possibility. Likewise, RNase T2-digested product originated from [γ-P]-transfer to an internal 2′-OH should generate N(2′-p*)pNp (−5), and therefore it was also ruled out as additional evidence to the ppp-dependent activity. Taken together, we could narrow the potential chemistry down to the phosphoryl transfer (p*pppGp; −7) or 5′-nucleotidyl transfer with 2′–5′ phosphate linkage (p*ppA(2′)p(5′)Gp; −7) onto the 5′-ppp site. To distinguish these two products, we decided to conduct direct molecular mass measurement of the reaction product using MALDI-TOF. For this experiment, we used a nucleotide containing thiophosphate, e.g. ATP-γS, as a substrate, since the 5′-thiophosphate adduct could be readily purified by APM–PAGE. Our preliminary attempt using the cis-acting system showed increased molecular mass changes, and from the mass change it was likely that 5′-nucleotidyl transfer, rather than phosphoryl transfer, had taken place. However, the observed resolution was not high enough to withdraw a solid conclusion for mass changes (data not shown). To observe mass changes with a higher resolution, we performed reaction () with guanosine 5′-α-thiophosphate (GMP-αS) instead of ATP-γS, and () using a -acting M4 system in which the catalytic core and the 5′-acceptor substrate were separated (C). Based on its secondary structure (C), we synthesized a 9-nt substrate and a -acting ribozyme via dissection of L1 loop in M4 (C, 9-nt sM4 and 34-nt rM4, respectively). These two pieces of RNA would form the complex via seven base pairs and react with GMP-αS to yield the 5′-GMP-αS adduct if the expected reaction took place. The MALDI-TOF analysis of reacted sM4 in comparison with unreacted sM4 showed the average mass change of approximately 202, which conformed to the expected value of 201 for the extension of sM4 by GMP-αS at the expense of pyrophosphate (D). Recall the result of RNase T2-digestion experiment in which the M4-catalyzed self-incorporation of [γ-P]-ATP increased negative charges in the product (A). Taken all results together, we concluded that M4 ribozyme catalyzes 5′-nucleotidyl transfer, i.e. 3′→5′ mononucleotide extension, forming a 2′–5′ linkage that is resistant against RNase T2 digestion (E). During the course of the above product determination, we found that M4 ribozyme was able to incorporate GMP-αS to its own 5′-terminus. This observation extended our curiosity to search a full spectrum of substrate specificity. Structures of substrates tested in our studies are shown in A. We first examined the substrate specificity of M4 toward canonical NTPs and dNTPs. Interestingly, both ATP and GTP could be a substrate while GTP was a superior substrate to ATP (B). On the other hand, neither CTP nor UTP were viable substrates, suggesting that purine base is critical for the recognition of M4 (B). Moreover, it was determined that the M4-catalyzed 5′-nucleotidyl transfer yielded the 2′–5′ linkage. This predicted that none of dNTPs would be the substrate, and this was indeed the case (B). This observation added our additional confidence for the selective formation of the 2′–5′ linkage. The promiscuous activity toward GTP and ATP motivated us to investigate a greater variety of purine NTPs (C). We found that inosine 5′-triphosphate (ITP) was a substrate as good as GTP. Similarly, purine-riboside 5′-triphosphate (PRTP) served as an active substrate, but less efficiently than GTP or ITP did; yet it was comparable to ATP (B and C). The above results indicate that the functional groups of ATP 6-amino group, GTP 4-amino and 6-keto groups do not play vital roles in the recognition by M4 ribozyme. Incorporation of GMP-αS to the 5′-terminus of M4 was readily detected by running an APM–PAGE (D). Using this methodology, we investigated if M4 could incorporate ATP-γS, ADP-βS, GTP-γS and GDP-βS to its own 5′-terminus. All substrates turned out to be active, indicating that the activity of M4 did not rely on the number of 5′-phosphates in the substrate. Along similar lines, we also tested whether NAD and m7GpppG could be incorporated into the 5′-end of M4. These nucleotides were also active substrates but less efficient than their canonical nucleotide counterparts, ATP and GTP, respectively (E). This suggests that some steric restriction may exist in the substrate recognition site. Lastly, we performed M4-catalyzed 5′-incorporation of fluorescein-12-GTP in which N7 on the guanine ring was modified with a linker that connected to the fluorescein molecule (A). We were interested in this molecule since the incorporation of fluorescein-12-GTP would allow us to detect the product by a direct fluorescent gel scanning and at the same time the importance of purine N7 could be verified. Although the reaction did take place (F), the efficiency was significantly low compared with the reaction with GMP-αS. It cannot be ruled out that the steric hindrance of the fluorescein linker interferes with the active site of M4, but this may suggest that the purine N7 plays some positive roles in catalysis. Here we have shown selection of a novel ribozyme C06 that catalyzes the 5′-nucleotidyl transfer reaction forming the 2′–5′ phosphodiester bond (E). The C06 sequence was found in pool 14-I as a minor population compared to possible internal kinases (B). Finding of such a sequence was made possible by executing selection under the layers of selection criteria and the careful sequence comparison in two pools at the round 14 generated by the procedures involving 5′-ppp or 5′-OH RNA for the selection. Especially, the latter counterselection strategy facilitated the confirmation of the desired activity without checking all clones for their activities. The 5′-nucleotidyl transfer reaction is considered as a 3′→5′ mononucleotide extension, and thus no natural counterpart is known to exist in the present enzyme world. Nonetheless, an RNA molecule reported here is able to perform such a unique chemistry in as well as in . Joyce . have reported a ribozyme capable of catalyzing both 5′- and 3′-nucleotidyl transfer reactions (). This ribozyme, called E278-19, is a variant of Class I ligase ribozyme isolated by a continuous evolution procedure aimed at evolving 5′→3′ mono- or di-nucleotide extension. Thus, the 3′→5′ nucleotidyl transfer activity was coincidently discovered in the collection of ligase or polymerase ribozymes, but it certainly kept the same signatures as the parental class I ligase ribozyme, where the incoming nucleotide formed a base pair with the template nucleotide and the 3′–5′ phosphodiester linkage was generated. In contrast, C06 and its miniaturized variant M4 promiscuously recognize the incoming purine nucleotides for the 3′→5′ extension (). Although it has not been ruled out if a nucleotide templating to the incoming purine nucleotide resides in the active site, the observed promiscuous activity toward various purine bases suggests that specific hydrogen bonding to purine bases is very unlikely. We rather think that the incoming substrate is guided by base-stacking interactions to the active site of ribozyme and somehow the ribozyme 5′-α-phosphate is projected to the 2′-OH of the incoming nucleotide for the nucleophilic attack. Even though we cannot identify a direct counterpart enzyme in the contemporary protein world, terminal deoxyribonucleotide transferase (TdT) catalyzes a similar reaction. This enzyme can add nucleotide(s) to the 3′-end of DNA in a template-independent manner. Also TdT is well-known to have a preference to dG over other deoxyribonucleotides even though it basically has a promiscuous ability to accept nucleotide substrates (). TdT apparently uses base-stacking interaction at least for its second nucleotidyl transfer step (). Although TdT shares some of key features of M4, distinct differences exist regarding directionality and substrate specificity of the nucleotidyl transfer event. As TdT plays an indispensable role in enhancing the diversity of the immunoglobulin repertoire (), we can imagine a beneficial role of terminal nucleotide transfer at either end of RNA in the RNA world with regard to a genetic diversity, where the recombination could have been the primary source to achieve an evolution or adaptation (,). During the course of mutation studies, we found that indispensable motifs in M4 reside in the regions of the 5′-overhang, P2–P3 junction and L3 loop ( and A). Since M4 catalyzes essentially the same chemistry as RNA ligase ribozymes except for the directionality of extension, it is of interest to compare the secondary structure of M4 with those of reported ligase ribozymes. Four representative ligases are selected for comparison; Bartel's class II ligase, Ellington's L1 ligase, Joyce's R3C and cytidine-free ligases (B–E) (), all of which form a 2′–5′ phosphodiester linkages like M4 ribozyme. These four ligases share the structural features of 5′-overhang and internal guide sequence for the incoming oligonucleotide. An obvious similarity found in all ribozymes including M4 is that the junction domain likely consisting of a part of catalytic core is dominated with A and G, occasionally containing U (note that G is dominated at the 5′-end because the consecutive Gs generally give higher transcription efficiency and therefore such RNA pools have been often used for selection). This suggests that such AG-rich motifs are suitable for composing the tertiary space where 5′-α-phosphate is positioned to the 2′-OH of the incoming nucleotide. On the other hand, a significant difference between M4 and these ribozymes is that M4 lacks the internal guide sequence for specific base pairing. Probably, the P2–P3 junction and L3 region would be located in close proximity, creating a 3D space for bringing the incoming purine nucleotides to the active site via base-stacking interactions. A similar discussion could be extended to a DNA enzyme 7S11 that ligates RNA sequences through 2′–5′ phosphodiester linkage. Unlike the ligase ribozymes discussed earlier, this DNA enzyme produces a branched or lariat RNA (,) using an internal unpaired nucleotide as a branch-site. Though 7S11 is a DNA enzyme that reacts with a RNA substrate, it shares a similarity in structure to M4 where the electrophilic 5′-ppp-end of substrate is needed to be unpaired (F). Moreover, it was shown that the branch-site nucleotide should be a purine nucleotide with 2′-hydroxyl even though adenosine was preferred to guanosine kinetically (), which makes this reaction analogous to the M4-catalyzed purine nucleotidyl transfer. The -acting activity of rM4 and the feasibility of designing the substrate strand guide sequence can lead us to the development of a new tool for the 5′-end modification using purine nucleotides. Particularly, the unnatural 2′–5′ phosphodiester linkage at the 5′-end makes the oligonucleotide resistant toward 5′-exonucleases. Unfortunately, the current system is not yet sustainable for such applications due to the poor activity (using GMP-αS as a substrate, k and K were 0.070 h and 34 mM, respectively). However, re-evolution of M4 ribozyme with appendix domains would potentially produce M4 variants with higher activities, so that a new tool for the 5′-nucleotide modification of RNA can be devised.

DNA sequences that contain four or more closely spaced G-tracts can fold to form intramolecular quadruplexes, which consist of stacked G-quartets that are linked by three loops between the four G-strands (). These structures are stabilized by monovalent cations (especially potassium) (,) and can adopt a variety of different folding patterns dependent on the relative orientation of the strands and the position of the loops. G-rich sequences with the potential to form quadruplex structures are common in genomic DNA and these have been identified in several biologically important regions (). The most widely studied is telomeric DNA, which in higher eukaryotes is composed of repeats of the sequence GGGTTA (,) and for which about 50–100 bases at the 3′-end are single stranded. A number of other non-telomeric G-rich DNA sequences may also form quadruplexes and these have been identified in the promoters of (), Ki- (), 2 (), (), VEGF gene () and HIF 1α (), as well as in fragile X-syndrome () and other trinucleotide repeat sequences (), the retinoblastoma susceptibility gene (), the chicken β-globin gene () and the insulin gene (). G-rich sequences are especially abundant in gene promoter regions () and there is an overabundance of G-rich sequences in the regulatory regions of muscle-specific genes (). For intramolecular quadruplexes, the four G-tracts are separated by loops. These are of various lengths and can be as short as a single nucleotide (). Genomic searches (,) have revealed many G-rich sequences which may be able to adopt these structures, the most common of which are successive G-tracts that are separated by single T or A residues. The loops can be arranged in several different ways; double-chain reversal (propeller) loops link two adjacent parallel strands (), while edgewise or diagonal loops link two antiparallel strands (). Some structures contain both edge-wise and propeller loops (). In the all-parallel (propeller) structures, the nucleotides are in the conformation, while the other structures have different combinations of or glycosidic bonds (,). It is known that loop length and sequence affect quadruplex stability and structure (,). Sequences with single nucleotide loops between the G tracts only adopt a parallel structure, while longer loops can also adopt an antiparallel arrangement of the strands. Quadruplex stability is also affected by the sequence of the loops (), and the bases that flank the quadruplex (). There is considerable variation in quadruplex structure, depending on the DNA sequence and the ionic conditions. The biological function of quadruplexes may well depend on the folded conformation that is adopted, especially if this involves interaction with specific proteins. Such an effect has been suggested for the NHE element of the c-myc promoter, which can in principle adopt multiple conformations. Since the loops can have a considerable effect on quadruplex folding and stability, we have examined how changes in loop length affect quadruplex properties. One very stable intramolecular quadruplex contains four G tracts that are linked by single T residues (,,) and this is known to be an inhibitor of HIV integrase. We have used variations on this sequence to examine the importance of loop length on quadruplex folding and stability. In this study, we have systematically replaced each of the single T loops with T and have used CD, fluorescence melting, 1D-NMR, gel electrophoresis and kinetic studies to examine the effect of loop length and position on quadruplex folding and stability. All oligonucleotides were synthesized on an Applied Biosystems ABI 394 automated DNA/RNA synthesiser on the 0.2 µmole scale using the standard cycles of acid-catalysed detritylation, coupling, capping and iodine oxidation procedures. Phosphoramidite monomers and other reagents were purchased from Applied Biosystems, Proligo and Link Technologies. The sequences of the oligonucleotides used in this work are shown in . Fluorescently labelled oligonucleotiodes were used in all the experiments. These were labelled at the 5′-end with 6-amidohexylfluorescein (FAM), and at the 3′-end with dabcyl using C7 dabcyl cpg (Link Technologies). Oligonucleotides were purified by gel filtration using Nap10 columns (GE Healthcare) and analysed by gel electrophoresis. The bases adjacent to the fluorophore and quencher were the same (T) for all the oligonucleotides to avoid any differences in their effects on quadruplex formation and stability. The thermal melting temperatures of the quadruplexes were determined using the fluorescence melting technique that we have developed () and have used previously for assessing the stability of related quadruplexes (,,,). When the sequence adopts a folded structure the quencher and fluorophore are in close proximity and the fluorescence is quenched. When the structure melts, these groups become separated and there is a large increase in fluorescence. Since the fluorophore and quencher are anchored on relatively long aliphatic tethers the quenching does not depend on the quadruplex topology and the fluorescence is quenched for both parallel and antiparallel complexes. Fluorescence melting experiments were conducted in a Roche LightCycler as previously described (,,,,) in a total reaction volume of 20 µl. Oligonucleotides (final concentration 0.25 µM) were prepared in 10 mM lithium phosphate pH 7.4, which was supplemented with various concentrations of potassium chloride or sodium chloride. The LightCycler has one excitation source (488 nm) and the changes in fluorescence were measured at 520 nm. For several of the oligonucleotides initial experiments revealed that there was considerable hysteresis between the heating and annealing profiles when the temperature was changed at 0.2°C.s, indicating that the process was not at thermodynamic equilibrium. Melting experiments were therefore performed at a much slower rate of heating and cooling (0.2°C.min) by changing the temperature in 1°C steps, leaving the samples to equilibrate for 5 min at each temperature before recording the fluorescence. Under these conditions, no hysteresis was observed (except for some experiments with GT). In a typical experiment, the oligonucleotides were first denatured by heating to 95°C for 5 min. They were then annealed by cooling to 30°C at 0.2°C.min and melted by heating to 95°C at the same rate. The fluorescence was recorded during both the annealing and melting steps. In some instances, the formation of intramolecular or intermolecular complexes was examined by determining the melting curves using a range of oligonucleotide concentrations (0.1–10 µM). Melting temperatures ( values) were determined from the first derivatives of the melting profiles using the Roche LightCycler software. values were obtained from the maxima of the first derivatives of the melting profiles using the LightCycler software or, together with Δ, from van't Hoff analysis of the melting profiles using FigP for Windows. The fraction folded was calculated as previously described () from the difference between the measured fluorescence and the upper and lower baselines. All reactions were performed at least twice and the calculated values usually differed by <0.5°C with a 5% variation in Δ. Since Δ = 0 at the , Δ was estimated as Δ/. Values for Δ at 310 K were then estimated from Δ = Δ – Δ. The van't Hoff analysis assumes that Δ is independent of temperature (i.e. Δ = 0), that the reaction is only a two-step process (i.e. that there are no significant reaction intermediates) and that there is only one folded form of the quadruplex. The number of specifically bound monovalent cations (Δ), was calculated from the slopes of plots of Δ against log[M] as previously described (,). Hysteresis between the melting and annealing profiles occurs when the reaction is not at thermodynamic equilibrium as a result of the slow folding and/or unfolding kinetics. Individual folding () and unfolding () rate constants can be derived from this hysteresis as previously described (,,,). The kinetics of quadruplex unfolding were also determined by measuring the rate of change of fluorescence after rapidly increasing the temperature (,). The quadruplexes were equilibrated at a temperature around the , which was then rapidly increased by 5°C at the fastest rate on the LightCycler (20°C.s). This temperature change causes the quadruplex to partially unfold, moving along the melting curve. Although the theoretical dead-time under these conditions is only 0.25 s, all fluorescence changes that occurred in the first 2 s were ignored, during equilibration to the new temperature. Successive temperature-jumps were then recorded on the same sample by further increasing the temperature by 5°C. Each experiment was repeated at least twice. The time-dependent changes in fluorescence were fitted by an exponential function = × (1 − e) + , using SigmaPlot 10, where is fluorescence at time is the initial fluorescence and is total change in fluorescence (the final fluorescence is + ). The relaxation rate constant () obtained from this analysis is equal to the sum of the folding () and unfolding () rate constants. Arrhenius plots of ln() against 1/ were constructed from these data and used to estimate the activation energy and pre-exponential factor [ = × exp(−/)]. Non-denaturing gel electrophoresis was performed using 14% polyacrylamide gels, which were run in TBE buffer that had been supplemented with 20 mM KCl. Bands in the gels were visualised under UV light. The oligonucleotide concentration was 20 µM. CD spectra were measured on a Jasco J-720 spectropolarimeter as previously described (). Oligonucleotide solutions (5 µM) were prepared in 10 mM lithium phosphate pH 7.4, containing either 200 mM potassium chloride or 200 mM sodium chloride. The samples were heated to 95°C and annealed by slowly cooling to 15°C over a period of 12 h. Spectra were recorded between 220 and 320 nm in 5 mm path length cuvettes. Spectra were averaged over 10 scans, which were recorded at 100 nm.min with a response time of 1 s and a bandwidth of 1 nm. A buffer baseline was subtracted from each spectrum and the spectra were normalized to have zero ellipticity at 320 nm. One-dimensional H NMR experiments were performed on a Varian Inova 600 MHz spectrometer. Oligonucleotides were prepared in 200 mM potassium phosphate pH 7.4 and were annealed by heating to 95°C before slowly cooling to 15°C. 300 µl of the oligonucleotide sample was mixed with 20 µl DO and placed in a Shigemi NMR tube. The final strand concentration was 100 µM. 1D proton NMR spectra were recorded at 25°C with a sweep width of 25 p.p.m., WATERGATE water suppression, an acquisition time of 0.5 s and 32 k scans. Data were processed using VNMR software (Varian Inc.) with zero filling and resolution enhancement. xref #text Circular dichroism is often used to indicate the folding topology of DNA quadruplexes (,,,). Antiparallel quadruplexes typically have a positive CD signal at around 295 nm, while parallel quadruplexes display a positive signal around 260 nm. These differences reflect both the arrangements of the strands and the orientations around the glycosidic bonds. Parallel topologies have all- glycosidic angles, while antiparallel ones have both and in varying ratios. However, it is clear that these spectral signatures are not necessarily an indicator of quadruplex folding as some exceptions have been noted (,). Nonetheless, CD spectra are useful indicators of changes in global quadruplex configuration for series of related oligonucleotides. In the presence of potassium, all the oligonucleotides that contain at least one single T loop exhibit a similar CD spectrum that is indicative of a parallel topology. It therefore appears that in potassium the presence of only one single T loop, in any position, is sufficient to promote all the other loops to form a fold-back propeller-like structure. In principle, these oligonucleotides could adopt several different folded configurations, yet the NMR and gel electrophoresis experiments suggest that only one predominates. When all three loops contain T, there is a dramatic change in the CD spectrum to a form that is consistent with antiparallel folding, though the details are dependent on the ionic strength suggesting that GT can adopt multiple configurations. This again is consistent with the NMR and electrophoresis experiments, which suggest the presence of multiple folded forms. Previous studies have suggested that the quadruplex formed by d(GTG) adopts an antiparallel hairpin dimer in the presence of both sodium and potassium (,). A similar effect is seen for GT, GT-T-T and GT-T-T in the presence of sodium ions and these display CD spectra that are consistent with parallel topologies. However, the greater propensity to form antiparallel structures in the presence of sodium is seen with T-T-T and T-T-T loops, which have CD spectra with peaks at both 260 nm and 295 nm. This may indicate the presence of multiple structural forms, but it is more likely due to the formation of a structure that contains both edgewise (T) and fold-back (T) loops, as observed with other sequences (). The fluorescence melting experiments show that the number of short loops, rather than their position, has the greatest effect on quadruplex stability. In the presence of potassium ions, GT is the most stable and in concentrations above 5 mM it does not display a melting transition. Substituting a T into either the first or second loop decreases the by about 20°C, with a further 20°C decrease on introducing a second T substitution. The same effect is seen in the presence of sodium ions though there is only a small decrease in stability on changing the third loop to T, consistent with the CD spectra which show that GT-T-T, GT-T-T and GT display some antiparallel characteristics in contrast to all the other oligonucleotides. It is noticeable that sequences with a single T loop in the central position are less stable and have lower gel mobilities than their sequence isomers with T in this position (i.e. compare T-T-T loops with T-T-T and T-T-T with T-T-T). It appears that folded structures with a central T loop are more compact and have higher thermal stability. There is then a further decrease in stability when all three loops are composed of T, which as noted above adopts a different configuration. The variation of Δ with ionic strength allows us to estimate the difference in the number of potassium ions specifically bound to the folded and unfolded structures. This value is close to two for GT as expected, since two potassium ions can bind between the three stacked quartets. Although the precise values of Δ should be interpreted with caution, it is noticeable that there is a steady increase in this value as the number of longer loops is increased. The value of Δ is similar for GT-T-T and GT-T-T and is lower than for both GT-T-T and GT-T-T. These results suggest that the longer loops are involved in cation binding. The larger value of Δ seen with GT may not be significant, as this sequence adopts multiple configurations. Comparing the kinetic parameters for the sequences with one or two T loops () reveals that the unfolding parameters are very similar, while there are clear differences in the folding reaction. Complexes with longer loops have higher (less negative) activation energies for the association reaction and larger values for the pre-exponential factor (which is related to the entropy of the transition state). In comparison, no hysteresis is observed with GT and temperature-jump experiments showed a very fast re-equilibration, while GT has slower folding and unfolding parameters, though not as slow as GT (). It is clear that the folding of intramolecular structures with only single T loops is fast, and we imagine that when one loop is composed of a single nucleotide the G-tracts on either side rapidly associate, forming a platform to which the other G-tracts can bind. The results with the oligonucleotides containing one or two T loops suggest that the position of the single-base loop has little effect on the kinetics and that the most important factor is the number of longer loops. It is interesting to note that none of these sequences show any hysteresis in the presence of sodium ions, even though GT and those with two single T loops appear to adopt a similar global structure. The higher stability and slower kinetics in the presence of potassium may therefore reflect conformational changes subsequent to the initial folding events (). p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Mitochondria are multifunctional organelles in eukaryotic cells involved in numerous metabolic activities, the production of ATP, and regulation of apoptosis. Consistent with their bacterial ancestry and absolutely critical for their function, is the mitochondrial DNA (mtDNA) housed in the matrix of the organelle. In humans, the 16 569-bp circular mtDNA molecule contains 37 genes encoding thirteen essential integral membrane proteins of the ATP-producing oxidative phosphorylation (OXPHOS) system, two rRNA subunits of the mitochondrial ribosomes, and 22 tRNAs required for mitochondrial translation in the matrix. Therefore, the vast majority of the ∼1500 proteins that localize and function in mitochondria are not mtDNA-encoded, but rather are products of nuclear genes that are imported into the organelles. An important consequence of this arrangement is that there must be coordination of nuclear and mitochondrial gene expression in order to maintain organelle homeostasis and to properly regulate mitochondrial activities. This is uniquely pertinent to the assembly of the OXPHOS system and mitochondrial ribosomes, which are composed of subunits encoded by both the nuclear and mitochondrial genomes. That is, the coordination of synthesis of subunits in mitochondria (the 13 OXPHOS subunits and two rRNAs) with the import of those from the cytoplasm (the ∼70 OXPHOS subunits and ∼80 ribosomal proteins) is presumably critical to regulate the biogenesis of these large complexes. In this context, it is also important to consider that all of the machinery required for transcription and replication of mtDNA is a subset of the nucleus-encoded factors that are targeted to mitochondria (). How the relative amounts and altered expression of these key regulatory factors influence expression of mtDNA-encoded genes and subsequent assembly of the mitochondrial OXPHOS complexes and ribosomes is largely unknown. The core protein components required for human mitochondrial transcription have recently been defined as a mixed three-component system comprising a T-odd bacteriophage-related mitochondrial RNA polymerase (POLRMT), the high mobility-group-box mitochondrial transcription factor A (h-mtTFA/TFAM), and two dual-function transcription factors (h-mtTFB1/TFB1M and h-mtTFB2/TFB2M) that are orthologs of an ancestral bacterial rRNA methyltransferase (). In terms of transcription, POLRMT/h-mtTFB1 and POLRMT/h-mtTFB2 complexes are each capable of initiation at the mitochondrial L-strand promoter (LSP) and H-strand promoter 1 (HSP1) in the presence of h-mtTFA, although POLRMT/h-mtTFB2 complexes are reported to be significantly more active than POLRMT/h-mtTFB1 complexes (). However, at specific h-mtTFA: DNA ratios the activities of the two complexes are altered to differing degrees. This leaves open the likely possibility that the rates of transcription can be dictated by the relative concentrations of mtDNA and each of the transcription components in different cell types (,). Estimates of h-mtTFA abundance in cells of human and have been reported by several groups, but with conflicting results (), and the relative or absolute amounts of POLRMT, h-mtTFA, h-mtTFB1 and h-mtTFB2 have not been systematically examined. Thus a basic definition of the mitochondrial transcription system is still lacking, a situation that limits our understanding of how mitochondrial gene expression is regulated. As already mentioned, the recently identified h-mtTFB1 and h-mtTFB2 transcription factors are homologous to rRNA methyltransferases. Specifically, they are related to a large family of site-specific methyltransferases that catalyzes the N6-dimethylation of two adjacent adenine residues in a conserved stem-loop found in small subunit rRNAs (). In bacteria this modification, while not essential, influences both stability and fidelity of ribosomes and is thought to modulate translational output (). In the original report describing h-mtTFB1, this transcription factor was postulated to have rRNA methyltransferase activity based on its ability to bind the methyl-group donating co-factor -adenosylmethionine (). Subsequent work showed that both h-mtTFB1 and h-mtTFB2 can functionally replace the homologous KsgA dimethyltransferase in (,), demonstrating that both proteins have retained this enzymatic activity. Interestingly, in this assay, h-mtTFB2 has significantly lower activity than h-mtTFB1 (). This, coupled to their differing relative activity with regard to transcription output and the concentration of h-mtTFA, again suggests a potentially interesting regulatory scenario in which differential effects on mitochondrial transcription and ribosome biogenesis/function (via differential rRNA methylation) are mediated by the relative abundance or activity of h-mtTFB1 and h-mtTFB2 (,). Interestingly, other functions for members of this class of rRNA methyltransferases have been assigned that are independent of the catalytic activity. For example, the yeast Dim1p cytoplasmic rRNA methyltransferase plays a role in rRNA processing (,). Likewise, we have shown that the transcription factor activity of h-mtTFB1 is unaffected by mutations that inactivate its rRNA methyltransferase activity (). Thus, whether h-mtTFB1 and h-mtTFB2 have additional roles beyond transcription and rRNA methylation remains a formal possibility. To date, the only studies regarding the function of metazoan mtTFB transcription factors were performed on the proteins using cultured Schneider cells. Consistent with a role for dm-mtTFB2 in transcription and transcription-primed mtDNA replication, RNAi knock-down or over-expression results in decreased or increased steady-state levels of mitochondrial RNA and mtDNA, respectively (). In contrast, no effect on mtDNA copy number or the steady-state levels of mitochondrial transcripts is observed when expression of dm-mtTFB1 is knocked down or over-expressed in this system (). However, RNAi knock-down of dm-mtTFB1 does result in a reduction in the rate of mitochondrial protein synthesis, indicating a role for this protein in regulating mitochondrial translation (perhaps via its rRNA methyltransferase activity) (). These results demonstrate that the functions of dm-mtTFB1 and dm-mtTFB2 are not fully redundant and suggest that the primary role of dm-mtTFB2 is to regulate transcription and/or mtDNA copy number and that of dm-mtTFB1 is to modulate translation. However, it is important to note that these studies do not discount the possibility of partially overlapping functions for each factor in transcription and rRNA methylation that one might predict based on the fact that both h-mtTFB1 and h-mtTFB2 have retained both of these activities over the course of evolution (). Furthermore, given the significant differences between mammals and flies in general and with regard to mitochondrial genome organization and dynamics (), it is important not to generalize any relationships described in this (or any other) model system without direct confirmation in human cells. Here, we describe our analysis of the human mitochondrial transcription system in HeLa cells that defines the relative steady-state levels of the entire human mitochondrial transcription machinery for the first time and elucidates novel properties and functions of h-mtTFB1 and h-mtTFB2 via their purposeful over-expression. Bacterial expression vectors for production of h-mtTFB1 and h-mtTFB2 in were created using pET21b (Promega). The h-mtTFB1 and h-mtTFB2 cDNAs were amplified using primers 5′-AACATATGGCTGCCTCCGG-3′ and 5′-AAGAGCTCGAGTCTGTAATTCTC-3′ or primers 5′-AACATATGTGGATCCCAGTGG-3′ and 5′-GCGGCCGCCCTATCTTCCAGGGTTC-3′, respectively, and the resulting PCR products were ligated into pGEMT-Easy (Promega). These plasmids were then digested with appropriate restriction enzymes for ligation in pET21b (I and I for h-mtTFB1 or I and I for h-mtTFB2). To generate plasmids for over-expression in human cells, the h-mtTFB1 and h-mtTFB2 cDNAs were both cut from previously described vectors () using EcoRV and NotI and ligated into pcDNA3.1 zeo (+) (Invitrogen) cut with the same enzymes. The vector used to express POLRMT in bacteria was pProEX-Htb (Invitrogen). A portion of the human cDNA encoding amino acids 41–1250 and the stop codon was cloned into the BamH1 and XhoI of this vector via a BamH1–Sal1 restriction fragment. Amino acids 1–40 were deleted since they compose the mitochondrial localization sequence (MLS) that is predicted to be removed by proteases during import into mitochondria (). However, in place of the MLS, there are 29 unnatural amino acids fused to POLRMT that include a initiator methionine, a His6 tag, a spacer of seven amino acids, a TEV protease cleavage site, and another spacer of six amino acids. The vector has an intact lacI gene allowing POLRMT expression from the promoter to be regulated by addition of isopropylthiogalactoside (IPTG; Sigma). BL21-Codonplus® (Stratagene) were transformed with pET21b vectors containing either h-mtTFB1 or h-mtTFB2. For purification of h-mtTFB1, an overnight culture of bacteria was diluted into 1 l of LB media and grown at 37°C until culture reached an OD of 0.8. IPTG was added to 1 mM for induction of expression and the culture was allowed to incubate at room temperature with shaking for 24 h. Cells were harvested by centrifugation, then pellets were resuspended in 80 ml of lysis buffer (50 mM sodium phosphate, pH 8.0; 0.3 M NaCl, 10% glycerol, 0.1% Tween-20 and 10 mM imidazole). Cells were lysed by sonication and the suspension was cleared by centrifugation. The clarified lysate was applied to a BD-Talon column (BD Biosciences). The column was washed with 10 column volumes of lysis buffer and eluted with three column volumes of lysis buffer plus 150 mM imidazole. Fractions were assayed for presence of mtTFB1 by SDS-PAGE followed by either coomassie staining or western blot. Peak fractions were combined, concentrated and dialyzed into storage buffer (20 mM Tris-HCl, pH 8.0; 0.5 mM EDTA, 0.25 M sucrose, 15% glycerol, 1 mM DTT and 1 mM PMSF). Protein concentration was determined with a protein assay kit from Bio-Rad and confirmed by SDS-PAGE and coomassie staining with comparison against a BSA standard curve. For purification of h-mtTFB2, the same procedure was followed for mtTFB1, except that the protein was found to be highly insoluble and present in the pellet after sonication. Pellets after sonication were resuspended in lysis buffer with the addition of 0.1% Triton, 5 mM BME, 1 mM PMSF and 6 M guanidine-HCl. The solution was stirred overnight at 4°C then clarified by centrifugation. The protein solution was added to a Talon column as before, then washed with 10 column volumes of the pellet resuspension buffer. The column was eluted with three column volumes of the same buffer plus 200 mM imidazole. Fractions were analyzed as above and peak fractions were collected. Concentrated samples were dialyzed extensively against storage buffer and protein concentration was determined. Purification of h-mtTFA was performed as described (), except the process was stopped after the BioRex-70 column (Bio-Rad). The expression and purification of POLRMT was performed as described (). For transfection of HeLa cells (Clonetech) with h-mtTFB1 or h-mtTFB2 over-expression plasmids, cell were seeded at 5 × 10 per 10 cm dish in DMEM (Sigma) plus 10% Bovine Growth Serum (Hyclone) and allowed to grow for 24 h. Cells were transfected with empty pcDNA3.1 vector, pcDNA3.1-mtTFB1 or pcDNA3.1-mtTFB2 using Effectene (Qiagen) according to manufacturer's suggestions. Cells were allowed to incubate with plasmid/reagent complexes for 24 h. Transfected cells were subcultured by diluting cells 1:50 or 1:100 in growth media with 400 μg/ml Zeocin (Invitrogen) and plated in 10 cm dishes. Cells were grown until individual colonies were visualized and 10–15 clones were individually selected with glass cylinders and transferred to 24-well plates. Cells were grown to confluence, subcultured into 6-cm dishes in growth media plus 100 μg/ml Zeocin and grown for 48 h. One dish for each clone was harvested and assayed for protein expression via western blot. Remaining plates were harvested and stored in growth media plus 10% DMSO at −80°C. Stably transfected HeLa lines were consistently plated at 5000 cells/cm, grown for 72 h at 37°C and 5% CO before subculturing again. All experiments were performed on cells passaged at most 10 times. Four polyclonal peptide antibodies (two for h-mtTFB1 and two for h-mtTFB2) were generated for us by Multiple Peptide Systems. The peptides used as antigens were as follows: TFB1-1 H-CVPKPEVDVGVVHFTPLIQPKIE-NH, TFB1-2 H-CREELKRRKSKNEEKEEDDAENYRL-NH, TFB2-1 H-CWIPVVGLPRRLRLSALAGA-NH, TFB2-2 H-CPQLWPEPDFRNPPRKASKASLD-NH. Multiple Peptide Systems performed the synthesis of peptides, injection of rabbits and collection of serum, and also provided a small batch of peptide-affinity purified antibody. We also used antibodies that were purified from provided crude serum using protein-A sepharose (Amersham). Specificity of each antibody was determined by immunoblots of 200 ng of recombinant full-length h-mtTF1 and h-mtTFB2 proteins run alongside one another on the same gel. Antiserum used to detect human POLRMT was the same as that described previously () and polyclonal antibodies for detection of h-mtTFA were generously provided by Dr David Clayton. For whole-cell extracts, 1 × 10 cells were suspended in 100 μl of cold lysis buffer (50 mM Tris-HCl pH 8.8, 150 mM NaCl, 1% NP-40, 0.5% sodium deoxycholate, 0.1% SDS, 2 mM EDTA, 10% glycerol, 5 mM DTT, 1 mM PMSF) and incubated at 4°C with rotation for 30 min. Protein concentration was determined using the BioRad Protein Assay Kit and indicated amounts of total protein were loaded on polyacrylamide gels for analysis. For mitochondrial extracts, mitochondria were harvested by differential centrifugation. Briefly, cells were resuspended in 10 pellet volumes of RSB buffer (10 mM NaCl, 1.5 mM MgCl, 10 mM Tris-HCl pH 7.5), swelled on ice for 10 min, homogenized with a motorized Teflon pestle, then 2.5 × MS Buffer (525 mM mannitol, 175 mM sucrose, 125 mM Tris-HCl pH 7.5, 2.5 mM EDTA) was added to 1 ×. The homogenate was centrifuged at 980  for 10 min twice to pellet nuclei and unbroken cells. The supernatant was transferred to a fresh tube and spun at 17 000  for 30 min to pellet mitochondria. The mitochondrial pellet was washed three times with 1 × MS buffer then stored at −80°C until further use. Mitochondrial pellets were resuspended in lysis buffer equal to one volume of the original cell pellet. The protein extract was then treated and quantified as described above. Mitochondria were harvested as described above but were purified additionally by centrifugation through a 1.0–1.5 M sucrose step gradient at 45 000 r.p.m. in a SW50.1 rotor for 2 h. Mitochondria were removed from the interface of the two sucrose solutions by pipetting and transferred to a fresh tube. Mitochondria were washed with 1 ml of 1 × MS Buffer and pelleted by centrifugation at 13 000 r.p.m. for 10 min. The final pellet was stored at −80°C until further use. Efficiency of mitochondrial harvest was measured by amount of HSP60 and h-mtTFA immunoblot signal obtained from fractions reserved throughout the process and compared to amounts from starting whole-cell lysates. Final mitochondrial pellets were lysed, extracts were quantified as described above, and samples ranging from 10 to 100 μg of total mitochondrial protein were separated alongside 50–150 µg of whole-cell lysate and a range of amounts of purified recombinant proteins (1–100 ng) on 6–20% SDS–polyacrylamide gradient gels. Proteins were transferred to PVDF and western blotting was performed as described above. A standard curve was generated from signal obtained from recombinant proteins and used to determine the amount of the respective protein in each of the sample lanes. Cells were harvested after 72 h of growth in conditions described above and counted. For each cell line, 1 × 10 cells were used for RNA extraction using the RNEasy Kit (Qiagen) according to included instructions. During the process, DNA was digested with RNAse-free DNAse according to manufacturer's suggestions. RNA was eluted in the final step using dHO, quantified and stored at −80°C until further use. Gel electrophoresis, blotting of RNA and probing of blots with 16S, 12S, ND2 and ND6 probes were performed as previously described (). For analysis of h-mtTFB1 transcript levels, first strand cDNA synthesis was performed by combining 2 μg of total RNA with 1 mM dNTPs and 8 μM oligodT in a 20 μl reaction volume. This mixture was heated to 70°C for 10 min then allowed to cool to 4°C over 10 min to anneal the oligodT primer to polyA tails of mRNAs. An equal volume of reverse transcriptase mix [200 units M-MuLV Reverse Transcriptase (NEB), 2X M-MuLV RT Buffer and 25 units of RNAseOUT (Invitrogen)] was added to each primer-RNA mix and incubated at 42°C for 1 h. The reactions were then heated to 70°C for 10 min to deactivate the reverse transcriptase. Finally, an equal volume of 3 mM Tris-HCl pH 8.5, 0.3 mM EDTA was added to each reaction and samples were stored at −20°C until further use. Samples of each cDNA were thawed and diluted from 16- to 32-fold in dHO. A 10 μl of a diluted cDNA sample was added to 14 μl of SYBR Green reaction mix () to each well along with 0.5 μl of both appropriate 25 μM primers. β-Actin primers, β-actin RT F 5′ TGGCACCACACCTTCTACAATGAGC 3′ and β-actin RT R 5′ GCACAGCTTCTCCTTAATGTCACGC 3′, were used as controls for total cDNA in each reaction. To amplify the h-mtTFB1 transcript, primers TFB1 RT-F 5′ TCTGCAATGTTCGACACATC 3′ and TFB1 RT-R 5′ ACCTATATAAGAAGCTCCAC 3′ were used. Reaction conditions for the BioRad iCycler were as follows: 95°C, 10 min, 1 cycle; 95°C, 30 s, 56°C, 30 s, 72°C, 30 s, 40 cycles. Fluorescence was measured after each 56°C step. Melt-curve analysis to ensure single products was performed immediately after completion of steps above by using a temperature step gradient from 55 to 80°C in 0.5°C increments with fluorescence measured after a 10 s incubation at each temperature. values for each reaction were obtained through the iCycler iQ software. h-mtTFB1 transcript levels were then normalized by β-actin transcript levels and compared to empty vector controls. Total DNA was extracted from 1 × 10 cells by addition of 500 μl of extraction buffer (50 mM Tris-HCl pH 8.5, 0.25% SDS, 1 mM EDTA, 5 mM DTT) and boiling for 10 min. After cell lysis, tubes were allowed to cool to room temperature, 100 μg of RNAse A were added, and tubes were allowed to incubate at 37°C. Following a 3 h incubation, 100 μg of proteinase K were added and samples were placed at 55°C overnight. Samples were heated to 95°C for 5 min and allowed to cool to room temperature. Total DNA concentration was measured and samples were stored at -20°C until further use. Relative and absolute mtDNA copy numbers were measured by SYBR Green fluorescence using a BioRad iCycler and accompanying software (v3.1). Total DNA samples were diluted to a range of DNA concentrations from 500 to 7.8 pg/µl and 10 μl were dispensed to appropriate wells of a 96-well PCR plate. 14 μl of SYBR Green reaction mix () were added to each well along with 0.5 μl of both appropriate 25 μM primers. For detecting relative levels of mtDNA, a region of the mitochondrial genome encompassing a portion of COX3 and a region of the multicopy nuclear 18S rDNA locus were used. Primers are as follows: RTQ COX3-F 5′ CACCCAAGAACAGGGTTTGT3′, RTQ COX3-R 5′ TGGCCATGGGTATGTTGTTAA 3′, RTQ 18S F 5′TAGAGGGACAAGTGGCGTTC3′ and RTG 18S R 5′CGCTGAGCCAGTCAGTGT3′. Reaction conditions for the iCycler were as follows: 95°C, 10 min, 1 cycle; 95°C, 15 s, 60°C, 1 min, 40 cycles. Fluorescence was measured after each 60°C step. Melt-curve analysis to ensure single products was performed immediately after completion of steps above by using a temperature step gradient from 55 to 80°C in 0.5°C increments with fluorescence measured after a 10 s incubation at each temperature. Standard curves to determine absolute copy number were constructed with known amounts (1 638 400 to 100 templates) of the plasmid pGEMT Easy (Promega) containing either one of the PCR products obtained using the mitochondrial and nuclear primers described above (COX3 or 18S). Labeling was performed as described () with modifications. HeLa cells were seeded and grown as described above in 10 cm dishes. After 72 h of growth, cells were washed three times with 5 ml of sulfur-free media (Gibco) without serum. Cells were allowed to incubate for 5 min at 37°C, 5% CO before removing media. After the last wash, 5 ml of sulfur-free media with 10% BGS plus 100 μg/ml emetine were added to each plate and allowed to incubate at 37°C, 5% CO for 5 min. After this incubation step, ExpreSS Protein Labeling Mix (Perkin Elmer) was added to each plate to a concentration of 125 μCi/ml and labeling was allowed to proceed at 37°C and 5% CO. After 1 h, the labeling media was removed and the plates were rinsed once with 5 ml of normal DMEM with 10% BGS. Plates were then washed twice with 10 ml of TD buffer (25 mM Tris pH 7.5, 137 mM NaCl, 10 mM KCl, 0.7 mM NaHPO), trypsinized and harvested by centrifugation at 1000 . Cell pellets were washed once with 10 ml of TD buffer, then transferred to a 1.5 ml tube with 1 ml of TD buffer and harvested again by centrifugation. Mitochondria were harvested as above and protein was quantified. 50 μg of total protein were added to each lane of a 10–20% linear gradient SDS–polyacrylamide gel. Samples were separated and then the gel was dried and exposed to film at −80°C. HeLa cells were plated and grown as described above. After 48 h of growth, kasugamycin was added at indicated concentrations and cells were allowed to incubate for 72 h. Cells were washed, trypsinized and treated with 0.08% Trypan Blue in PBS. The total number of viable cells at each drug concentration was compared to that of zero-drug controls to yield a percentage of viable cells that is plotted as a function of the drug concentration. HeLa cultures were plated and grown as described above. After 72 h of growth, cells were stained in culture media with 70 nM Mitotracker Green and 90 nM Mitotracker Red CMXRos, both from Molecular Probes, for 30 min at 37°C, 5% CO. After staining, dyes were removed and cells were washed three times with PBS. Cells were trypsinized and collected by centrifugation. Cell pellets were resuspended in 1 ml of PBS and analyzed on a BD FACS calibur instrument with accompanying software. Histograms and means for these data were obtained using FlowJo (v 8.0.1) Full-length sequences of h-mtTFB1, h-mtTFB1 and sc-mtTFB were obtained from NCBI. These sequences were submitted to cleavage site prediction by the SignalP program (v3.0, ) (). Hidden Markov modeling option was selected and the first 70 amino acids for each sequence were used for signal and cleavage site prediction (). To fully understand any transcription system, a critical parameter to know is the relative amounts of the basal transcription machinery . Our approach was to use immunoblotting to establish the levels of the mitochondrial transcription machinery on a total cell and mitochondrial basis. This method requires antibodies that are specific for their target, especially for homologous proteins that are very similar in size. Since h-mtTFB1 and h-mtTFB2 share a large degree of sequence similarity and are predicted to migrate similarly on polyacrylamide gels, it was unclear if antibodies we had generated would cross react with the paralogous protein and confound our analysis. Therefore, we first generated peptide antibodies against h-mtTFB1 and h-mtTFB2 (see Materials and Methods section) and determined their specificity by assessing reactivity toward purified full-length recombinant h-mtTFB1 and h-mtTFB2 by immunoblotting. We found that these antibodies were indeed highly specific for their targets showing no reactivity with the paralogous protein (A). Having antibodies capable of specifically detecting each of the four human mitochondrial transcription proteins, as well as known amounts of corresponding recombinant proteins, allowed us to determine for the first time their relative abundance. We purified mitochondria from logarithmically growing HeLa cells and performed quantitative western blot analysis of POLRMT, h-mtTFA, h-mtTFB1 and h-mtTFB2 on known amounts of total mitochondrial protein. For each protein analyzed, multiple dilutions of total mitochondrial protein or total whole-cell lysate were probed in parallel with a dilution series of known amounts of recombinant protein to allow the amount of each transcription component relative to the total amount of mitochondrial lysate to be determined. Representative western blots and standard curves showing linearity of the assays employed are shown in Supplementary Figure S1. Based on this analysis, we calculated the following amounts of each transcription component/100 μg of total mitochondrial protein: 45.8 ± 9.7 ng of h-mtTFA, 4.03 ± .20 ng of h-mtTFB1, 14.9 ± .97 ng of h-mtTFB2 and 32.3 ± 2.1 ng of POLRMT. For whole-cell lysates, we were only able to accurately measure h-mtTFA levels and found 11.1 ± 2.4 fg of protein per cell (data not shown). Next, we determined the number of molecules/cell based on the amount of total mitochondrial protein isolated from 4.5 × 10 cells. We found there to be 25.2 ± 2.51 pg of mitochondrial protein per cell with recoveries of 75–80% of total mitochondria for three separate experiments based on HSP60 and h-mtTFA immunoblots (data not shown). These measurements result in 2.52 × 10, 1.35 × 10, 4.29 × 10 and 3.12 × 10 molecules per cell for h-mtTFA, h-mtTFB1, h-mtTFB2 and POLRMT, respectively (). This amount of h-mtTFA from mitochondrial samples agreed well with our measurement in whole-cell lysates at 2.78 × 10 molecules per cell (data not shown). Finally, we also determined the mtDNA copy number in the cells to be 5010 ± 386 using quantitative real-time PCR (Supplementary Figure S2). From these data, we calculated the relative levels of the four transcription proteins on a molecule/cell basis and as function of the number of mtDNA molecules/cell (). Interestingly, there is ∼3-fold more h-mtTFB2 than h-mtTFB1 in these cells, which results in a close to 1:1 relationship between h-mtTFB2 and POLRMT (1.37:1), but an excess of POLRMT to h-mtTFB1 (2.31:1). Furthermore, our measurements place POLRMT at an ∼6-fold molar excess over the number of mtDNA molecules and between 42 and 58 h-mtTFA molecules/mtDNA molecule in HeLa cells. Our h-mtTFA to mtDNA ratio is consistent with that observed by Wiesner and colleagues (), but more than 30-fold lower than that proposed by Kang and colleagues (). With a determination of relative levels of the mitochondrial transcription machinery in hand, we next set out to examine the consequence of altering the relative amounts of these proteins on mitochondrial gene expression and to determine if their expression levels are coordinately regulated. To do this, we created stable HeLa cell lines that over-express either h-mtTFB1 or h-mtTFB2. From our initial analysis of these lines, we immediately made two novel observations. First, the mobility of h-mtTFB2 isolated from cells was significantly faster than that of the corresponding recombinant protein (B). Since many matrix-localized mitochondrial proteins have an N-terminal localization sequence (MLS) that is often removed by proteases upon import, we hypothesized that this was the case for h-mtTFB2. Consistent with this, using SignalP (a MLS and cleavage site prediction program), a single high-probability mitochondrial peptidase cleavage site for h-mtTFB2 was found between amino acids 30 and 31 (data not shown). No such cleavage site was predicted for h-mtTFB1, consistent with no obvious change in mobility of this protein relative to its recombinant control (B). Furthermore, mapping amino acids 5–22 onto a helical-wheel diagram reveals a pattern consistent with a putative amphipathic helix (data not shown), which is another common attribute of an MLS. Finally, using a peptide antibody (TFB2-1) that was made against amino acids 2–20 of h-mtTFB2 (which are predicted to be removed by SignalP), we were unable to successfully detect endogenous h-mtTFB2 by western blotting (data not shown). However using the peptide antibody (TFB2-2) generated against amino acids 49–70, which is used throughout this study, we readily detect endogenous h-mtTFB2 (B). Taken together, these data suggest that amino acids 1–30 likely comprise a significant portion of the MLS for h-mtTFB2 and are removed upon import. In addition to the increased mobility of h-mtTFB2 described above, we also found that in the cell lines that over-express h-mtTFB2, there was a corresponding increase in the steady-state level of h-mtTFB1 (B, compare lanes 6, 7 and 8 to lane 4). However, the converse was not true. That is, there was no increase in h-mtTFB2 steady-state levels when h-mtTFB1 was over-expressed ∼10-fold (B, compare lane 5 to lane 4). We also analyzed the steady-state levels of the remaining transcription machinery in the face of h-mtTFB1 or h-mtTFB2 over-expression. No obvious changes in the amounts of POLRMT or h-mtTFA per mitochondrion were observed in either case (D). To begin to determine the mechanism of increased h-mtTFB1 levels due to h-mtTFB2 over-expression we employed reverse transcriptase real-time PCR to measure the steady-state amounts of h-mtTFB1 mRNA. In the h-mtTFB2 over-expression cell lines, h-mtTFB1 mRNA was increased ∼2.5-fold (C and Supplementary Figure S3), in good correspondence with the observed increase in h-mtTFB1 protein levels (B). As expected there was a large (∼64-fold) increase in h-mtTFB1 mRNA in the h-mtTFB1 over-expression lines (C). However, this did not correspond with the only ∼10-fold over-expression of the protein (B), suggesting that post-transcriptional mechanisms may limit h-mtTFB1 expression or accumulation. To address how altered levels of h-mtTFB1 and h-mtTFB2 affect mitochondrial gene expression, we next examined the h-mtTFB1 and h-mtTFB2 over-expression HeLa cell lines for changes in the steady-state levels of mtDNA-encoded transcripts and proteins, as well as for alterations in mtDNA copy number. Northern analysis of the mitochondrial 16S and 12S rRNAs and of ND2 and ND6 transcripts (representing mRNAs transcribed from each strand of mtDNA) revealed a ∼2-fold increase in their steady-state levels in the h-mtTFB2 over-expression cell line, but no change in the h-mtTFB1 over-expression line (A). Similar results were obtained when immunoblots of the mtDNA-encoded COX1 and COX2 proteins was performed and mtDNA copy number was measured. That is, over-expression of h-mtTFB2, but not h-mtTFB1, led to a significant increase in the steady-state levels of COX1 and COX2 proteins (B) and a doubling of the mtDNA copy number (C). Altogether, these results are consistent with a role for h-mtTFB2 in transcription and in transcription-primed mtDNA replication. However, the fact that h-mtTFB1 is up-regulated in the h-mtTFB2 over-expression lines makes it difficult to assign these functions to h-mtTFB2 acting independently of h-mtTFB1. Both h-mtTFB1 and h-mtTFB2 have retained rRNA methyltransferase activity that is postulated to modulate mitochondrial ribosome biogenesis and/or translation (). This, coupled to the fact that a function for mtTFB1 in mitochondrial translation has been documented (), we next measured mitochondrial translation rates in the h-mtTFB1 and h-mtTFB2 over-expression HeLa cell lines using an -radiolabeling approach. Similar to the analysis of the other mitochondrial gene expression parameters (), we again saw major differences only in the h-mtTFB2 over-expression cell line, where a significant increase in the overall rate of translation was observed (A). Interestingly, however, the labeling of specific products was not uniform (e.g. COX1, COX2 and ATP6), suggesting that increasing the amount of h-mtTFB2 and/or mitochondrial mRNAs is leading to increased translation of some mRNAs to the exclusion of others. In all of the analyses described thus far, no obvious consequences of over-expressing h-mtTFB1 alone were observed. We were somewhat surprised that over-expression of h-mtTFB1 did not influence mitochondrial translation rates given that it is more active as a rRNA methyltransferase () and that its ortholog has been implicated in translation efficiency (). It remained a formal possibility that h-mtTFB1 is affecting mitochondrial ribosome function or biogenesis in manner that is not read out as an increase in the overall rate of translation. In bacteria, methylation of the small subunit rRNA by the h-mtTFB1 ortholog KsgA results in sensitivity to the aminoglycoside antibiotic kasugamycin and we have shown that h-mtTFB1 and h-mtTFB2 can functionally replace KsgA in (). Furthermore, the human mitochondrial 12S rRNA and the bacterial 16S rRNA are highly conserved at the site that is methylated (a stem-loop at the 3′ end). We therefore determined whether over-expression of h-mtTFB1 alters the sensitivity of human mitochondrial ribosomes to kasugamycin. We found that over-expression of h-mtTFB1 resulted in a significant decrease in viability of HeLa cells grown in the presence of high concentrations of this drug (B). Similar results were obtained in the h-mTFB2 over-expression line (B). However, given that h-mtTFB1 is also up-regulated when h-mtTFB2 is over-expressed (B), whether this effect is due to h-mtTFB1 or h-mtTFB2 or perhaps both cannot be distinguished. Given that over-expression of h-mtTFB2 (and as a result also h-mtTFB1, B) results in an increase in multiple mitochondrial gene expression parameters ( and ), we next determined whether this signaled cells to increase overall mitochondrial biogenesis. In parallel, we also analyzed the h-mtTFB1 over-expression line. Surprisingly, we found that both cell lines exhibited a ∼50% increase in mitochondrial mass as measured by Mitotracker Green staining and FACS analysis (A). Thus, despite the fact that over-expression of h-mtTFB1 does not increase mitochondrial gene expression in any manner examined thus far ( and ), remarkably, it does invoke a mitochondrial biogenesis response. The results of the Mitotracker Green staining were largely confirmed by immunoblots, where mitochondrial biogenesis was assayed as the ratio of the amount of the mitochondrial outer membrane protein VDAC to that of tubulin (B). We also analyzed the h-mtTFB1 and h-mtTFB2 over-expression cell lines for mitochondrial membrane potential using Mitotracker Red staining and FACS analysis. Here we observed an ∼80% increase in membrane potential in the h-mtTFB2 over-expression line, but no statistically significant change as the result of h-mtTFB1 over-expression alone (A). These results suggest the interesting possibility that the simultaneous up-regulation of both h-mtTFB1 and h-mtTFB2 (as is the case in the h-mtTFB2 over-expression lines) is required to fine-tune mitochondrial activity in accordance with changes in mitochondrial biogenesis and gene expression. In this study, we have performed a detailed analysis of the human mitochondrial transcription machinery in HeLa cells. Specifically, we have determined for the first time the relative steady-state levels of the four core components of this system: the human mitochondrial RNA polymerase, POLRMT; the mtDNA binding transcription factor, h-mtTFA; and the two dual-function transcription factors/rRNA methyltransferases, h-mtTFB1 and mtTFB2. In addition, we have clearly demonstrated that the two h-mtTFB paralogs have unique attributes and distinct functions with regard to mitochondrial gene expression and biogenesis. We will discuss the primary findings and the main conclusion we reach based on the results of this study below. The first goal of this study was to define the relative amounts of the human mitochondrial transcription machinery . Previously, groups have estimated the amounts of h-mtTFA (), but a simultaneous assessment of the entire core transcription system had not been described. We now have antibodies that recognize the four core human mitochondrial transcription components and, importantly, we generated peptide antibodies that are capable of distinguishing the two related h-mtTFB paralogs, h-mtTFB1 and h-mtTFB2 (A). Using these antibodies, we were able to detect each component in mitochondrial extracts containing known amounts of total mitochondrial protein by western blotting and compare these to the signals obtained from known amounts of cognate recombinant protein analyzed in parallel. Since we also quantified the total number of cells from which the extracts were derived and the copy number of mtDNA, we are able to express the results for each individual component as the number of molecules/cell or the number of molecules/molecule of mtDNA (). Our results revealed a number of novel and salient points about the relative abundance of the mitochondrial transcription system in HeLa cells (see ). First, relative to the mtDNA copy number, which we found to be 5010 ± 386 (Supplementary Figure S2), POLRMT is in ∼6-fold excess of the mtDNA on a per-molecule basis. In principle, this is sufficient to allow all templates to be engaged in transcription simultaneously. The second main conclusion that we reach is that there is roughly three times more h-mtTFB2 than h-mtTFB1 molecules/cell. This difference is perhaps most relevant when compared to the amount of POLRMT, from which it becomes clear that there is an excess of h-mtTFB2 to POLRMT (1.3:1), but a limiting amount of h-mtTFB1 relative to POLRMT (0.43:1). While the relevance of these differences is difficult to predict, it is noteworthy that for both h-mtTFB1 and h-mtTFB2 the levels relative to POLRMT are not too far removed from 1:1, which is consistent with the predicted optimal stoichiometry in the core transcription complexes for transcription (). Finally, we arrive at a value of ∼25 000 molecules of h-mtTFA per cell, which places it in ∼5–18 fold excess of the other core transcription components (). Given that h-mtTFA has been postulated to have an mtDNA-packaging role in addition to its transcription factor function (,), its abundance relative to mtDNA is important to discuss. Based on our measurement of mtDNA copy number, the ratio of h-mtTFA:mtDNA molecule we observe is 50 ± 8:1. This value is in good agreement with that of 35:1 reported by Wiesner and colleagues (), but substantially lower than that proposed by Kang and colleagues, who suggest a ratio of ∼1700:1 (). Kang and colleagues cite the ratio of mtTFA to mtDNA in at 2000:1 () to support of their findings (). However, ratios of mtTFA (xl-mtTFA):mtDNA are greatly up-regulated during maturation ranging from a resting immature oocyte level of ∼200:1 () to the noted ratio of ∼2000:1 (), which occurs only in mature oocytes. Given that xl-mtTFA binds mtDNA as a tetramer (), there is effectively ∼50 xl-mtTFA complexes per genome in immature oocytes, which we argue is a cell type that is more relevant for comparison to mammalian cell types than a mature oocyte, which has dramatically up-regulated mitochondria and mtDNA in preparation for fertilization and development. Finally, if h-mtTFA levels were indeed high enough to completely coat the mtDNA genome as suggested by Kang and colleagues (,,,), this would seem incompatible with any significant transcriptional output based on transcription studies (,). Furthermore, it becomes difficult to explain why over-expression of h-mtTFA increases mitochondrial transcription and mtDNA copy number (,) if, in fact, mtDNA is already fully saturated with h-mtTFA (). It is noteworthy that we observe less than one-sixth the amount of h-mtTFA and five times more mtDNA per cell than Takamatsu . measured in HeLa cells (). These differences likely begin to account for the apparent overestimation of the mtTFA:mtDNA ratio by Kang and colleagues compared to that reported herein and by Wiesner and colleagues, which largely corroborate each other. With the new knowledge of the relative levels of the human mitochondrial transcription system in HeLa cells, we went on to analyze the consequences of over-expressing each of the h-mtTFB paralogs on mitochondrial gene expression and biogenesis . We established stable HeLa cell lines that over-express h-mtTFB1 or h-mtTFB2 by ∼10-fold or up to ∼3-fold, respectively (B). Characterization of these lines by western blotting immediately revealed a salient difference between these two factors; that h-mtTFB2 is processed (B). Several additional lines of evidence strongly indicate that the processing of h-mtTFB2 is via cleavage by mitochondrial proteases upon import of the protein into mitochondria, including a strong predicted mitochondrial protease cleavage site between amino acids 30 and 31, the presence of a predicted amphipathic alpha helix (stereotypical of mitochondrial localization sequences) spanning amino acids 5–22, and an inability to detect h-mtTFB2 with a peptide antibody that was directed against amino acids 2–20, which would be removed by the predicted cleavage event (data not shown). No obvious mobility differences were observed between recombinant h-mtTFB1 and that isolated from cells. Furthermore, SignalP does not predict a mitochondrial cleavage site for h-mtTFB1 (data not shown). Thus, h-mtTFB2 and h-mtTFB1 are handled quite differently upon import, which is consistent with their distinct evolutionary history subsequent to the putative gene duplication event that created the two protein families early in eukaryotic lineage (,). It is also tempting to speculate that the use of different modes of import for these two transcription factors could provide a mechanism to control their relative levels in the organelle in response to different conditions. A second important observation that came from the initial analysis of the h-mtTFB1 and h-mtTFB2 over-expression lines is that there is an increase in h-mtTFB1 when h-mtTFB2 is over-expressed (B). However, the converse was not true. That is, in the h-mtTFB1 over-expression line, there is no change in the steady-state amounts of h-mtTFB2 per mitochondrion (B). Furthermore, there are no obvious changes in the levels of POLRMT or h-mtTFA per mitochondrion in either of the h-mtTFB factor over-expression lines (D). We conclude that there is some form of one-way communication between h-mtTFB2 and h-mtTFB1. The fact that this upregulation occurs, at least in part, via increased levels of the h-mtTFB1 mRNA (C), suggests that this involved a retrograde signal transduction mechanism from the mitochondria to the nucleus (). It is tempting to speculate that this response is initiated via signals generated by alterations in mitochondrial transcriptional output or the amount of 12S rRNA methylation. We next examined multiple mitochondrial parameters in the h-mtTFB1 and h-mtTFB2 over-expression cell lines. In the h-mtTFB2 over-expression cell lines (where it is important to keep in mind that there is also a compensatory increase in mtTFB1 as discussed above; B), there is a ∼2-fold increase in overall mitochondrial transcript levels as evidenced by northern analysis of the two rRNAs (12S and 16S) and two mRNAs (ND2 and ND6) encoded by mtDNA and representing transcripts derived from both strands (A). This was accompanied by a corresponding doubling of the mtDNA copy number (C). Given the documented roles for the h-mtTFB factors in directing transcription initiation efficiency (), these changes most likely represent an increase in the rate of mitochondrial transcription that is also driving a increase in transcription-primed mtDNA replication (). However, it remains a formal possibility that the increase in steady-state levels of mitochondrial transcripts is due to enhanced RNA stability and/or similar rates of transcription from the increased number of mtDNA templates. Also evident in the h-mtTFB2 over-expression lines was an increased rate of mitochondrial translation of most, but not all mitochondrial gene products (A) that, at least in the case of COX1 and COX2, results in significantly increased steady-state amounts of protein (B). However, the rates of mitochondrial translation were not uniform for all of the subunits. For example, there is apparently a reduced rate of translation of ATP6 and no apparent change in the rate of synthesis of ND3 or ND4L (A). Thus, artificially raising the levels of h-mtTFB2 (and in response, also h-mtTFB1) does increase mitochondrial gene expression and mtDNA replication, but may come at the cost of imbalanced relative rates of mitochondrial protein synthesis. We suspect that most of the described effects on mitochondrial gene expression just listed are driven primarily by the increased levels of h-mtTFB2 since no major changes in mtDNA copy number, transcript levels, translation rates or steady-state levels of COX1 and COX 2 were observed in the cell line in which h-mtTFB1 alone was over-expressed ( and ). However, it is also just as likely that up-regulation of both factors is necessary to coordinately mount all of the responses we observe. The lack of a significant ‘mitochondrial gene expression’ response in the h-mtTFB1 over-expression lines is somewhat surprising given the documented ability of this protein to activate transcription (,). One potential reason for this may be explained via the relative levels of the transcription system we established in these cells. For example, the ratio of h-mtTFA to h-mtTFB1 required for optimal transcription is ∼1:2 (). Thus, the corresponding ratio we determined of 18:1 would likely not allow much of a contribution of h-mtTFB1 to the total transcriptional output. Furthermore, to reach the optimal ratio of 1:2 would require nearly 40-fold over-expression of h-mtTFB1. We were only able to increase h-mtTFB1 by ∼10-fold, which likely explains why no alterations in mitochondrial transcripts were observed in the h-mtTFB1 over-expression cell lines (A). In contrast, the ratio of h-mtTFA to h-mtTFB2 of 5:1 () is well within the range of 2:1–40:1 that is optimal for mitochondrial transcription (). Increasing this by ∼3-fold in the h-mtTFB2 over-expression cell line would keep this ratio within this optimal range, entirely consistent with our results. Whether the relative amounts of the transcription components vary in different cell or tissue types such that h-mtTFB1 also contributes to mitochondrial transcription remains an open question. While over-expression of h-mtTFB1 does not show any major effects on the mitochondrial transcription and translation parameters we measured, it did result in increased sensitivity of HeLa cells to the aminoglycoside antibiotic kasugamycin (B). Sensitivity to this antibiotic in is modulated by dimethylation of two adenine residues in the 3′ terminal stem loop of the small subunit ribosomal RNA by KsgA (,), the bacterial homolog of h-mtTFB1 and h-mtTFB2. We have shown previously that h-mtTFB1 and h-mtTFB2 are able to methylate the homologous stem-loop in bacteria and restore sensitivity to kasugamycin (,). Thus, we interpret the ability of increased levels of h-mtTFB1 to sensitize HeLa cells to kasugamycin to indicate that h-mtTFB1 is increasing the number of methylated ribosomes that are targets for inhibition by the drug. This provides the first confirmation that h-mtTFB1 is likely the 12S rRNA methyltransferase in human mitochondria as we predicted from our earlier studies (,). Interestingly, we observed a similar sensitivity to kasugamycin in the h-mtTFB2 over-expression lines (B). While this is most likely due to the fact that h-mtTFB1 is also up-regulated in this cell line, we cannot exclude the possibility that h-mtTFB2 is also contributing to the methylation of the 12S rRNA since it too possesses rRNA methyltransferase activity, albeit at a much lower level (). In this regard it is interesting to note that others have shown that different methylation combinations on the 16S rRNA (adenines in the stem loop being monomethylated, dimethylated or mixed) result in different levels of aminoglycoside resistance (). Thus, the idea that h-mtTFB1 and h-mtTFB2 might orchestrate different states of 12S rRNA methylation with different functional outcomes is an intriguing possibility. Finally, we have elucidated a new role for h-mtTFB1 and h-mtTFB2 in mitochondrial biogenesis. Over-expression of h-mtTFB1 alone results in a ∼50% increase in mitochondrial membrane as judged by Mitotracker Green staining and confirmed by western analysis of a mitochondrial marker protein, porin (a.k.a. VDAC, ). Thus, even in the absence of a direct effect on mitochondrial transcription or translation, increased h-mtTFB1 somehow signals a change in mitochondrial biogenesis. It will be of interest to determine the nature of this signal and whether it involves, for example, a sensing of the amount of ongoing mitochondrial ribosome assembly through the rRNA methylation status. A similar increase in mitochondrial biogenesis is observed in the h-mtTFB2 over-expression cell line (), in which both h-mtTFB1 and h-mtTFB2 are elevated. However, here, unlike h-mtTFB1 over-expression alone, it was accompanied by a significant (∼80%) increase in mitochondrial membrane potential (A). Taken together, these results suggest that, while h-mtTFB1 can alone induce a mitochondrial biogenesis response, an increase in both h-mtTFB1 and h-mtTFB2 is needed to coordinate an increase in mitochondrial gene expression with the increase in mitochondrial mass. We propose that the lack of an increase in membrane potential in the h-mtTFB1 over-expression cell line (A) is a result of a defect in this response that leads to more mitochondria with fewer OXPHOS complexes per mass of organelle. This study provides the first analysis of the relative levels of the human mitochondrial transcription system and functional roles of the h-mtTFB1 and h-mtTFB2 paralogs . Our results revealed distinct, but possibly coordinated functions of each of these factors in mitochondrial gene expression, biogenesis and activity that provide a new framework for future studies aimed at understanding the regulation of human mitochondrial transcription system and its potential as a therapeutic target for human diseases and aging. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

RNA interference (RNAi) has become an extremely useful genetic tool to study gene function in mammalian cells. The discovery that short double-stranded RNAs, known as short interfering (si)RNAs, avoid an interferon response and the global shutdown of translation has enabled the wide use of transient gene silencing in cultured cells and specific tissues of mice upon local administration (). To elicit permanent gene silencing, short hairpin (sh)RNA expression vectors can be used. These vectors consist of an RNA polymerase III promoter producing short RNA fragments, which form hairpin structures. These shRNAs are processed by the RNAi machinery in the same way as linear double-stranded RNAs such that sequence-specific gene silencing occurs (). Mice transgenic for shRNA vectors produce an all-over knockdown phenotype, similar to conventional knockout mice (). To overcome the embryonic lethality of many mutants and to investigate gene functions in specific tissues or in a time dependent manner, conditional vectors have to be used. ShRNA expression can either be regulated by an inducing compound-like doxycycline acting on artificial regulatory sequences in the polymerase III promoter () or shRNA production is blocked by a transcriptional stop element that can be deleted through Cre mediated recombination. The latter Cre/loxP approach is similar to conditional knockout or knock in strategies where it is widely used. Various vector designs for Cre/loxP regulated RNAi have been described (,). Here, we show a fast and highly reproducible system to generate mice expressing shRNAs under the control of Cre recombinase. This tool can be applied to a great variety of biological questions since a large collection of mouse strains that express Cre recombinase in specific cell types is available and can be used to activate conditional shRNA vectors at different developmental stages and in selected cell types of mice (). Mitogen activated protein kinases (MAPKs), also called extracellular signal-regulated kinases (ERKs), are a group of serine/threonine terminal protein kinases evolutionarily conserved from yeast to human. They form an intracellular signaling cascade regulating fundamental cellular functions including proliferation, cell survival and differentiation (). The MAPK pathway plays also an important role in neurons as it is involved in synaptic plasticity, neuronal survival and regeneration (). A regulatory function of MAPK signaling for anxiety and depression-like behavior of adult mice has been proposed () but not tested with genetic models . Here, we show that a conditional knockdown of and and in the adult murine brain is possible with our RNAi approach, which gives the possibility to test the role of these genes in anxiety and depression-like behavior . italic #text The pSHAG plasmid containing the human U6 promoter () was opened with BseRI/BamHI and ligated with a lacZ specific shRNA oligonucleotide pair (5′-gcgttacccaacttaatcgccttggaagcttgcaaggcgattaagttgggtaacgccttttttggaaa-3′, 5′-gatctttccaaaaaaggcgttacccaacttaatcgccttgcaagcttccaaggcgattaagttgggtaacgccg-3′; targeting nt 61–85 of lacZ gene) to generate U6-shLacZ or ligated with a humanized F-Luciferase specific shRNA oligonucleotide pair (5′-tgcgctgctggtgccaacgaagcttggttggcaccagcagcgcacttttttggaaa-3′, 5′-gatctttccaaaaaagtgcgctgctggtgccaaccaagcttcgttggcaccagcagcgcacg-3′) to generate U6-shLuc. To create U6-shLacZ-loxP3, the U6-shLacZ segment was recloned into a modified pBluescript plasmid, opened with SgrAI at the transcriptional start site and ligated with an oligonucleotide pair (5′-ccggataacttcgtatagcatacattatacgaagttatatatactagtcgac-3′, 5′-ccgggtcgactagtatatataacttcgtataatgtatgctatacgaagttat-3′). For U6-shLacZ-loxP4, U6-shLacZ was opened in the shRNA loop region with HindIII and ligated with a pair of oligonucleotides (5′-agctataacttcgtatagcatacattatacgaagttatggatcc-3′, 5′-agctggatccataacttcgtataatgtatgctatacgaagttat-3′) that introduce a loxP sequence and a 5′ BamHI site. For U6-shLacZ-loxP5, U6-shLacZ was opened with HindIII and ligated with a pair of oligonucleotides (5′-agctataacttcgtatagcatacattatacgaagttat-3′, 5′-agctataacttcgtataatgtatgctatacgaagttat-3′) that inserts a loxP sequence into the shRNA loop in a symmetric fashion. To generate the conditional lacZ and Luciferase specific shRNA vectors U6-lox-lox-shluc and U6-lox-lox-shLacZ, plasmids U6-shLuc and U6-shLacZ were opened in the loop region with HindIII, the ends were filled with Klenow fragment and ligated with a 338 bp MlyI fragment serving as loxP flanked stop cassette. This MlyI fragment was cut from a pNEB193 based vector that contained a 19 bp R-Luciferase antisense and two polythymidine sequences, serving as termination signals, and that has been flanked with loxP sequences from oligonucleotide pairs that were ligated into the HindIII site (5′-agctgagtcgactgataacttcgtatagcatacattatacgaagttatggatcc-3′, 5′-agctggatccataacttcgtataatgtatgctatacgaagttatcagtcgactc-3′) and NdeI site (5′-tattttttggatccataacttcgtatagcatacattatacgaagttatgactggactc-3′, 5′-tagagtccagtcataacttcgtataatgtatgctatacgaagttatggatccaaaaaa-3′). The sequence of the 338 bp stop cassette fragment is (loxP-sites underlined): 5′-ggatccagcttggtagcgcggtgtattatactttttggaaagaattcactggccgtcgttttacaacgtcgtga ctgggaaaaccctggcgttacccaacttaatcgccttgcagcacatccccctttcgccagctggcgtaatagcgaagaggcccgcaccgatcgcccttcc caacagttgcgcagcctgaatggcgaatggcgcctgatgcggtattttctccttacgcatctgtgcggtatttcacaccgcatattttttggatcc-3′. To delete the loxP flanked stop cassette from U6-lox-lox-shluc and U6-lox-lox-shLacZ, the plasmids were transformed into Cre expressing bacteria (294cre; ) and recombined subclones (U6-loxP5-shLuc, U6-loxP5-shLacZ) were retransformed into strain DH5α. All plasmids were grown in DH5α, isolated with Qiagen plasmid Maxiprep columns and the integrity of the promoter and shRNA regions was confirmed by DNA sequencing. The pSHAG plasmid containing the human U6 promoter () was opened with BseRI/BamHI and ligated with a specific shRNA oligonucleotide pair (5′-gagaggagttacatgttgaagaagcttgttcaacatgtaactcctctccttttttggaaa-3′, 5′-gatctttccaaaaaaggagaggagttacatgttgaacaagcttcttcaacatgtaactcctctccg-3′, targeting 5′-ggagaggagttacatgttgaag-3′ in Exon 5) to generate pU6-shBraf or ligated with a specific shRNA oligonucleotide pair (5′-acggcgagatcagcatctgcatgaagcttgatgcagatgctgatctcgccgtcttttttggaaa-3′, 5′-gatctttccaaaaaagacggcgagatcagcatctgcatcaagcttcatgcagatgctgatctcgccgtcg-3′, targeting 5′-gacggcgagatcagcatctgcatg-3′ in Exon 2 and Exon 2) to generate pU6-shMek1/2. To create pU6-shBraf-flox and pU6-shMek1/2-flox, the vector pU6-shBraf or pU6-shMek1/2, respectively, was opened with HindIII in the loop region of the hairpin sequence, ends were filled with Klenow fragment, followed by ligation with the above described MlyI fragment containing the loxP-flanked stop cassette. Mouse F1 ES cells (IDG3.2) were used for transient and stable transfections. Cells were grown in DMEM medium (Gibco) containing 15% fetal calf serum, 20 mM HEPES, 1× non-essential aminoacids, 0.1 mM β-mercaptoethanol and 1.5 × 10 U/ml leukemia inhibiting factor on gelatin coated culture dishes for transient transfections or on mitomycin c treated embryonic fibroblasts for stable transfections and blastocyst injections. Transient transfections with the lacZ and Luciferase specific shRNA vectors were performed with Fugene6 transfection reagent following the manufacturer's protocol (Roche Diagnostics). One day before transfection 60 000 ES cells were plated into each well of a 24 well culture plate. 150 ng of each supercoiled plasmid, a total amount of 450 ng, were complexed with 2 µl Fugene and transfected in duplicate into each well in 500 µl growth medium. For β-Galactosidase and Luciferase measurements 150 ng of the β-Galactosidase expression vector CMVβ (Promega) were transfected together with 150 ng shRNA vector and 150 ng of the F-Luciferase expression vector pCMV-hLuc. pCMV-hLuc was generated by ligation of the CMV promoter (770 bp XhoI/EcoRI fragment) from pUHD17-1 () into the XhoI/HindIII sites of pGL3 basic (Promega) that contains a codon optimized F-Luciferase gene. The medium was changed after 24 h and the cells were lysed and chemiluminescence was detected 48 h after transfection. The preparation of lysates and measurement of β-Galactosidase activity were performed with the β-Galactosidase analysis kit (Roche Diagnostics) following the manufacturer's protocol. Fifty microliter of freshly prepared lysates were measured for 5 s in an Orion I plate luminometer (Berthold). For the detection of F-Luciferase activity, 20 µl of each lysate were mixed with 100 μl Luciferase assay buffer (25 mM glycylglycine, KHPO, 4 mM EGTA, 2 mM ATP, 1 mM DTT, 100 µM Coenzyme A, 75 µM Luciferin, pH8) and measured for 5 s in the Orion I plate luminometer. The Rosa26 gene targeting vector pRosa26.5 was assembled from Rosa26-1 () by insertion of a splice acceptor element (), the β-Galactosidase (lacZ) coding region and polyA signal from CMVβ (Promega), and the hygromycin resistance gene cassette from plasmid pgk-hygro-pA (a gift from P. Krimpenfort) into the XbaI site. The Rosa26 gene targeting vector pRosa26.9 was assembled from Rosa26-1 by insertion of a splice acceptor element, a SV40 late polyA signal, a neomycin resistance cassette from pgk-neo-pA (a gift from P. Soriano) and the conditional shRNA vector U6-lox-lox-shLacZ into the XbaI site; further details on cloning are available on request. The linearized vectors were electroporated into mouse F1-ES cells followed by selection with G418 (140 µg/ml) or hygromycin (125 U/ml) for 7 days. Resistant colonies were analyzed by Southern blotting of EcoRV digested genomic DNA using the Rosa26 5′-probe. One of the Rosa26.5 ES cell clones was further transfected with pRosa26.9 to introduce the conditional shRNA vector into the second Rosa26 locus. To activate the shRNA vector through Cre mediated recombination, double-targeted Rosa26.5/26.9 ES cells were transiently transfected with the Cre expression plasmid pCAG-cre-bpA. It was generated from pgk-cre-bpA (a gift from W. Müller) by EcoRI/PstI digestion to remove the pgk promoter and by insertion of the CAGGS promoter from CAGGS-FLPe () as a 1673 bp SalI/AlwNI fragment. Transfected subclones were expanded and analyzed by Southern blotting of BamHI digested genomic DNA for Cre mediated deletion of the stop cassette. The Rosa26 5′-probe detects a 6.5 kb band from the Rosa26.5 allele and a 5.8 kb band from the non-recombined Rosa26.9 allele, the same size is produced by the wild type locus. Cre mediated deletion can be recognized by the shift of the 5.8 kb Rosa26.9 band to a size of 9 kb since the deleted stop cassette contains a BamHI site (data not shown). Cells were washed with phosphate buffered saline (PBS) and fixed for 5 min at room temperature in 4% paraformaldehyde in PBS. Next, the cells were washed twice with PBS and incubated in X-Gal staining solution (5 mM K(Fe(CN)), 5 mM K(Fe(CN)), 2 mM MgCl, 1 mg/ml X-Gal (BioMol) in PBS) for 24 h at 37°C. Images were taken using an AxioCam HRc camera and the AxioVision program (Carl Zeiss). The cassette exchange acceptor vector pRosa26.10 was assembled from Rosa26-1 () by insertion of a splice acceptor sequence and a modified hygromycin resistance gene derived from pgk-hygro-pA (a gift from P. Krimpenfort) into the XbaI site. One 50 bp attP recognition site (5′-gtagtgccccaactggggtaacctttgagttctctcagttgggggcgtag-3′) for φC31 integrase () was placed between the pgk-promoter and the hygro-coding region. A second attP site was cloned in the same orientation downstream of the hygro-polyA region. Homologous recombination of pRosa26.10 in ES cells was achieved as reported above and hygromycin resistant colonies were analyzed for homologous recombination by Southern blotting of EcoRV digested genomic DNA using the Rosa26 5 ‘-probe (see above). Homologous recombination of the pRosa26.10 vector is indicated by a 4.5 kb band in addition to the 11.5 kb band from the wild type Rosa26 allele. The ES cell clone IDG26.10-3 was selected for further work, and the ability of these cells to colonize the germline in chimeric mice was confirmed (data not shown). The expression vector pCAG-C31Int was generated by PCR amplification of the 1900 bp coding region of φC31 integrase from phage DNA (DSM-49156 received from DSMZ—Deutsche Sammlung von Mikroorganismen und Zellkulturen GmbH, Braunschweig, Germany) with primers C31N (5′-taagtctagaccgatatgacacaaggggttgtgaccggggtg-3′) and C31C (5′-cgactctagactaaaccttcctcttcttcttaggcgccgctacgtcttccgtgccgtcctgg-3′), digestion with XbaI and cloning into the XbaI site of pNEB193. Primer C31C introduces a SV40 derived nuclear localization signal to the C-terminus of φC31 integrase as described (). Upon sequence confirmation, the C31Int coding region was cloned as PacI-PstI fragment into pNEB-CAG downstream of the CAGGS promoter resulting in pCAG-C31Int(NLS). pNEB-CAG was generated by cloning the CAGGS promoter as 1673 bp blunt ended SalI/AlwNI fragment from pCAGGS-FLPe () into the filled BamHI site of pNEB193. A bovine polyA region from pgk-neo-bpA was added into the PmeI site of pCAG-C31Int(NLS) as 300 bp XbaI/NotI fragment resulting in pCAG-C31Int(NLS)-bpA. Further details on cloning strategies are available on request. The sequences and maps of pRosa26.10 and pCAG-C31(NLS)-bpA and a protocol for cassette exchange in ES cells are available at . and shRNA constructs were cloned as NotI/EcoRV fragment from pU6-shBraf-flox or pU6-shMek-flox, respectively, into pRMCE to generate the shRNA donor vector for RMCE. The donor vector was coelectroporated with a C31Int expression vector into mouse acceptor ES-cells (clone IDG26.10-3) followed by selection with G418 for 7 days. Resistant colonies were isolated and analyzed for cassette exchange by Southern blotting of ScaI digested genomic DNA using the Rosa26 5′-probe (see above). A 9.2 kb band in addition to the 6.1 kb band from the wild type Rosa26 locus indicated correct cassette exchange, whereas the original RMCE locus gives rise to a 7.6 kb band. Positive ES cells containing the shBraf-flox or the shMek-flox construct, respectively, were injected into C57BL/6 blastocysts. Resulting chimeras were bred to C57BL/6 mice and offspring were tested for germline transmission. Heterozygous mice for the modified Rosa26 allele were crossed to CamKII-cre and Nestin-cre mice (on C57BL/6 background), respectively (,), to obtain shBraf/CamKII-cre mice and shMek1/2/Nestin-cre mice. Experiments on animals were carried out in accordance with national and institutional guidelines. Cre transgenes were genotyped by PCR using the primer pair 5′-ATGCCCAAGAAGAAGAGGAAGGT-3′ and 5′-GAAATCAGTGCGTTCGAACGCTAGA-3′ to amplify a 447 bp fragment from the gene. The shBraf and shMek1/2 constructs in the Rosa26 locus were genotyped by Southern blotting of EcoRV digested genomic DNA using the Rosa26 5′-probe (see above). ShRNA-alleles are indicated by a 5.2 kb band in addition to the 11.5 kb band of the wildtype Rosa26 allele. DNA was extracted from brain tissue with the Wizard genomic DNA extraction kit (Promega). Ten micrograms of genomic DNA were digested with BamHI, run on a 0.8% agarose gel and blotted on a nylon membrane (Hybond N, Amersham). Hybridization was performed overnight at 65°C with the Rosa26 5′-probe, labeled with P (5′-[α-P]dCTP, Amersham). After washing, the membrane was exposed to a Kodak BioMax MS film. Whole brains or half brains (with forebrain regions only, for ) were used for RNA extraction with TriReagent (Ambion). Northern blots for mRNA were performed with the NorthernMax-Gly system (Ambion). Membranes were hybridized overnight at 65°C with a P-labeled DNA probe (5′-[α-P]dCTP, Amersham) against the 3′UTR of or (Ambion). After washing, the membrane was exposed to a Kodak BioMax film. Quantification of band intensities was performed with an Imaging Plate and the FLA-8000 instrument (Fujifilm Life Science). For analysis of siRNAs, forebrain regions of shBraf/CamKII-cre and half brains of shMek1/2/Nestin-cre were used for extraction of small RNAs with the mirVana miRNA isolation kit (Ambion). Two or three micrograms of small RNA were separated on a Novex® TBE-urea gel (15%, Invitrogen). As positive control 0.05 pmol of an oligonucleotide corresponding to the antisense region of the hairpin was used (5′-ATGCAGATGCTGATCTCGCCGTC-3′ for shMek1/2 and 5′-TTCAACATGTAACTCCTCTCC-3′ for shBraf). RNA was blotted onto a nylon membrane (Hybond N, Amersham), which was hybridized at 40°C with an oligonucleotide probe (5′-GACGGCGAGATCAGCATCTGCAT-3′ for shMek1/2 and 5′-GGAGAGGAGTTACATGTTGAA-3′ for shBraf), labeled with P (5′-[γ-P]dATP, Amersham). After washing, the membrane was exposed to a Kodak BioMax film. Mice were asphyxiated with CO and perfused intracardially, after a brief rinse with ice-cold PBS, with ice-cold 4% paraformaldehyde in 0.1 M phosphate buffer (PB, pH 7.5). Brains were dissected and post-fixed for 2 h in 4% paraformaldehyde-PB. For embedding, the brains were dehydrated through an ascending ethanol scale (30%, 50%, 70%, 95%, 2× 100%, 1 h each passage), clarified in Rotihistol (Roth) 2× 45 min, equilibrated in 50% Rotihistol/50% paraffin for 1 h, and then transferred into paraffin at +65°C, 2× 1 h; they were equilibrated to room temperature in the last paraffin passage and kept at +4°C until cutting. Eight-micrometer sections were cut coronally and mounted on SuperFrost® Plus slides (Menzel-Glaeser); slides were dried overnight at 37°C and put at +4°C until proceeding with hybridization (ISH). For preparation of the riboprobe, a DNA template was prepared by RT-PCR on total RNA extracted from adult brains of C57BL/6J mice. Total RNA was extracted using Trizol (Invitrogen) following manufacturer's instructions. Reverse transcription was performed using random hexamers and SuperScriptII (Invitrogen Kit) following manufacturer's instructions. PCR amplification was performed with the following oligonucleotides: 5′-GTCTGAGAGGGAGCCTTGTG-3′ and 5′-GCCAGCATCTGAGCCTTTAG-3′ (from NCBI acc #BC051137, nt 1303–2151). After linearization with the appropriate enzyme and purification (PCR purification Kit, QIAgen), two micrograms of DNA template were used for the synthesis of radiolabeled transcript by transcription with S-UTP (Amersham). After 20 min of DNase I (Roche) treatment, the probe was purified by the RNeasy Clean up protocol (QIAgen) and measured in a scintillation counter. For hybridization, sections were dewaxed, pretreated and prehybridized as described previously (). Subsequently, they were hybridized overnight with a probe concentration of 7 × 10 c.p.m./ml at 57°C and washed at 65°C in 0.1× SSC and 0.1 mM DTT. The hybridized slides were dipped in autoradiographic emulsion (Kodak, NTB2), developed after 3 weeks, counterstained with cresyl violet, dehydrated in an ascending ethanol scale and Xylol, and lidded with DPX. Images were taken using an AxioCam MRc camera and the AxioVision program (Carl Zeiss). Detailed protocols for the preparation of radiolabeled probes and ISH procedures are given at . Total protein was extracted from brain tissue. Tissue was homogenized in RIPA buffer (50 mM Tris-HCl pH 7.4, 1% NP-40, 0.25% sodium deoxycholate, 150 mM NaCl, 1 mM EDTA, protease inhibitor), sonificated and centrifuged. Fifty micrograms protein of each sample were run on a 10% Tris-HCl gel (Biorad) and blotted on a PVDF membrane (Pall). After blocking with 4% skim milk the membrane was incubated with the first antibody (1–3 h), washed with TBST, incubated with the second horseradish-peroxidase-conjugated antibody (1 h) and washed with TBST. The detection reaction was initiated with ECL detection reagents (Amersham) and the membrane was exposed to Hyperfilm (Amersham). For quantification of band intensities ECL plus was used instead of ECL detection reagent and chemifluorescent bands were detected with the FLA-8000 fluorescent image analyzer (Fujifilm Life Science). The antibodies used were anti-β-ACTIN (AC-15, #ab6276, Abcam, 1 : 100 000, 1 h incubation), anti-MEK1 (sc-219, Santa Cruz Biotechnology, 1 : 1000, 1 h incubation), anti-MEK2 (# 610235, BD Transduction Laboratories, 1:2500, 1 h incubation) and anti-BRAF (sc-166, Santa Cruz Biotechnology, 1 : 600, 3 h incubation). To control the activity of shRNA vectors through Cre/loxP mediated recombination, their transcription must be initially aborted by a loxP flanked stop element. The shRNA vector should be activated upon Cre mediated deletion of the stop segment. Since a single 34 bp loxP site remains in the activated vector after Cre recombination, we first tested whether this additional sequence disturbs RNAi efficiency. Therefore, we used vector configurations with different positions for the remaining loxP site, where the shRNA is driven by the human U6 or H1 promoter (A and B). We found that the loxP site in all configurations of H1 promoter driven constructs strongly diminished the efficiency of gene silencing whereas the loxP site in all constructs driven by the U6 promoter did not significantly affect RNAi-mediated gene silencing (C). Thus, the U6 but not the H1 promoter is of use for Cre mediated control of shRNA expression and we chose the loxP5 configuration (A) as one effective configuration that includes the loxP sequence within the loop region. According to this design we inserted a loxP-flanked 270 bp DNA segment, which includes termination sequences and serves as transcriptional stop element, in our parental U6 driven shRNA vector. The presence of the stop element in the ‘off ’ configuration of the conditional shRNA vector allows only the transcription of the sense region that does not induce RNA interference, provided that the shRNA construct follows the order 5′-sense-loop-antisense-3′ (A). Upon Cre mediated deletion of the stop cassette, the shRNA vector acquires the configuration of loxP5 with the remaining loxP site in the loop sequence. To assess the functionality of the stop element, it was first inserted into the loop region of U6-shRNA vectors against the β-Galactosidase or F-Luciferase reporter genes (A). Upon transient cotransfection with reporter gene expression plasmids into mouse ES cells, the efficiency of the stop cassette in the ‘off ’ state and RNAi efficiency in the ‘on’ state were analyzed and compared to the non-modified parental vectors U6-shLuc and U6-shLacZ (B). The measurement of F-Luciferase and β-Galactosidase activity from transfected cells showed that the ‘off ’ versions did not induce gene silencing whereas the ‘on’ versions induced the same level of RNAi as the parental, non-conditional vectors (B). These results show that the stop element and our loxP5 configuration provide a way to control shRNA expression through Cre mediated recombination. For the genomic integration of conditional shRNA vectors, a single-copy approach is preferable since multi copy integrations could undergo unpredictable and non-functional rearrangements upon Cre mediated recombination. We tested the efficiency of a single shRNA vector copy in the Rosa26 locus () of murine ES cells with β-Galactosidase as reporter gene. We first inserted a splice acceptor sequence and the β-Galactosidase coding region by gene targeting downstream of the first exon of one Rosa26 allele of murine ES cells. These modified R26.5 ES cells express β-Galactosidase from the endogenous, ubiquitously active Rosa26 promoter. Next, the second Rosa26 allele of R26.5 ES cells was targeted with a single copy of the conditional shRNA vector U6-lox-lox-shLacZ giving rise to R26.5/R26.9 ES cells. In the double-targeted R26.5/R26.9 cells, the shRNA vector stop cassette was deleted by transient transfection with a Cre expression plasmid (A). Three ES cell clones harboring the non-activated and two clones carrying the recombined shRNA construct (R26.5/R26.9Δ cells) were isolated and analyzed for β-Galactosidase activity by histochemical staining (X-Gal) and by a chemiluminescence assay using cell lysates. X-Gal staining revealed a strong reduction of β-Galactosidase activity in R26.5/R26.9Δ ES cells as compared to non-recombined R26.5/R26.9 cells (B). The quantitative analysis revealed that the β-Galactosidase activity in R26.5/R26.9Δ cells is reduced by 90% (C). Thus, a single genomic shRNA vector copy within the Rosa26 locus is sufficient to induce RNAi in ES cells and the extent of gene silencing can reach a similar level as obtained by transient transfections (compare to B). To facilitate the insertion of shRNA vectors into the Rosa26 locus of ES cells for the fast generation of multiple knockdown mouse lines, we developed an approach for recombinase mediated cassette exchange (RMCE) using the integrase of phage φC31 (C31Int). For this purpose, ES cells with a Rosa26 acceptor allele were generated. These cells harbor a pgk promoter driven hygromycin resistance gene, of which the coding and polyA region is flanked by attP recognition sites, in the Rosa26 locus (B). The acceptor allele is designed such that the attP-flanked segment can be replaced by an attB-flanked construct from a donor plasmid (pRMCE) upon transfection of ES cells together with a C31Int expression plasmid (A). The donor vector contains a promoterless neomycin resistance coding and polyA region and the shRNA expression cassette between the attB recognition sites. The shift from hygromycin to neomycin resistance in ES cells selects for RMCE events (C), such that correct cassette exchange occurs at a frequency of 40–60% among the neomycin resistant ES cell colonies (data not shown). To study the effects of RNA interference against components of the MAPK cascade , we generated two conditional knockdown mouse lines. For the first mouse line, shBraf-flox, expressing an shRNA against , we generated six shRNA vectors and tested their knockdown efficiency by transient transfection in cell culture (data not shown). In the vector containing the most efficient shRNA sequence, we inserted the stop cassette to block shRNA transcription before Cre recombination (pbs-U6-shBraf-flox). The shRNA expression cassette of this vector was in turn inserted into the donor vector pRMCE to generate recombinant ES cells via RMCE (A). The second mouse line, shMek-flox, was generated in the same way as shBraf-flox. The shRNA sequence was chosen in a way that it targets as well as , in a region where both mRNAs show 100% homology (in Exon 2 of both genes). Due to the restricted targeting region, here only two different shRNA vectors were generated and tested for efficiency. Recombinant ES cells were injected into blastocysts to generate shBraf-flox and shMek-flox mice. Upon germline transmission, heterozygous shBraf mice were crossed to CamKII-cre mice to activate RNAi in forebrain neurons of adult mice (). Mice carrying one modified Rosa26 allele with the shMek-flox construct were crossed to Nestin-cre mice, which express Cre recombinase in neuronal and glia cell precursors, to activate RNAi in the nervous system (). All animals, containing the conditional shRNA construct or the activated shRNA together with the Cre transgene, were viable, fertile and showed no obvious phenotype. Cre recombinase under the CamKII promoter activates the shRNA only in neurons of the adult forebrain (). To show the deletion of the stop cassette in the U6-shBraf-flox element, a Southern blot analysis with genomic DNA from different brain regions was performed. A shift of the 5.4 kb band with the stop cassette to an 8.6 kb band after Cre recombination indicates activation of RNAi (A). This larger band is visible only in forebrain regions of mutant mice, like the olfactory bulb, the hippocampus, striatum, cortex and very weak in the thalamus, but not in more posterior parts of the brain of these mice or in control mice. As expected, the 5.4 kb band from the inactive shRNA allele is still visible in all the forebrain regions of mutant mice, resulting from non-neuronal cells in the forebrain expressing no Cre recombinase. Consistent with the situation on the genomic level, we could show expression of the shRNA in mutant mice, but not in control mice without Cre transgene, on a Northern blot with short RNAs (B). At the protein level, reduction of BRAF protein is visible in brain regions of mutant mice where Cre recombination occurred (C). Slight protein knockdown of BRAF in midbrain and cerebellum, where no Cre recombination on the DNA level is visible, may be due to recombination in a small fraction of cells that remains undetected by Southern blot analysis (A). Since the BRAF expression in these regions is low and restricted (), the limited Cre activity could still result in visible protein knockdown if a substantial fraction of Braf expressing neurons also coexpress Cre recombinase. No protein knockdown is visible in the brainstem. Estimation of levels of protein reduction in forebrain regions revealed a knockdown of ∼70% compared to protein expression in control mice. In contrast to the CamKII promoter, the Nestin promoter drives the expression of Cre recombinase in all nervous tissue (). To show the deletion of the stop cassette from the U6-shMek-flox vector in the whole brain of adult mice, a Southern blot was performed in the same way as for the shBraf-flox mice. Also here the band from the shRNA allele is shifted from 5.4 kb with the stop cassette to 8.6 kb after Cre recombination (A). The larger band is visible in all brain regions of mutant mice, but not in control mice. The 5.4 kb band from the inactive shRNA allele is absent in all brain regions from mutant mice, indicating complete Cre mediated recombination and therefore shRNA activation in the whole adult brain. Hence, we could show high expression of the shMek transcript in mutant mice but not in control mice on a Northern blot with short RNAs from whole brains (B). Since the shRNA of these mice is specific for both and mRNA, knockdown should occur at the mRNA and protein level of both genes. But due to the very low expression level of mRNA in the adult brain (), detection of mRNA is possible only for (C). Quantification of the Northern signal revealed a 65% reduction of mRNA in mutant brains as compared to wild type controls. the reduction of mRNA can be observed in the whole brain of mutant mice compared to control mice (D and E). On the protein level, reduction of MEK1 and MEK2 protein was observed in the whole brain of mutant mice (F). Estimation of levels of protein reduction revealed a knockdown of about ∼65% for MEK1—consistent with the mRNA knockdown—and ∼50% for MEK2 compared to control expression each. Here, we describe a novel technique to inactivate one or two related genes in the adult murine brain with RNA interference in a tissue-specific manner. We developed conditional shRNA expression vectors that can be activated upon Cre mediated recombination. After testing various configurations for the positional effect of a transcriptional stop cassette within H1 or U6 promoter driven shRNA vectors, we selected one construct with high knockdown efficiency after Cre recombination. Due to the position of the stop cassette, the loop region of the shRNA transcribed from this expression construct is elongated by 34 nt. In contrast to former studies asserting a 9 nt loop sequence being the most efficient configuration (), our elongated loop sequence does not interfere with shRNA efficiency. Essential for this technique, we could show that the insertion of a loxP flanked stop segment into the loop region of shRNA vectors disrupts RNAi induction and that such vectors can be activated through Cre mediated recombination. An obvious application for conditional RNAi are shRNA vector transgenic mice since a large collection of mouse strains that express Cre recombinase in specific cell types is available and can be used to activate conditional shRNA in different developmental stages and cell types (). For the genomic integration of such expression vectors, different approaches are available (). Transgenic RNAi mice have been generated by pronuclear injection (), by lentiviral infection () or electroporation of ES cells (). All these approaches result in random, multicopy integrants of the shRNA vector and therefore require a laborious analysis of multiple lines due to the influence of the genomic environment and the vector copy number on transgene expression (,). Apart from the time and resources needed for this initial screening of mouse lines, it is not applicable to conditional shRNA vectors since multicopy integrations could undergo unpredictable and non-functional rearrangements through Cre mediated recombination. Hence, a single-copy approach using a defined and well-characterized genomic locus is preferable. The HPRT and the Rosa26 locus are frequently used for the genomic integration of expression vectors and have also been used for shRNA vectors (). Since the HPRT locus is affected by X-inactivation in female mice, we chose the Rosa26 locus for integration of our shRNA vectors. Using a reporter gene, we showed in homologous recombinant ES cells that our conditional shRNA expressed from one vector copy in the Rosa26 locus gives rise to highly efficient gene knockdown after Cre mediated recombination, comparable to transient transfections of multiple vector copies. Thus, the Rosa26 locus allows effective and reproducible U6 promoter driven transcription of shRNA and the amount of shRNA transcribed from one single vector integrant is sufficient for efficient gene knockdown. The functionality of the U6 promoter within a defined chromosomal locus, like Rosa26, lays the basis for conditional RNAi in mice using a single vector copy that is recombined by Cre recombinase into a single, predictable product. To circumvent the laborious and inefficient homologous recombination step for mouse generation, we further used RMCE to produce ES cells harboring one copy of the shRNA expression vector in the Rosa26 locus. With this approach, the frequency of properly recombined ES cell clones among selection positive clones rises from ∼1% with homologous recombination at Rosa26 to 40–60% with RMCE. Our RNAi mice generated with these techniques harbor a conditional shRNA vector either against or and at once. These mice are viable, fertile, show no overt phenotype, and do not express the specific shRNA before Cre mediated recombination. We used two different Cre expressing mouse lines—both expressing Cre recombinase in a brain-specific manner—to activate shRNA transcription. We showed that in shBraf/CamKII-cre mice Cre mediated recombination occurs only in neurons of forebrain regions of the adult brain and in shMek/Nestin-cre mice in all cells of the adult brain, as described previously (,). Consequently, shRNA expression and reduction of mRNA and protein is detectable in these tissues from mice expressing Cre recombinase. The extent of specific mRNA and protein knockdown reaches ∼70% in case of and and ∼50% for . Important to mention here is, that shMek/Nestin-cre mice produce one shRNA targeting and at once. Although the efficiency of RNAi is not identical for both genes, we show that it is feasible to knockdown several related genes simultaneously. So, this technique will facilitate functional studies of gene families where the loss of one gene may be compensated by other family members (,). Addressing such questions with conventional knockout strategies implies an enormous effort of breeding and genotyping to obtain double or even triple knockout animals. With RNAi, generating multiple knockdown mice is not more effort as compared to single knockdown mice. Furthermore, the production of shRNA expressing mice is much faster and easier than that of knockout mice, especially with the use of RMCE. Thus, in 3–4 months even conditional knockdown mice can be generated whereas at least 12 months are required for the production of a classical unconditional knockout strain. Moreover, the effort of breeding shRNA mice is significantly decreased since only one shRNA allele is required to exert the knockdown of the targeted gene and breeding for homozygosity (as for knockout mice) is not required. In the conditional shRNA mice against MAPKs reported here, we reached knockdown levels of up to 70% after Cre mediated recombination. In the reporter gene experiments, we showed that in ES cells a higher knockdown level of up to 90% is possible with the single-copy approach in the Rosa26 locus. Using shRNAs against the (corticotropin releasing hormone receptor 1) and (leucine-rich repeat kinase 2) genes we indeed obtained knockdown efficiencies of 80 and 90%, respectively, in the adult brain (R. Kühn, unpublished data). Therefore, the level of knockdown that can be reached with the Rosa26-shRNA approach is not limited to 70% but rather depends on the intrinsic efficacy of the specific target sequence of an individual shRNA, able to elicit either higher or lower levels of gene silencing. Using public available siRNA prediction programs as paradigm for the design of five shRNA vectors for each gene, we usually find one vector that induces a knockdown of 90%. This average efficiency should improve in future with the further development of siRNA and shRNA prediction algorithms. To obtain higher knockdown levels for and the selection of new, more efficient shRNA sequences would be necessary. Another possibility to enhance the efficiency of less potent shRNA sequences may be to increase the number of active siRNAs in the cell. To test whether increased shRNA levels result in improved knockdown, we compared the residual target gene expression in single copy heterozygous and double copy homozygous shRNA mice. In the three tested mouse lines we found in homozygous shRNA mice a moderate (∼5–10%) but not dramatic increase of knockdown efficiency (C. Hitz, P. Steuber-Buchberger, unpublished data). But even a less potent protein knockdown can result in a phenotype giving interesting insights to the mode of function of the gene. For example Shalin . () showed impaired fear conditioning in mice with a 40% reduction in ERK1/2 activation. Furthermore, constitutive knockdown with our shRNA against Mek1/2, indeed, does not lead to a placental defect as described for the constitutive knockout of (), but it leads to dwarfism and death at the age of 6 weeks (data not shown). All in all, the doubling of the shRNA copy number and the use of different shRNA target sequences offer the option to exploit differences in efficiency of shRNA sequences to produce allelic series as refined models of genetic diseases beyond the ‘all or nothing’ principle of knockout mice (). We demonstrate that RNAi is a powerful tool for the generation of conditional mouse mutants to study the function of single or related genes . Recently, Seibler . () described a new approach for conditional RNAi using the tetO/tetR system. In contrast to our cell type-specific Cre/loxP regulated system the tetR regulated system is reversible and can be switched on and off optionally. However, the tetO/tetR system is not cell type specific and acts simultaneously in all organs and cell types. Since more than 150 strains of tissue specific Cre transgenic mice are available, we expect that our conditional shRNA approach can be applied to a wide range of biological questions in a variety of tissues. Here, we used this approach to generate brain-specific knockdown mice for members of the MAPK pathway to gain insight into the function of this signaling cascade that has been proposed to play a role in mood disorders (). In a first behavioral characterization of knockdown mice, we observed a contribution of these kinases to the expression of exploratory and anxiety related behavior (data not shown). Further, more detailed studies using conditional knockdown mice are in progress to establish the relation of this phenotype to human anxiety disorders.

The ribosome is a ribonucleoprotein machine that catalyzes protein synthesis (,). It consists of a small (30S in prokaryotes) and a large (50S) subunit (). The ribosome dynamically interacts with tRNAs, mRNA and translation factors and shows large-scale dynamical movements (,). There are three tRNA binding sites (A-aminoacyl, P-peptidyl and E-exit) localized across both subunits (). The small subunit binds mRNA, mediates the codon—anticodon interaction between mRNA and tRNA and is responsible for proofreading of the codon–anticodon helix. The peptide bond is formed at the peptidyl transferase center located at the large subunit. After the reaction the A- and P-site tRNAs (and mRNA) translocate by one codon to the P- and E- sites. Finally the E-site deacylated tRNA leaves the ribosome and a new aminoacyl tRNA binds to the A-site (,). There is a growing number of structural studies of subunits and the whole ribosome, such as cryo-electron microscopy (cryo-EM) (,) and X-ray crystallography studies (,). Cryo-EM (,) reveals conformational changes of particular parts of the ribosome during different phases of protein synthesis. Coarse-grained modeling (,) also provides crude estimates of the ribosome dynamics and correlation between motions of its most flexible parts (e.g. the L1 and L7/L12 stalks). These methods, however, do not provide atomic resolution insights. The X-ray studies (,) show basically static structural snapshots and some presumably flexible regions are often not resolved. Thus, many key structural and dynamical features of the ribosome remain unclear. Translocation involves movement of tRNAs and mRNA through the central cavity between the subunits and is accompanied with large movements of the L1 stalk. Recent studies reveal also high mobility of the L7/L12 stalk (, ). Translocation is accompanied with conformational changes coupled by GTP hydrolysis () which is catalyzed by translation factor EF-G (in bacteria) and triggered by its interaction with the L7/L12 stalk domain. Translation factor EF-G interacts with the rRNA portion of GTPase-associated center (rGAC; helices 43 and 44) known also as the factor binding site. This causes EF-G-induced conformational changes in the ribosome (). Here we complement the preceding studies of the L7/L12 stalk dynamics by explicit solvent molecular dynamics (MD) simulations focused on the rGAC. MD simulations have often been used to provide qualitative information about the structure and dynamics of RNA molecules (). MD simulations can be also used for fitting the low-resolution cryo-EM maps with the aim to provide better structural information. For example, recent study of L11-rGAC complex (i.e. the complete GAC) () focused on A1067 flipping and improved overlap of fitted MD structures with available cryo-EM maps. Thus, they found out possible structural candidates matching the functional L11-rGAC conformations captured by cryo-EM. It is obvious, that the MD simulation method is limited by the force field approximations and the short simulation time scale [for a review see ()]. Nevertheless, the simulations are well poised to capture, e.g. qualitative differences in the basic intrinsic dynamical flexibility of various RNA segments and motifs (). Due to the limited simulation time scale, the simulated rRNA segments remain in the ribosome-like geometries even for molecules that would unfold in solution experiments in the absence of the adjacent ribosomal parts. In simulations, before any unfolding, the molecule first extensively samples the conformational space around its ribosome-like geometry (). Simulations thus suggest regions of easily accessible conformations that are available for the motions inside the ribosome. MD simulation captures also quasiharmonic and anharmonic contributions which often are of primary importance and are not included with methods like normal mode analysis (NMA) that are based on harmonic approximation. When qualitatively addressing the RNA flexibility, the outcome of simulations is less sensitive to force field approximations compared to majority of other MD applications often dealing with quite subtle structural details. Recently, we have shown that rRNA kink-turns (K-turn; Kt) () show profound elbow-like intrinsic flexibilities around the ribosome-like geometries, without disruption of any single structural feature characteristic of a folded K-turn (). The K-turn oscillatory dynamics is pivoting at the A-minor interaction () mediating the contact between the - and , is associated with a dynamical water insertion and the motion is very anharmonic (). Anharmonic structural elements are well suited to passively mediate large motions due to their very wide and flat free energy minima. We speculated that Kt-42 of helix 42 of the large subunit could contribute to the dynamics of factor binding site (i.e. rGAC) () seen in experiments (,,). Also other studies (,) revealed K-turn flexibility. In this work, we show that the GTPase-associated center rRNA contains a second flexible region formed by the helix 42—helix 43/44 junction. The direction of preferred motion at this junction roughly coincides with the direction of the elbow-like motion of the Kt-42 and both motions preferably shift the rGAC towards or outwards the body of the subunit. The two consecutive flexible elements create a highly versatile RNA limb characterized by a complex set of bending and twisting essential dynamical modes. In other words, MD technique shows that individual rRNA building blocks have contrasting intrinsic dynamical predispositions and consecutive rRNA segments can further create molecular structures with characteristic patterns of internal flexibilities (intrinsically preferred low-energy motions). We assume that the basic physico-chemical properties of the RNA motifs as characterized by MD method can often be maintained in the RNA assemblies, and thus are worth to analyze. The starting geometry of helices 42–44 of Domain II of 23S rRNA of was taken from the X-ray structure of the 50S subunit of (PDB file 1JJ2) (). Helix 42 forms the kink-turn motif (Kt-42) () while helices 43 and 44 form the GTPase-associated center rRNA (rGAC). The whole rRNA system named as Kt-42+rGAC was simulated as single-stranded RNA molecule containing 84 nucleotides (nt); residues 1140–1223 using numbering (). The system can be roughly divided into two parts: Kt-42 with the attached stems (residues 1140–1157 and 1210–1223) and the rGAC (helices 43/44; residues 1158–1209, C). The V-shaped Kt-42 contains the canonical stem (), internal loop () and non-canonical stem (). The rGAC comprises a very complex pairing pattern described in detail in A using the standard nomenclature (). The starting structures of helices 42–44 of 23S rRNA of were taken from PDB files 2AW4 and 2AWB () (residues 1036–1119 using numbering). Simulations were carried out using the Sander module of AMBER-6.0 with the Cornell . () force field (30.5 ns of standard and 11 ns of control restrained MD—details available in Supplementary Data). The control MD simulations of (2 × 10 ns) were carried out using AMBER-8.0 (). The RNA molecules were neutralized by Na monovalent cations, initially placed using the Xleap module of AMBER at the most negative solute positions. The counterions were displaced away from the solute to improve sampling [see () for justification] and these starting coordinates are available in Supplementary Data. Na radius was 1.868 Å and well depth 0.00277 kcal/mol (). Solute molecules were solvated by a water box with periodic boundary conditions using ∼20 000 TIP3P water molecules. Prismatic water box was added around the rRNA to a depth of 16 Å () and 17 Å (). The actual size of water box was ca. 112 × 76 × 88 Å for and 115 × 86 × 79 Å for systems. Due to large motion occurring in the first 5 ns of MD and the length of the simulated molecule (see ‘Results’ section), we monitored its position inside the box after every ns of simulation. No contacts with the periodic image structures occurred in the simulations. Equilibration was carried out in the following way. First, the RNA structures were kept rigid while only solvent molecules with counterions were allowed to move. Then the RNA structures were relaxed, the systems were heated gradually from 50 to 300 K and simulations were initiated under periodic boundary conditions. A 2 fs time step was used and the PME (particle mesh Ewald) () method was employed to calculate electrostatic interactions. Structures were visualized using VMD (). Solute-solvent contacts were monitored over the entire trajectories using the Carnal and modules of AMBER. All cation binding sites with inner-shell occupancy >40% were analyzed. Histograms were calculated from the original MD data, with bin widths 0.1 Å and 1.0°. Bending and displacement of rGAC pivoting at (C) was quantified via angle between the rGAC and the Kt-42 , defined using three centers of mass: the of Kt-42 (residues 1152–1157 and 1210–1214), nucleotides involved in base triples forming the (residues 1158, 1159, 1208, 1209, 1188 and 1189) and the rest of rGAC (residues 1160–1187 and 1190–1207). Twisting and coupled shift of rGAC was described as virtual torsion angle formed by four centers of mass. The 1st one contains the of Kt-42 (residues 1142–1146 and 1217–1221). The 2nd one contains C = G pair of the type I A-minor interaction of Kt-42 (residues 1147 and 1216), the 3rd one contains its A/G pair (residues 1152 and 1214) and the 4th one comprises the rest of rGAC (residues 1161–1185 and 1191–1206). Analogous centers of mass were also defined for the structure. The essential dynamics analysis [EDA, known also as principle component analysis and related to quasiharmonic analysis (QHA)] was performed using trjconv, g_covar and g_anaeig modules of GROMACS (). EDA filters out unessential motions (noise) and decomposes the overall motion into individual modes (directions of motions), which belong to individual eigenvectors with particular eigenvalues, derived by diagonalization of the covariation coordinate matrix from the atomistic MD trajectory (,). EDA, as applied in our study, is more general compared to related QHA which, after diagonalization, would approximate the free energy surface by harmonic modes (,). EDA thus reflects the full-scale motion sampled during the simulation including harmonic, quasiharmonic and anharmonic contributions. Nevertheless, the distributions along individual modes (cf. Supplementary Figure S2) seem to be reasonably well approximated by gaussian functions which would be assumed by QHA (). The NMA, based on the harmonic approximation of potential energy surface around the minimum energy conformation, was performed using ElNemo () web server (). It allows the analytic solution of the equations of motion by diagonalizing the Hessian matrix. The eigenvectors of this matrix are the normal modes, and the eigenvalues are the squares of associated frequencies. The macromolecule movement can then be represented as a superposition of normal modes, fluctuating around a minimum energy conformation (). Instead of using the atomistic force field, the potential function is simplified with single-parameter ‘coarse grained’ Hookean potential (elastic network mode analysis—ENM) (). We standardly used cut-off radius 12 Å, force constant 10 N/m and 1 nt as the block. Note that since the spring constant is arbitrary, ENM does not predict the absolute values of the fluctuations. We carried out 31 ns MD simulation of the Kt-42+rGAC rRNA of . During the first 5 ns the molecule gradually changed its initial (X-ray) arrangement to a new stable geometry. The relaxation changes the initial position of helices 43/44 by ca. 25° with respect to helix 42 (A). After the initial transition is completed, the overall RMSD fluctuates in the range of 5–11 Å and 1–5 Å versus the starting and averaged structures, respectively (Figure S1). This initial structural transition is pivoting around the junction between helices 42 and 43/44, specifically base triples G1158 = C1209/A1188 and G1159 = C1208/A1189. We suggest that this structural transition reflects the relaxation of the system in the absence of the adjacent ribosomal elements. The initial relaxation does not lead to any changes in base pairing or isostericity of the simulated molecule. The simulated system consists of three rigid segments (one of them shows breathing, see subsequently) that are inter-linked by two flexible segments, leading to a double-elbow intrinsic dynamics (C). The first flexible segment (, ) includes nucleotides 1147–1152 and 1214–1216, i.e. the Kt-42. It shows anharmonic elbow-like oscillatory dynamics correlated with insertion of long-residency waters into its A-minor type I base pair (C1147 = G1216/A1152) between the - and . The Kt-42 dynamics is described elsewhere (,) while the present article is aimed at description of the rGAC helices 43/44, the junction between helices 42 and 43/44 and coupling of all motions. is localized at the junction between of Kt-42 and helices 43/44 (residues 1158–1160, 1207–1209, 1186 and 1188–1190, ) and is responsible for the initial shift of the during the first 5 ns (A). After the initial relaxation it shows substantial oscillations around its averaged geometry, preferably in direction back towards or further away from the starting structure but with a smaller amplitude (D). The most rigid segment is the of Kt-42 (1140–1146 and 1217–1223, with internal RMSD of ca. 1.0 Å). The other rigid segment is the arm of Kt-42 (1153–1157 and 1210–1213) with RMSD 1.7 ± 0.5 Å. The compact helices 43/44 of the rGAC (1161–1185 and 1191–1206) do not contribute to the global motion but show internal breathing (see subsequently) with internal RMSD 1.9 ± 0.8 Å. The EDA finds essential motions occurring during the simulation (), disregarding the first 5 ns. The ratio of EDA eigenvalues of modes 1–4 is ca 1.0: 0.20: 0.18: 0.12 (plots of displacement along the individual eigenmodes are given in Supplementary Data, Figure S2). The first mode (ca. 60% of the overall motion) corresponds to the hinge-like oscillatory global motion of the upper part of the structure with respect to the at the base of helix 42 with a range ∼20 Å (B) and pivoting around the Kt-42 (). The second mode represents internal breathing of the compactly folded rGAC not associated with displacements of distant parts of the simulated system (C). It can be described as reversible expansions and compactions of the within the range of ca. 5 Å (Figure S3). The third mode (D) is oscillation around the relaxed geometry of the helix 42–helices 43/44 junction in direction of the initial structural rearrangement. Thus, the initial position of the in the crystal structure can be interpreted as a large amplitude deflection along the EDA mode 3. Interestingly, this direction roughly coincides with the preferred direction of the elbow-like motion of the Kt-42. Thus, the overall flexibility is highly anisotropic, with both hinges shifting the rGAC either towards or outwards the body of the subunit. The mode 3 oscillations produce movement of rGAC on a scale of ∼10 Å (displacement of U1170(P)) due to ∼10° oscillation of the angle between rGAC and Kt-42’s (see ‘Materials and Methods’ section). The fourth mode represents twisting fluctuations of the rGAC with respect to the helix 42 (E) due to combined twisting around the of Kt-42 and the internal twisting of Kt-42. Note that although the Kt-42 is a genuine elbow-like element, it is also associated with non-negligible twisting components of its low-energy modes, cf. Figures S10–11 in Razga (). Mode 4 is associated with ∼13 Å range of U1170(P) atom fluctuations and ∼25° range of fluctuations of the fictive dihedral angle between the of rGAC and the of helix 42 (see subsequently). The junction between helix 42 and helix 43/44 comprises the bases localized at the border between the extended Kt-42 and the rGAC, namely base triads G1158 = C1209/A1188 and G1159 = C1208/A1189 and tetrad A1207/G1160 … G1190/C1186 ( and ). This represents the pivoting point of the initial displacement of rGAC. Then it becomes the center of the directional (anisotropic) oscillations (D) of the angle between the upper part of helix 42 and rGAC. The intrinsic bendability is not localized as in case of Kt-42, where the motion is pivoting around a single H-bond of the A-minor interaction (). For , the structural dynamics of the A-minor interaction, triad and tetrad (see Supplementary Data) are not coupled with either the initial displacement of rGAC or the subsequent EDA mode 3 dynamics. The movement of rGAC (the initial displacement and the EDA mode 3) correlates with change of major groove width of the RNA duplex formed by the upper part of helix 42 and bottom of rGAC. Spontaneous initial straightening of this rRNA duplex from its bent form (X-ray) to straight canonical-like form is indicated by increase of the following inter-phosphate distances: G1210(P)-A1152(P) from 9.5 Å (X-ray) to 16.0 Å (relaxed MD), C1209(P)-C1153(P) from 9.2 Å to 19.1 Å, C1208(P)-A1154(P) from 12.4 Å to 22.9 Å and A1207(P)-G1155(P) from 11.3 Å to 23.0 Å (Figure S4). The 11 ns restrained MD simulation confirms (see Supplementary Data) that motion around is fully independent of the dynamics of the triads in this region and the initial displacement of rGAC can be best described as a spontaneous straightening of an initially bent duplex to a straight conformation. Additionally, two free MD simulations of corresponding rRNA segment in structures reveal similar initial dynamical behavior of the position. Full details about the complex pairing patterns and rich local dynamics at are present in Supplementary Data. We compared the sequence, base pairing and 3D-architecture of the region in crystal structures of large ribosomal subunits of , and (codes 1JJ2, 1NKW, 2AW4 and 2AWB). We also examined sequence alignments obtained from the latest release of the European Ribosomal Database () and isostericity analysis was carried out by Ribostral (). The data show that the 3D-structure of this region is universally conserved in all organisms even if individual nucleotides are not the same. Full details are given in Supplementary Data. The X-ray study of 70S ribosome () reveals modest conformational changes in the position of the rGAC. The authors concluded that there is a twisting of the of Kt-42 with no structural change in the K-turn itself. However, we suggest that the two X-ray structures of (2AW4 and 2AWB) could within the limits of the resolution equally well reflect the dynamics of the A-minor interaction of the Kt-42. As shown earlier, the elbow-like flexibility of Kt-42 ( involves also substantial twisting motions, see Supplementary Data in Razga . (). Thus, some Kt-42 twisting can easily occur even without a visible elbow-like bending. The key inter-atomic distance C1043(O2′)-A1048(O2′) in the dynamical A-minor type I interaction of the Kt-42 is 2.6 and 3.8 Å in the 2AW4 and 2AWB structures while the corresponding C1147(O2′)-A1152(O2′) distance in structure (1JJ2) is 3.0 Å. The range of dynamics of this distance in simulations is 2.6–4.0 Å. The dynamical inter-phosphate distance C1043(P)-A1048(P) is 13.4 and 14.4 Å in 2AW4 and 2AWB structures. The corresponding C1147(P)-A1152(P) distance is 14.7 Å in and 13.6–16.7 Å in simulations. Thus, if these X-ray distances are accurate enough, they match the typical internal K-turn dynamics (). Similarly, the virtual torsion angle describing the overall helix 42–44 twisting (see ‘Materials and Methods’ Section and Figures S5 and S6) is −59° and −83° in the 2AW4 and 2AWB structures and −67° in the crystal structure. MD reveals broad distribution of this angle with peaks around ca. −65° and −90° (Figure S6), similar to what is seen in the X-ray structures. In contrast, RMSD of 0.7 Å characterizes the base of the helix 42 containing the of Kt-42 for the two structures when residues 1036–1041 and 1114–1119 are overlaid. Comparing structures of and the helical properties of the base of helix 42 look different (RMSD = 1.15 Å). However, sequence contains two non-canonical WC/WC base pairs (A1039/G1116 and A1040/G1115) instead of the standard base pairs in the . This may contribute to different helical parameters and mechanical properties of this region (), and thus detailed comparison of and regions is not straightforward. We, therefore, tentatively suggest that the modest X-ray variability of the rGAC positions can be reasonably explained as a combination of twisting around of Kt-42 and twisting component of Kt-42 dynamics. It resembles the EDA mode 4 analyzed above, with participation of motion of the stemming from . To investigate the substantial internal breathing of rGAC (helix 43/44, EDA mode 2) we carried out detailed analysis of the dynamics of all base pairs. The non-canonical interactions reveal considerable structural dynamics and primarily contribute to the internal breathing of the rGAC. Many rGAC non-Watson–Crick interactions are mediated by complex long-residency hydration sites and we also identified several sites with a substantial occupation by monovalent cations. We noticed a dynamical A1192 … C1182 interaction involving base-sugar A1192(C8)-C1182(O2′) H-bond, A1192(C2′)-C1182(O2) and A1192(C3′)-C1182(O2) contacts, and some sugar–sugar contacts (C and S7). This interaction is seen as planar in the crystal structure but prefers a perpendicular orientation of bases in the simulation (Figure S8). An analogous tertiary contact is also seen in the between G1160 and G1190, with bifurcated H-bonds between O2′ of G1160 and O2′ and C8 of G1190. This interaction remains coplanar in the simulations. This base pairing pattern does not belong to any characterized types of base pairs according to Leontis classification and may represent a new recurrent type of tertiary interactions. This is supported by structural motif search in available ribosome structures () which in addition reveals that this type of interaction is frequently found as a perpendicular contact in the ribosome. Full details about this interaction and about structural dynamics of all other interactions in the helix 43/44 area are given in Supplementary Data. The motion of the two in the Kt42+rGAC region is highly anisotropic and actually both hinges have preferable motions in the direction towards and outwards the body of the subunit. Variation of the length in any of the stems of helix 42, on which the rGAC region is based, would considerably change the direction of flexibility. This is not likely to occur if the motions are of biological significance. We thus investigated the stem lengths of helix 42 in available sequence alignments (), to see if they are conserved throughout evolution. The area studied corresponds to numbers 1030–1055 (5′ strand) and 1104–1124 (3′ strand). After filtering out repeated sequences and obvious alignment or sequencing errors, we ended up with an alignment that includes 34 archaeal, 751 bacterial and 154 eukaryal unique sequences. With the help of Ribostral () we found out that the length of helix 42 is completely conserved across all three domains. The only exception was found in about 15% of bacteria that have the 3′ strand longer by 3–5 nt compared to the remaining sequences. This increase in the 3′ strand length is not complemented by any change in the 5′ strand length, and occurs near the edges of the kink turn internal loop area (Figure S9). Therefore, the extra nucleotides do not seem to lengthen the helix, and could be probably bulged out near the internal loop area without causing any major change in the overall structure and flexibility. If we compare the conservation of length of helix 42 to that of other large ribosomal subunit helices in the three domains of life, we conclude that helix 42 is among the ∼50% most conserved in length (Figure S10). Even more striking is that when the two organelles, mitochondria and chloroplasts, are considered, helix 42 is among ∼30% of the most length-conserved helices (Figure S10). The two organelles are under substantially different evolutionary pressure than members of the three domains of life, and thus the subunits are smaller in size, with some helices shortened or even missing. Helix 42 has, therefore, been suggested to be an integral part of a minimal functional ribosome (). Coarse-grained methods were used earlier to evaluate the dynamics of ribosome (,). Such calculations typically perform a normal mode analysis (NMA) within the elastic network mode (ENM) approximation. ENM uses a simplified potential to create a network of harmonic springs that connects atoms or pseudoatoms within a given cut-off distance. It works well for, e.g. globular proteins where interatomic interactions are quite homogenous. It is less likely to work for something as specific and non-globular as the stalk elements of ribosome. We performed the ENM NMA analysis on Kt-42+rGAC RNA system (starting from the X-ray structure) using ElNemo () web server () and analyzed the four lowest-frequency modes of Kt-42+rGAC. The NMA (NME) mode 1 with frequency (f) equal to 1 and collectivity (c) equal to 0.5467 () represents the bending motion with the origin in the area. This motion is similar to the initial displacement of the (observed during the first 5 ns) or EDA mode 3 (A and D). The mode 2 (f = 1.65 and c = 0.6994) represents the overall twisting of the rGAC with respect to the helix 42, stemming from a combined twisting around the of Kt-42 and the internal twisting of Kt-42, similar to our EDA mode 4 (E). The mode 3 (f = 2.15 and c = 0.3350) represents mainly the internal twisting of Kt-42. The mode 4 (f = 3.16 and c = 0.1754) shows some breathing of the Kt-42’s and is already rather insignificant. In conclusion, the ENM ElNemo analysis captures only two out of four dominant motions of the Kt-42 + rGAC system observed in MD. Most importantly, the key elbow-like dynamics of the Kt-42 appears to be missed and also the breathing of the (EDA mode 2) is not observed. Another assessment of the applicability of normal mode type of calculations applied to RNA can be found in the literature () and for additional information see also (). Our ENM results are given in Supplementary Data. In conclusion, the atomistic simulations with explicit inclusion of solvent followed by EDA have no alternative for the present system, where we need to describe details of H-bonding dynamics including the dynamical water insertion. Further, we deal with a system, which has a very broad free energy minimum, and thus is anharmonic. The present simulations were carried out in presence of a net-neutralizing set of Na ions. This can be justified in the following way. The non-polarizable pair additive force field relies on a quite primitive approximation, representing the ions as simple van der Waals spheres with atom-centered point charges of +1 or + 2. Due to this crude approximation, the force field is unlikely to exactly mimic the ‘experimental’ ion conditions and it is quite justified to run the simulations using the net-neutralizing set of monovalent ions only. A meaningful description of divalent cations is fairly outside the applicability of the force field while sampling of divalent cations in simulations is terribly insufficient. Therefore, inclusion of divalent ions into nucleic acid simulations may cause artifacts and basically is not advised. Fortunately, the simulations are usually too short to exhibit instabilities stemming from inexact salt conditions. We have shown recently that the K-turn dynamics is independent of the type of ions used in simulation (). Further discussion of the ion issue can be found in our recent papers (,,). We carried out 31 ns MD simulation on Kt-42+rGAC rRNA (complete rRNA helices 42–44 including kink turn 42 and GTPase-associated center rRNA), starting from the X-ray structure of the 50S subunit of (). Simulation was supplemented by restrained control simulation, control simulations of helix 42–44 rRNA of X-ray structure, and sequence, isostericity and motif search analyses. The aim was to highlight the basic intrinsic dynamical flexibility of this rRNA segment in the ribosome-like geometry. The simulation trajectory can be divided into two parts. During the first 5 ns, we observed a smooth initial rearrangement which changes the initial position of helices 43/44 by ca. 25° with respect to helix 42 (A). This initial rearrangement brings no visible local changes in the base pairing and isostericity of the interactions. The structural transition is roughly pivoting around the junction between helix 42 and helices 43/44, specifically around the type II A-minor base triple G1158 = C1209/A1188 and its neighboring triad G1159 = C1208/A1189. It can be best described as straightening of rRNA duplex from its bent geometry (X-ray) to straight, canonical-like RNA. We suggest that some other component of the ribosome (this area is substantially disordered in the X-ray structures) may push the rGAC towards the large subunit. This simulation in any case clearly indicates that the junction between helices 42 and 43/44 is easily deformable. After the initial transition is completed, the simulated molecule shows no further structural development, but exhibits profound stochastic fluctuations revealing that this rRNA region possesses a unique internal flexibility. The Kt-42+rGAC rRNA consists of three rather rigid (one of them internally ‘breathing’) segments linked by two flexible ones. The basically rigid segments are the - and of Kt-42 and the compact helices 43/44 of the rGAC. The first flexible segment () is the region localized between - and of Kt-42. The Kt-42 shows anharmonic elbow-like oscillatory dynamics correlated with insertion of long-residency waters into its A-minor type I base pair between the - and (). is localized at the junction between the of Kt-42 and helices 43/44 (). Bendability of this region is, however, not localized, in contrast to Kt-42. is responsible for the initial shift of the (helices 43 and 44) (A). After the initial relaxation it shows substantial oscillations around its averaged geometry, preferably in direction towards or away from the starting structure but with smaller amplitude. The intrinsic flexibility of the helix 42–44 23S rRNA segment is visualized by EDA filtering out unessential motions and noise. EDA (disregarding the initial displacement of the of the GTPase-associated center rRNA) reveals several leading motions. The first EDA mode represents ca. 60% of the overall motion and corresponds to the hinge-like oscillatory global motion of the GTPase-associated center rRNA with respect to the at the base of helix 42, stemming from the Kt-42. There are three additional modes that substantially contribute to the dynamics. The second mode represents internal breathing of the compactly folded rGAC. The third mode represents oscillations (around the relaxed geometry) of the rGAC at the junction between helix 42 and helices 43/44 (D) in direction of the initial structural rearrangement. Thus, the initial position of the in the crystal structure can be interpreted as a large amplitude deflection along the EDA mode 3. Interestingly, this direction also coincides with the preferred direction of the elbow-like motion of the Kt-42. The dynamics does not stem from any specific oscillation of base pairs, triples or quadruples but correlates with changes of major groove width of the RNA duplex formed by the upper part of helix 42 and helix 43. The fourth mode represents twisting fluctuations of rGAC with respect to the helix 42 stemming from combined twisting around the of Kt-42 and the internal twisting of Kt-42. The simulation results were compared with NMA using the ENM approximation that is commonly used for coarse-grained modeling. As expected, the leading EDA modes 1–2 were missed by the ENM approach. Note that these include mainly the elbow-like dynamics of the Kt-42. The present system therefore requires full-scale atomistic explicit solvent simulation to be properly described. The dynamics should be interpreted in the following way. The free MD simulation samples spontaneously the available space of low-energy geometries of the studied system. In other words all geometries that are significantly populated in such simulation are intrinsically very easily accessible for the studied RNA. The helix 42–44 region is thus an internally highly flexible (deformable) and anisotropic RNA modulus with preferred motions towards and outwards the 50S subunit, localized at two hinges and complemented by twisting motions. The motion of the two in the Kt42+rGAC region is highly anisotropic. Variation of the length of stems of helix 42 would considerably change the direction of flexibility. This is not likely to occur if the motions are of biological significance. Indeed, we found that helix 42 length is entirely conserved and when mitochondria and chloroplasts are considered, helix 42 is among ∼30% of the most length-conserved helices (Figure S10). Although there are many conceivable structural and functional constraints that would require essential helices to be conserved in length, such as preservation of local or tertiary interactions (), it is likely that helix 42 has the additional constraint of housing two important hinges whose movement would be asynchronized and re-directed, if helix length is changed. Base pair substitutions do occur in the helix 42 stems and may subtly modify the directionality of motions (). The entire conservation of the length of both rigid arms flanking the Kt-42 indicates that the profound anharmonic and anisotropic double-elbow flexibility of the Kt-42+rGAC rRNA segment may be important in tRNA selection and translocation. For example, it may be involved in positioning the L7/L12 stalk with respect to the 50S ribosomal subunit. The flexible rRNA Kt-42+rGAC segment is in a close contact with the highly dynamical L7/L12 complex, whose exact position is yet to be determined. The Kt-42+rGAC rRNA segment flexibility is very different from the anisotropic spring-like stiffness of the tRNA (), since the Kt-42 is a highly anharmonic element suitable to act as a passive and adjustable elbow to mediate motions of the surrounding structural elements. Analysis of the sequence conservation and isostericity of the junction between the helix 42 and the rGAC shows that all observed base substitutions appear to fully keep conserved isosteric structure of this region. We, however, made two interesting observations. The second triple in the junction region (G1159 = C1208/A1189 in , G1066 = C1115/A1096 in and G1055 = C1104/A1085 in forms a typical A-minor interaction in the latter two species. In , however, the adenine is flipped to and forms a rather unexpected Hoogsteen (H)/Sugar edge (SE) interaction between A1189 and G1159 (Figure S11). Since the overall compactness of the triad is almost the same for both arrangements, both geometries are compatible with the sequence analysis. There is a tertiary contact in the adjacent tetrad between G1160 and G1190, with bifurcated H-bonds between O2′ of G1160 and O2′ and C8 of G1190. This is replaced by identical G … A contacts in both bacterial organisms while sequences belonging to the three domains are GA, AG, GG and AA. There is an almost identical base pair also in U1164–A1192/C1182 base triad between A1192 and C1182 (C). This base pairing pattern does not belong to any characterized types of base pairs () and may represent a new recurrent type of tertiary interactions in ribosome, as supported by the structural motif search () (Figure S8, Table S1). The compact region () formed by helices 43 and 44 contains an intricate set of base pairing patterns including triads and tetrads. Many of these interactions are very dynamical, conferring a substantial structural plasticity to the shape of the rGAC. The shows visible inflation/deflation dynamics and also the local RNA structure on its surface is variable. The structural dynamics of this region is intimately associated with long-resideny hydration and cation binding sites (,,). It forms a typical family C three-way junction with extensive interactions between the P1 and P3 stems and between J31 and the shallow groove of P2 (). The ‘breathing’ described above originates in the dynamics of the P1/P3 inter-stem interactions and the bending is localized in the area of J31/P2 interactions. The recent crystallographic study of 70S ribosome () indicates modest conformational changes in the position of the rGAC in two independent structures of the ribosome. The structural difference was attributed to twist of the region below the Kt-42 at the base of helix 42. However, we also found a 1.2 Å variability of the key inter-atomic distance C1043(O2′)-A1048(O2′) in the dynamical A-minor type I interaction of the Kt-42 (corresponding to the C1147(O2′)-A1152(O2′) distance in structures). This indicates that the crystal data could also by interpreted assuming the flexibility of the Kt-42, because its dynamics is pivoting around this inter-atomic distance. Since the K-turn low-energy conformational space has non-negligible twisting components, K-turn twisting motions could occur without any substantial bending, and can be combined with twisting motions of the of the rGAC (E). In summary, we demonstrate that rRNA building blocks posse contrasting intrinsic flexibilities (flexibility signatures) and can be combined to form larger architectures with complex patterns of preferred low-energy motions and geometries. MD simulation technique appears to be particularly suitable to capture the qualitative differences in intrinsic flexibilities of rRNA building blocks since it captures atomic resolution dynamics while the molecules do not unfold on the simulation time scale away from the ribosomal geometries. In contrast to, e.g. the NMA method the MD simulations include the key anharmonic contributions. We suggest that the basic intrinsic physico-chemical properties of the RNA motifs can in many cases be maintained in the RNA assemblies, and thus are worth to analyze. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Alternative splicing of pre-mRNAs is recognized as the most important source of protein diversity in vertebrates (,) and defective regulation of pre-mRNA splicing has been identified as cause of several genetic diseases and various forms of cancer (). Alternative splicing is regulated by the presence of enhancer/silencer elements, the strength of splicing signals and additional presence of protein factors. In addition, the structure and conformation of the pre-mRNA also has an influence on the efficiency of splicing (). This is demonstrated by the recently identified riboswitches in the genomes of eukaryotes (). Riboswitches are regulatory elements which can adopt a defined structure to directly sense a metabolite. Ligand-binding then leads to changes in the conformation which influences gene expression. In contrast to bacterial riboswitches, which mainly interfere with transcription termination or translation initiation, eukaryotic riboswitches preferentially seem to tackle mRNA processing steps (). A recent report unravels the mechanisms of riboswitch-controlled alternative splicing in the filamentous fungi The expression of genes involved in thiamine pyrophosphate metabolism is regulated by riboswitches which are located in introns in the 5′untranslated region (5′UTR) (). They restructure upon metabolite-binding therewith forcing alternative splice site (SS) usage. These findings not only expand the scope of gene regulation by direct RNA ligand interaction, but also demonstrate that eukaryotic cells also employ riboswitches to control certain metabolic pathways by targeting pre-mRNA splicing. These findings prompted us to develop a synthetic riboswitch able to control pre-mRNA splicing in yeast. For that, a small molecule-binding, selected RNA aptamer has been used. Aptamers display high binding affinity and specificity and adopt a unique conformation only upon ligand-binding with the ligand becoming an integral part of the complex (). This inducible conformational change has already been used to develop conditional gene regulation systems. Inserting an aptamer into the 5′UTR of a eukaryotic mRNA led to interference of the aptamer-ligand complex with initial stages of translation initiation (). We have identified a tetracycline (tc)-binding aptamer which confers tc-dependent gene regulation in yeast (). The tc-aptamer complex inhibits the small ribosomal subunit joining when the aptamer is placed close to the cap structure and interferes with formation of the 80S ribosome when inserted directly in front of the start codon, probably by blocking scanning (). We describe here a conditional gene expression system , which exploits tc-aptamer mediated inhibition of pre-mRNA splicing (). This expands the applicability of the tc-binding aptamer to regulate gene expression. In addition, we significantly enhanced the efficiency of tc-aptamer based regulation by combining the regulatory elements that influence both translation initiation and pre-mRNA splicing. We used the yeast 2 µ plasmid pWHE601 to constitutively express the gene from an promoter (). The actin- and the U3-intron were PCR amplified and inserted into a NheI restriction site directly downstream of the start codon. The resulting vectors were named pWH601_A and _U, respectively. To allow insertion of aptamer sequences at different intron positions, unique restriction sites for Acc65I and Bsu36I were introduced by PCR mutagenesis, thereby deleting the first NheI restriction site. A schematically displays the positions of the respective restriction sites. For aptamer insertion, vectors were digested either with AflII/Acc65I or Bsu36I/ClaI. Double-aptamer constructs were generated by inserting the complete minimer-containing actin-intron either into the NheI restriction site or next to nucleotide 142 into either pWHE601AN32 () or pWHE601_A e min via PCR mutagenesis. Primer and vector sequences are available upon request. For all measurements, strain RS453α was transformed according to the protocol supplied with the frozen EASY yeast transformation II kit (Zymo Research, Orange, CA, USA). Yeast cells transformed with the respective constructs were grown at 28°C for 48 h in minimal medium [0.2% (w/v) yeast nitrogen base, 0.55% ammonium sulfate, 2% (w/v) glucose, 12 µg/ml adenine, MEM amino acids, Gibco BRL] in the absence or presence of 250 µM tc in a final volume of 5 ml. Cells were harvested by centrifugation and re-suspended in 2 ml phosphate-buffered saline (PBS). For each construct, three independently grown cultures were analyzed. Fluorescence measurements were carried out at 25°C on a Fluorolog FL3-22 (SPEX. Industries Inc.) with the excitation wavelength set to 484 nm and an emission wavelength of 510 nm. Additionally, we determined the optical density (oD) at 600 nm to ensure homogeneous cell growth. The vector pVTU102 without gene was analyzed in parallel as a blank and its value subtracted from all data. All measurements were repeated at least twice. Yeast cells transformed with the respective constructs were grown overnight at 28°C in minimal medium in the absence or presence of 250 µM tc. Total RNA was isolated according to the protocol supplied with the NucleoSpin RNA II kit (Macherey-Nagel, Düren, Germany). RT–PCR was performed according to manufacturer′s protocol using the SuperScript ™ III One-Step RT–PCR kit (Invitrogen, Karlsruhe, Germany) and analyzed on 3% agarose gels. -cDNA was specifically amplified using primers adh_in (CAACTCCAAGCTAGATCTC) and GFP_rev_Julia (CCACTGACAGAAAATTTGTGC). 3-cDNA, which is encoded as auxotrophy marker on all plasmids, was used as internal control and amplified using primers URA3_in (CAGCCTGCTTTTCTGTAACG) and URA3_out (GGAAGAGATGAAGGTTACG). An RNA construct used for in-line probing was transcribed from PCR-amplified DNA, dephosphorylated, and 5′-P-labeled as previously described (). In-line probing was performed as described (). In detail, 5′-P-labeled RNA (100 pM) was incubated for 70 h at 25°C in 50 mM Tris–HCl pH 8.3, 20 mM MgCl and 100 mM KCl containing a tc concentration defined for each reaction. After incubation, 10 µl of 10 M urea was added and the products were separated using denaturating 10% polyacrylamide gel electrophoresis. Gels were dried and analyzed using a Storm Phosphoimager (GE Healthcare). The concentration of ligand needed to cause half-maximal modulation of spontaneous cleavage yields the K for the RNA-ligand interaction. The introns of the actin (A) and U3 (U) genes from yeast were inserted directly downstream of the start codon of a constitutively expressed gene resulting in the plasmids pWH601_A and _U, respectively (A). GFP fluorescence only occurs when the mRNA has been correctly spliced. Unspliced mRNA should not be exported. If it does escape from the nucleus, several premature stop codons present in the intron sequence would either lead to rapid mRNA degradation via the nonsense-mediated decay pathway or to translation of a truncated, non-functional protein. We then introduced a set of unique restriction sites into the introns (A) to allow easy insertion of the tc-aptamer next to the 5′ SS and the branch point (BP). Furthermore, we varied the distance between the aptamer and the 5′SS (shown in as positions a–e) or to the BP (positions f, g). Inhibition of splicing by the tc-aptamer complex can be monitored by a decrease in GFP fluorescence. Yeast cells were transformed with the respective constructs and the expression of GFP was measured in the presence and absence of 250 µM tc ( and ). The aptamer responds in a dose-dependent manner to tc with maximum repression level of 250 µM (). Insertion of the tc-aptamer two nucleotides downstream of the 5′ SS (position a) leads to a negligible 1.3-fold decrease in GFP expression. Shifting the aptamer closer towards the 5′SS increases regulation to 2.5- and 4.3-fold for positions b and c, respectively. Next we attempted to increase the efficiency of regulation by varying aptamer stability as it had already been shown for translational regulation (). Therefore, we constructed P1 stems with six and nine base pairs. This was accompanied by further decreasing the distance between aptamer and 5′SS to include the complete 5′consensus sequence into P1 (, positions d and e). Both constructs increase regulation to 5.5- and 6.2-fold, respectively. To further analyze the effect of stability on regulation, we replaced the aptamer by a minimer variant in which its complete head region is replaced by a GAAA tetraloop [C ()]. The minimer exhibits the same K as the full-length aptamer and is proposed to adopt a more compact and stable conformation () and indeed, regulation was enhanced to 16-fold. Aptamer insertion close to the BP leads to negligible regulation, but caused a drastic decrease in fluorescence, already in the absence of tc. Regulation was also detected with a minimer inserted into the U3 intron at the 5′SS. We observed only slightly changed regulatory properties, the aptamer sequence itself exerted less influence on basal expression of GFP in the absence of tc, however, regulation was also reduced slightly. This indicates that the aptamer-mediated regulation is universal and is not dependent on a specific intron. Control constructs either lacking an aptamer or containing an aptamer mutant A9U unable to bind tc show no regulation at all [A: mut, ()]. These findings indicate that the regulatory effect we observe is indeed tc-aptamer mediated. To verify that the tc-mediated decrease in fluorescence is due to altered pre-mRNA splicing we performed RT–PCR. shows that the amount of spliced mRNA decreases in the presence of tc [exemplarily shown for an aptamer at position a (), position c () and for the minimer at position e ()], whereas control constructs show the same ratio of spliced and unspliced mRNA in the absence and presence of tc [pWH601_A without inserted aptamer () or with the mutant aptamer A9U ()]. A mutation of the consensus sequence AUGU leads to tc-independent inhibition of splicing. In addition, mRNA as internal control shows no influence of tc on its expression for all constructs tested (, six lanes at the right side of both blots). Previous studies showed that riboswitch activity of the tc-aptamer can inhibit different steps of translation initiation. It interferes with initial binding of the 43S subunit when located close to the cap structure but also inhibits 80S formation when inserted at a more downstream position, probably by impeding the scanning ribosome (). Thereby, a 6-fold regulation was obtained when inserting the aptamer directly adjacent to the start codon. We speculated that combining translation with splicing control might further increase regulation. We used the intron construct with the highest regulation factor (minimer at position e with the 9 nt long stem) and combined it with an aptamer inserted directly in front of the start codon resulting in D1 (). The construct shows a slightly reduced regulation compared to the single intron construct, however the two aptamers are separated by only 12 nt so that an interaction cannot be excluded. Therefore, we constructed a gene with an intron at a position further downstream (at amino acid 47) resulting in D2. Both double constructs result in a remarkable increase in regulation (11- and 29-fold, respectively), however, regulation is accompanied by a more pronounced drop in basal activity. We then inserted a second aptamer-containing intron at distances of either 9 or 141 nt downstream of the 3′SS of the first intron resulting in the double-intron constructs D3 and D4 (). Both constructs lead to an increase in regulation with factors of 20- and 32-fold. Interestingly, the construct with the widely separated introns shows a less decrease in overall activity. Taken together, construct D4 containing two independently regulated introns has a regulatory efficiency of 32-fold which is the highest reported so far for aptamer-mediated conditional regulation of gene expression. Previous studies indicated that the tc-aptamer has a pre-formed structure with the two single-stranded regions (B1-2 and L3) responsible for ligand-binding and the stem structures P1 and P2 forming the scaffold of the aptamer and proposed to be already present in the absence of tc. Since the 5′SS has to be accessible for efficient splicing, it was astonishing that the construct with the SS completely buried within P1 is still active. To analyze tc-mediated changes in the aptamer structure with a special focus on P1, we performed in-line probing of the minimer within the mRNA context. This method allows analysis of changes in the pattern of spontaneous RNA cleavage that occur upon ligand association (). The aptamer exhibits substantial changes in spontaneous cleavage at several locations in response to increasing tc concentrations (). These tc-induced structural modulations mainly occur in regions proposed to be involved in ligand-binding (B1-2 and L3) and fit the probing data of the aptamer in the context of the 5′UTR (). We quantified the level of spontaneous cleavage at five positions (, blue arrowheads) and used it to estimate the apparent dissociation constant (K) for tc-binding (B). The K value of 800 pM determined independently at all five positions is in agreement with the value obtained by fluorescence titration spectroscopy and isothermal calorimetry [() and unpublished data] and represents one of the tightest binding of a small molecule by an aptamer. In addition to the expected signals at the tc-binding sites within B1-2 and L3 (,), spontaneous cleavage also occurs in the lower part of stem P1 which resembles the 5′ SS. The signal intensity is not as strong as in the region proposed to be single stranded and involved in ligand binding. However, clear spontaneous cleavage can be observed at positions A15 and U16 which belong to the conserved nucleotides of the 5′SS (GUGU). For both positions, a pronounced tc-dependent decrease is seen concomitant with occurring changes in the tc-binding pocket. This clearly indicates that stem P1 is not completely formed in the absence of tc, so that the 5′SS can yet be recognized. Addition of tc then leads to a tightening of the complete aptamer structure including P1 which then efficiently interferes with recognition. Pre-mRNA splicing is recognized as a critical step in mRNA maturation. Disturbances in this highly regulated process can lead to various severe diseases. Therefore, several systems which aim to prevent missplicing have been developed which either target the protein components involved (,) or the corresponding pre-mRNA (). One approach makes use of antisense oligonucleotides which hybridize to distinct regions of the pre-mRNA and restore correct splicing. Antisense oligonucleotides have been employed to alternate splicing patterns for β-thalassemia, Duchenne muscular dystrophy and cancer (). However, susceptibility to nuclease digestion, off-target effects and delivery problems prompted the development of active switches controllable by non-toxic small molecules with good cell permeability. We have shown in the current report that control of pre-mRNA splicing using a synthetic small molecule-dependent riboswitch is possible. Regulation is efficient when the regulatory element is inserted near the 5′SS. This is in agreement with recently discovered TPP riboswitches in the filamentous fungus involved in regulation of alternative splicing. All riboswitches identified in this organism are exclusively located within the intron close to the 5′SS (). Insertion of the tc-aptamer close to the BP did not result in remarkable regulation. No aptamer was inserted between BP and 3′SS since the aptamer would introduce several alternative 3′SS. In addition, it has been shown that a 66 nt insertion at this position leads to a >70% drop in splicing efficiency of the actin intron (). Gaur and co-workers () targeted this region by inserting the theophylline aptamer previously shown to be active as artificial riboswitch for translational regulation (). In their report, the 3′SS is located within the theophylline aptamer sequence and a theophylline-dependent regulation of pre-mRNA splicing is shown using an splicing assay. However a transfer to any given mRNA would be difficult since parts of the theophylline aptamer are located within the coding region of the 3′exon. The influence on splicing is position-dependent. Best regulation is obtained when the 5′SS is completely buried within the aptamer stem. This is surprising since previous data indicated that the P1 stem of the aptamer is already formed in the absence of tc (). However, in-line probing revealed that P1 is not completely structured without ligand and tc-binding forces tightening of the stem. Thereby, the 5′SS is masked for recognition by U1 snRNP-binding which is discussed as the prime step of spliceosome formation. Further on, our data demonstrate that the efficiency of splice regulation is dependent on aptamer stability with stable constructs showing higher regulatory efficiency. In contrast to previous findings (), in which gradually stabilized secondary structures next to the 5′SS lead to more pronounced splicing inhibition, stabilization of the tc-aptamer has no influence on splicing efficiency in the absence of ligand. First, upon tc addition the aptamer stabilization results in a marked decrease in splicing efficiency. A remarkable enhancement of regulation was achieved by combining more then one regulatable aptamer leading to the highest reported efficiency for aptamer-mediated gene regulation so far. In comparison, the tc-regulated activator/repressor-mediated gene expression system in yeast shows an at least 10-fold higher dynamic range of the protein-based system (). The disadvantage, however, of these systems is their complexity since they depend on the heterologous expression of the respective activator and repressor proteins and the manipulation of the target promoter. A further limitation of the system is that in many cases tight regulation is not possible due to leaky expression caused by not always predictable enhancer elements (). Therefore, an individual choice between the two systems highly depends on the requirements of the specific experiment. Taken together, our data demonstrate that the artificial tc-riboswitch is not only capable of regulating translation initiation, but can also modulate gene expression in eukaryotic organisms by facilitating ligand-dependent splicing of pre-mRNA. We therefore provide a valuable tool for alternative splicing which is a highly important step in regulation of gene expression. This work further highlights the importance of RNA structures and their capacity to influence alternative splicing. It underlines the need for simple model systems to study how secondary structures affect SS choice and opens up a wide field of applications for conditional gene expression systems based on regulatory active RNAs.

In eukaryotes, repair of mismatches generated during replication is initiated when a complex of either Msh2 and Msh3 or Msh2 and Msh6 binds to mismatched DNA. These MutS proteins also have other functions in cells (), one of which is binding to damaged DNA to initiate events that result in damage-induced cytotoxicity (). Msh2, Msh3 and Msh6 share five conserved protein domains (designated I–V) in common with bacterial MutS. The crystal structures of these five domains in and MutS are known (,), and their roles in binding to mismatched DNA, ADP/ATP binding and hydrolysis, heterodimerization, conformational changes and partnerships with proteins downstream in the mismatch repair pathway have been extensively studied and are partially understood (). However, unlike bacterial MutS or Msh2, Msh6 and Msh3 have an additional evolutionarily conserved region preceding domain I comprised of from ∼100 to more than 600 amino acids, depending on the organism. These N-Terminal Regions (NTRs) of Msh3 and Msh6 contain short, conserved PIP (PCNA interacting protein) boxes near the N-terminus that interact with PCNA (), the sliding clamp that participates in both DNA replication and DNA mismatch repair. Non-conservative amino acid replacement of residues in these PIP boxes partially reduces Msh2-Msh3-dependent and Msh2-Msh6-dependent mismatch repair in yeast (,), and deletion of the PIP box in human Msh6 partially inactivates mismatch repair (). In addition to residues important for binding to PCNA, other residues in the NTR, diagramed in , could also be functionally important. This importance is suggested by the evolutionary conservation of NTRs in both Msh6 and Msh3 (albeit of different lengths and sequence), by the identification () in human Msh6 NTR of a PWWP domain characteristic of proteins associated with chromatin, and by the presence in the human Msh6 NTR of missense mutations that are associated with cancer (). In the present study, we examine the possibility that residues in the Msh6 NTR other than those in the PIP box are functionally important. We demonstrate that recombinant yeast and human Msh6 NTRs bind to duplex DNA, identify amino acids in yeast Msh6 that contribute to this binding, and characterize mutants that concomitantly reduce DNA binding and reduce Msh6-dependent mismatch repair and sensitivity to killing by MNNG . We also show that substituting alanine for residues in a previously unrecognized, highly conserved motif at the extreme C-terminus of the Msh6 NTR also reduce Msh6-dependent mismatch repair and sensitivity to MNNG treatment. These results suggest that the Msh6 NTR has multiple roles in Msh6-dependent mismatch repair and in cellular response to alkylation damage. The coding sequences of the NTRs of yeast and human Msh6 were amplified from plasmids pRDK439 (gift from Richard Kolodner) and pAC61.2 () using primers to create restriction sites at the 5′ and 3′ ends. The amplified fragments were digested and ligated into pET28a(+). The yMsh6 and human proteins contained a TAG stop codon immediately after codon 299 and 394, respectively (see description in Results). The yeast DNA fragment was ligated into an NdeI and HindIII digested vector placing a His tag at the N-terminus of the protein. The human Msh6 N-terminal fragment was ligated into the NcoI and XhoI sites of pET28a(+) to place the His tag at the C-terminus of the protein. The yeast and human NTRs were expressed in upon IPTG induction of a log phase culture followed by a 3-h incubation. Cells were collected from 12-l cultures, washed with 20 mM Tris pH 8, centrifuged, frozen in liquid nitrogen and stored at −80°C. Frozen cell pellets were thawed and re-suspended in an equal volume (milliliter) of 20 mM Tris pH 8, 150 mM NaCl, 1 mM β-mercaptoethanol (Ni-A buffer) per weight (gram) of pellet, and lysed either by sonication or with a French press. The lysate was applied to a 4-ml nickel NTA (Qiagen) column, washed with 40 ml of NiA buffer + 5 mM imidazole, 40 ml of NiA buffer + 30 mM imidazole and eluted with 25 ml of NiA buffer + 100 mM imidazole. The eluted fraction was diluted 2-fold with 20 mM Tris pH 8, 1 mM EDTA, 1 mM β-mercaptoethanol, 10% glycerol (HepQ-A buffer) and loaded onto a 5-ml HiTrap heparin column (GE). The NTR proteins were eluted by a 100 to 700 mM NaCl gradient over 150 ml. The yeast protein eluted near 400 mM NaCl and the human NTR eluted near 600 mM. The proteins were diluted to 100 mM NaCl with HepQ-A buffer, loaded onto a 1-ml HiTrap Q column (GE), and eluted with a 20-ml gradient of NaCl from 100 mM to 1 M. Both yeast and human proteins elute at about 400 mM NaCl concentration. The eluted proteins were dialyzed against 20 mM Tris pH 8, 150 mM NaCl, 1 mM DTT, 5% glycerol, frozen in liquid nitrogen and stored at −80°C. Fifty micrograms of yeast or human NTR was diluted to 1 ml with 20 mM Tris pH 8, 15 mM NaCl, 1 mM EDTA, 1 mM β-mercaptoethanol, 10% glycerol (column buffer) and applied to a 0.1-ml volume of dsDNA cellulose (Sigma) packed into Poly-Prep chromatography columns (Bio-Rad). Columns were washed with 1 ml of column buffer and step eluted with 0.2 ml of column buffer with NaCl concentrations from 25 to 300 mM. DNA filter binding assays were performed as previously described (). DNA electrophorectic mobility shift assays were performed as previously described (). The yeast NTR was diluted to 50 µg/ml in 1 ml of 20 mM Tris pH 8, 100 mM NaCl and 1 mM β-mercaptoethanol and digested with 200 µl of 1 µg/ml chymotrypsin for 15 min at 25°C. This condition was selected because it produced a partial proteolysis of the protein resulting in a wide range of fragments. After digestion, a 100 µl aliquot was removed and the remainder of the reaction was diluted 10-fold with 20 mM Tris pH 8, 1 mM EDTA, 1 mM β-mercaptoethanol and 10% glycerol and loaded onto a 0.1-ml dsDNA cellulose column. After loading, the column was washed with loading buffer containing 10 mM NaCl and step eluted with 0.2 ml of loading buffer containing a range of NaCl concentrations from 25mM to 300 mM. A 25 µl sample from each elution was loaded on a 4–12% SDS polyacrylamide gel and fragments were resolved by electrophoresis. Two bands from the 200 mM NaCl elution were excised from the gel, cut into small pieces, and transferred into a 96-well microtiter plate. Gel pieces were subjected to automatic tryptic digestion using an Investigator™ Progest protein digestion station (Genomic Solutions, Ann Arbor, MI) as previously described (). Briefly, gels were sequentially washed twice with 25 mM ammonium bicarbonate buffer (pH 7) and acetonitrile, dehydrated, rehydrated with 25 ml of the enzyme solution and digested at 37°C for 8 h. The enzyme solution used was sequencing grade, modified trypsin (Promega Corporation, Madison, WI) at a concentration of 0.01 mg/ml in 25 mM ammonium bicarbonate buffer (pH 7). Resulting tryptic peptides were extracted from the gel, lyophilized and stored at −80°C. Prior to matrix assisted laser desorption ionization mass spectrometry (MALDI/MS) analysis, the peptides were reconstituted in 10 ml of a 50:50 solution of acetonitrile:water (0.1% formic acid). MALDI analyses were performed on the digested fragments using a Voyager-DE STR (Applied Biosystems, Framingham, MA) delayed-extraction time-of-flight (TOF) mass spectrometer in the positive ion reflector and/or linear modes. The instrument is equipped with a nitrogen laser (337 nm) to desorb and ionize the samples. A close external calibration, using two points to bracket the mass range of interest, was used. A 0.5 µl aliquot of the tryptic peptide solution was spotted with 0.5 µl MALDI matrix on a stainless steel sample target and allowed to dry at room temperature. A saturated solution of α-cyano-4-hydroxycinnamic acid in 45:45:10 ethanol:water:formic acid (v/v) was used as the MALDI matrix. Spectra were obtained over the mass range of 800–4000 Da in the reflector mode and 1000–25 000 in the linear mode with 50–100 laser shots per spectrum. For the in-gel digest analyses, ions corresponding in mass to trypsin autolysis products were used to internally calibrate mass spectra when possible. Msh6 NTR truncations were made in the pET28a construct expressing the 299-amino acid yMsh6 NTR. Truncations were constructed by designing a phosphorylated primer that included a termination codon to anneal to vector sequence at the 3′ end of the truncation and paired with phosphorylated primers annealing to the site of the desired truncation. The phosphorylated primer p- AAG CTT GCG GCC GCA CTC GAG CAC CAC CAC CAC was paired with p-TGA AGA ATG CGA AGT GTT GTA TGA AAA TTT TTT CTT AG for truncation N227 (after S227), p-TCT ACT TGG TGC CTG ATT TGG CCT GCT TTT CTT TTT TG for N251 (after R251), p-TTT GCT AGA CTT AGA AGT CGC TGA TGG TTG ACT ATG for N268 (after K268), p-ACG GCG CTG AGC ATC TCG TTC ATC CAC TAA CCA TTG for N289 (R289). pET-28a yMsh6NTR plasmid was amplified by Pfu turbo polymerase (Stratagene) in 20 cycles of a 3-step amplification reaction (95° × 30 s/55° × 1 min/68° × 10 min). After amplification, products were digested with DpnI at 37° for 1 h. Digested products were dialyzed against TE buffer 2x using a microcon (Amicon). During the final centrifugation, the sample was concentrated to a 10–20 µl volume. An aliquot was ligated and used to transform . Plasmid DNA was isolated and constructs were verified by DNA sequencing. Sensitivity to MNNG was measured in AC711a () that was derived from E203 (). was deleted by delito perfetto. and was amplified from pGSHU with MGT1 ends for recombination using amplification primers 5′yMGT1-P.IIS: ACAAAAAAAAAATTGAAAACGGTCGCATTTTTGATCTAAATGGACCAACG ccgcgcgttggccgattcat and 3′yMGT1-P.I: ATACATAACTATTTCTTATGTTTATTTTCCTAAAATCCTTTATCCAACTAttcgtacgctgcaggtcgac. Capitalized letters represent MGT1 sequence. Italicized letters represent an SceI unique restriction endonuclease site. Lower case letters represent either or sequence. E203 was transformed with amplified DNA and Ura+ and Hyg colonies were selected for transformation with IRO oligos. MGT1 IRO oligos were yMGT1-IRO-S: ACAAAAAAAAAATTGAAAACGGTCGCATTTTTGATCTAAATGGACCAACGTAGTTGGATA AAGGATTTTAGGAAAATAAACATAAGAAATAGTTATGTAT and yMGT1-IRO-A: ATACATAACTATTTCTTATGTTTATTTTCCTAAAATCCTTTATCCAACTACGTTGGTCCATT TAGATCAAAAATGCGACCGTTTTCAATTTTTTTTTTGT. FOA and Hyg IRO transformed colonies were selected, screened for by PCR and confirmed by sequencing (). Colonies deleted for MGT1 were selected for deleting RAD52 by transformation with SalI digested pΔRAD52blast and selecting on plates lacking uracil (). Ura+ Δ transformants were screened for pΔRAD52blast containing sequences by PCR and sensitivity to MMS. Ura and MMS transformants confirmed by PCR were patched onto FOA plates for deleting URA3 sequence. Deletions of were confirmed by DNA sequencing. These rad52Δ strains contained a copy of hisG sequence from pΔRAD52blast. Mutation rates and 95% confidence intervals were determined as described (), using 9–16 individual cultures. The strain AC711a was transformed with wild-type and mutant alleles to measure sensitivity to MNNG. Overnight cultures were grown at 30°C from single cell isolates. The cultures were diluted 10-fold and grown for three additional hours. Each culture was divided and either a 100x stock of MNNG in DMSO or DMSO alone was added. The cultures were incubated for an additional hour, washed with water, diluted and plated to determine survival. When the structure of MutS protein was solved, a structure-based amino acid sequence alignment was provided [ in ()]. This alignment suggested that domain I of Msh6 may begin at approximately residue 300 in yeast Msh6 and at approximately residue 395 in human MSH6. On that basis, and absent structural information on Msh6 proteins , here we studied and refer to the preceding residues as Msh6 NTRs. As a first step towards determining if these Msh6 NTRs interact with macromolecules other than PCNA, we expressed and purified the yeast Msh6 NTR comprised of residues 1–299 and the human Msh6 NTR comprised of residues 1–394. Both proteins were expressed in with a 6-His tag. This tag was placed at the N-terminus of the yeast NTR, but at the C-terminus of the human NTR to avoid perturbing the PIP box that is at the extreme N-terminus. Both NTRs were purified using three chromatographic steps, one of which involved binding to a heparin column. Both NTRs were obtained in highly purified form (see lanes labeled ‘load’ in A). Because heparin is a negatively charged resin to which many DNA binding proteins bind, we tested whether the Msh6 NTRs could bind to a dsDNA cellulose column. Indeed, the NTR of yeast Msh6 bound, and peak fractions eluted from the column at 125–150 mM NaCl (A). The NTR of human Msh6 also bound, and the peak fraction eluted from the column at 225 mM NaCl (A). Thus both proteins can bind to dsDNA via ionic interactions, and the NTR of human Msh6 appears to bind more tightly than the NTR of yeast Msh6. When DNA binding capacity was measured using a filter-binding assay (), both the yeast and human NTRs bound to double-stranded plasmid DNA with an affinity similar to yMutSα (B) The yMsh6 NTR also bound to single-stranded M13 DNA (open circles in B), but with lower affinity. Using an electrophoretic mobility shift assay (EMSA), both yeast and human Msh6 NTRs were observed to bind similarly to homoduplex DNA and to heteroduplex DNA containing a G–T mismatch (C). As a step towards determining if DNA binding by the NTR is important for Msh6 functions , we partially proteolyzed the yeast Msh6 NTR and tested the ability of the resulting polypeptide fragments to bind to the dsDNA cellulose column. Limited digestion with chymotrypsin generated a ladder of polypeptides (lanes labeled ‘Digested’ in A), several of which bound to the dsDNA cellulose column and eluted at high NaCl concentrations (A). To determine the identity of strongly bound chymotryptic fragments, two bands (boxed in A) were excised from the SDS–PAGE gel, digested and analyzed by MALDI-TOF mass spectrometry. In the MALDI mass spectra of the tryptic digestion of band 1 (B, upper panel), several ions corresponding in mass to predicted theoretical tryptic peptides of yeast Msh6 are observed (peptides labeled T44, T36, T41, T35–36 and T37). In addition, an autolysis product of trypsin ( 2211.11) was observed (labeled with an asterisk in B). The corresponding residues for these peptides are amino acids 292–299, 243–251, 277–284, 242–251 and 252–265, respectively. Similar data were obtained for band 2 except that peptide T44 was not observed (data not shown). To establish the amino termini of the chymotryptic fragments eluted from the dsDNA column, the column eluents were also analyzed by MALDI/MS analysis in the positive ion linear mode (B lower panel). Ions were observed corresponding in mass to theoretical chymotryptic peptides consisting of amino acids 270–277, 270–279, 254–269, 280–299, 231–253, 278–299 and 231–269. These ions are labeled as Y13, Y13–14, Y12, Y15–16, Y11, Y14–16 and Y11–12, respectively. Collectively, these data indicate that the chymotryptic fragments that bound strongly to the dsDNA cellulose column contain residues 231 through 299. Next, we expressed, purified and examined the DNA binding properties of yeast NTRs containing C-terminal truncations of residues in the DNA binding region of the yeast Msh6 NTR. Deletion of residues 290–299 did not significantly reduce DNA binding (, closed boxes), while deletion of more residues progressively diminished DNA binding (269–299, closed triangles; 252–299, closed diamonds). Deletion of all residues of the DNA binding fragment (228–299, open circles) completely eliminated DNA binding (). Next, we tested whether the reduced DNA binding capacity of the mutant Msh6 NTR proteins correlated with reduced DNA mismatch repair activity , as measured by elevated spontaneous mutation rates in haploid yeast strains. We constructed Msh6 alleles with in-frame deletions and measured mutation rates for resistance to canavanine, which results from a wide variety of mutations that inactivate the gene encoding arginine permease, and for lysine prototrophy resulting from deletion of a single base pair from a run of 14 A-T base pairs in the gene. An yeast strain lacking functional Msh6 has mutation rates at these two loci that are elevated by 10- and 400-fold, respectively (, line 2), in comparison with the rates in this same strain into which we introduced an ARS-CEN vector expressing Msh6 from its natural promoter (line 1). When we examined an mutant with an in-frame deletion of residues 3–289 of yeast Msh6, mutation rates remained as high (line 3) as for the complete absence of (line 2). This complete lack of complementation is consistent with one or more functions for the yeast NTR in addition to its interactions with PCNA. This interpretation is based on the fact that an derivative with alanine replacing two residues in the PIP box strongly reduced Msh2–Msh6 interactions with PCNA (), but nonetheless exhibit partial complementation (, line 4 and ()). Moreover, several different mutants with in-frame deletion of residues that reduce DNA binding () but do not change the PIP box all have mutation rates that are significantly elevated, and combining a PIP box mutation with an in-frame deletion results in a higher mutation rate (, last line) than observed in either single mutant alone. The correlations between loss of DNA binding () and elevated mutation rates in yeast () are consistent with the interpretation that DNA binding by the yeast Msh6 NTR contributes to mismatch repair activity . As a further test of the importance to mismatch repair of DNA binding by the Msh6 NTR, we examined complementation of the mutant strain with full-length containing negatively charged glutamate substituted for positively charged lysines and arginines within the DNA binding peptide. For this purpose, we measured mutation rates at the locus, which gives a greater mutator response when mismatch repair is defective. Three individual single residue replacements (R232E, K271E, R289E) had no apparent effect on mutation rates (, lines 3–5). However, mutator effects of 2–5-fold were detected in mutants containing either two or three replacements (lines 6–8), mutator effects of 7–12-fold were detected in mutants containing four replacements (lines 9–11), and five replacements yields 16-fold increases in mutation rate at the locus (lines 12 and13). We then expressed and purified one NTR derivative that contained glutamates replacing Arg232, Lys271 and Arg289. Compared with the normal yeast Msh6 NTR, the DNA binding affinity of this mutant protein was somewhat reduced (open squares in ). These observations further support the interpretation that DNA binding by the yeast Msh6 NTR contributes to mismatch repair activity. The DNA binding polypeptides identified by mass spectrometry are in a region of Msh6 that is not present in bacterial MutS or eukaryotic Msh2 proteins. To determine if residues in this region are conserved in multiple Msh6 proteins, a sequence alignment was performed that excluded other MutS proteins and focused only on residues in Msh6, specifically those residues within the yeast Msh6 DNA binding fragment through amino acid Phe311. Phe311 was used as an endpoint for this alignment because Thr312 was previously aligned with the first residue in helix 1a of domain 1 of MutS [ in ()]. This focused alignment reveals that the amino acid sequence in this region is conserved (). Within a span of nine residues, two residues, corresponding to yMsh6 T300 and P304, are invariant and six others are highly conserved, yielding a consensus sequence of ‘YDPxTLYI/V/LP’. Additional residues proximal to these are also conserved, through an invariant residue corresponding to E275 in yMsh6. Because this conservation suggests functional significance, we examined the consequences of substituting alanine for one, two or four of the conserved residues in this region. Each of these mutants was partially defective in complementing the mutator phenotype of an strain (). These partial defects were similar to that resulting from alanine substitutions in the PIP box (line 3). Moreover, combining the alanine substitutions in the PIP box with a quadruple YDTL-AAAA mutant resulted in a much stronger defect (line 9), approaching that of the mutant. These results are consistent with the interpretation that the conserved residues at the C-terminus of the yeast NTR contribute to mismatch repair function in a manner different from PCNA binding. In addition to DNA mismatch repair, Msh6 also participates in cellular processes that determine sensitivity to killing upon exposure to agents that damage DNA (), and reviewed in references 2–4. Yeast strains defective in Rad52-dependent homologous recombination are highly sensitive to killing by treatment with the methylating agent MNNG, and resistance to MNNG-induced killing is conferred by a loss of DNA mismatch repair (). An strain lacking Msh6 function is resistant to MNNG-induced killing () and complementation with Msh6 transforms it into an MNNG-sensitive strain. A similar degree of resistance was observed with mutants that lack residues 3–289 or residues 228–289 and with mutants that have alanine substituted for Phe337 in domain I that binds to the mismatched base, or for conserved residues in the consensus motif at the C-terminus of the NTR (). Thus, in addition to contributing to DNA mismatch repair, NTR residues involved in DNA binding and residues in the conserved motif near the C-terminus of the NTR all contribute to cellular response to MNNG treatment . This contribution to cellular response to DNA damage is not by specific recognition of MNNG induced O-methyl guanine (O-meG) base pairs in DNA by the NTR. The yeast NTR displayed similar binding affinities for O-meG-T, O-meG-C, G-T and G-C containing duplex oligonucleotide substrates (EMSA data not shown). In conjunction with previous studies, the results presented here suggest that the NTR of Msh6 makes several contributions to DNA mismatch repair and cellular response to alkylation damage to DNA. The first role to be identified for the Msh6 NTR was in mismatch repair (,), and involved PCNA binding via the PIP box at the N-terminus (boxed in green in ). The DNA binding data in and , the elevated mutation rates in and and the resistance to MNNG in suggest that DNA binding by the Msh6 NTR also contributes to mismatch repair, as well as to cellular response to alkylation damage. Residues near the C-terminus of the NTR (boxed in blue in ) contribute to this DNA binding, and at least some of these residues are likely to be lysines and arginines that may interact with the phosphate backbone of DNA (). A conserved motif () near the C-terminus of the NTR (boxed in black in ) also appears to be functionally important for mismatch repair and response to MNNG (, ). Residues in this conserved region could contribute to DNA binding, especially since they are located immediately proximal to core domain I that is already known to interact with mismatched DNA [reviewed in ()]. However, it cannot be excluded that this conserved motif may have a role other than or in addition to DNA binding. This possibility is consistent with the fact that an in-frame deletion of residues 290–299 had no detectable affect on DNA binding (), but did result in a mutator phenotype (). Double mutants that concomitantly reduce PCNA binding and DNA binding (), or double mutants that concomitantly reduce PCNA binding and alter the consensus motif at the C-terminus of the NTR (), yield mutator effects that are greater than observed with single mutants, suggesting that these multiple functions are combined for ensuring the full efficiency of mismatch repair. However, their relative contribution to cellular response to MNNG treatment is somewhat different. For example, in considering the contribution of DNA binding, in-frame deletion of residues 228–299 yields a modest MMR defect (), but results in resistance to MNNG similar to that of an null mutant (). Likewise, whatever the function of the consensus motif at the C-terminus of the NTR, a quadruple alanine mutant only partially inactivates MMR () but strongly inactivates the damage response (). The evidence that the Msh6 NTR has multiple functions leads one to wonder how many of these functions might be shared by Msh3. One shared function is PCNA binding via a PIP box near the N-terminus (,). Initial attempts to express the yeast Msh3 NTR in resulted in cell lysis, precluding an examination of DNA binding capacity. The NTRs of Msh3 proteins are generally shorter than those of Msh6, and initial alignments have not suggested the presence of a PWWP domain. As mentioned in the introduction, the human Msh6 NTR does contain a PWWP sequence motif located distal to the PIP box, and Slater and colleagues () have predicted that this is part of a PWWP domain, a module of unknown function that is sometimes found in proteins associated with chromatin. While the yeast Msh6 NTR lacks a PWWP motif , certain Msh6 NTRs have been suggested to contain a structurally related Tudor domain, again just distal to the PIP box. Just beyond this, we further note that 45% (31 of 69) of residues between amino acids 144 and 212 of yeast Msh6 are negatively charged (boxed in red in ). Similarly, 49% (19/39) of residues from 192 through 230 in human Msh6 are negatively charged. This feature is reminiscent of proteins that act as DNA mimics [reviewed in ()], wherein side chain carboxylates of aspartates and glutamates mimic the charge pattern of phosphates in a DNA backbone. Putnam and Tainer () have suggested that DNA mimics provide ‘an elegant mechanism by which interfaces can be reused to force sequential rather than simultaneous complex formations such as seen in systems involving polar protein assemblies and DNA repair machinery.’ If the highly negatively charged regions of yeast and human Msh6 NTRs are indeed DNA mimics, their location (boxed in red in ) immediately adjacent to the region of the Msh6 NTR that binds to DNA is intriguing. This juxtaposition suggests a model wherein the putative DNA mimic (red in ) and the DNA binding region (blue in ) might cooperate to regulate sequential steps in mismatch repair. Such regulation could involve interactions with other proteins such as PCNA or MutLα, or perhaps the proposed transition from an initial mismatch recognition complex containing bent DNA to an ultimate recognition complex in which the DNA is unbent ().

Uracil is a common base lesion in DNA and is introduced into the genome by deamination of cytosine and misincorporation of dUMP instead of dTMP during replication. Spontaneous deamination of cytosine has been estimated to occur at a rate of 60–500 events per day in human cells (). In addition, recent research has revealed that enzymatic deamination of cytosine at the Ig loci by activation-induced cytosine deaminase (AID) initiates the antigen-dependent antibody diversification processes (). Uracil generated by deamination of cytosine is 100% miscoding, and result in C:G to T:A transition mutations if not repaired prior to replication. Misincorporated uracil is not directly miscoding, but it appears to be a critical source of spontaneously generated AP-sites (apurinic/apyrimidinic-sites) in the genome (). Uracil and some uracil analogs generated by oxidation of cytosine are excised from the genome by uracil-DNA glycosylases (UDGs). Mammalian cell nuclei contain at least four UDGs; UNG2, SMUG1, TDG and MBD4. Current evidence suggests that UNG2 (racil--lycosylase 2) and SMUG1 (ingle-strand-selective onofunctional D) are the major enzymes responsible for repair of spontaneously deaminated cytosine (), while post-replicative excision of misincorporated dUMP (U:A) and excision of AID-generated uracil (U:G) are performed mainly by UNG2 alone (). Consistent with the role of UNG2 in replication associated repair, UNG2 binds PCNA and RPA, is localized to replication foci, and is cell cycle regulated with the highest levels in S-phase (,). Conversely, SMUG1, is not cell cycle regulated and is evenly distributed in the nucleoplasm (). SMUG1 excises uracil from DNA with a much lower efficiency than UNG2, but has broader substrate specificity. Only SMUG1 excises thymine with an oxidized methyl group (,). UNG and SMUG1 belong to a superfamily that has apparently evolved from the same ancestral gene (). Comparison of crystal structures of human UNG and SMUG1 has revealed that these enzymes share a common fold and that the SMUG1 active site is a mosaic of features from UNG and MUG/TDG enzyme families (). UNG is widely distributed in bacteria, eukaryotes and even some large DNA viruses, while SMUG1 has previously been reported to be present in vertebrates and insects only (,). Here we report the existence of bacteria that contain SMUG1 as their only identified UDG. Interestingly, identification of these bacterial SMUG1 orthologs shed new light on the origin of SMUG1 and UNG. Vertebrates contain both SMUG1 and UNG, but their distinct roles in base excision repair (BER) of deaminated cytosine are still not fully defined. We have compared the repair mechanisms of human SMUG1 (hSMUG1) and human UNG2 (hUNG2) on deaminated cytosine, by using replicating cells containing AID-induced U:G lesions, and using purified enzymes (hSMUG1, hUNG2 and hAPE1) including a panel of hSMUG1 mutants. We find that only hUNG2 can complement Ung in repair of U:G mismatches, whereas hSMUG1 inhibits cell proliferation in the same system. analyses reveal that hSMUG1 and hUNG2 coordinate the initial steps of BER by distinct mechanisms. Furthermore, we characterize a specific motif in hSMUG1 that confers U:G-substrate preference and stabilizes the product AP-site binding. Finally, we propose a model for how SMUG1 and UNG2 initiate and coordinate repair of deaminated cytosine (U:G) by distinct mechanisms. This model is consistent with a role for the slow-acting SMUG1 in repair of deaminated cytosine in non-replicating chromatin, and efficient and highly coordinated repair by UNG2 of uracil (both U:G and U:A) in replicating DNA. To generate pAID an NcoI site flanking the ATG start codon of hAID cDNA (Image Clone: 4853069) was made by site-directed mutagenesis. The complete reading frame was then cloned into the NcoI–PstI sites of the expression vector (Amersham Biosciences). Cloning of pUNG2 (p658kan-UNG2) was published previously (). pSMUG1 (p658kan-SMUG1) was constructed by cloning the complete reading frame of hSMUG1 cDNA (Image Clone: 726197) as an NdeI–BamHI fragment into the vector (,). Cloning of 6 x His-tagged hSMUG1 (pET28a–SMUG1) and hUNG2 (pET28a–UNG2) was published previously (,). The hAPE1 expression vector () () was a gift from Dr Ian Hickson (Cancer Research UK Laboratories, Oxford, UK). Site-directed mutagenesis was carried out using the Quick-Change™ kit (Stratagene), and the mutants were confirmed by sequencing. Ung-deficient strain NR8052 (Δpro-lac, thi-, ara, trp9777, ung1) or Ung-proficient strain NR8051 (Δpro-lac, thi-, ara, trp9777) () were transformed with the IPTG inducible constructs; -AID [encoding AID wild-type and ampicillin resistance (amp)] or -AID-C87A (encoding catalytically inactive AID as control) followed by transformation with the toluic acid inducible p658kan-hSMUG1 or p658kan-hUNG2 constructs [encoding hSMUG1 and hUNG2, respectively and kanamycin resistance (kan)]. Empty vector () was used as control. Single amp + kan colonies were picked from plate and grown in 3 ml LB containing 100 µg/ml amp, 30 µg/ml kan, 1 mM IPTG and 1 mM toluic acid at 30°C over night. Aliquots were mixed with 3 ml soft agar and plated on LB-amp + kan plates and LB-amp + kan plates containing 100 µg/ml rifampicin (rif) 100 µg/ml. The numbers of (amp + kan) colonies were counted after incubating the plates at 37°C for 24 h, while the numbers of (amp + kan + rif) colonies were counted after 48 h. Mutation frequencies were calculated as the number of (amp + kan + rif) colonies per 10 (amp + kan) colonies. Expression of hUNG2 and hSMUG1 in was confirmed by western analysis. Five microgram soluble cell lysate from cultures induced over night were separated by SDS–PAGE (NuPAGE®, Invitrogen) and electro-blotted onto Immobilon PVDF membranes (Millipore). hUNG2 and hSMUG1 were detected using the primary antibodies PU101 () and PSM1 (), respectively, followed by HRP-conjugated swine anti-rabbit secondary antibody (DakoCytomation) and Super Signal West Femto substrate (Pierce Chemical Co.) The western blots were analysed on a Kodac Image station 2000R. hSMUG1, hUNG2 and hAPE1 were expressed in BL21-CodonPlus (DE3)-RIL (Stratagene) and purified as described (,). Protein concentrations were measured by the Bradford protein assay (BioRad) using BSA as standard, and stored at –80°C in 50% glycerol and 1 mM DTT. The hSMUG1 mutants were confirmed by trypsin digestion followed by MALDI–TOF mass spectrometry. Standard UDG assays were performed as previously described (). Briefly, 10 nM SMUG1 and 1.8 µM [H]dUMP-containing calf thymus DNA (U:A) with specific activity 0.5 mCi/μmol were incubated in a 20 μl assay mixture containing (final) 20 mM Tris–HCl pH 7.5, 10 mM NaCl, 7.5 mM MgCl, 1 mM EDTA, 1 mM DTT, 0.5 mg/ml BSA (UDG assay buffer), for 10 min at 30°C. The amount of released uracil was measured as described (). Limited turnover oligonucleotide assays were performed by using equimolar amounts of enzyme and [P]-5-end-labelled oligonucleotide (U141: CATAAAGTGAAAGCCTGG). dsDNA substrates (U141A and U141G) were prepared and assays were performed as previously described (). Enzyme 20 nM and substrate 20 nM were incubated in 10 µl UDG assay buffer at 30°C for 0, 5, 30 or 60 min. Substrate and product were quantified by phosphor imaging. Multiple turnover oligonucleotide assays were performed as described () by using 0.4 nM enzyme and 20 nM oligonucleotide substrates in 10 µl UDG assay buffer at 30°C for 15 min. Substrate and product were quantified by phosphor imaging. Kinetic assays for the determination of K and k were performed under Michaelis–Menten conditions, using high molar excess of substrate: 20 nM [P]-labelled and 1–20 µM non-labelled U141A, U141G and U141 oligonucleotide substrates in 10 µl samples were incubated at 30°C for 10 min. In this assay, enzymes (SMUG1-WT and SMUG1-P245A) were adjusted to give <30% uracil excision at each substrate concentration. 1 nM, 20 nM and 5 nM enzyme were used together with the U:G, U:A and Uss substrates, respectively. Kinetic parameters were calculated according to the method of Wilkinson using the Enzpack for Windows version 1.4 (Biosoft). UDG activity in bacterial extracts were measured using 1 µg soluble cell lysate (from cultures induced with 1 mM toluic acid at 30°C over night) and 5 nM U141G oligonucleotide in 10 µl total volume containing 20 mM Tris–HCl pH 7.5, 35 mM NaCl, 3 mM EDTA and 1 mM DTT with or without 0.1 µg Ugi. The samples were incubated at 30°C for 15 min and analysed as described (). AP-site inhibition assays: AP-site inhibitors were generated from an uracil-containing oligonucleotide (U93:TGAAATTGTATCCGCTCA). Fifteen nanomol U93 were incubated with 1 µg (40 pmol) hUNGΔ84 () in a total volume of 300 µl containing (final) 20 mM Tris–HCl pH 7.5, 10 mM NaCl, 1 mM DTT, 1 mM EDTA, 0.5 mg/ml BSA at 37°C for 2 h. UNG was inactivated at 65°C for 5 min, followed by addition of the specific UNG protein inhibitor Ugi (160 pmol). To prepare dsAP-DNA the AP-oligonucleotide was annealed with 50% molar excess of complementary oligonucleotides containing either A (93A) or G (93G) opposite the AP-site, generating AP:A and AP:G, respectively. The inhibitory effects of AP-sites were analysed by standard UDG assays using 1.8 µM [H]dUMP-containing calf thymus DNA substrate, 7.5 nM hSMUG1 (WT or mutants) and 62.5 nM AP:G or 500 nM AP:A inhibitor. SMUG1 (0.05–0.50 µM) or UNG2 (0.05–0.50 µM) was incubated with 4 nM [P]-end-labeled oligonucleotide (U141G, U141A, T141A, or U141ss) in 10 μl UDG assay buffer containing 2.5% glycerol for 30 min at 30°C to generate AP-sites. Complete excision of U from the oligonucleotides was confirmed in parallel samples by piperidine cleavage and denaturing PAGE as described (). After uracil-excision, binding of enzyme to end-products (AP-sites) was analysed by non-denaturing 8% PAGE (containing 2.5% glycerol) in 0.5 × TAE pH 8 buffer at room temperature for 15 min at 100 V followed by 30 min at 150 V. The gels were fixed, dried and analyzed by phosphorimaging. The amount of bound AP-site oligonucleotide was quantified and plotted using a sigmoid curve fit model in GraphPad Prism®. K values (concentration of enzyme giving 50% of maximum binding) were calculated for SMUG1 on AP:A and AP:G. Exonuclease III was purchased from New England Biolabs (#M0206S). AP-site substrate was prepared by incubating 3 pmol [P]-end labelled U141-oligonucleotide with 37.5n g (1.2 pmol) hUNGΔ84 () in a total volume of 30 µl containing (final) 20 mM Tris–HCl pH 7.5, 10 mM NaCl, 1 mM DTT, 1 mM EDTA, 0.5 mg/ml BSA at 37°C for 1 h. The UNG enzyme was inactivated by heating at 65°C for 10 min. To generate dsAP-DNA the AP-oligonucleotide was annealed with 50% molar excess of complementary oligonucleotide containing G (141G) opposite the AP-site. AP-endonuclease assays were performed with 0.025 nM hAPE1 or ExoIII and 2 nM [P]-labelled AP:G substrate in final volume of 10 µl UDG assay buffer and incubated at 30°C for 10 min. AP-endonuclease cleaved products and uncleaved AP-site substrates were separated by denaturing PAGE and quantified by phosphorimaging. SMUG1 was suggested to be a relatively new evolutionary offspring in the UDG superfamily found only in vertebrates and insects (,). However, a BLAST search using the sequence of hSMUG1 protein as query revealed SMUG1 orthologs both in prokaryotes (proteobacteria and planctomycetes) and in marine non-vertebrates such as sea urchin and sea squirt (). Remarkably, the vertebrate SMUG1 has highest similarity to sequences identified in bacteria, showing 51.1% identity and 69.9% similarity between human SMUG1 and SMUG1 from (). We did not find genes in the SMUG1-containing non-vertebrate organisms identified here except in sea urchin, neither is it present in insects. Moreover, the prokaryotes encoding SMUG1 also lack orthologs of other members of the UDG family [MUG, UDG 4 () and UDG 5 ()], indicating that SMUG1 may be the only uracil-DNA glycosylase in these species. The observation that SMUG1 is apparently the sole UDG in some bacteria, prompted us to examine whether human SMUG1 can act as a functional homolog of Ung in that lacks SMUG1. Human SMUG1 has been reported to complement Ung1 in (). In the yeast study, antifolate agents were used to increase misincorporation of dUMP, generating U:A base pairs. However, mammalian SMUG1 is more likely involved in removal of deaminated cytosine rather than misincorporated uracil (), thus U:A pairs may not represent the most relevant substrate for SMUG1. Here we used an system in which the cytosine deaminase AID was expressed in Ung-deficient to specifically generate promutagenic U:G mispairs (). In this background, we expressed hSMUG1 or hUNG2. Mutation frequencies were monitored by the rifampicin resistance (rif) assay (), and cell growth was analysed by counting viable cells. Expression of SMUG1 and UNG2 in the cells was verified by western blot analysis and uracil-excision activity (U:G substrate) in clarified lysate from induced cultures. The results are summarized in . Expression of AID in Ung-deficient yields a mutator phenotype, which is reversed by co-expression of UNG2. SMUG1 did not suppress the mutator phenotype, but markedly inhibited cell growth in AID-expressing Ung-deficient cells. Notably, hSMUG1 was neither growth inhibitory nor mutagenic in cells. The level of genomic uracil is enhanced more than 30-fold (to 31 uracil per 10 nucleotides) in Ung-deficient cells, probably mostly U:A base pairs caused by replicative incorporation of dUMP (). Importantly, SMUG1 did not influence growth or mutation frequency in cells expressing the inactive AID-C87A mutant (), neither did induction of only SMUG1 in Ung-deficient cells inhibit cell growth (data not shown). This demonstrates that the growth inhibitory effect of SMUG1 is dependent on U:G-lesions or U:G-repair intermediates. Taken together these results reveal that SMUG1 does not act as a functional homolog of Ung in U:G repair in proliferating cells. To elucidate the molecular mechanisms underlying the observed effects of SMUG1 and UNG2, we analysed the product (AP-site) binding subsequent to uracil-excision by purified human SMUG1 and UNG2 using electrophoretic mobility shift assays (EMSA). The results, illustrated in A, demonstrate that SMUG1 readily binds to AP-sites in dsDNA (AP:G and AP:A), while no binding to AP-sites in single-stranded DNA (APss) or dsDNA without AP-site (T:A) was detected. In contrast, we did not observe binding of UNG2 to the same set of oligonucleotides. SMUG1 binds AP:G with slightly higher affinity than AP:A with K values (concentration of enzyme yielding 50% of maximum binding) calculated to 0.125 ± 0.022 µM and 0.183 ± 0.007 µM, respectively (B). The sigmoid curve plotted in B represents the EMSA data in A. The binding experiments were, however, repeated several times and consistently revealed higher affinity for AP:G than for AP:A. To gain more insight into the coordination of the first and the second step of BER of deaminated cytosine, we investigated the effect of the major human AP endonuclease, APE1, on uracil-excision from U:G substrate by SMUG1 and UNG2. The excision rate from U:G substrate by SMUG1 was 2–3-fold stimulated by APE1, while no stimulatory effect was observed with Uss substrates (C). In contrast, uracil-excision by UNG2 was only weakly stimulated with both substrates (D). This confirms our previous quantitation of APE1 stimulation of SMUG1 and UNG2 measured with U:A-containing substrate (), and is in accordance with previously published data on SMUG1 (). Together with the EMSA results (A and B), this indicates that uracil-turnover by SMUG1 is increased by alleviation of product-binding after cleavage of the AP-site. In support of this, we observed that bacterial AP endonuclease Endo IV also stimulates SMUG1 activity and that SMUG1 does not bind to nicked AP-sites (data not shown). Based on the different AP-site binding properties of SMUG1 and UNG2 we analysed their effects on the activity of human APE1 and bacterial ExoIII. A molar excess of hSMUG1 inhibited both hAPE1 and ExoIII activity (E), indicating that SMUG1 and AP endonucleases compete for binding to AP-sites. In contrast, hUNG2 markedly stimulated the activity of hAPE1 but had no effect on ExoIII (F). Thus, SMUG1 probably binds strongly to the product until it is displaced by an AP endonuclease that cleaves the AP-site, whereas UNG2 may physically interact with APE1 to coordinate and facilitate the first and the second step in BER. We have previously characterized substrate specificities and kinetic constants of human SMUG1 and UNG2 (). In addition, mechanisms of uracil recognition, substrate binding and catalysis by human UNG have been extensively studied (,). The active site of SMUG1 is a mosaic of features from UNG and MUG/TDG enzyme families () (A). To further explore the molecular mechanisms of SMUG1, active-site mutants were generated (A), purified and activities were measured by standard UDG assays (see ‘Materials and Methods’ section) using 1.8-µM [H]dUMP-containing DNA substrate (U:A). Residual activities of corresponding hSMUG1 and hUNG mutants, analysed by identical activity assays, are compared and listed in . In the structures of SMUG1 (xSMUG1) and hUNG an asparagine at the bottom of the catalytic pocket binds to O4 and N3 of uracil (B). Mutating this residue to aspartic acid in hSMUG1 (N163D) resulted in 11% residual activity, the residual activity of the corresponding UNG mutant (N204D) was only 0.04% (). H268 in hUNG is believed to have a critical function in stabilization of the transition state intermediate (), and accordingly the hUNG-H268L mutant displayed only 0.32% activity compared with WT. Interestingly, the equivalent mutation in hSMUG1 (H239L) retained 28.6% residual activity. To rule out a possible contribution to activity from contaminating UNG, we analysed the H268L mutant in the presence of the specific UNG inhibitor Ugi. Ugi did not have any inhibitory effect on activity, demonstrating that the hSMUG1-H268L mutant was not contaminated by UNG (data not shown). Mutation of the asparagine (hSMUG1-N85A) proposed to coordinate the active water molecule in SMUG1 () resulted in 3% residual activity, whereas mutation of the corresponding residue in hUNG (D145N) reduced the activity to 0.04% compared with WT. In UNG, Tyr147 blocks the entrance of thymine to the active site pocket (,,). SMUG1 has glycine in the corresponding position, explaining its ability to accept uracil with hydrophilic substitutions at C5 position (A and B). Substitution of this glycine with the much larger tyrosine (SMUG1-G87Y) resulted in a catalytically dead enzyme, suggesting a side-chain orientation of this residue that completely blocks entrance of all substrates to the binding pocket. However, except from this latter substitution, corresponding active-site mutations have generally less effect on SMUG1 activity than on UNG activity, when measured on a U:A substrate. To quantify residual activity of the hSMUG1 mutants towards U:G, we used a multiple turnover oligonucletide assay (see “Materials and Methods” section) with a 50-fold molar excess of U:G substrate. Interestingly, the residual activities of the mutants were even higher when analyzed with U:G substrate as compared with U:A substrate (). To minimize the effect of product binding, the SMUG1 mutants were also analysed using equimolar amounts of enzyme and oligonucleotides with uracil in either U:G, U:A or Uss contexts. Under these conditions, the U:G preference of the SMUG1 mutants was even more pronounced (). A stable enzyme-substrate complex with a long residence time will increase the possibility for catalysis to occur, thus these results indicate that SMUG1 binds U:G substrate (not only AP:G product) with higher affinity than U:A and Uss substrates. The crystal structure of the xSMUG1–DNA complex reveals a DNA-helix penetrating wedge formed by a loop followed by a short α-helix (). This structural architecture suggests a more invasive interaction with dsDNA than observed for UNG, and the wedge motif apparently also contacts base pairs adjacent to the flipped out lesion (A and B). Thus, we examined to what extent the nature of these bases influences the catalytic activity SMUG1 and UNG2. As demonstrated in C, SMUG1 activity was markedly influenced by the bases flanking uracil, with a preference for A:T base pairs. Conversely, no such preference was observed for UNG2 using the same dsDNA substrates (D). These results support the hypothesis that SMUG1 interacts with the base pairs flanking the lesion in the DNA helix, and possibly disrupts the Watson–Crick hydrogen bonds. The N-terminal part of the wedge (239-HPSP-243) faces the uracil-containing strand, and resembles the intercalating leucine loop in UNG (268-HPSP-272) (), except that SMUG1 has arginine in the position corresponding to hUNG-Leu272 (A and B). This residue aids expulsion of the uracil residue from the DNA-helix and subsequently fills the gap left by the flipped out nucleotide (). Similar to SMUG1, UNG encoded by human cytomegalo- and vaccinia virus has arginine at this position (A). To analyse this uracil-flipping residue in SMUG1 in more detail we mutated the Arg243 to leucine and alanine. The R243L mutation reduced the activity to about 3% on U:A substrate and 7% on U:G substrate, while the R243A mutation had little effect on enzyme activity (). Notably, there is alanine at this position in SMUG1 from bacteria and sea squirt and isoleucine in sea urchin ( and A). Thus, different UNG and SMUG1 family members may either have a large hydrophilic (Arg) or a large hydrophobic side chain (Leu or Ile) or surprisingly even a small side chain (Ala) as the residue expelling uracil. The C-terminal part of the wedge is completely different in SMUG1 and UNG (A and B). Whereas no direct interaction between enzyme and the bases in the distal strand was observed in the hUNG-DNA crystal structure (), the xSMUG-DNA structure indicates that the C-terminal part of the wedge in SMUG1 interacts with the distal strand. To investigate this in more detail, a series of site-specific alanine mutants were generated in the C-terminal part of the wedge motif (243-RNPQANK-249) (A and E). Strikingly, mutations in the 244-NPQANK-249 region of hSMUG1 significantly increased the U:G activity (), an effect that was most pronounced for the SMUG1-P245A mutant. To investigate this phenomenon in more detail we performed kinetic analysis of SMUG-WT and SMUG1-P245A on U:G, U:A and Uss oligonucleotide substrates. The effect of mutating Pro245 to Ala turned out to be specific (7-fold increase in k) to U:G substrates, since no significant changes were observed using U:A and Uss substrates (). This strongly suggests that SMUG1-P245 is involved in making specific interactions with guanines opposite uracil-lesions probably by pushing into the base-stack opposite the lesion as suggested by Wibley and colleagues () (E). Thus, mutation of Pro245 most likely increases turnover by destabilizing binding to the AP:G product. Supporting this view we have previously reported that AP-sites in dsDNA, but not in ssDNA, are inhibitors of hSMUG1, and that AP-sites opposite guanine are much more potent inhibitors than AP-sites opposite adenosine (). The activities of the SMUG1 wedge mutants were therefore analysed in the presence of oligonucleotides containing AP-sites opposite adenine (AP:A) or guanine (AP:G) (). SMUG1 mutated in residues pointing towards the distal strand, SMUG1-R243A, SMUG1-P245A and SMUG1-K249A (E), were less inhibited (∼47%) by AP:G than WT (70%), indicating a weaker binding to AP:G. As expected from the U:A activity, the wedge mutations had a less pronounced effect on the inhibition by AP:A (). The side chain of SMUG1-Asn248 is not oriented towards the distal strand, but stabilizes the wedge by making hydrogen bonds to hSMUG1 residues Pro242 and Arg141 (E). Mutating this residue (SMUG1-N248A) increased both U:G and U:A activity. Inhibition by AP-sites was, however, only marginally reduced compared with wild-type. This indicates that the increased enzyme activity of the SMUG1-N248A mutant probably is the result of a more flexible wedge, and not due to reduced AP-site binding. The existence of SMUG1 orthologs in bacteria suggests that SMUG1 is of older origin than previously assumed. Notably, non-vertebrates (except from sea urchin) appear to have UDGs of either the SMUG1- or the UNG-type, while vertebrates contain both. In vertebrates, SMUG1 and UNG2 have probably evolved to carry out different and specialized functions in processing of genomic uracil (and some uracil analogous) in the most appropriate way depending on uracil context, gene locus (e.g. Ig-genes), cell type, proliferative status, cell cycle phase, sub-nuclear localization and mutagenic potential of the lesion. In the present work, we have compared and characterized some of these specialized molecular functions of hSMUG1 and hUNG2 by both and experiments, and demonstrated that they coordinate the initial steps in BER by different molecular mechanisms. We show that hUNG2, but not hSMUG1, can repair U:G lesions in proliferating cells . In contrast, hSMUG1 expression inhibits cell growth in this system. Interestingly, it was reported that hSMUG1 can functionally compensate for Ung1 in yeast cells treated with antifolate agents to increase misincorporation of uracil in the genome (). In WT cells, antifolate treatment results in S-phase arrest and cellular toxicity due to uracil excision and single-strand breaks. Antifolate-treated ▵ cells are, however, able to complete DNA replication, but when hSMUG1 is expressed the cells are arrested in S-phase like the WT. This indicates that SMUG1 can target misincorporated uracil in the yeast genome and generate cytotoxic AP-sites. However, complete repair of the lesion was not monitored in the yeast study. Since the authors use S-phase arrest and cellular toxicity as a measure of complementation, their results are in agreement with our results, although their conclusion is different. In conclusion, SMUG1 and UNG2 can both target uracil residues, but only UNG2 can initiate complete BER in rapidly growing cells. It cannot be excluded, however, that SMUG1 can compensate for Ung in U:A repair both in yeast and in bacteria. Furthermore, it would be interesting to find out whether SMUG1 from a prokaryote can complement Ung-deficient bacteria. We have previously characterized the catalytic domain of hUNG in detail (,,). Here we generated active site mutants of hSMUG1 and compared their activities with those of the corresponding mutants of hUNG. Interestingly, using the same standard UDG assay protocol, single active-site mutations in SMUG1 have less effect on catalytic activity than the corresponding mutations in UNG. Analysis of hSMUG1 active-site mutants has also been published by another group (), and they report a more dramatic reduction in activity of several of the mutants. This is probably because they measured SMUG1 activity using a very high molar excess of enzyme (up to 670-fold) and substrate concentrations about three orders of magnitude below the K value of hSMUG1-WT () (A). Such assay conditions will, however, mainly reflect substrate affinity and not catalytic turnover, because the substrate is the limiting factor. By using a high substrate concentration (1.8 µM), we here focused on catalytic turnover and less on substrate affinity. The relatively high residual uracil-excision activity (measured with molar excess of substrate) in the SMUG1 active-site mutants, especially against U:G substrates, may in part be explained by the specific helix-inserting motif that probably is important for binding to dsDNA substrate and product. When SMUG1 is bound to the substrate, the DNA substrate itself may be important to drive the reaction forward by so-called ‘substrate autocatalysis’. For glycosylases in the UDG superfamily, it has been reported that the substrate itself is a major contributor to lowering of the activation energy, thus explaining residual activity in mutants lacking catalytic key residues (). This ‘substrate autocatalysis’ phenomena may also explain the discrepancy between residual activity of SMUG1 active-site mutants measured at very low substrate concentrations () and those presented here measured at high substrate concentrations. Interestingly, one of the active-site SMUG1 mutants was, however, catalytically dead. Introducing a UNG like thymine expulsion residue in SMUG1 (SMUG1-G87Y) abolished the activity completely. The superimposed structures of xSMUG1 and hUNG reveal that the thymine expulsion loops do not follow the same path, bringing the side-chains of SMUG1-G87Y and UNG-Y147 in different orientation in the substrate binding pockets of the enzymes. Additionally, in SMUG1 this residue is sandwiched between two prolines that restrict conformational flexibility in this loop segment. Thus, a large-residue in this position will most likely block the entrance of the substrate in the active site pocket of SMUG1. We find that hSMUG1 binds to product AP-sites in dsDNA , with the strongest binding to AP:G, while no product binding was observed for UNG2. The growth-inhibitory effect of SMUG1 in cells containing U:G lesions is most likely explained by this product binding. SMUG1 attached to AP-sites may probably interfere with replication, and thereby prevent cell division, a situation that is especially prominent when SMUG1 is over-expressed. Under these circumstances, the endogenous level of AP-endonuclease activity is likely insufficient to alleviate the product binding and replication is blocked. In support of this, we find that SMUG1 inhibits activity of both the human APE1 and the bacterial ExoIII AP-endonucleases (D). Thus, a high level of SMUG1 likely interferes with the downstream processing of the AP-sites and prevents complete repair. This is in agreement with the observation that expression of hSMUG1 in Δ yeast cells did not suppress the spontaneous mutator phenotype, but rather caused an increased mutation frequency (). It is well known that APE1 stimulates the excision activity of many DNA glycosylases (,,). However, stimulation of APE1 by a DNA-glycosylase has to our knowledge not previously been reported. Here we find that UNG2 stimulates the cleavage activity of APE1, indicating that UNG2 may physically interact with APE1. In support of this, we have previously isolated UNG2-associated complexes containing all factors required for complete BER of uracil, including APE1, from human cell extracts (). We have examined whether sequence context of the substrate influenced on uracil-excision activity of SMUG1 and UNG2. Surprisingly, the nature of the bases flanking the uracil had only impact on uracil-excision activity of SMUG1 and not of UNG2, measured on the same double-stranded oligonucleotide substrates. The latter was rather unexpected since sequence preference of UNG from several sources [a truncated form of UNG purified from calf thymus, Ung and the catalytic domain of human UNG (UNGΔ84)] has previously been demonstrated in our laboratory (,,) and by others (herpes simplex virus UDG) (). However, all these enzymes lack the regulatory N-terminal sequence. In the present study we have analysed the full-length human UNG2 enzyme in presence of Mg, which has a strong stimulatory effect particularly on UNG2 (,,). It is possible that the N-terminal domain of UNG2 diminishes the sequence specificity observed for the truncated forms of UNG in order to obtain the most efficient repair of uracil in all contexts at the replication fork. Taken together, it is clear that SMUG1 and UNG2 have evolved distinct mechanisms for the coordination of the second step in BER. Based on previous results and the new data presented here, we propose a model for how SMUG1 and UNG2 initiates and coordinates repair of deaminated cytosine (U:G) by distinct ‘hand-over’ mechanisms (). This model is consistent with a role for SMUG1 in repair of deaminated cytosine in non-replicating chromatin and repair of uracil (U:G and U:A) by UNG2 in replication foci. The catalytically highly efficient and context-independent UNG2 enzyme is probably important in rapidly dividing cells to remove deaminated cytosine in front of the moving replication fork (pre-replicative repair), in addition to post-replicative repair of misincorporated uracil (). This pre-replicative repair of U:G by UNG2 is supported by the observed 5.2-fold increased mutation frequency in Ung-deficient mouse embryonic fibroblasts (MEFs), mostly G:C to A:T transitions (). SMUG1, on the other hand, is not designed to rapidly repair uracil during replication, and is probably more important in non-replicating chromatin, outside S-phase and in resting cells where the level of UNG2 is low (). However, SMUG1 counteracts mutations also in cycling mouse cells (MEFs). Knocking down Smug1 by siRNA in MEFs resulted in 2.4-fold increased mutation frequency at the HPRT locus (). Thus, the slow-acting, product-binding SMUG1 may efficiently recognize deaminated and some oxidized cytosine derivatives in non-replicating dsDNA (especially in A-T rich regions where the cytosine deamination rate is expected to be higher due to increased DNA breathing), excise the lesion and remain attached to the cytotoxic AP-site product until APE1 arrives and initiates further repair. Notably, in mouse SMUG1, the residue corresponding to the conserved Pro245 in the hSMUG1 wedge motif is alanine (A). Kinetic analysis of the hSMUG1-P245A mutant, mimicking mouse SMUG1, revealed that this mutant has more than a 7-fold increased turnover number (k) on U:G substrate compared with WT (A). The increased U:G activity of this mouse SMUG1 mimicking mutant could thus provide a mechanistic explanation for the apparently higher SMUG1 activity in extracts from mouse cells than from human cells (,). This observation should be kept in mind when using mice as model organisms for uracil repair in mammals. The presence of at least one family member of the uracil-removing glycosylases in all known organisms points to the importance of this repair mechanism. The present article demonstrates new distinct properties of SMUG1 and UNG2 that point to different mechanisms for coordination of the initial steps in BER. Considering functional differences, SMUG1 still seems to be able to compensate for UNG-deficiency in most somatic tissues (), and is apparently sufficient to maintain genomic stability in some organisms. However, from mice and human UNG-deficient patients it is evident that SMUG1 is not able to compensate for UNG2 in Ig diversification in B-cells (,,,). Furthermore, old mice develop B-cell lymphomas (,). Whether human individuals lacking UNG will develop malignancies remain unknown since they are yet too few identified and too young for conclusions to be made (). A more comprehensive knowledge of the short-term and long-term consequences of deficient uracil removal require further studies of the mice and generation and characterization of mice and double knockout mice.

Alternative RNA splicing of a single gene transcript is a common strategy to generate multiple protein isoforms with different functional properties (). A type of alternative splicing is the inclusion/exclusion of exon between competing 5′ splice sites (). The regulation of this type of alternative splicing has been shown to be quite complex in genes expressed in a variety of tissues and in artificial chimeric genes (,). In the central nervous system (CNS), much of the transcript and protein complexity is generated by alternative splicing, which is regulated in a cell- and development-specific manner (,). Proteolipid protein (PLP), a major CNS myelin protein comprises two protein products, PLP and DM20, which are generated by alternative splicing of two competing 5′ donor sites resulting in either inclusion or exclusion of exon 3B (). In oligodendrocytes (OLs), the myelin-producing cells of the CNS, the PLP 5′ splice site is preferentially utilized, while in OLs progenitor cells (OPCs) and in other cell types, DM20 is the preferred site (,). Although DM20 is the ancestral gene and PLP did not appear until relatively recently in evolution (,), PLP is the more abundant of the two protein isoforms in the mammalian post-natal brain and is uniquely expressed in myelin. The inclusion of exon 3B confers PLP unique signaling functions in axo-glial interactions that maintain axonal integrity (,). In differentiated OLs and in the post-natal brain, the ratio of PLP to DM20 transcripts is 3 : 1 and accounts for the preponderance of the PLP protein isoform. Tight control of the PLP/DM20 ratio is critical for normal brain development and function. Mutations that impair the PLP/DM20 ratio cause a spectrum of developmental and degenerative disorders in humans (). Thus, the identification of regulatory sequences and factors that control the PLP/DM20 ratio is of critical importance for the understanding of brain development and human disease and may also provide insight into the evolutionary history of alternative splicing. A PLP splicing construct, in which the alternative inclusion of PLP exon 3B is reconstituted was used to characterize the effect of disease-causing mutations on exon 3B splicing (). An analysis of mutations at and around the PLP and DM20 splice site showed that the intrinsic strength of the competing 5′ splice sites contributes to the final PLP/DM20 ratio (). Although the DM20 5′ site is weaker than the PLP 5′ site (), it is the preferred site in OPCs and non-glial cells, suggesting that either an enhancer of DM20 or a silencer of PLP 5′ splice site may regulate PLP alternative splicing in these cells. Analysis of mutations occurring in patients have identified enhancers of PLP 5′ splice site: a G-rich intronic splicing enhancer (ISE) () and an exonic splicing enhancer (ESE) containing an ASF/SF2 motif in exon 3B (). In other genes, the exon between competing 5′ splice sites has been shown to contain enhancers and silencers that regulate the alternative splicing selection (,). In this study, we investigate the role of regulatory sequences in PLP exon 3B by systematically mutating exon 3B. We characterize a novel G-rich enhancer of DM20 5′ splice site and its relationship with a G-rich ISE of PLP 5′ splice selection using functional and biochemical approaches. The hnRNP H and F bind to both enhancers. A reduction in the hnRNPH/F expression levels correlates temporally with the inclusion of PLP exon 3B in differentiated OLs. We investigate the role of hnRNPH and F in the regulation of PLP/DM20 ratio by knocking down their expression. This study defines a novel synergistic regulation of PLP/DM20 mediated by hnRNPH and F and establishes a role of a novel G-rich enhancer in PLP alternative splicing. Immunoselected OPCs were cultured in B104-conditioned medium and differentiated into OLs in T3 medium (40 ng/ml) for 72 h (,). Oli-neu cells (kind gift of Dr Trotter) were cultured in SATO medium with 1% horse serum and differentiated in dbcAMP, 1 mM for 3–10 days (). L cells were grown in DMEM containing 10% FBS. OPCs plated at 7 × 10 cells/well, Oli-neu cells at 5 × 10 cells/well and L cells at 3 × 10 cells/well were transfected for 6 h with 0.5 µg plasmid DNA using Qiagen Effectene Transfection Reagents (Qiagen, Valencia, CA, USA). After 18 h in growth medium, the cells were cultured in differentiation medium and harvested 72 h after transfection. Mutant PLP-neo constructs were generated by site-directed mutagenesis using the QuikChange Site-Directed Mutagenesis Kit and the QuikChange Multi Site-Directed Mutagenesis Kit (Stratagene, La Jolla, CA, USA) (). Undifferentiated Oli-neu cells were transfected with siRNA (50–100 nM) alone or in combination with reporter plasmids (0.5 μg) in siPORT amine transfection reagent using the neofection protocol (Ambion, Austin, TX, USA). To keep constant the amount of siRNA, Silencer® Negative Control #1 siRNA was added to transfections in which a single siRNA was used. Western blots of cell lysates (10 μg proteins) were performed using the One-Step ™ Complete Western Kit (Genescript, Piscataway, NJ, USA). Blots were probed with 10 μg primary antibody to hnRNP F [rabbit polyclonal, generous gift of Dr Milcarek, (,)] and hnRNP H (Bethyl Laboratories). Bands were quantitated with Kodak 440CF Digital Image station using 1D analysis software with actin as a standard for loading. Total RNA was extracted with RNeasy Mini Kit (Qiagen, Valencia, CA, USA) and treated with the DNA-free Kit (Ambion, Austin, TX, USA). Total RNA (0.5 μg) was reverse transcribed using random hexamer primers (BD Biosciences, Palo Alto, CA, USA). The PLP and DM20 PCR products and PLP+DM20 product derived from PLP-neo were amplified and quantitated as described () (A and A). Nine micrograms of proteins of nuclear extracts (NEP, Pierce) were separated by 10% SDS-PAGE, blotted and reacted with MAb104 (ATCC), antibodies to β-tubulin (Sigma, T 4026), PCNA (Santa Cruz), hnRNPH and F [rabbit polyclonal, generous gift of Dr Milcarek, (,)], A1 (MAb9H10, generous gift of Dr Dreyfuss) and L (AbCam) diluted 1 : 2000 and QKI5 (Bethyl Laboratories) diluted 1 : 1000, HRP-conjugated secondary antibody (Jackson ImmunoResearch Laboratories) diluted 1 : 5000 and developed with enhanced chemiluminescence (ECL, Amersham) (). Blots were quantified with Kodak 440CF Digital Image Station using 1D analysis software with tubulin as a standard for loading and quantification. Biotinylated RNA oligonucleotides (500 pmol) (Integrated DNA Technologies, Inc) were incubated with 200 μg proteins of nuclear extracts in buffer containing 4 mM creatine, 2 mM ATP, 1.5 mM MgCl, 1.2 mM DTT for 20 min at 30°C, heparin was added and the reactions were exposed to 254 nm UV light with a Spectronics XL-1000 UV crosslinker at a setting of 1.8 J/cm on ice (about 10 min). RNA-binding proteins were precipitated with streptavidin beads (Pierce, Rockford, IL, USA). Fifty microliters of prewashed immobilized streptavidin beads, 800 μl binding buffer and 10 μl Halt Protease Inhibitor Cocktail (Pierce, Rockford, IL, USA) were added to the UV-crosslinked reactions, incubated at 4°C in a rocking platform, and washed at least eight times with binding buffer prior to gel electrophoresis. Proteins from half of the entire mixture were separated by 10% SDS-PAGE and either visualized by silver staining (Bio-Rad) or detected by Western blot. Bands separated in 1D gel stained with silver stain were excised, destained and digested with trypsin. Samples were prepared and loaded on the MALDI plates and analyzed by Qstar XL with the MALDI source (Proteomics and Mass Spectroscopy Core Facility, University of Kentucky). The three most abundant peptide ions were automatically selected to perform MS/MS to obtain sequence information and help increase the confidence of a match with a known protein. Measured peptide masses obtained with MALDI–MS/MS and LC–MS/MS were compared with peptide masses from an in silico digestion of the protein database using MASCOT search engine for protein identification and modification detection. To identify novel regulatory sequences that control the PLP/DM20 ratio, we have systematically mutated PLP exon 3B in a PLP splicing construct and assessed splicing in transfected Oli-neu cells (A). In published work, we transfected primary OLs (). In the present study, we have used Oli-neu cells, immortalized OPCs that are induced to differentiate by dbcAMP (,). Because the efficiency of transfection in Oli-neu cells is 36% compared with 6% in primary OLs (data not shown), changes in plasmid-derived PLP/DM20 ratio are more easily quantified. We have validated the use of Oli-neu cells as a model for the OLs in splicing studies. We have established that various PLP-neo constructs previously tested in primary OLs expressed the same PLP and DM20 products when transfected into Oli-neu cells (data not shown). Expression results for some of the critical mutations characterized in the present study in Oli-neu cells were confirmed in primary OLs (). Finally, we show that differentiated Oli-neu cells replicate the trend of changes in hnRNP's expression that we detect in differentiated OLs (). We made 10 PLP-neo constructs in which sequences of exon 3B (M1-M10) were replaced by a poly-T linker (). In order to not affect sequences that are necessary for recognition of the 5′ splice site by the spliceosome, we began replacing bases after position +8 relative to the DM20 5′ splice site (349, A) up to the fourth base preceding the PLP 5′ splice site (453, A) (). Ten bases were replaced except for M6 and M10, which both contain seven base changes. M6 spans an ASF/SF2 binding motif that regulates PLP alternative splicing () (B). Following transfection and expression of PLP-neo constructs carrying the mutated M1–M10 (M1-MT–M10-MT) in Oli-neu cells differentiated in medium containing dbcAMP for 72 h, the PLP and DM20 products were simultaneously amplified by RT-PCR (A). Replacement of a G-rich sequence in the 5′ end of exon 3B (M2) resulted in 5.4-fold increase in PLP/DM20 ratio compared with the WT (A). In contrast, replacement of the other exon 3B sequences caused a reduction in the PLP/DM20 ratio, with sequences replaced in M1, M8 and M10 having less of an effect (A). Similar changes in the PLP/DM20 ratio were detected in Oli-neu cells transfected with M1-MT/M10-MT and differentiated for 6 days (data not shown). To determine whether replacing M2 affects the PLP/DM20 ratio in non-glial cells, we transfected L cells with M1-MT through M10-MT and quantified the plasmid-derived PLP and DM20 products by RT-PCR. Although the PLP/DM20 ratio is low in L cells consistent with the cell-specific regulation of PLP 5′ splice site recognition (), replacement of M2 resulted in 7.8-fold increase in the PLP/DM20 ratio, suggesting that M2 is active in L cells (B). To confirm the functional relevance of M2, we replaced M2 with an IgM sequence (-GCATGACTCT-) (M2-MT2) (A) (). In M2-MT2-transfected Oli-neu cells, the PLP/DM20 ratio was 6.4-fold higher than in WT-transfected cells (B), confirming that M2 regulates the PLP/DM20 ratio. These data show that G-rich exonic sequences (M2) regulate the PLP/DM20 ratio in favor of the DM20 5′ splice site in both glial and non-glial cells, although the efficiency of exon 3B inclusion is greater in Oli-neu cells, suggesting a contribution of cell-specific context in the regulation of the PLP/DM20 ratio. The remainder of exon 3B favors PLP 5′ splice site selection. Interestingly, every substitution appears to affect the PLP/DM20 ratio, suggesting that multiple regulatory elements may be present in exon 3B. In support of this interpretation is the finding that one of these sequences M6 contains a previously characterized ASF/SF2 motif (B, ESE) and M6-MT reduces the PLP/DM20 ratio consistent with results obtained by introducing patients’ mutations at this ASF/SF2 motif (). In an artificial α-globin minigene construct containing G triplets downstream of duplicated 5′ splice sites, the G triplet was the basic functional unit (). To determine the functional relevance of G triplets within M2, PLP-neo constructs in which either the G triplets at the 5′ end (G2) or 3′ end (G3) or the internal GAG triplet (G2–G3) were replaced with Ts were transiently expressed in Oli-neu cells (A). The PLP/DM20 ratio was increased 4-fold by mutation of G2 (G2–MT) and 29-fold by mutation of G3 (G3-MT) compared with WT (B). In contrast, the PLP/DM20 ratio was greatly reduced by mutation of G2–G3 (G2–G3-MT) and the PLP product was not detectable (B). The data suggest that M2 is functionally complex and is composed of G triplets of different strengths separated by a triplet that has an opposite effect. We next sought to assess the function of M2 in primary OLs because these cells more closely replicate the developmental increase in the endogenous PLP/DM20 ratio observed (A). OPCs were transfected with WT, M2-MT, M2-MT2 and G3-MT and differentiated into OLs for 72 h. The PLP/DM20 ratio was 9.4-fold higher in M2-MT-transfected OLs than in WT-transfected OLs and in M2-MT2- and G3-MT-transfected OLs only the PLP product was amplified. The results replicate those obtained with Oli-neu cells (compare B with 3C), however, they also show that the increase in PLP/DM20 ratio in OLs is greater than in Oli-neu cells in keeping with the higher degree of differentiation reached by OLs versus Oli-neu cells. The data suggest that differentiation may affect the efficiency of PLP alternative splicing derived from the construct. We next sought to determine whether M2 functions as a silencer of PLP 5′ splice site or an enhancer of DM20 5′ splice site. We inactivated the DM20 5′ splice site by changing the invariant G (349 in exon 3B) to a T (A). If M2 suppresses PLP 5′ splice site, inactivating the DM20 5′ splice site should not increase the PLP/DM20 ratio. The G>T change caused a complete loss of the DM20 product and a corresponding increase in the PLP product in differentiated Oli-neu cells (B). When both the G>T and the M2-MT mutations were introduced in PLP-neo, the results were similar to those obtained with the G>T mutation alone (B). A PCR product of ∼100 bp, which was consistently amplified from the G>T and G>T-M2-MT-transfected cells (B) was sequenced and it is the spliced product in which exon 3 is completely skipped. These data suggest that M2 is an enhancer of DM20 5′ splice site. We next inactivated the PLP 5′ splice site by mutating the invariant +2T (c.453+2T>C) in the M2-MT construct. This mutation was identified in humans and abolishes the PLP 5′ splice site (). If M2 is an enhancer of DM20 5′ splice site, we would expect to detect a reduction of the DM20 product derived from c.453+2T>C-M2-MT compared with the WT and c.453+2T>C. The DM20 product is greatly reduced in c.453+2T>C-M2-MT-transfected Oli-neu cells, while it is amplified in approximately the same amount in c.453+2T>C as in WT-transfected Oli-neu (C). No PLP product is detected in either construct. The plasmid-derived PLP+DM20 PCR product amplified with a forward primer in the neo gene and reverse primer in PLP exon 3A is drastically reduced in c.453+2T>C-M2-MT compared with WT and c.453+2T>C (C). This reduction could be explained by skipping of exon 3 due to inefficient recognition of the DM20 5′ splice site caused by the M2-MT or less likely inefficient 3′ splice site utilization. Alternatively, the reduction in the combined PLP+DM20 product may be explained by retention of intron 3 resulting in an unstable transcript. However, PCR products of the predicted size for exon 3 skipping and intron 3 retention were not detected. Collectively, these results support the interpretation that M2 is an enhancer of DM20 5′ splice site. M2 and ISE contain G-rich sequences that have nearly identical composition, raising the possibility that a balance between these enhancers regulates PLP/DM20 ratio (A). To evaluate their functional relationship, we asked whether M2 and ISE enhance the upstream 5′ splice site when they are exchanged. The G-rich core and flanking sequences of the 19-bp ISE contribute to the enhancer's function (). In order to move the ISE and not change the distance of G sequences from the 5′ splice site, we replaced M2 and the five bases 5′ and four bases 3′ of M2 (19 nt, M2F) with the ISE and vice versa. We have made the following constructs: ISE-ISE, M2F-M2F and M2-MT/ISEdel (A). The naturally occurring ISE deletion (ISEdel) reduces PLP 5′ splice site selection (). PLP and DM20 PCR products were amplified in Oli-neu cells transfected with these constructs and with PLP-neo, M2-MT and ISEdel (B). The PLP/DM20 ratio derived from ISE-ISE and M2F-M2F was lower than the ratio derived from the WT, suggesting that the ISE is a strong enhancer of DM20 5′ splice site and M2F is a weak enhancer of PLP 5′ splice site. The PLP/DM20 ratio derived from M2-MT/ISEdel was increased compared with WT, although it was not as high as that derived from M2-MT (B and C). These data suggest that although the ISE and M2F can replace each other, they differ in the strength by which they enhance the 5′ splice site when removed from their natural position. To further examine M2F and ISE, we analyzed proteins that bind to M2F and ISE in differentiated Oli-neu nuclear extracts by RNA-affinity precipitations. Biotinylated 19 bp RNA oligoribonucleotides containing the M2F and ISE sequences and templates in which the Gs were replaced by poly-U, M2F-MT and ISE-MT (A) were incubated with nuclear extracts. A reaction without the oligoribonucleotide was used as control for non-specific binding to the matrix. The UV-crosslinked protein–RNA complexes were precipitated with streptavidin–agarose beads and the proteins were separated by 10% SDS-PAGE and visualized by silver staining. Representative RNA-affinity precipitates ( = 2) are shown in B. While a number of protein bands of similar molecular weights (MWs) in the >100 to 50–55 kDa and 45–37 kDa range were detected in both ISE and M2F precipitates, there were bands uniquely present in either precipitate (shown by asterisks, B). The protein bands detected in non UV-crosslinked precipitates were similar in gel mobility and pattern to those detected in UV-crosslinked precipitates, suggesting that the differences in protein bands between ISE and M2F precipitates are likely to reflect real differences in protein binding rather than shifts in mobility caused by UV-crosslinking ( = 2, data not shown). The data suggest that proteins of similar and distinct MWs bind to M2F and ISE. In precipitates with the ISE-MT, high MW bands were absent while protein bands in the ∼45–37 kDa range were still detected (B). In contrast, nearly all protein bands were not detected in precipitates with M2F-MT (B). The generalized reduction in protein binding to M2F-MT is not due to inefficient synthesis and/or amount of the M2F-MT template, since all RNA templates were quantified prior to the RNA-affinity precipitations (data not shown). In addition, the differences in binding are not due to unequal protein loading to each reaction. Similar levels of PCNA were detected by Western blot analysis of each RNA–protein mixture separated by SDS-PAGE prior to streptavidin-beads precipitation (B, upper panel). These data support the interpretation that the presence of the G-rich sequences is necessary for binding of all proteins to M2F, whereas it is required for binding of some, but not all proteins to the ISE. To identify proteins whose binding depends on the presence of G-rich sequences, we have analyzed the affinity-precipitated proteins by Western blot with antibodies to known RNA binding proteins. On the basis of MW, binding motifs in the G-rich sequences, and published data on proteins binding to G-rich elements (,,), we anticipated that proteins in the MW range 37–65 kDa were likely to be hnRNPs. The RNA-affinity precipitates were separated by SDS-PAGE, blotted and probed with antibodies to hnRNPs. The hnRNPA1, H, F and L were detected in the precipitates with M2F and ISE, but were not detected in precipitates with M2F-MT and ISE-MT (C). While very strong binding motifs are present in M2F and ISE, only partial motifs for hnRNPL are identified. To test the specificity of the RNA affinity precipitations, we have probed Western blots of the RNA affinity precipitates with an antibody to QKI5, an RNA binding protein that contains a heterogeneous nuclear ribonucleoprotein K homology (KH) domain () and is not expected to bind to M2F and ISE and an antibody to PCNA, an unrelated nuclear protein. We found that both QKI5 and PCNA do not bind to either the wild-type or mutated RNA templates (C). These data demonstrate the specificity of the RNA affinity assay. We next demonstrated that hnRNPH, F and A1 are present in the RNA–protein complexes by LC/MS/MS of three discrete bands in the 37–65 kDa MW range of the ISE precipitates. The peptide sequence data were matched with the protein database and confirmed to be hnRNPF, H and A1 (block arrows, B). The results demonstrate that hnRNPs bind to the ISE and M2F in G sequence-dependent manner and suggest that these factors may participate in the enhancer's function. We next investigated whether hnRNP A1, H, F and L are differentially expressed in OLs versus OPC by Western blot analysis of nuclear extracts prepared from OPC and OLs differentiated for 72 h using antibodies specific for these hnRNPs (B). Significant reductions in hnRNP F (47%), hnRNPH (45%) and hnRNPA1 (37%) levels were detected in OLs compared with OPC extracts ( = 5) (B). No change in the hnRNPL level was detected (B). Expression of the SR proteins detected with the MAb104 antibody did not change in OLs versus OPC (data not shown) and is in keeping with previous expression studies of ASF/SF2 in OLs (). CNPase, a marker of OLs differentiation was expressed in cytoplasmic extracts of OLs, but not in OPC, indicating that the cells have differentiated (B). Changes in hnRNPs expression are associated with the developmental switch in the endogenous PLP/DM20 ratio. The PLP/DM20 ratio was 3 : 1 in OLs compared with 1.5 : 1 in OPC ( = 3) (A) and replicates the developmental increase in PLP/DM20 ratio in the post-natal brain (). The data show a temporal association between decrease in hnRNPs expression levels and increase in the PLP/DM20 ratio. We next assessed whether changes in the expression of hnRNPs in differentiated versus undifferentiated Oli-neu cells replicate the pattern observed in primary OLs. Oli-neu cells were either grown in serum-containing medium or differentiated for 3, 7 and 10 days in dbcAMP-containing medium. Oli-neu cells reach a higher degree of differentiation after 7 and 10 days of culture in dbcAMP-containing medium (). Significant reductions in hnRNP F and A1, but only modest decreases in hnRNPH were detected in Oli-neu differentiated for 7 and 10 days compared with undifferentiated and 3 day-differentiated Oli-neu cells (C). CNPase expression was used as an internal marker of differentiation (C). These results demonstrate that changes in hnRNPs associated with differentiation of Oli-neu cells generally replicate the results obtained with OLs. Together, the data show that a decrease in hnRNP H, F and A1 expression occurs in differentiated OLs and is temporally associated with differentiation-dependent increase in PLP/DM20 ratio in OLs. The coordinate decrease in the expression of hnRNP H, F and A1 in OLs suggests that these splicing factors may regulate the changes in the PLP/DM20 ratio. We have tested whether RNAi-mediated removal of either hnRNPH or F is sufficient to increase the PLP/DM20 ratio in undifferentiated Oli-neu cells in which the endogenous expression of H and F is high. Oli-neu cells were treated with siRNA that either target hnRNPH (siH1, H2, H3) or hnRNPF (siF1, F2 and F3) (Ambion, see Materials and Methods section) and hnRNPH and F levels were quantified by Western blot of cell lysates prepared 72 h after transfection. hnRNPH was reduced by 70% with siH2, 50% with siH3 and 40% with siH1 (A). hnRNPF was reduced by 60% with siF2 and 40% with siF3, while was unaffected by siF1 (A). Expression of hnRNP A1 and L was not changed by treatment with the siRNAs, confirming the specificity of the knock down (data not shown). Subsequent experiments were carried out with siH3, H1 and siF2, F3. To determine whether knock down of either hnRNPF or H affects the PLP/DM20 ratio, we transfected PLP-neo and siRNAs into Oli-neu cells. The plasmid-derived PCR products were amplified in RNA prepared from transfected Oli-neu cells cultured for 72 h in growth medium (B). Knock down of hnRNPH resulted in >2-fold increase in the PLP/DM20 ratio in cells treated with siH3 compared with mock treated cells (0.98 versus 0.36) (similar results were obtained with H2, data not shown), a smaller increase was induced by siH1 (0.56 versus 0.36), consistent with lower silencing efficacy of siH1 on hnRNPH expression (B, upper panel). The PLP/DM20 ratio was not increased by treatment with siF2 and siF3, although the levels of hnRNPF are significantly reduced by the treatment (B, upper panel). These data show that reduction in hnRNPH is sufficient to increase the PLP/DM20 ratio, while knock down of hnRNPF does not have an effect on the PLP/DM20 ratio. To determine whether knock down of both hnRNP H and F cooperatively regulates the PLP/DM20 ratio, we co-transfected Oli-neu cells with siF3+siH3 and PLP-neo. The simultaneous knock down of hnRNPF and H caused 10-fold increase in the PLP/DM20 ratio compared with mock treated cells (3.5 versus 0.36) (B). siF2 combined with either siH2 or siH3 treatments gave similar results (data not shown). The data suggest that hnRNPH and F cooperatively regulate PLP/DM20 ratio and their knock down results in a synergistic effect. These findings were confirmed by knock down of hnRNPH and F with a custom-made siRNA that targets both hnRNPF and H (siF/H) (Ambion, see Methods and Materials section). siF/H reduced the expression of hnRNPF by 60% and hnRNPH by 50% (A) and caused a 10-fold increase in the PLP/DM20 ratio (B, upper panel). We have also determined that knock down of hnRNPH and hnRNPH+F increased exon 3B inclusion in GloPLP-derived products similarly to PLP-neo (data not shown), suggesting that exon 3B and intron 3 contain regulatory elements that mediate the effects of hnRNPH and F. We next evaluated whether knock down of hnRNPH and F increases the endogenous PLP/DM20 ratio. Knock down of both hnRNPH and F caused 2-fold increase in the endogenous PLP/DM20 ratio compared with mock treated cells (1 versus 0.5), while removal of hnRNPH increased the PLP/DM20 ratio modestly (0.8 versus 0.5) (B, lower panel). These results demonstrate a role of hnRNPH and F in the regulation of the endogenous and plasmid-derived PLP alternative splicing. Because hnRNPH and F bind to M2 and ISE, we tested whether the presence of M2 and ISE is required for the hnRNPH and F-mediated effect on PLP/DM20 ratio. If either M2 or ISE is necessary for hnRNPH and F transactivation, mutation of these enhancers would reduce the increase in the PLP/DM20 ratio induced by removal of hnRNPH and F. Knock down of hnRNPH did not change the PLP/DM20 ratio derived from M2-MT compared with mock treated cells, while knock down of both hnRNPH and F caused <3-fold increase in PLP/DM20 ratio (5.5 versus 2), which is much lower than the 10-fold increase in PLP/DM20 ratio derived from the WT (A). Although the PLP/DM20 ratio derived from M2-MT is relatively high (), the smaller increase in PLP/DM20 ratio is not due to limitation in detection and quantification of the PCR products, as we can accurately quantify PLP/DM20 ratio of up to 10 (). The PLP/DM20 ratio derived from the ISEdel was increased 3-fold by removal of hnRNPH (0.37 versus 0.13) (B). Knock down of hnRNPH and F with siF3+siH3 increased the PLP/DM20 ratio 7-fold (0.93 versus 0.13) and with siF/H increased the ratio 10-fold (1.32 versus 0.13) (B). Although the efficiency of PLP 5′ splice site selection is low in the ISEdel, the fold increase in the PLP/DM20 ratio induced by knock down of hnRNPH and hnRNPH+F is similar to that derived from the PLP-neo. When both M2 and ISE are mutated (M2-MT/ISEdel) knock down of hnRNPH did not change the PLP/DM20 ratio (C) similar to the results obtained with M2-MT (A). The PLP/DM20 ratio derived from M2-MT/ISEdel was increased >2-fold by treatment with siF3+siH3 (1.43 versus 0.6) and >3-fold by siH/F (2.13 versus 0.6) (C) similar to the results obtained with M2-MT (A). These data suggest that disabling M2 interferes with hnRNPH-mediated effect on the PLP/DM20 and greatly reduces, although does not completely eliminate the synergistic effect induced by removal of hnRNPH and F. The latter result suggests that sequences other than M2 and ISE and contained in PLP exon 3B and intron 3, also participate in mediating the synergistic effect. We next determined whether the other G-rich sequences present in exon 3B may mediate the increase in PLP/DM20 ratio derived from M2-MT/ISEdel after knock down of hnRNPH and F. We have replaced G runs in M1 (G1), M6 (G4) and M8 (G5) with T's in the M2-MT/ISEdel (A) and determined the PLP/DM20 ratio derived from this construct in transfected Oli-neu cells (B). The PLP/DM20 ratio derived from M2-MT/ISEdel/G1-G4-G5-MT was 0.22 which is similar to the ratio derived from the WT (A). The PLP/DM20 ratio derived from M2-MT/ISEdel-G1-G4-G5MT in Oli-neu cells treated with siF3 + siH3 did not change compared with mock treated cells (0.27 versus 0.22), while ∼1.6-fold increase was detected after treatment with siF/H (0.4 versus 0.22) (B). The data show that mutations of G1, G4 and G5 in addition to M2 almost completely abolish the PLP/DM20 increase induced by knock down of hnRNPH/F and suggest that some or all of these G runs participate in mediating the synergistic effect. Regulation of PLP alternative splicing and maintenance of the PLP/DM20 ratio are critical for brain function and are relevant to neurological disorders in humans (). In this study, we have identified a novel G-rich enhancer (M2) of DM20 5′ splice site selection and show that it is active in OLs and non-glial cells. However, the overall efficiency of PLP alternative splicing and the impact that mutation of M2 has on the PLP/DM20 ratio is dependent on cell-specific (compare Oli-neu with L cells) and differentiation-dependent factors (compare OLs versus Oli-neu). The presence of M2 and ISE downstream of DM20 and PLP 5′ splice sites suggests that the PLP/DM20 ratio may be regulated by a balance between these enhancers. We have addressed this question both functionally and biochemically. When switched, M2 and ISE differ in the strength with which they enhance the upstream 5′ splice site. Changes in the distance of the G-rich sequences from the 5′ splice site in the switched position and the presence of putative ESE in the exonic context and their potential interaction with the G-rich enhancer may account for the difference in strength. However, we favor the hypothesis that sequences immediately adjacent to the G runs may influence the enhancer's strength. In the ISE, the flanking sequences enhance the PLP 5′ site selection (), while sequences adjacent to M2 reduce the DM20 5′ splice site selection (, M1-MT and M3-MT). In addition, the biochemical studies show that some proteins exclusively bind to either M2 or ISE and mutations of the G-rich sequences have a different impact on the binding of proteins to M2F and ISE. Together, these data would support the interpretation that sequence-dependent differences in protein binding may play a role in defining the function of M2 and ISE. Identification and characterization of these proteins in future studies will allow to test this hypothesis. In this report, we show that the G runs are necessary for binding of hnRNPs to both enhancers. Potential binding motifs for hnRNPH and F are poly G sequences in M2 and ISE (,) while hnRNPA1 may bind to a partial consensus sequence -AGGG- () () present in single copy in ISE and in double copy in M2. An unexpected finding is that hnRNPL binds to M2 and ISE, since the high affinity site for L is not present in M2 and ISE (,) and only partial motifs, CCA in M2F and ACA in the ISE are found. It is possible that hnRNPL does not bind directly to M2 and ISE, but is a part of a molecular complex formed by sequence-specific binding of hnRNPH, F and A1 to these enhancers. Unlike previous findings that the G triplet is the basic functional unit of G-rich enhancers (), in M2 the 3′ G triplet exerts a dominant enhancing effect, while the bases separating the G triplets reduce DM20 5′ splice site selection. The functional complexity of M2 may modulate the overall strength of the enhancer in response to changes in RNA-binding proteins during OLs differentiation. The developmental decrease in hnRNPF, H and A1 in differentiated OLs would support the hypothesis. The hnRNPs are differentially expressed in various tissues () and hnRNPH is linked to differentiation of muscle cells (), however, to our knowledge, coordinate decrease in hnRNPH, F and A1 associated with differentiation was not previously reported. Removal of both hnRNPH and F has uncovered a novel synergistic effect on PLP alternative splicing regulation mediated by hnRNPH and F, and points to a critical role of hnRNPH in this process. The hnRNPH is sufficient to affect the PLP/DM20 ratio and is necessary for hnRNPF to affect the PLP/DM20 ratio. Although knock down of both hnRNPH and F reduces the shorter of the two products derived by alternative splicing of competing 5′ splice sites in the bclx gene more strongly than knock down of either hnRNPH or F alone, a synergistic effect was not identified in these studies (). The broader significance of hnRNPH and F synergistic effect in regulating alternative 5′ splice site selection of other genes remains to be investigated. Our findings raise important questions as to whether the synergism of hnRNPF and H is cell-specific and limited to OLs or represents a more general way of regulating 5′ alternative splicing and whether the synergism is gene context specific. We have started to investigate -acting elements that may participate in the hnRNP's mediated regulation and have assessed whether M2 and ISE are necessary for hnRNPH and F to affect the PLP/DM20 ratio. The presence of M2 appears to be required for the effect of hnRNPH, since removal of hnRNPH does not increase the PLP/DM20 ratio when M2 is mutated. In contrast, the synergistic effect of hnRNPH+F knock down depends only in part on the presence of M2, since mutation of M2 either alone or in combination with ISEdel greatly reduces the fold increase in the PLP/DM20 ratio, but does not completely abolish it. However, when the other G runs in exon 3B are mutated the synergistic effect is almost completely eliminated. We interpret these results to indicate that M2 in part contributes to the effect of hnRNPH+F on the PLP/DM20 ratio and that G runs in exon 3 B sequences other than M2 and ISE also play a role. The finding that replacement of G runs in different positions in exon 3B affects the hnRNPH/F-mediated regulation of PLP/DM20 ratio corroborates the data obtained with the M1–M10 substitutions and support that multiple regulatory sequences are present in PLP exon 3B. The role and relative contribution of each G run to the synergistic effect remain to be elucidated. How M2 may participate in hnRNPH- and F-dependent regulation of PLP alternative splicing remains to be determined. Our data do not demonstrate that hnRNPH and F act directly through M2. For instance, M2 may enhance DM20 5′ splice site selection through a mechanism that is independent from hnRNPH and F such as formation of stacked G structures (). Loss of the stacked G structure may impair DM20 5′ splice site recognition reducing the efficiency of splicing rather than affecting hnRNPH- and F-mediated -activation directly. Alternatively, loss of G stacked structure may indirectly reduce the efficiency of hnRNPH and F binding to other cognate sequences. Our data suggest that ISE does not participate in the hnRNPF- and H-dependent regulation of the PLP/DM20 ratio at least in undifferentiated OLs. However, it is possible that a role of ISE is overlooked in our studies if reduction of hnRNPH and F needs to be coupled to differentiation-induced changes in other cell-specific and ubiquitous -acting splicing factors (,). In summary, our studies have identified a novel G-rich enhancer and a synergistic effect of hnRNPF and H in regulating the PLP/DM20 ratio and suggest that these splicing factors, the G-rich enhancer and other G runs may together regulate PLP/DM20 ratio in OLs. Elucidation of the mechanism that mediates the synergistic interactions of hnRNPH and F may reveal a novel role of these splicing factors in 5′ alternative splicing. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R o n l i n e .

Homologous recombination (HR) is a template-dependent, high-fidelity DNA damage repair and tolerance pathway that negotiates complex DNA damage, such as double-stranded DNA breaks (DSBs), gaps, and interstrand crosslinks, as well as stalled or collapsed replication forks (). Defects in HR lead to sensitivity to genotoxic agents and genomic instability (,). The process of HR can be conceptually divided into three stages. In presynapsis, the lesion is processed to a single-stranded DNA (ssDNA) intermediate. , ssDNA is likely bound by the ssDNA-binding protein RPA, requiring mediator proteins, such as Rad52 and the Rad55-Rad57 heterodimer, to allow formation of the presynaptic filament of Rad51-ATP-ssDNA (6–9). During synapsis, the presynaptic filament mediates homology search and DNA strand invasion to form a joint molecule or D–loop (). In postsynapsis, Rad51 is assumed to dissociate from the product heteroduplex DNA to allow DNA synthesis from the invading 3′ end. After DNA synthesis, different sub-pathways (Synthesis-Dependent Strand Annealing, DSB Repair) generate contiguous chromosomes (). Rad51 and Rad54 are two key proteins in HR (). Like RecA, Rad51 forms a helical filament on DNA in the presence of ATP and catalyzes homology search and DNA strand invasion (10–13). This reaction produces heteroduplex DNA, a central recombination intermediate. DNA-binding and DNA-strand exchange activity by budding yeast Rad51 requires ATP-binding but not ATP hydrolysis as revealed by the analysis of mutants in the invariant lysine residue within the Walker A box (K191 in budding yeast), as well as by using slow or non-hydrolyzable ATP analogs (10,14–19). However, ATP hydrolysis appears to be required for protein function (). Rad51-K191R encodes a protein that can presumably bind ATP, although this was never directly demonstrated, but not hydrolyze ATP (). The mutant protein when expressed at native levels in haploid cells conferred severe sensitivity to IR and a DSB-repair defect (). Similarly, the equivalent RecA mutant (RecA-K72R) cannot complement the UV sensitivity of a RecA-deficient strain (,). Other ATPase defective RecA mutants show less severe defects (,). When expressed at high levels, budding yeast Rad51-K191R complements the MMS- and IR-sensitivity of a Rad51-deficient strain to near completion (,). The importance of ATP hydrolysis for the function of RecA-like proteins is underlined by the dominant-negative effect exerted by the equivalent human Rad51-K133R mutant protein when expressed in mouse, chicken or human cells (25–28). The and results regarding the requirement of ATP hydrolysis for Rad51 (RecA) function present an apparent contradiction. However, DNA strand invasion assays typically sidestep the requirement for product release by Rad51 (or RecA), which is achieved by SDS/proteinase K treatment. Indeed, biochemical and electron microscopic analysis of the human Rad51 protein showed that Rad51-DNA complexes are stabilized under conditions that inhibit ATP hydrolysis (,), suggesting a need of ATP hydrolysis for Rad51 turnover from dsDNA. Also experiments with RecA protein have demonstrated the need for ATP hydrolysis to release the heteroduplex product (). These experiments suggest that ATP hydrolysis by Rad51 is needed to enhance filament dynamics. Rad54 protein is a member of the Swi2/Snf2 family of dsDNA-dependent ATPases (,) and processively translocates along the dsDNA lattice at up to 300 bp/sec (). Deletion of the gene in budding yeast confers as strong a mitotic recombination defect or IR sensitivity as a deficiency in the gene (,). function of Rad54 protein is largely compromised in the mutant, which encodes a mutant Rad54 protein that is defective in ATP hydrolysis (33–36). Rad51 and Rad54 act in a single molecular pathway and physically interact (37–39). Rad54 was found to stabilize Rad51-ssDNA filaments in the presynaptic phase of recombination in an ATPase-independent fashion (40–42). Rad54 also stimulates DNA strand exchange at the synaptic stage (35,43–46). The mechanism of this stimulation is not fully understood but requires the motor function of Rad54 and depends on its ATPase activity. Possible mechanisms include the sliding of the target dsDNA during homology search, opening the target dsDNA by translocating on duplex DNA, or removing Rad51 from dsDNA that inhibit Rad51-mediated DNA strand exchange (,). Moreover, Rad54 was demonstrated to be able to remodel nucleosomes , although no effect of Rad54 on nucleosome positioning during DSB repair has been identified in budding yeast (42,47–49). During postsynapsis, Rad54 was shown to enhance branch migration during DNA strand exchange (,), but it is unclear whether Rad54 acts as a junction-specific motor protein like RuvB in bacteria (). Overexpression of Rad54 leads to a decrease in conversion tract length () not increase as predicted, if Rad54 were to drive branch migration. Genetic analysis suggests that Rad54 acts at a step after Rad51 protein, consistent with cytological data showing that is not required for Rad51 focus formation (54–58). Together with data from chromatin immunoprecipitation experiments (,), these results suggest that the critical function of Rad54 is either during synapsis or postsynapsis but the results are not able to distinguish between both possibilities. Since Rad54 is a dsDNA motor protein that interacts with Rad51, we have focused on the interaction between Rad54 and the Rad51-dsDNA filament (). We discovered that Rad54 remodels the Rad51-dsDNA filament, resulting in the dissociation of Rad51 from dsDNA (). Using electron microscopic analysis, Rad54 was found to preferentially localize to one terminus of the Rad51-dsDNA filament, consistent with Rad54 translocating along dsDNA and then docking at the end of the Rad51 filament or directly binding the filament end from solution (). These data support a model, in which Rad54 acts as a turnover factor for the Rad51-heteroduplex DNA product complex, resulting in product release by Rad51 to allow access of the DNA synthesis machinery to the invading 3′ end. This model also rationalizes significant differences between RecA and Rad51. RecA exhibits ∼200-fold higher ATPase activity on dsDNA and releases from DNA after ATP hydrolysis, whereas Rad51 appears to be much less dynamic in filament assembly/disassembly. The function of Rad54 as a Rad51 turnover factor is consistent with the absence of a Rad54 homolog in bacteria, because RecA is capable of turnover using its intrinsic ATPase activity. Here, we analyzed the roles of both the Rad51 and Rad54 ATPase activities in turnover of the Rad51-dsDNA filament. The results revealed a number of surprising features of the ATPase-deficient Rad51-K191R and Rad51-K191A proteins. First, unexpectedly Rad51-K191A protein bound ATP, albeit with reduced affinity, similar to human Rad51-K133A (). Second, the Rad51-K191R mutant protein displayed a DNA binding defect. Third, once formed the Rad51-K191R-dsDNA complexes were exceptionally stable. We demonstrate that the Rad54 ATPase activity was specifically enhanced in its interaction with the Rad51-dsDNA filament. Efficient Rad51 turnover from dsDNA required both the Rad51 and Rad54 ATPase activities. These results show that Rad51 is not a passive remodeling target of Rad54 and uncovered an intricate cooperation between the Rad51 and Rad54 ATPase activities in Rad51 turnover. Rad51, Rad51-K191R and Rad51-K191A proteins were purified after overexpression in the cognate host (,). Rad54 protein was overexpressed as an N-terminal GST fusion, and the tag was released by PreScission protease cleavage. The over-expression of Rad54 and its initial purification by glutathione-Sepharose 4B (Pharmacia) chromatography was performed as described (), except that the sodium chloride concentration was changed to 500 mM. After GST-Rad54 protein was bound on the GST-column, purified PreScission protease, itself a GST fusion protein, was loaded on the column at an enzyme to substrate molar ratio of about 1:1. The column was washed with PBS buffer (140 mM NaCl, 2.7 mM KCl, 10 mM NaHPO, 1.8 mM KHPO) containing 500 mM NaCl to remove unbound protease. Cleavage of the GST-tag from GST-Rad54 was achieved by sealing the column outlet, re-suspending the matrix in buffer A [(20 mM Tris–HCl pH 7.5, 1 mM EDTA, 10% glycerol, 10 mM β-mercaptoethanol, 1 mM phenylmethanesulfonyl fluoride (PMSF, Fluka), 1 μM pepstatin A, 2 μM benzamidin and 1 μM leupeptin] containing 500 mM NaCl, and incubation for 12 h at 4°C (). The untagged Rad54 was eluted from the column with buffer A. Fractions containing the untagged-Rad54 were pooled and applied to a hydroxylapatite (HAP) column. The HAP column was eluted by a KHPO buffer gradient from 20 to 500 mM (containing 500 mM NaCl), at a flow rate of 0.3 ml/min. The fractions containing Rad54 protein were collected, concentrated and dialyzed to storage buffer and stored at −80°C as described (). Single-stranded φX174 virion DNA and double-stranded φX174-RFI DNA were purchased from New England Biolabs. The 600-bp dsDNA substrate was prepared by standard PCR with φX174-RFI template using primer olWDH427 5′- TTATCGAAGCGCGCATAAAT-3′, and primer olWDH431 5′-GTCTTCATTTCCATGCGGTG-3′. The 621 nt ssDNA substrate was prepared by one-sided PCR from PstI-linearized φX174-RFI using primer olWDH431. Both DNA substrates were purified using the Qiagen gel extraction kit. Protein concentrations are given in moles of monomers, ssDNA concentrations are given in moles of nts, and dsDNA concentrations are given in moles of basepairs. Binding was performed in 10 µl reactions in ATPase reaction buffer [25 mM triethanolamine pH 7.5, 1.8 mM DTT, 13 mM magnesium acetate and 100 μg/ml bovine serum albumin (BSA)] containing 2.0 μM Rad51, Rad51-K191R or Rad51-K191A. For each reaction, ATP from a stock solution spiked with ∼30 μCi α-P-ATP was added to the indicated final concentration. After incubating 15 min at 30°C, the reactions were placed on a piece of parafilm on top of ice and irradiated at 254 nm in a UV-crosslinker (XL-1000, Spectronics) at about 10 cm from the light source. The total exposure intensity was ∼1260 μJ/cm. The reactions were incubated at 100°C for 5 min, and the proteins resolved by 10% SDS–PAGE. The gels were dried, and analyzed by PhosphoImager and quantified by ImageQuant software (Molecular Dynamics, Inc., Sunnyvale, CA, USA). For the azido-ATP labeling experiments, an ATP stock solution (400 µM) was spiked with α-P-8-Azido-ATP (MP Biomedicals) to 0.36 µCi/µl, and the experiment was conducted as described above. Salt titrations of the protein–DNA complex formation was performed by incubating Rad51 or Rad51-K191R (2 µM) with DNA (6 µM of 600- bp dsDNA or 621-nt ssDNA) in the presence of the indicated sodium chloride concentration for 15 min at 30°C. The reactions were then fixed with glutaraldehyde (0.25%). Nucleoprotein gel electrophoresis was conducted in 1% agarose in TAE buffer (40 mM Tris–Base, 20 mM acetic acid, 1 mM EDTA) for 2 h at 4 V/cm. The dried gels were analyzed by PhosphoImager and quantified using ImageQuant software (Molecular Dynamics, Inc., Sunnyvale, CA, USA). Salt titrations of the protein–DNA complex stability were performed as above but with incubation in the absence of sodium chloride for 15 min at 30°C to allow the formation of the protein–DNA complexes. Sodium chloride was then added to the indicated final concentrations, and the protein–DNA complexes were incubated for another 30 min at 30°C. The reactions were then fixed with glutaraldehyde (0.25%). Nucleoprotein gel electrophoresis and analysis was performed as described above. ATPase assays were performed in the ATPase reaction buffer [25 mM triethanolamine (pH 7.5), 1.8 mM dithiothreitol, 13 mM magnesium acetate and 100 µg/ml BSA], with 5 mM ATP, 2.5 µM of Rad51 (or Rad51-K191R, or Rad51-K191A), with or without 30 µM of ssDNA (poly-dA), for 10 min at 30°C. By supplementing 20 U/ml lactate dehydrogenase (Sigma), and ATP regeneration system [20 U/ml pyruvate kinase (Sigma), 3 mM phosphoenolpyruvate (Sigma) and NADH (Sigma) to give an absorbance of 1.6–1.8 (about 0.2 mg/ml)], the ATP hydrolysis rates were measured by a NADH-coupling microplate photometric assay for 60 min at 30°C as described (). ATPase assays were performed in ATPase reaction buffer with the indicated concentrations of ATP and 6 µM supercoiled φX174-RFI DNA for 15 min at 30°C. To form sub-saturated protein–DNA filaments (1 Rad51 per 37 bp), 0.16 µM of Rad51 or Rad51-K191R was added into the reactions. To form saturated protein–DNA filaments (1:3 bp), 2 µM of Rad51 or Rad51-K191R was added. The reactions were incubated for 60 min at 30°C, with the addition of Rad54 (3.3 nM) and the ATP regeneration system (1.5 mM of phosphoenolpyruvate, 30 U/ml of pyruvate kinase, 30 U/ml of lactate dehydrogenase and 0.2 g/ml of NADH) to initiate the reaction. The absorbance data were collected using an 8453A diode array spectrophotometer equipped with UV-visible ChemStation software (Agilent). The initial rates were calculated as described (). The calculated initial rates were fitted into the Michaelis–Menten kinetic equation by software PRISM (Graphpad). The reaction was performed as described (). Rad51 protein (7.5 μM) was bound to dsDNA (30 µM pUC19) in 10.5 μl reactions containing 30 mM Tris acetate pH 7.5, 1 mM DTT, 50 μg/ml BSA, 20 mM magnesium acetate, 20 mM ATP, 2.5 mM spermidine and an ATP regeneration system (20 mM phosphocreatine and 0.1 μg/μl creatine kinase) for 30 min at 23°C. Rad54 (0.375 μM) was added, followed after 5 min at 23°C by addition of scavenger DNA (63 µM PstI-linearized M13mp19 dsDNA). After incubation at 23°C for 2 h, 10 U of wheat germ topoisomerase I (Promega) was added, and dsDNA was allowed to be relaxed for 1 h at 23°C. Stop buffer (final concentrations 60 mM EDTA, 0.7 mg/ml proteinase K and 0.1% SDS) was added, and reactions were deproteinized for 20 min at 23°C. Two-dimensional gel electrophoresis was performed on 1.2% agarose gels at 45 V for 22 h in the first dimension (without chloroquine). The lanes were cut out of the gels and soaked in TBE buffer (89 mM Tris–Base, 89 mM boric acid, 2 mM EDTA) with 4 μg/ml chloroquine for 2 h. New 1.2% agarose gels containing 4 μg/ml chloroquine were poured for the second dimension, and electrophoresis was performed for 22 h at 65 V in TBE containing 4 μg/ml chloroquine. Gels were stained with ethidium bromide. Protein–DNA filaments were formed by incubating either Rad51 or Rad51-K191R (at a concentration of 1.5 µM) with calf thymus dsDNA (Sigma) in 25 mM Triethanolamine–HCl buffer (pH7.2) with 1.25 mM ATP and 10 mM magnesium acetate at 37°C for 15 min. The Rad51: DNA ratio was 40: 1 (w/w). Samples were applied to carbon-coated grids and stained with 2% uranyl acetate (w/v). Grids were imaged with a Tecnai 12 electron microscope operating at 80 keV with a nominal magnification of 30 000×. The Rad51 ATPase activity appears to be critical for its biological function (12,20,25–28). Yet, the purified Rad51-K191R (or human Rad51-K133R) protein can catalyze recombination similar to the wild-type proteins (,). Substitution of the Walker A box residue K191 to R in the Rad51 protein (and K133 in human Rad51) has been assumed to block ATP hydrolysis but leave ATP binding unaffected, whereas the K to A substitution is assumed to abolish ATP binding, based on previous data (,). These assumptions have not been experimentally tested for Rad51. To understand the function of ATP hydrolysis by Rad51, in particular during the disassembly of the Rad51-dsDNA product complex by Rad54, we purified Rad51-K191R and Rad51-K191A proteins from the cognate host to apparent homogeneity (A). In the absence of DNA substrate, wild-type Rad51 and the mutant Rad51-K191R, and Rad51-K191A proteins display a similar and very low background ATP hydrolysis rate. Addition of the poly-dA substrate to the reaction strongly stimulated the ATP hydrolysis rate by Rad51, while the ATPase activity of both Rad51-K191R and Rad51-K191A proteins remained at the background level (B). These results are consistent with a previous report on Rad51-K191R protein (). We determined the apparent ATP binding affinities for the Rad51 wild-type and Rad51-K191R/A proteins using an UV-cross-linking assay (). Wild-type budding yeast Rad51 protein displayed non-cooperative ATP binding with an apparent K of 8.8 µM (C, ). This value is similar to the one determined for the human Rad51 protein with ATP-γ-S (3–5 µM) (). The Rad51-K191R protein exhibited a reduced affinity to ATP (K = 52.5 µM) as compared with that of wild-type (C, ). Surprisingly, Rad51-K191A protein was found to bind to ATP with similar affinity as the Rad51-K191R protein (K = 34.2 µM). The binding of ATP is specific, as the BSA present in the reaction was not labeled (C). To further confirm these observations, the UV-cross-linking analysis was carried out with a photo-affinity analog of ATP, azido-ATP. The photosensitive azido group has been used as a tool to study the corresponding specific binding site for biological ligands (). Here, we use it as an alternative to study ATP binding by Rad51 protein. The results (E and F; ) were consistent with the previous data with α-P-ATP, showing that compared with Rad51, Rad51-K191R and Rad51-K191A exhibited reduced ATP binding affinity. We note that the Rad51-K191R showed better binding to azido-ATP than to ATP, whereas the Rad51-K191A protein showed reduced binding to azido-ATP compared with ATP (). The reasons for this difference remain unknown. The decreased affinities of the mutant proteins were also observed in the presence of ssDNA (data not shown). Azido-ATP led to minimal unspecific binding to BSA (E). Taken together, these results challenge the often-stated assumption that the lysine to alanine change in the Walker A box motif abolishes ATP binding. Both, Rad51-K191R and Rad51-K191A proteins bound DNA in a nucleotide-dependent manner (data not shown). We utilized primarily electrophoretic analysis of glutaraldehyde-fixed Rad51-DNA complexes, a method normally used with RecA (,) and human Rad51 (,). We have verified in previous studies that the electrophoretic mobility of yeast Rad51-DNA complexes corresponds within the limits of resolution to filament saturation using electron microscopy, topological assays (), and nuclease protection assays (,,). An earlier analysis failed to detect Rad51-K191A binding to DNA (), but using fixation prior to electrophoretic analysis stabilized the Rad51-K191A-filaments sufficiently to allow their visualization (data not shown). The Rad51-K191A-dsDNA filament displayed a significant defect in its interaction with Rad54 protein. Since the purpose of the study is to understand the interaction between the Rad51 and Rad54 proteins, we focused on the Rad51-K191R protein and its interaction with Rad54 for the further analysis. The DNA binding properties of the Rad51-K191R protein were analyzed by nucleoprotein gel assays and electron microscopy. Rad51-K191R protein formed protein–DNA filaments similar to wild-type Rad51 protein but with notable differences (). First, we titrated increasing amounts of wild-type or Rad51-K191R proteins to a 600-bp dsDNA substrate. Consistent with a previous determination (), Rad51 wild-type protein approached maximal binding at a stoichiometry of ∼3–4 bp per Rad51 subunit in the filament (A), as judged by the disappearance of free DNA. The Rad51-K191R protein required slightly higher protein concentrations, approaching maximal binding at a stoichiometry of 2 bp per subunit (A). The electrophoretic mobility of the Rad51-K191R-dsDNA complexes tended to be more homogeneous than those formed by the wild-type Rad51 protein. It appeared that the Rad51-K191R protein did not form the partially saturated complexes that are evident at sub-stoichiometric amounts of wild-type Rad51 protein (A, lanes 12 and 13). These observations suggest that the Rad51-K191R protein displays a partial dsDNA-binding defect either diminishing nucleation or enhancing binding co-operativity or both. Electron microscopic analysis of the protein–DNA complexes formed by wild-type Rad51 (E left panel) or Rad51-K191R (E right panel) proteins demonstrated that the Rad51-K191R mutant protein formed helical filaments on dsDNA with no discernible difference to the filaments formed by wild-type Rad51 protein. Salt titration experiments were employed to analyze formation and stability of the protein–DNA filaments formed by either Rad51 or Rad51-K191R protein, determining the salt titration midpoints (STMs), which represent the sodium chloride concentrations at which one-half of the substrate is protein-free (). Formation of stable Rad51-K191R-ssDNA filaments precipitously dropped at sodium chloride concentrations over 150 mM (B, lanes 14 and 15), while the wild-type protein efficiently formed stable filaments up to about 250 mM NaCl (B, lanes 6–8). This indicates that formation of Rad51-K191R-ssDNA filaments (STM = 228 mM) was impaired compared with that of the Rad51-ssDNA filament (STM = 305 mM) (B and C; ). In contrast, once formed, Rad51-K191R-ssDNA filaments (STM = 1900 mM) were much more stable than Rad51-ssDNA filaments (STM = 780 mM) (C, ). Rad51-ssDNA filaments disassembled completely at about 1M NaCl, while Rad51-K191R-ssDNA filaments were partially resistant up to 2 M NaCl. Like with ssDNA, the formation of Rad51-K191R-dsDNA filaments was also more sensitive to salt (STM = 78 mM) than with wild-type Rad51-dsDNA filaments (STM = 192 mM) (D, ). Rad51-K191R-dsDNA filaments were also extremely salt-resistant and remained largely intact in the presence of 2 M NaCl, while wild-type Rad51-dsDNA filaments largely disassembled (STM = 680 mM) when the concentration of sodium chloride was over 1 M (D, ). In the presence of free dsDNA substrate, Rad54 exerts a basic mode of ATPase activity, hydrolyzing about 1000 ATP per minute (). In the presence of saturated Rad51-dsDNA filaments (3 bp/Rad51), Rad54 displays a reduced mode of ATPase activity; while in the presence of sub-saturated Rad51-dsDNA filament, Rad54 displays an enhanced mode with a 5 to 6-fold increase in ATPase activity (). In these experiments, wild-type Rad51 protein had been utilized, leaving open the possibility that activation of the Rad51 ATPase activity contributed to the observed effect. Although the Rad51 ATPase activity is vastly lower (0.4 mol/mol/min; B) than the ATPase activity of Rad54 (∼1000 mol/mol/min), the excess of Rad51 protein in these reactions (160 nM Rad51, 3.3 nM Rad54) and the possibility that Rad51 activity is stimulated under our conditions required to control for this possibility. We measured the kinetic parameters and K of Rad54 in the presence and absence of fully and partially saturated Rad51-dsDNA filaments. To exclude a contribution by the Rad51 ATPase activity, the experiments were performed with both wild-type and Rad51-K191R proteins. In the presence of sub-saturated Rad51-dsDNA filaments (1/37 bp), the ATP turnover rate of Rad54 was stimulated over 4- fold (70.5 µM/min versus 16.3 µM/min), while the ATP affinity was not significantly affected within the error of the data (482.5 µM versus 730.7 µM) (A, ). In contrast, when Rad54 was incubated with saturated Rad51-dsDNA filament (1/3 bp), the ATP turnover rate was reduced from 16.3 µM/min to 10.8 µM/min (B, ). Surprisingly, the K for ATP of Rad54 decreased almost 15-fold to 49.4 µM from 730.7 µM (B, ). To formally exclude an effect of the Rad51 ATPase activity, the kinetic measurements were repeated with dsDNA filaments formed by the Rad51-K191R protein. With sub-saturated Rad51-K191R-dsDNA filaments (1/37 bp), the ATP turnover rate of Rad54 was enhanced more than 4-fold (71.1 µM/min versus 16.3 µM/min), with no significant effect on its K (637.9 µM versus 730.7 µM) (A, ). These values are not significantly different from those obtained with the wild-type Rad51 filaments (). With saturated Rad51-K191R-dsDNA filaments, the K was significantly reduced about 8-fold (93.02 µM versus 730.7 µM), similar to the experiments with wild-type Rad51. However, the ATP turnover rate was slightly stimulated (20.47 µM/min versus 16.3 µM/min) with Rad51-K191R (1/3 bp) (B, ). This latter effect differed from the wild-type Rad51 protein and is likely due to the partial DNA binding defect we observed above with Rad51-K191R. When Rad51-K191R was incubated with the dsDNA substrate at 1/3 bp stoichiometric ratio, most of the dsDNA substrate was completely covered by protein, but a small fraction of substrate remained protein-free (A, lane 7), or possibly partially covered by protein. These substrates are likely responsible for the observed slight increase in the k. Taken together, these results suggest that the interaction of Rad54 with subsaturated Rad51-dsDNA filaments stimulated the ATP turnover rate of Rad54, while the interaction with fully saturated Rad51-dsDNA filaments enhanced the affinity of Rad54 for ATP. While some data displayed a certain scatter, all conclusions are based on differences that are clearly significant as demonstrated by the non-overlapping errors. Rad54 disassembles Rad51-dsDNA filaments, and we suggested this function to be critical during postsynapsis to turnover the Rad51-heteroduplex DNA product complex (,). This disassembly was shown to require the Rad54 ATPase activity (), but the requirement for the Rad51 ATPase remained unaddressed. To test a possible contribution of the Rad51 ATPase activity, we tested disassembly of Rad51-K191R-dsDNA filaments. As shown before, the Rad51-DNA filament disassembled gradually (A, lanes 2–5). A slight accumulation of free dsDNA and reduction of the saturated filament was observed after 90 min of incubation (A and B). Visualizing disassembly was dependent on the presence of scavenger DNA (lanes 14–17), showing that disassociated Rad51 can rebind the dsDNA. Addition of Rad54 at a sub-stoichiometric ratio (1 Rad54 per 15 Rad51) promoted nucleoprotein filament disassembly (A, lanes 6–9). A significant accumulation of free dsDNA and reduction of the saturated filament was observed after 90 min of incubation (A and B). Again, the presence of scavenger DNA was required, since Rad51 can rebind to dsDNA even in the presence of Rad54 (lanes 10–13). In contrast, Rad51-K191R-dsDNA filaments were extremely stable even in the presence of scavenger DNA (C, lanes 2–5, D), consistent with the results from the salt titrations (D). Upon addition of Rad54, the Rad51-K191R-dsDNA filament was partially disassembled, and a gradual increase in the mobility of the nucleoprotein filament was observed during the time course (C, lanes 6–9, D). The presence of scavenger DNA was required to prevent Rad51-K191R to rebind the substrate (C, lane 10–17). Furthermore, Rad54-K341R, an ATPase-deficient mutant, has no effect on either Rad51- or Rad51-K191R-dsDNA filament disassembly [(), D]. Together, these results suggest that the intrinsic ATPase activity of Rad51 and the ATPase activity of Rad54 are both required for the efficient turnover of Rad51 from dsDNA substrate. To corroborate the results from the nucleoprotein gel assays, we employed a topological assay to examine the nucleoprotein filament disassembly ()(A). Rad51-dsDNA filaments partly disassembled after 2 h incubation in the presence of scavenger DNA (B, upper left panel), indicated by the disappearance of form X DNA species and the appearance of negatively supercoiled (sc) and relaxed (re) DNA. Incubation of Rad54 with Rad51-dsDNA filament increased filament disassembly significantly, as most DNA substrate had been converted to fully relaxed DNA species after 2 h of incubation (B). Rad51-K191R protein remained bound to the DNA substrate even in the presence of scavenger DNA, indicated by the continued presence of form X DNA after 2 h of incubation (B). We reason that the relaxed DNA species observed in the Rad51-K191R reaction is likely to represent initially unbound DNA, because of the subtle DNA binding defect documented before (D) rather than evidence for nucleoprotein filament disassembly. This interpretation is consistent with the absence of intermediate negatively sc DNA species. Upon incubation of Rad54 with Rad51-K191R-dsDNA filaments, the X-species disappeared and was accompanied by the appearance of a group of negative sc species, which represent partially disassembled protein–DNA filaments (B). Taken together, the results of topological assays are consistent with the results of nucleoprotein gel assays, that the ATPase activity of Rad51 is required for efficient disassembly of Rad51-dsDNA filament by the Rad54 motor. ATP hydrolysis by Rad51 is crucial for intrinsic and Rad54-mediated nucleoprotein filament disassembly (, and ). We tested whether the type of Rad51-bound nucleotide (ATP versus ADP) has an effect on the dissociation reaction. Rad51-dsDNA filaments containing ADP-bound subunits were assembled . ADP by itself cannot support the formation of Rad51-dsDNA filaments (). Therefore, we chose to mix ADP and ATP at different ratios for efficient nucleoprotein filament assembly. Our data suggest that Rad51-dsDNA filaments can be efficiently assembled even in the presence of only 20% ATP (data not shown). Rad51-dsDNA filaments assembled in the presence of 20% ADP were disassembled with faster kinetics than those assembled exclusively in the presence of ATP (B, lanes 6–9, lanes 14–17, C). Rad51-dsDNA filaments assembled in the presence of 80% ADP were disassembled even faster (A, lanes 6–9, lanes 14–17, C). This difference is not due to the inherent instability of Rad51-dsDNA filament assembled in the presence of ADP, since in the absence of Rad54, the amount of free DNA did not differ in reactions with 0%, 20% and 80% ADP. Rather, the filaments containing ADP-bound subunits displayed slower disappearance of the saturated complexes (A, lanes 2–5, lanes 10–13; B, lanes 2–5, lanes 10–13, C). In these reactions, the ATP regeneration system was supplemented after filament assembly but before Rad54 addition to ensure optimal Rad54 activity. Control experiments confirmed that the Rad54 ATPase activity remained unaffected by the presence of ADP under these conditions (data not shown). Together, these results suggest that Rad54 prefers disassembling filaments containing Rad51 subunits bound to ADP. The observed effect represents a lower estimate as possible exchange of Rad51 bound ADP by ATP may occur under these reaction conditions. If the low efficiency of Rad51-K191R-dsDNA filament disassembly by Rad54 is due to its ATPase defect ( and ), one prediction is that Rad51-dsDNA filaments assembled with the slowly hydrolysable ATP analog, ATP-γ-S, would have the same effect. Rad51-ATP-γ-S-dsDNA filaments were assembled in the presence of 50-µM ATP-γ-S. To initiate disassembly, Rad54 and 5-mM ATP were added together with the ATP regeneration system. Control ATPase assays showed that Rad54 maintained about 98% of its original ATPase activity under these conditions (data not shown). Compared with the nucleoprotein filament assembled with ATP, the Rad51-ATP-γ-S-dsDNA filament was only partially disassembled by Rad54 during the time course with no significant increase of protein-free DNA (D, lanes 2–5, lanes 10–13, E). In the absence of Rad54, filaments assembled with ATP-γ-S were extremely stable and resistant to the challenge by scavenger DNA (D, lanes 6–9 and lanes 13–17). This is very reminiscent of the observation with the Rad51-K191R-ATP-dsDNA filaments (C). In addition, in the presence of ATP-γ-S, Rad51 displayed a similar dsDNA-binding defect as the Rad51-K191R protein did with ATP, as indicated by the presence of free DNA (D, lanes 2–9, A). Taken together, these results support the idea that the Rad51-dsDNA filament formed in the presence of ATP-γ-S resembles the Rad51-K191R-ATP-dsDNA filament, and that intrinsic ATP hydrolysis by Rad51 is critical for its dissociation from dsDNA substrates. The resistance of the Rad51 filament to dissociation by Rad54 in the absence of Rad51 ATP hydrolysis is probably greater than observed here, because of residual ATP-γ-S hydrolysis and possible impurities in the ATP-γ-S preparation. d 5 1 a n d R a d 5 4 a r e k e y D N A - d e p e n d e n t A T P a s e s f u n c t i o n i n g i n H R i n e u k a r y o t e s . O u r a n a l y s i s u n c o v e r e d u n e x p e c t e d b i o c h e m i c a l p r o p e r t i e s o f t h e A T P a s e - d e f i c i e n t R a d 5 1 - K 1 9 1 R a n d R a d 5 1 - K 1 9 1 A m u t a n t p r o t e i n s . K i n e t i c a n a l y s i s a n d e x p e r i m e n t s w i t h d i f f e r e n t n u c l e o t i d e c o f a c t o r s r e v e a l e d a n i n t r i c a t e r e l a t i o n s h i p b e t w e e n t h e R a d 5 1 a n d R a d 5 4 A T P a s e a c t i v i t i e s , d e m o n s t r a t i n g t h a t b o t h A T P a s e a c t i v i t i e s a r e r e q u i r e d f o r e f f i c i e n t t u r n o v e r o f t h e R a d 5 1 - d s D N A f i l a m e n t .

Selenium (Se) is an essential dietary trace element that benefits many aspects of human health (). Se is believed to exert its biological effects through its incorporation into selenoproteins as the amino acid, selenocysteine (Sec). To date, 25 selenoproteins have been identified in humans and all but one of these exist as selenocysteine-containing proteins in mice (). The functions of members of the selenoprotein family elucidated to date include roles in thyroid hormone metabolism, intra- and extra-cellular antioxidation, redox regulation, glucose metabolism, and sperm maturation and protection. However, more comprehensive details regarding tissue distribution and physiological roles for the entire selenoproteome are necessary for a better understanding of the mechanisms by which Se affects human health. Selenoprotein P (Sel P) is a unique member of the selenoprotein family in that it contains multiple Sec residues per protein molecule. In particular, human and mouse Sel P both contain 10 Sec residues. Culturing cells in Sel P-depleted media was found to reduce activity of the glutathione peroxidase (GPX) family of selenoproteins, and activity was restored following reconstitution of the media with Sel P (). These facts combined with the finding that a majority of plasma Se is contained within Sel P suggests that this protein plays an important role in transporting Se throughout the body for use in synthesis of other selenoproteins. This notion was supported by two independently developed Sel P-knockout mouse models, both of which exhibited altered tissue distribution of Se (,). Hill . found that, in terms of Se content, the tissues most affected by Sel P-deficiency were testes and brain, while kidney and heart were less affected. The phenotype of the knockout mice was consistent with these findings and included neurological problems and male sterility. Sel P-knockout mice injected with Se had lower levels of isotope in brain and testes, but higher accumulation in livers compared to controls. These findings suggested that the liver is a tissue that readily takes up Se and incorporates it into Sel P, which is then secreted into the plasma for transport of Se to other tissues. This was supported by studies of Schweitzer . demonstrating that liver-specific inactivation of the gene encoding Sec tRNA () resulted in decreased plasma and kidney GPX activity (). In contrast, GPX activity in brain tissue in these mice remained unaffected, suggesting that liver Sel P is not required to the same extent for expression of different selenoproteins in different tissues. Synthesis of selenoproteins is regulated at the levels of mRNA transcription and stability, and protein translation (). Turnover of mRNA is a particularly important point of regulation in that selenoprotein mRNAs require recoding of UGAs within the coding regions from stop codons to Sec-insertion sites. Under circumstances of low selenium, some selenoprotein mRNAs are degraded through nonsense-mediated decay (NMD), a pathway that degrades mRNAs containing premature stop codons (). Se-deficiency decreases the efficiency with which Sec incorporation is directed by the Sec incorporation machinery and increases the efficiency with which the codons are recognized as nonsense, thus eliciting NMD (,). Sensitivity to NMD is determined by the location of termination codons in the mRNA in relation to exon-junction complexes, which are deposited upstream of exon–exon boundaries during mRNA splicing and export. In higher eukaryotes, a nonsense codon is usually recognized as premature if it is located more than 50–55 nt upstream of the last intron in the pre-mRNA (,). Interestingly, Se-deficient conditions result in degradation of different selenoprotein mRNAs to different extents. Several studies have shown that, within the GPX family of proteins (GPX1 through 4), Se-deficiency leads to degradation of GPX1 and GPX3 mRNAs to lower levels than GPX2 and GPX4 mRNAs (). Other studies have supported a similar notion that GPX1 mRNA levels are decreased by Se-deficiency while other GPX mRNAs are relatively unchanged (,). However, how expression of other members of the selenoprotein family or the protein factors involved in their synthesis is affected by Se availability in different tissues remains unknown. Given the important roles that Se transport and metabolism play in selenoprotein expression, we investigated the relative levels of mRNA abundance in eight different mouse tissues and how these levels are affected by genetic deletion of Sel P. Our results describe mRNA levels for selenoproteins in different mouse tissues including brain, heart, intestine, kidney, liver, lung, spleen and testes. In addition, deletion of Sel P resulted in lower levels of many selenoprotein mRNAs and some proteins in brain and testes, but not in heart and lung. Taken together, the results provide insight into how the presence of Sel P influences the different selenoproteins in different tissues. Sel P mice on a C57Bl/6 background were kindly provided by Drs Raymond Burk and Kristina Hill and were used to generate a colony of mice here at the University of Hawaii. Genotyping of the mice was carried out using methods previously described () and heterozygous breeding pairs were used to produce litters consisting of pups homozygous for Sel P deletion (Sel P) as well as wild-type littermate controls (Sel P). Male weanlings were fed standard mouse chow until 8 to 10 weeks of age, when they were sacrificed and eight tissues quickly harvested. This was carried out prior to onset of any apparent neurological disorders. Each tissue was immediately washed in PBS and frozen in liquid nitrogen. Tissues were ground into powder on dry ice, which was then divided into separate tubes for RNA or protein extraction. Frozen, powdered tissue samples were thawed and RNA extracted using RNeasy Mini kit and RNase-free DNase I (all from Qiagen, Valencia, CA, USA). Concentration and purity of extracted RNA was determined using A/A measured on an ND1000 Spectrophotometer (NanoDrop Technologies, Wilmington, DE, USA). Synthesis of cDNA was carried out using Superscript III (Invitrogen, Carlsbad, CA, USA) and oligo dT primer, with 2 μg RNA per 50 μl reaction. For real-time PCR, 1 μl of the cDNA was used in 10 μl reactions with Platinum SYBR Green qPCR SuperMix-UDG (Invitrogen). Reactions were carried out in a LightCycler 2.0 thermal cycler (Roche Applied Biosystems, Indianapolis, IN, USA). Oligonucleotides used for qPCR are listed in . Cycling conditions were used as suggested in the SYBR Green kit instructions and results analyzed using Relative Quantification Software (Roche). Protein was extracted from tissue by homogenizing 0.5 g of tissue on ice in 10 ml of CellLytic MT buffer (Sigma) containing 1 mM DTT, 1X protease inhibitor cocktail (Calbiochem, San Diego, CA, USA), and 5 mM EDTA. Homogenate was centrifuged at 12 000 r.p.m. (13 000 ; Beckman Coulter microcentrifuge) for 10 min and supernatant removed and stored at −70°C. Bradford assay was carried out using Bradford Reagent (Bio-Rad, Hercules, CA, USA) and 30 μg total protein was combined with reduced Laemeli buffer, boiled at 95°C for 10 min, cooled on ice, and loaded into wells of 10–14.5% polyacrylamide gels (Bio-Rad). Protein was transferred to PVDF membranes, which were blocked for 1 h with 5% BSA and then probed for 1 h with primary antibodies, including rabbit polyclonal anti-GPX1 and anti-GPX4 (Lab Frontier, Seoul, Korea), rabbit anti-Sel W raised against the peptide: SKKRGDGYVDTESKFRK (custom synthesized and affinity purified by ProSci, Inc., Poway, CA, USA), mouse monoclonal IgG anti-β-actin (Sigma) and anti-α-tubulin (Novus Biologicals, Littleton, CO, USA). Appropriate HRP-conjugated secondary antibodies were purchased from Jackson Immunolabs (West Grove, PA), incubated with the membranes for 45 min and detected using ECL Plus (GE Healthcare). In further attempts to detect Sel W in heart and lung, immunoprecipitation reactions were carried out to concentrate any Sel W that might be present in protein extracts from these tissues, followed by electrophoresis and western blotting. However, these and other methods did not result in detection of Sel W in these two tissues except for questionable bands upon overexposure of the film. For densitometry, digital images of autoradiographic film were captured using Gel Logic 200 and Kodak MI software (Kodak Scientific Imaging Systems, Rochester, NY, USA). This software was used to measure mean intensity from regions of interest (ROI) that corresponded to bands to be measured. The intensity of the target bands (e.g. GPX1 band) was normalized to that of the loading control band (e.g. β-actin band) to obtain normalized levels of target proteins. All statistical tests were performed using GraphPad Prism version 4.0 for Windows (GraphPad Software, San Diego, CA, USA). Means of two groups were compared using a student's -test and significance was considered at < 0.05. We first set out to evaluate mRNA levels for the entire murine selenoproteome in eight different mouse tissues: brain, testes, liver, kidney, lung, heart, intestine and spleen. To achieve this goal, we utilized real-time PCR and included oligonucleotides specific for all 24 murine selenoproteins, three non-selenoprotein factors involved in their synthesis (sbp2, secp43 and sps1), and four housekeeping genes: hypoxanthinephosphoribosyltransferase (hprt), ubiquitin c (UBC), β-actin (actin) and glyceraldehyde-3-phosphate dehydrogenase (gapdh) (). Because levels of hprt were most consistent between different tissues and different mice, this housekeeping gene product was used to normalize the abundance of other mRNAs. Results for the complete list of mRNA levels in the eight different tissues are shown in . The mRNAs for each tissue were divided into low (<0.5), medium (0.5 − 10) and high (>10) copy groups, based on levels of mRNAs relative to the stable hprt housekeeping mRNA. Use of these cutoff values facilitated comparative analyses of selenoproteome and housekeeping mRNA levels that ranged from rare to highly abundant transcripts. While the specific pattern of mRNA abundance differs between tissues, there are some common features of selenoproteome mRNAs when comparing the eight different tissues. First, gpx4 is among the mRNAs detected at the highest levels for nearly all tissues examined. This reinforces the notion that this antioxidant enzyme may be essential for nearly all cell types and tissues, consistent with the embryonic lethal phenotype displayed by the gpx4 knockout mouse model (). Another mRNA near the top of the abundance hierarchy is sel P, except in the spleen and testes, which both contain sel P as a medium copy mRNA. Sel P mRNA is detected at particularly high levels in the liver. This finding is consistent with other studies demonstrating the liver as a source of a majority of sel P found in plasma (). However, it is interesting to note the relatively high levels of sel P message in most tissues, including the kidney, which was shown to be a tissue particularly dependent on hepatically derived Sel P protein as a Se source (). Another common feature of mRNA abundance shared among the different tissues is the relatively low levels of all three deiodinase enzymes (dio1, 2 and 3). The lone exception to this is in liver, where dio1 is a medium copy mRNA. Also, sel V expression is low in most tissues, except in testes where it is detected at very high levels, a finding that is consistent with previous results (). The selenoprotein-synthesis factors that we included in this study were detected at relatively low levels, suggesting high specific activity, low requirement or rapid mRNA turnover. The only notable exception was sps2, which is both a selenoprotein and a factor involved in synthesis of other selenoproteins. Sps2 mRNA was detected at relatively high levels in the liver, where it was the fifth most abundant selenoprotein mRNA. After evaluating the results in for similarities, we next turned our attention to the differences found between the various tissues. One result that is apparent is the high abundance of gpx3 mRNA in the kidney, which is consistent with previously reported results (). A surprising finding is that the heart also displays a relatively high level of gpx3 mRNA. The liver has high sel P mRNA levels compared to other tissues and confirms previous findings suggesting that this tissue produces high quantities of Sel P protein for secretion into blood plasma. Both liver and kidney are relatively high in abundance of mRNA for sel R, a selenoprotein that has been characterized as a zinc-containing stereo-specific methionine sulfoxide reductase (). Two selenoprotein mRNAs present at high levels in the testes compared to other tissues are sel K and tr3. We next investigated how deletion of Sel P in mice affects the mRNA levels of the selenoproteins and non-selenoprotein synthesis factors in four different tissues: brain, testes, heart and lung. These particular tissues were chosen due to the phenotype of the Sel P mice. These mice have been shown to demonstrate neurological disorders and sterility, but have no apparent physiological problems related to the lung or heart (,). In this manner, we set out to compare mRNA abundance for the selenoproteins in two tissues involved with the phenotypical problems (brain and testes) and two seemingly unaffected tissues (heart and lung). Results from our comparisons are displayed in and reveal several interesting features regarding how genetic deletion of Sel P affects different tissues. First, both testes and brain displayed overall decreases in selenoprotein mRNA abundance as well as decreases in housekeeping mRNAs. Although statistical significance was not achieved for most of the mRNAs due to variability and low numbers of mice per group ( = 3 − 6), the brain and testes were clearly more affected than heart and lung. This was particularly evident for mRNAs for the medium and high copy genes and may reflect apoptotic or necrotic cellular damage, which may be involved in or lead to the physiological problems in these tissues. In contrast, in heart and lung mRNA abundance was decreased only in a small number of selenoproteins expressed in the low copy group. The medium and high copy selenoprotein mRNAs in heart and lung are much less affected by genetic deletion of Sel P. Other interesting features that emerge from these results include the changes in sel W and gpx4 mRNA levels. In brain and testes, the levels decreased dramatically in Sel P compared to Sel P mice, while in lung and heart they increased. Other mRNA levels also increased in response to genetic deletion of Sel P. For example, sps2 increased in three tissues (brain, testes, and heart) in Sel P versus Sel P mice. Both sps2 and sep15 are high copy mRNAs in testes, and both demonstrate high resistance to effects of Sel P knockout in this tissue. Lastly, the brain and testes tissue from Sel P mice appear to have low, but detectable levels of sel P mRNA (C and D). Levels of sel P mRNA transcripts should be near zero considering the oligonucleotides used for real-time PCR (exons 4 and 5) are located downstream of the cassette insertion site (exon 2) (). Insertion of the neo cassette results in a premature stop codon, which leads to degradation of the transcript. Thus, for some tissues the rate at which the transcript containing the neo cassette is degraded may be slower than others, resulting in detection of transcript en route to NMD degradation. As shown in A, levels of all three proteins are decreased in brain and testes. Using densitometry to quantify the levels of proteins, we found significant reductions in GPX1 in brain and both GPX4 and Sel W in testes. These results are consistent with the mRNA results described above. The results from lung and heart tissue differ from brain and testes in several ways. First, the antibody that detects Sel W in brain and testes does not detect Sel W in lung and heart. This could be due to different isoforms of Sel W expressed in the different tissues, as mRNA levels for this selenoprotein were relatively high in heart and lung. A second feature that is evident from the western blot analyses is that the decreased levels of GPX1 and GPX4 detected in Sel P brain and testes tissues are not detected in lung and heart, which is consistent with the mRNA levels described above. In fact, GPX4 levels are significantly increased in heart tissue from Sel P mice and GPX1 levels show a trend toward an increase in the knockout mice. These results are similar to those found at the mRNA level and raise the question of why these two members of the GPX family may increase in heart with the deletion of Sel P. The study presented herein provides a comprehensive description of selenoprotein mRNA abundance for eight murine tissues including brain, testes, kidney, liver, spleen, intestine, lung, and heart. This multi-tissue analysis allows comparisons to be made within and between tissues for relative levels of the different selenoprotein mRNAs. Our results suggest that certain selenoprotein mRNAs (e.g. gpx4 and sel P) are found at high levels in most of these tissues, while others are generally expressed at very low levels (e.g. deiodinase enzymes, dio1-3). Differences in mRNA levels may be due to multiple factors acting individually or in combination. Stability of mRNA has been shown to play a key role in regulation of selenoprotein synthesis (). Selenoprotein mRNA stabilities are affected differentially by limiting availability of Se or selenoprotein synthesis factors (). The degradation of these and other mRNAs is believed to be carried out by NMD (,,). Differences in susceptibility to degradation may lie in the presence of -elements in the selenoprotein mRNA that either lead to recognition by the NMD machinery, affect their ability to compete for stabilizing factors such as SBP2, or alter subcellular localization (,). In addition to mRNA stability, differential transcription may contribute to varying mRNA levels for the various selenoproteins. Preliminary analysis of the promoters of selenoprotein genes suggests that several of these may be under control of shared regulatory pathways, including those upregulated under conditions of stress (). In particular, the genes for several selenoprotein mRNAs exhibiting increased expression in Sel P tissues share several common transcription factor motifs that may result in their upregulation (manuscript in preparation). In addition, multiple transcripts are annotated in the mouse genome for many of the selenoprotein genes, and the tissue expression patterns of many of these transcripts are not well characterized. These are relatively unexplored but undoubtedly important areas for future investigation. The main goal of this study was to determine the relationship between Sel P and the expression of other selenoprotein family members in different tissues. Because of the Se transport role that Sel P has been demonstrated to play, the Sel P knockout mouse model provides a useful tool to investigate how the absence of this protein affects expression of the different selenoprotein family members in various tissues. Our data suggest that brain and testes are similarly affected by genetic deletion of Sel P in mice. In both tissues, Sel P deletion leads to a generalized reduction in abundance of all medium and high copy selenoprotein mRNAs. The substantial decrease in levels of most selenoprotein mRNAs in these two tissues in Sel P mice suggests that Sel P is important for transport of Se within or to these tissues and is consistent with the neurological and sterility problems observed in the Sel P mice. Recent findings by the Burk lab further support the role of Sel P in transporting Se to the brain and testes (,). Olson . have proposed a mechanism in the testes by which Sel P is taken into cells of this Se-sensitive tissue via the apolipoprotein E receptor (). In contrast to the brain and testes, heart and lung tissue from Sel P mice did not demonstrate decreased selenoprotein mRNA levels. In fact, gpx4 and sel W mRNA levels were increased in heart and lung from Sel P mice compared to Sel P controls. In heart, GPX4 protein was also found to be significantly increased in mice lacking Sel P. This may be due to a feedback mechanism by which transcription of these two family members is upregulated during increased oxidative stress in these two tissues. When comparing our results involving Sel P deletion to those involving dietary Se-deficiency, some interesting differences are evident. For example, studies in rats have demonstrated that Se-deficiency results in decreased expression of GPX-1 in several tissues including liver, kidney, heart and colon (). In contrast, GPX-4 levels were found to be unaffected or less affected by Se-deficiency in most of these tissues, except in colon where GPX-4 was decreased at nearly equivalent levels as GPX-1. Our results suggest that Sel P deletion in mice affected GPX-4 and GPX-1 similarly in that they were lowered in brain and testes, but increased in heart. Furthermore, Se-deficiency in rats has been shown to decrease mRNA levels for Sel W in colon, similar to our results showing a decrease in Sel W mRNA and protein levels in brain and testes from Sel P-deficient mice compared to wild-type controls. Differences between species may account for some of these differences and further investigation is needed to fully understand how deficiencies in dietary Se compare with decreases in Sel P levels. An interesting finding of this study was the detection by western blot of Sel W in brain and testes, but the conspicuous absence of this selenoprotein in the heart and lung. The rabbit antibody used for this assay was specific for a 17 amino acid region in the C-terminal portion of the protein. Despite several different approaches including immunoprecipitation concentration of protein extracts prior to western blots, the antibody failed to detect Sel W in the heart and lung. This suggests that, despite high mRNA levels in heart and lung, the protein is not expressed at detectable levels in these tissues. This is consistent with previous findings that Sel W was detected by western blot in muscle, spleen, brain and testes only (). Interestingly, we used this antibody raised against murine Sel W to detect human Sel W on a commercially purchased human tissue blot and found that it detected the protein in nearly all tissues except for ovaries and testes (data not shown). This occurred despite a two amino acid mismatch between the antigenic region in mouse and human Sel W. It remains unclear whether different isoforms of Sel W are expressed in different tissues, but differences between rodents and primates have been previously noted (). The selenoprotein-synthesis factors analyzed in this study included sbp2, secp43, sps1 and sps2. These factors were detected at relatively low to moderate levels, suggesting high specific activity, low requirement or rapid mRNA turnover. The only notable exception was sps2, which is both a selenoprotein and a factor involved in synthesis of other selenoproteins. Sps2 mRNA was detected at relatively high levels in the liver, where it was the fifth most abundant selenoprotein mRNA. This may be an adaptation to relatively high levels of selenoprotein synthesis occurring in this tissue. Given that testes would seemingly require high levels of selenoprotein synthesis, it was surprising that this tissue exhibited only low to moderate abundance for the synthesis factors. Importantly, our findings are based on comparisons between target mRNAs and the most stable housekeeping mRNA, hprt. Using other housekeeping mRNAs as comparison transcripts produces different results. However, results from our protein analysis described above correlate best with the mRNA data obtained with the stable hprt as the comparison transcript. Overall, our study provides an extensive comparison of the selenoprotein mRNA abundance in terms of tissue distribution in the mouse under normal conditions and, in four tissues, under conditions of sel P deletion. Determining exactly how tissues cope differentially with changes in Se availability, metabolism and transport will lead to a better understanding of how expression of the selenoproteome throughout the organism contributes to its overall health status.

Nucleases, which cleave the phosphodiester bonds of nucleic acids, are very important not only for nucleic acid metabolism (,) but also for a variety of biotechnologies (). For this reason, many studies have examined their structure, stability and function (). Most of these studies, however, have been carried out under relatively dilute conditions. These are quite different to those inside a cell, where up to 40% of the total volume is taken up by macromolecules (,). As numerous studies have shown, molecular crowding dramatically increases the association between biomolecules and substantially affects biomolecular reaction rates (). For example, Zimmerman and Pheiffer () showed that DNA ligase activity is enhanced by molecular crowding conditions. Wenner and Bloomfield (), in contrast, found that crowding agents have negligible effects on EcoRV activity. Thus, molecular crowding has different effects on different types of nucleic acid metabolizing enzymes. For this reason, systematic studies of the effects of molecular crowding on enzyme activity, structure and stability for a series of nucleases are needed to predict the nuclease behavior in cellular conditions and to understand how molecular crowding affects protein functions in the presence of nucleic acids. In the current study, we systematically investigated the effects of molecular crowding on DNA hydrolysis by various nucleases. We found that the cleavage of both a large DNA (plasmid) and a short DNA oligonucleotide by DNase I was increased in the presence of polyethylene glycol (PEG). An increase in activity by molecular crowding was found for endonucleases (DNase I and S1 nuclease) but not for exonucleases (exonucleases I and III). Kinetic analyses showed that molecular crowding affected the maximal velocity () but not the Michaelis constant () for the nucleases. Overall, our results indicate that molecular crowding has different effects on the catalytic activities of exonucleases and endonucleases. PEG 4000, PEG 8000 and PEG 20000 [average molecular weight = 3000, 8000 and 20000, respectively ()] were purchased from Wako Pure Chemicals (Osaka, Japan) and used without further purification. DNase I from bovine pancreas was purchased from Invitrogen (Carlsbad, CA, USA). Exonuclease I from , S1 nuclease from and exonuclease III from were purchased from Takara Bio (Tokyo, Japan). Purified DNase I for the circular dichroism (CD) studies was purchased from Roche (Mannheim, Germany). All other reagents were of reagent grade. Plasmid [pcDNA 3.1(+)] was purchased from Invitrogen. strains were grown aerobically in Luria–Bertani medium in the presence of 100 μg/ml ampicillin at 37°C with shaking at 250 r.p.m. Top 10 cells (Invitrogen, CA, USA) were used for plasmid propagation. HPLC-grade oligonucleotide substrates of 29-mer sequences, 5′-ACGATATCTCCCTATAGTGAGTCGTATTA-3′ and 5′-TAATACGACTCACTATAGGGAGATATCGT-3′, were purchased from Hokkaido System Science (Sapporo, Japan). These sequences were designed using mfold to avoid folding into undesired structures (). For quantitative analysis of the hydrolysis reaction of nuclease, the sequence 5′-ACGATATCTCCCTATAGTGAGTCGTATTA-3′ was labeled with 6-carboxyfluorescein at its 5′-end, although the 6-carboxyfluorescein could affect the rate of cleavage by a nuclease. A 29-mer double-stranded DNA (dsDNA) was prepared by annealing (incubation at 90°C for 5 min, followed by cooling at 1°C/min to the reaction temperature) of 5′-ACGATATCTCCCTATAGTGAGTCGTATTA-3′ with 5′-TAATACGACTCACTATAGGGAGATATCGT-3′. The 29-mer dsDNA was used as substrate in assays of DNase I and exonuclease III. The sequence 5′-ACGATATCTCCCTATAGTGAGTCGTATTA-3′ was used as a single-stranded DNA (ssDNA) substrate in the assays of S1 nuclease and exonuclease I. Because the optimal condition (cations, pH and temperature) for DNA hydrolysis by each nuclease varies, it is difficult to study all the enzymes using a single-standard buffer. DNase I was assayed in a buffer of 2.5 mM MgCl, 0.5 mM CaCl and 10 mM Tris–HCl (pH 7.5) at 25°C. The reactions were performed with 0.01–0.1 U of DNase I and 0.5 μg of plasmid [pcDNA 3.1] or 20 μM dsDNA (29-mer) as a substrate. S1 nuclease was assayed in 280 mM NaCl, 1 mM ZnSO and 30 mM sodium acetate buffer (pH 4.6) at 37°C. The reactions were performed with 0.15 U of S1 nuclease and 10 μM ssDNA. Exonuclease III was assayed in a buffer of 5 mM MgCl, 10 mM 2-mercaptoethanol and 50 mM Tris–HCl (pH 8.0) at 37°C. The reactions were performed with 1–10 U of exonuclease III and 20 μM dsDNA. Exonuclease I was assayed in a buffer of 6.7 mM MgCl, 10 mM 2-mercaptoethanol and 67 mM glycine-KOH (pH 9.5) at 37°C. The reactions were performed with 0.5–5 U of exonuclease I and 10 μM ssDNA. The reactions were terminated by adding gel loading buffer containing 100 mM EDTA, 45% (w/v) sucrose and 0.03% (w/v) bromophenol blue, followed by cooling on ice. The mixtures were resolved by 20% native polyacrylamide gel electrophoresis (PAGE) for short DNA oligonucleotides (29-mer dsDNA and ssDNA) or 0.8% agarose gel electrophoresis for the plasmid DNA reactions. The voltage for the native PAGE and for agarose gel electrophoresis was 200 and 100 V, respectively. The short DNAs were visualized using a Fuji Film FLA-5100 phosphorimager (Fuji Film Co., Tokyo, Japan). The plasmid DNA was visualized by staining with ethidium bromide. The fluorescence intensities of bands (LAU/mm) were calculated using Multi Gauge ver. 2.2 software (Fuji Film Co.). DNA hydrolysis was evaluated according to the amount of residual substrate (%) because the product bands were too diffuse to analyze quantitatively. The amount of the residual substrate was estimated as follows (): Residual substrate (%) = [fluorescence intensity (LAU/mm) of the substrate band after reaction] × 100/[fluorescent intensity (LAU/mm) of the substrate band before reaction]. All experiments were repeated at least three times. The experimental error was <2%. For kinetic studies, the dsDNA and ssDNA concentrations were 0.1–20 μM for DNase I and 0.1–10 μM for exonuclease I. The initial rates () were estimated from the data for the first 10% of the reaction. The values of were plotted versus the DNA (substrate) concentration to allow calculations of the kinetic parameters ( and ). The kinetic parameters were calculated from the non-linear best fit of the data to the Michaelis–Menten equation = []( + []), where [] indicates substrate concentration, using Origin software (Microcal Software Inc., Northampton, MA, USA). The structure and stability of DNase I in the absence and presence of PEG was studied by CD analysis. CD spectra were recorded using a Jasco J-820 spectrometer equipped with a Peltier temperature control system (Jasco, Tokyo, Japan). The cuvette-holding chamber was flushed with a constant steam of dry N gas to avoid water condensation on the cuvette exterior. Data were collected from 195 to 255 nm with a 1-s response time and a 1-nm bandwidth using a 0.1-cm quartz cuvette. The CD measurements for 0.5 μM DNase I were carried out in 10 mM Tris–HCl (pH 7.5) in the absence and presence of 20% (w/v) PEG 4000 at 25°C. Each spectrum shown is the average of five individual scans and is corrected for the spectrum of the buffer. For thermal denaturation, the CD signal at 222 nm was monitored for a 1°C/min temperature rise from 5 to 95°C. The thermodynamic stability of the nucleases was studied by performing stability assays in the absence and presence of PEG. The thermal stability of DNase I was assayed in 10 mM Tris–HCl (pH 7.5) in the absence and presence of PEG 4000, and that for exonuclease I was assayed in 10 mM 2-mercaptoethanol and 67 mM glycine-KOH (pH 9.5) in the absence and presence of PEG 8000. The proteins were incubated at 60°C, and hydrolysis was assayed as described above (‘Assay of DNA hydrolysis by nucleases’). After 10 min of hydrolysis, the reaction mixture was resolved and visualized as described above. The thermal stability of DNase I and exonuclease I at 60°C was assessed according to the residual activity, which was calculated as follows (): Residual activity (%) = [Fluorescence intensity (LAU/mm) of the substrate band before the reaction with heat treatment − Fluorescence intensity (LAU/mm) of the substrate band after the reaction with heat treatment] × 100/[Fluorescence intensity (LAU/mm) of the substrate band before the reaction without heat treatment − Fluorescence intensity (LAU/mm) of the substrate band after the reaction without heat treatment]. All experiments were repeated at least three times. The experimental error was <8%. DNase I, an endonuclease, can cleave single- and double-stranded DNAs in a non-sequence specific manner (). Plasmid DNA is widely used as a substrate in studies of DNase I. In the current studies, we examined the effect of molecular crowding on nucleases. To mimic the aqueous cellular environment, we used PEG because it is inert and because it is commercially available in a wide range of molecular weights (). We first examined the effect of 20% (w/v) PEG on the hydrolysis of supercoiled plasmid DNA [pcDNA 3.1] by DNase I at 25°C. A shows the DNA fragments after electrophoresis on a 0.8% (w/v) agarose gel. The two bands in lanes 2 and 3 correspond to supercoiled and open circular plasmid DNA (). During the reaction, two distinct bands were observed in the absence of PEG 20000 (lanes 3–7). According to previous observations (), the slower and faster migrating bands correspond to open circular and linear plasmid DNA, respectively. Although open circular and linear plasmid DNAs were observed, diffuse bands corresponding to products degraded by DNase I were not observed. These results show that, in the absence of PEG, 0.01 U of DNase I was not sufficient for hydrolyzing the plasmid DNA within 5 min. In contrast, in the presence of PEG 20000, the plasmid DNA was cleaved rapidly after 1 min (lane 9), indicating that it was already degraded. These results demonstrate for the first time that molecular crowding significantly enhances the hydrolysis of supercoiled plasmid DNA by DNase I. To quantitatively evaluate the effect of molecular crowding on DNA hydrolysis by DNase I, we used a short (29-mer) dsDNA oligonucleotide as a substrate. B shows native PAGE of the 29-mer dsDNA after reaction with DNase I in the absence and presence of 20% (w/v) PEG 20000 at 25°C. Before the reaction, the migration of the substrate DNA was the same in dilute (lane 1) and crowded conditions (lane 5). This result supports previous studies showing that PEG does not significantly affect the thermodynamic stability of short DNA duplexes (). We did not observe a substantial change in the migration of the dsDNA before or after the DNase I reaction under dilute conditions (lane 1–4), showing that, in the absence of PEG, 0.01 U of DNase I is not sufficient to hydrolyze the 29-mer dsDNA within 10 min. In contrast, bands migrating faster than that of the substrate DNA are clearly visible in the presence of PEG 20000 (lane 5–8). The result shows that dsDNA hydrolysis by DNase I was greatly enhanced by the addition of 20% (w/v) PEG 20000. Therefore, it appears that molecular crowding increases the cleavage yield of DNase I not only for the large DNA (plasmid) but also for the short (29-mer) DNA oligonucleotide. In addition, we observed the accumulation of product DNA (lanes 6–8). We found that the size of the product DNA was ∼10 bp (Supplementary Figure S2). Suck and Oefner () reported that both sides of the DNA double helix at the cleavage point contact DNase I over a span of 10 bp. Thus, 10-bp dsDNA regions on both sides of the cleavage point are required for higher activity. This can explain the accumulation of ∼10-bp dsDNAs in our experiments. We next examined the influence of molecular crowding on the two main types of nucleases (exo and endo), including S1 nuclease, an endonuclease for ssDNA (); exonuclease III, an exonuclease for dsDNA (,); exonuclease I, an exonuclease for ssDNA () and DNase I, an endonuclease for both ssDNA and dsDNA (). In the case of DNase I, the amounts of residual dsDNA after 10 min in the presence of 20% (w/v) PEG 200, 4000, 8000 and 20000 was estimated to be 12, 4, 10 and 10%, respectively. Thus, PEG 4000 was the most effective of the tested PEGs at enhancing the hydrolysis reaction (Supplementary Figure S3A). For this reason, we further investigated the amount of residual 29-mer dsDNA in the presence of 0–20% (w/v) PEG 4000 versus the time using DNase I at 25°C (A). In the absence of PEG 4000, the amount of residual dsDNA after 10 min was estimated to be 87%. In contrast, the amount of residual dsDNA after 10 min in the presence of 5, 10, 15 and 20% (w/v) PEG 4000 was estimated to be 37, 12, 4 and 4%, respectively. These results show that hydrolysis of dsDNA by DNase I is greatly enhanced by PEG. In the case of S1 nuclease, amongst PEG 200, 4000, 8000 and 20000, PEG 4000 was the most effective at enhancing the rate of hydrolysis (Supplementary Figure S3B). Thus, we further investigated the amount of residual 29-mer ssDNA in the presence of 0–20% (w/v) PEG 4000 versus time using S1 nuclease (B). Similar to the results for DNase I, PEG 4000 enhanced the hydrolysis of ssDNA by S1 nuclease. Although the substrates for these two nucleases are different, molecular crowding had similar effects, indicating that the effect is not due to changes in substrate structure. Instead, the results indicate that molecular crowding generally enhances DNA hydrolysis by endonucleases regardless of the substrate structure. Next, we investigated the effects of molecular crowding on DNA hydrolysis by exonucleases. C and D show the amount of residual DNAs versus time for exonucleases III and I, respectively, in the presence of 0–20% (w/v) PEG 8000 (Supplementary and D shows the effect of molecular weight of PEG on the activities of exonucleases III and I). There was little hydrolysis of the substrate DNAs by exonucleases III and I at low enzyme concentrations (0.5–1 U) in both the dilute and molecular crowding conditions. This indicates that the effects of molecular crowding on the hydrolysis reactions of exonucleases were relatively small compared with the effects on the endonuclease reactions. Therefore, to further investigate the effects of molecular crowding on hydrolysis by exonucleases, we increased the amount of enzyme to 10 U for exonuclease III and to 5 U for exonuclease I. Surprisingly, PEG did not enhance the activity of exonuclease III, even at the high enzyme concentration (inset in C): after 10 min, the amounts of residual 29-mer dsDNA using 10 U of exonuclease III at 0, 5, 10, 15 and 20% (w/v) PEG 8000 were estimated to be 8, 3, 5, 10 and 13%, respectively. The results show that DNA hydrolysis by exonuclease III is not enhanced by molecular crowding. Interestingly, DNA hydrolysis by exonuclease I was strongly inhibited by molecular crowding with PEG 8000 (inset in D): after 10 min, the amount of residual 29-mer ssDNA at 0, 5, 10, 15 and 20% (w/v) PEG 8000 after 10 min was estimated to be 7, 4, 10, 32 and 41%, respectively. Thus, it appears that molecular crowding generally enhances endonuclease activities but does not affect or inhibit exonuclease activities. Although we found that molecular crowding enhances endonuclease activity rather than exonuclease activity, the origin of the difference in the effects remains unclear. Because we already demonstrated that the structure and stability of the substrate DNAs are not critical factors in the molecular crowding effect, we examined whether molecular crowding affects the thermodynamic stability of the nucleases. We first compared the structure and stability of an endonuclease (DNase I) in dilute and molecular crowding conditions by CD. A shows CD spectra of DNase I in the absence and presence of 20% (w/v) PEG 4000 at 25°C. Both CD spectra have negative peaks around 215 and 208 nm and a positive peak around 198 nm, showing that α-helices are the dominant structure in DNase I in dilute and molecular crowding conditions. These results are identical with those reported by Ajitai and Venyaminov (). In addition, PEG 4000 did not cause a peak shift, indicating that molecular crowding did not affect the secondary structure of DNase I. On the other hand, molecular crowding strongly affected the stability of the DNase I structure. B shows melting curves for the α-helical DNase I structure in the absence and presence of 20% (w/v) PEG 4000. The melting temperature () of the DNase I structure in the absence of PEG 4000 was estimated to be 60°C. Surprisingly, the of the DNase I structure was >80°C in the presence of 20% (w/v) PEG 4000. This demonstrates that molecular crowding stabilizes the structure of DNase I. We did not measure the CD spectra of exonuclease I because purified exonuclease I is not commercially available. Thus, to investigate how molecular crowding affects the stability of exonuclease I, we carried out a heat stability test for both exonuclease I and DNase I in dilute and molecular crowding conditions. In these experiments, DNase I and exonuclease I were incubated for 0–30 min at 60°C in the absence and presence of PEG (PEG 4000 or 8000). After the incubation, we carried out a 10-min hydrolysis reaction. shows the residual DNase I and exonuclease I activities versus time of incubation at 60°C. After a 30-min incubation, the residual activities of DNase I and exonuclease I in the absence of PEG were estimated to be 56 and 26%, respectively. The inactivation of the nucleases by incubation for 30 min at 60°C agrees with the observations of Chow and Resnick (). On the other hand, the residual activities of DNase I and Exonuclease I in the presence of PEGs (PEG 4000 or 8000, respectively) remained around 100%. These results show that PEG preserves the active structures of DNase I and exonuclease I. The stabilization of DNase I by molecular crowding observed in the heat stability assay agrees with the CD results showing that molecular crowding with PEG increased the of DNase I from 60°C to >80°C. Therefore, these results indicate that the structure of both exonuclease I and DNase I are stabilized by molecular crowding. Although elucidation of the mechanism of the stabilization by molecular crowding is interesting and demands further study, it appears that the difference in the effects of molecular crowding on the endonucleases and exonucleases is not due to effects on their thermodynamic stability. We further investigated the origin of the different effects of molecular crowding on the endonucleases and the exonucleases by determining the kinetic parameters for the hydrolysis reactions in dilute and molecular crowding conditions. We measured the initial velocity () of the hydrolysis reaction at various concentrations of DNA substrate in the absence of PEG and in presence of PEG 4000 at 25°C for DNase I or in the presence of PEG 8000 at 37°C for exonuclease I. The value of for DNase I increased as the concentration of PEG 4000 increased (A). In contrast, the value of for exonuclease I decreased as the concentration of PEG 8000 was increased (B). lists the and of DNA hydrolysis by DNase I and exonuclease I at the various concentrations of PEG. The values of DNase I and exonuclease I for the substrate DNA in the absence of PEG were estimated to be 5.0 and 2.9 μM, respectively, and in the presence of 20% (w/v) PEGs, they were estimated to be 4.5 and 2.3 μM, respectively. The values of DNase I and exonuclease I were not significantly affected by the addition of PEG. On the contrary, the of DNase I increased from 0.1 to 2.7 μM/min as the concentration of PEG 4000 was raised from 0 to 20% (w/v), whereas that of exonuclease I decreased from 2.2 to 0.4 μM/min as the concentration of PEG 8000 was raised from 0 to 20% (w/v). These effects of molecular crowding on the are consistent with the results of equilibrium analyses of DNase I and exonuclease I (A and D). These kinetic parameters reveal that molecular crowding influences the hydrolytic activity of the nucleases by directly affecting catalysis as indicated by the changes in the . Our equilibrium studies of DNA hydrolysis indicate that molecular crowding enhances the activity of endonucleases but has little effect on exonuclease activity. Moreover, we found that molecular crowding affected neither the thermodynamic stabilities of the nuclease structures nor their binding to substrate DNAs. Kinetic analyses demonstrated that molecular crowding affected catalysis. Therefore, the difference in the reaction mechanism and/or type of reaction (endo or exo) may account for the different effects of molecular crowding on endonucleases and exonucleases. Because of the importance of nucleases and , many studies have examined their reaction mechanisms. For DNase I, Glu75 in the active site accepts a proton from His131, which in turn accepts a proton from a water molecule positioned close to the substrate DNA. Then, the nucleophile attacks the phosphorous of the substrate DNA, cleaving the P-O-3′ bond (). This proton acceptor–donor chain (Glu75–His131–water), which is critical for the catalytic activity, is also found in the active site of exonuclease I as Glu17–His181–water (). Since the reaction mechanisms for DNase I and exonuclease I are almost identical, it is surprising that molecular crowding had opposite effects on endonucleases and exonucleases. Thus, it is difficult to determine the reason for the opposite effects on the basis of the cleavage mechanisms. On the other hand, the reaction types mediated by DNase I (endo) and exonuclease I (exo) are distinct. Endonucleases can randomly hydrolyze internal sites in DNA substrates, whereas exonucleases remove terminal nucleotides. We therefore suspect that the origin of the different effects of molecular crowding is the different reaction type (endo versus exo) rather a difference in the reaction mechanism. Wenner and Bloomfield () reported that molecular crowding increases the of EcoRV (endo-type). Conversely, the of T7 DNA polymerase activity and the nick-translation reaction of DNA polymerase I (both exo-type) are reduced by molecular crowding (,) (Supplementary Table 1). Although the effects of other factors, such as the dielectric constant, viscosity of the solution and flexibility of the protein structure on enzyme activities should be examined, our current findings and these previous results suggest that the different effects of molecular crowding on endonucleases and exonucleases are related to the reaction type. These results and considerations lead us to conclude that molecular crowding generally increases the catalytic activity of endonucleases and decreases or does not affect the catalytic activity of exonucleases. These findings should help clarify in general how molecular crowding affects protein functions. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Transcription elongation plays an important role in gene expression in both prokaryotes and eukaryotes. In recent years, several lines of evidence have indicated that transcription elongation is not only a rate-limiting step in gene expression, but also a dynamic and highly regulated process that also impacts downstream events, such as mRNA processing, and RNA surveillance and export (,). The importance of such regulation is illustrated by the discovery that a variety of elongation factors contribute to development, differentiation and disease progression (). Intensive study of transcription elongation factors has identified a number of regulatory mechanisms by which RNAPII efficiently elongates RNA, regardless of impediments (,). DSIF is a heterodimer protein complex composed of Spt4 and Spt5, and is conserved among eukaryotes. DSIF exerts both negative and positive effects on elongation by directly binding to RNAPII through the KOW domain of Spt5 (). DSIF negatively regulates transcription by acting in concert with NELF to mediate promoter proximal pausing of RNAPII (,). The positive regulatory activity of DSIF has been shown to stimulate transcription processivity through an as-yet undefined mechanism. DSIF progresses along with RNAPII to downstream regions of transcribed genes, and phosphorylation of the C-terminal repeat (CTR) region of Spt5 plays a key role in converting DSIF from a repressor to a positive regulator (), suggesting possible mechanisms of regulation of the activator functions of DSIF. Analysis of embryo polytene chromosomes using immunostaining demonstrated that Spt5 localizes to active sites of transcription, and co-localizes with the phosphorylated form of the large subunit of RNAPII during elongation (,). Saunders . () reported that Spt5 tracks with the RNAPII elongation complex along chromatin . In addition to its interaction with RNAPII, genetic and biochemical evidence indicates that DSIF also associates with other components of the transcriptional machinery, such as TFIIF, TFIIS, CSB, Spt6, FACT, Chd1 and the Paf complex, and with factors involved in mRNA maturation and surveillance, such as the mRNA capping enzyme, cap methyltransferase, and the nuclear exosome (,). Although the biochemical characteristics of DSIF are consistent with its function as a general transcription elongation factor, recent studies suggest that it also has a role in the regulation of development, and in gene-specific regulation. For example, zebrafish carrying a point mutation or deletion of display a pleiotropic but highly neuron-specific pattern of defects, suggesting the involvement of DSIF in development (). In embryos, an missense mutation has locus-specific effects on transcription, suggesting that Spt5 affects gene expression selectively (). Moreover, microarray analysis of both zebrafish and human knockdown cells showed changes in expression of only a small subset of genes (unpublished data). The above discrepancies may be explained by assuming that there is a stronger requirement for DSIF during high-levels of transcriptional activity (). This idea is supported by studies of and HIV genome activation. Induction of heat shock gene transcription causes massive recruitment of Spt5 to loci (). and zebrafish carrying null alleles show defects in their heat shock response (,). Knockdown of in human cells causes a significant defect in transcriptional activation in response to epidermal growth factor, while having a negligible effect on expression under basal conditions (). DSIF has also been implicated in Tat-mediated transactivation of HIV genome transcription. Tat is a viral activator that binds -acting TAR elements in nascent RNAs, and stimulates elongation of HIV genes. Knockdown of in human cells decreases Tat-mediated transactivation and HIV-1 replication, but does not significantly affect cell viability (). DSIF cooperates with Tat by preventing premature RNA release at terminator sequences, suggesting a possible mechanism of action of DSIF in regulating HIV transcription (). The transcription of most cellular genes, however, is thought to be activated by DNA-binding activators. It is not clear whether DSIF exerts similar effects when working with DNA-binding activators. In this report, we used transcription assays of Gal4-VP16, a DNA-binding transcriptional activator, to investigate the requirement for DSIF in transcriptional activation. Gal4-VP16 interacts with general transcription factors and the Mediator complex to stimulate initiation (). It has also been implicated in the stimulation of elongation, probably through its interaction with TFIIH (). We demonstrated that in the absence of DSIF, Gal4-VP16-mediated transcriptional activation causes more pausing during elongation than that which occurs during basal transcription. DSIF supported full transcriptional activation by reducing pausing of RNAPII during elongation. We also showed that transcriptional activity requires DSIF . In cultured HeLa cells, Gal4-VP16-induced expression of a reporter gene was significantly decreased upon knockdown. In the absence of the VP16 activation domain, reporter gene expression was at basal levels, and was not affected substantially by knockdown. Co-expression of the DNA-binding competitor of Gal4-VP16, Gal4DBD, which blocked transcriptional activation of the reporter gene, diminished the requirement for DSIF. These results suggest that DSIF regulates transcription elongation in response to transcriptional activation by DNA-binding activators. In addition, we showed that DSIF exerts its positive effect within a short time-frame from initiation to elongation, and that NELF is not involved in the positive regulatory effect of DSIF. An expression plasmid encoding recombinant Histidine (His)-tagged DSIF (His-DSIF) was constructed by combining sequences for His-tagged human Spt4 (hSpt4) and hSpt5 in a single expression plasmid. The co-expression construct was generated using pET-hSpt4 and pET-hSpt5 (). pET-hSpt5 was digested by II and I to generate the coding sequence fragment. pET-14b was digested using I and II to eliminate the His-tag sequence, and then ligated to the fragment to generate pT7hSpt5. pET-hSpt4 was digested using I and I and inserted into pT7hSpt5 that had been digested with I and II. Recombinant His-DSIF (His-hSpt4/hSpt5) was expressed in BL21-CodonPlus (DE3)-RIL (Stratagene). After induction with 1 mM IPTG for 4 h at 30°C, cells were harvested and lysed, and then lysates were loaded onto a Ni-NTA column (Qiagen). Recombinant His-DSIF was purified under native conditions according to the protocols in the QIAexpressionist handbook (Qiagen). Proteins eluted from the Ni-NTA column were loaded onto a 1 ml Mono Q column and eluted with a linear gradient of 100 to 1000 mM HGKEDP [20 mM HEPES (pH 7.9), 20% glycerol, 100–1000 mM KCl, 0.2 mM EDTA, 1 mM DTT, 1 mM PMSF]. The fractions were analyzed by SDS–polyacrylamide gel electrophoresis (PAGE), and fractions containing recombinant His-DSIF were dialyzed against 100 mM HGKEDP, and stored at −80°C until use. Coexpression of hSpt4 and hSpt5 was done to address the formation of insoluble aggregates, and avoid the denaturation/renaturation process used in a previous purification protocol (). His-GAL4 (1–94)-VP16 (413–490) was expressed in and purified as described by Reece . (). Flag-NELF was purified as previously described (). Cdk9 and Cyclin T1 subunits of P-TEFb were coexpressed in Sf9 cells with baculoviral vectors and purified as previously described (). The plasmid pG5MLPDG was generated by replacing the G-free cassette of pG5MLP () with the double G-free cassettes fragment from the plasmid pSLG402 (). Concentrated P1.0 fractions were prepared as described previously (,). transcription reactions using the concentrated P1.0 fraction and plasmid DNA templates were carried out as described previously (,). Briefly, in reactions using pG5MLP as a template, 12.5 μl reaction mixtures containing 125 ng DNA () and the concentrated P1.0 fraction were prepared in the presence or absence of recombinant DSIF and Gal4VP16 in TRX buffer [25 mM Tris–HCl (pH 7.9), 10% (v/v) glycerol, 50 mM KCl, 0.5 mM DTT and 0.5 mM EDTA]. Reactions were incubated for 40 min at 30°C. NTPs and 80 μM 3′-OMe-GTP in TRX buffer were then added, and the mixture was incubated for the indicated times. Where indicated, 1.5 mM each of ATP, UTP and CTP were added and reactions were incubated for an additional period of time. In D, pG5MLPDG was used as a template. Transcription reaction was allowed to proceed for 20 min in the presence of 60 μM ATP, 600 μM GTP, 600 μM CTP, 5 μM UTP and 5 μCi of [α-P]UTP (800 Ci/mmol). G-free RNA fragments derived from transcripts were isolated after RNase T1 treatment, deproteinized, precipitated with ethanol and analyzed using 8% acrylamide denaturing gels, as previously described (). In F and E, transcripts were quantified by a phosphorimager (Molecular Dynamics, Storm 860). Immunodepletion was performed by incubating 100 μl concentrated P1.0 fraction containing 0.2% NP-40 and 350 mM KCl with 3.5 μg of anti-CDK9 antibodies at 4°C for 30 min, followed by incubation with 30 μl protein-G Sepharose beads (GE Healthcare). After removal of the beads, the fraction was incubated with 3.5 μg fresh anti-CDK9 antibodies again for 30 min at 4°C, followed by three rounds of incubation with 30 μl protein-G Sepharose beads. The depleted fraction was dialyzed against 100 mM HGKEDP prior to analysis in Western blotting and transcription assays. The following sequences were inserted into pBluescript SK+ (Stratagene) carrying the mouse U6-promoter () to generate the shRNA expression plasmids pBS-U6-hSpt5 (No. 1) and pBS-U6-hSpt5 (No. 2), respectively: U6-P160-1 (No. 1), 5′-GAACTGGGCGAGTATTACAttcaagagaTGTAATACTCGCCCAGTTCtt-3′ [sequences in uppercase correspond to nucleotides (nts) 406–424 of mRNA]; U6-P160-2 (No. 2), 5′-GGCTATATCGGTG TGGTGAttcaagagaTCACCACACCGATGTAGCCtt-3′ (sequences in uppercase correspond to nts 2155–2173 of mRNA. HeLa S3 cells were maintained in DMEM (Invitrogen) supplemented with 10% fetal calf serum and -glutamate. Cells (8 × 10 cells) were plated in 24-well plates and transfected with a total of 552 ng of DNA [50 ng of reporter plasmid, 2 ng of pCG-GAL4DBD (1–94) or pCG-GAL4 (1–94)-VP16 (413–490), kindly provided by Dr W. Herr () and 500 ng of pBluescript carrying the U6-promoter] using Lipofectamine 2000 (Invitrogen). In B, the cells were harvested at the indicated times post-transfection. In C, the indicated amounts of Gal4DBD expression plasmid were used. HeLa cells were transferred to fresh medium in a 12-well plate 24 h post-transfection and harvested 72 h post-transfection and luciferase activity was measured. Luciferase activity from triplicate experiments was normalized to protein amounts; data represents normalized values. We employed an transcription assay developed previously in our lab to analyze the stimulatory activity of DSIF and Gal4-VP16. The system was based on previous work showing that DSIF stimulated elongation under conditions of limited nucleotide triphosphates (NTPs) (). A phosphocellulose eluate (P1.0) derived from HeLa cell nuclear extracts (NEs) was generated, as previously described, containing RNAPII, general transcription factors and P-TEFb, but lacking DSIF and NELF (A). A supercoiled plasmid DNA template containing five Gal4-binding sites upstream of an adenovirus major late promoter (MLP), and a 380-bp G-free cassette downstream of the promoter, was used as the template (B). In the absence of DSIF, synthesis of full length transcripts was efficiently carried out in P1.0 in the presence of 60 μM ATP, 600 μM CTP, 5 μM UTP and 5 μCi [αP] UTP (C, lane 1). Reducing the concentrations of UTP and CTP resulted in inefficient synthesis of the full length transcripts, and the appearance of short transcripts ranging in size from 100 to 200 nt (C, lanes 2 to 4). Further reducing the concentrations of UTP and CTP to 1 μM significantly decreased the amount of shorter transcripts (C, lane 5). This was consistent with the previous results obtained using a different template (). In order to determine whether these results reflected transcriptional pausing or termination, the reaction was carried with the addition of a 1.5 mM ATP, CTP and UTP chase (C, lanes 6 to 10). Under these conditions, short transcripts were extended and gradually disappeared, and full-length transcripts appeared after 2 min incubation with the high concentrations of NTPs. These results suggested that the short transcripts generated in this assay resulted from pausing of RNAPII, rather than transcription termination, and that under conditions of limited concentrations of NTPs, RNAPII is prone to pausing . Upon the addition of increasing amounts of recombinant histidine-tagged DSIF (His-DSIF) (D) to the reaction, there was a reduction in short transcripts, and full-length transcripts were efficiently produced (E, lanes 1 to 4). This result was also consistent with the previous results (), and suggested that DSIF stimulates elongation by reducing pausing. In contrast to DSIF, His-Gal4-VP16 (D) increased the synthesis of full-length transcripts as well as short transcripts of various lengths (E, lanes 5 to 7). Quantification of the transcripts using a phosphorimager (F) revealed that the addition of His-DSIF resulted in a 3- to 6-fold increase in the amount of full-length transcripts and a concomitant decrease in the amount of shorter transcripts. Transcripts of 120 nt, for example, decreased by approximately 2.5-fold (F, lanes 1 and 4). The addition of His-Gal4-VP16 resulted in a 2- to 8-fold increase in the amount of full-length transcripts, as well as an increase in the amount of shorter transcripts (i.e. 120 nt transcripts increased 3-fold, F, lanes 1 and 7). These results suggested that DSIF and Gal4-VP16 regulate different steps of transcription. To determine whether the short transcripts generated in the presence of Gal4-VP16 were the products of paused RNAPII, we performed a chase experiment in the presence of Gal4-VP16 (A). Short transcripts generated in the presence of His-Gal4-VP16 disappeared after the 2-min chase period (A, lanes 3 and 4), demonstrating that they were due to paused RNAPII. To study the effect of DSIF and Gal4-VP16 in more detail, we varied the incubation time after NTP addition (B). When neither DSIF nor Gal4-VP16 was present, there was a negligible difference between the products of an 11.5- and 15-min reaction, wherein RNAPII paused between nts +100 and +200. After a 30-min reaction, more transcripts of various sizes, including full-length transcripts, were generated (B, lanes 1, 5 and 9). These results suggested that longer incubation times permit more initiation events. In the absence of DSIF, synthesis of full-length transcripts took over 15 min, whereas in the presence of DSIF, less than 11.5 min were needed (B, lane 2). More full-length products appeared after 30 min incubation in the presence of DSIF compared to the absence of DSIF, which was likely due to a higher level of processivity (B, lanes 2, 6 and 10). When His-Gal4-VP16 was added to the reaction instead of His-DSIF, in an 11.5-min reaction, transcripts were generated in the size range of 100–200 nt; full-length transcripts first appeared after 15 min, which was later than their appearance in the DSIF-containing reactions (B, lanes 3, 7 and 11). In addition, in contrast to DSIF-containing reactions, small-sized transcripts increased significantly over time, suggesting that in the presence of His-Gal4-VP16, initiation occurred efficiently, and the elongation step was rate-limiting. Concomitant addition of His-DSIF and His-Gal4-VP16 increased the amount of full-length products compared to addition of either His-DSIF or His-Gal4-VP16 alone at all time points (B, lanes 4, 8 and 12), and the size distribution of the transcripts was more like that seen in reactions containing His-DSIF, rather than His-Gal4-VP16. These results showed that DSIF and Gal4-VP16 act cooperatively to stimulate a high-level of transcriptional activation. When a DNA template lacking Gal4-binding sites was used in the reaction, there was no stimulation of transcription (data not shown). Together, these results indicated that the cooperative function Gal4-VP16 and DSIF requires that both molecules act on the same DNA molecule. DSIF associates with its target genes upon transcriptional induction, such as during heat-shock induction in (), possibly through recruitment by activators to elongation complexes. On the other hand, we have shown here that DSIF reduces transcriptional pausing equally well in the presence or absence of an activator, suggesting that activators may be dispensable for the recruitment and function of DSIF. We next examined the mechanism of recruitment of DSIF to elongation complexes, and whether activators have any effect on its activity. As shown earlier (C), when high concentrations of cold NTP were added 15 min after transcription was allowed to initiate, paused transcripts were efficiently extended (A, lanes 8 to 10). When DSIF was instead added at the same time point, no stimulatory effect was observed (A, lanes 5 to 7), suggesting that DSIF is unable to exert its stimulatory activity on the elongation complex at this stage. We therefore changed the time points at which DSIF was added to the reaction, to determine the stage at which DSIF exerted its positive effect (B). Compared to reactions in which DSIF was added 40 min before the addition of NTPs (B, lane 1), the positive activity of DSIF was attenuated when it was added concomitantly with NTPs, and was abolished when added afterward (B, lanes 2 to 6). This indicated that DSIF exerts its positive effect within a short window of time from initiation to elongation. The presence of Gal4-VP16 increased the amount of transcripts, but did not change the effect of DSIF, indicating that the activator does not alter the time-frame in which DSIF functions (B, lanes 7 to 12). Although DSIF is thought to regulate transcription elongation, the above result suggests that DSIF may in fact affect a step before elongation, such as promoter clearance. To examine this possibility, we carried out transcription reactions by using a DNA template that produces long transcripts containing double G-free cassettes (C). RNase T1 treatment of the transcripts allows simultaneous quantification of promoter-proximal and -distal regions. The level of promoter-proximal transcripts can generally be equated with the level of transcription initiation, while the ratio of promoter-distal to promoter-proximal transcripts reflects the efficiency of elongation. As reported previously (,), DSIF led an increase of efficiency of transcription elongation by 3.8-fold, while it had negligible effect on the synthesis of promoter-proximal transcripts (D, lanes 1 and 2), suggesting that DSIF does not influence transcription initiation. On the other hand, Gal4-VP16 activated the synthesis of both promoter-proximal and -distal transcripts with only a marginal effect on the distal-to-proximal ratio, suggesting that Gal4-VP16 mainly affects transcription initiation (D, lane 3). Gal4-VP16 and DSIF showed selective effects on initiation and elongation, respectively, even at the highest concentrations examined (data not shown). Collectively, these results are consistent with the previous view and suggest that Gal4-VP16 and DSIF together enhance overall transcription by accelerating different steps in transcription. It has been established that DSIF negatively regulates transcription elongation by acting in concert with NELF (,). In contrast, how DSIF stimulates transcription elongation is largely unknown except that P-TEFb-mediated phosphorylation of the Spt5 subunit of DSIF is responsible for converting DSIF from a repressor to a positive regulator (). One possible model is that promoter-proximal pausing leads to processive elongation thereafter, possibly serving as a checkpoint for this subsequent process. We therefore examined the roles of NELF and P-TEFb in the positive activity of DSIF. We purified Flag-epitope-tagged NELF (FLAG-NELF) (A) from a Flag-NELF-E-expressing HeLa cell line derivative, and added it to the transcription assay (B). There was no appreciable effect of addition of NELF on DSIF-activated transcription, or basal transcription levels in the absence of DSIF (B, lanes 3 to 6). In the presence of DRB, an inhibitor of the P-TEFb kinase, the positive activity of DSIF was significantly inhibited (B, lanes 2 and 8), suggesting that P-TEFb is critical for the positive regulatory effect of DSIF. Addition of NELF resulted in a much stronger inhibitory effect and the generation of transcripts of less than 150 nt (B, lanes 9 and 10). Based on the results of previous studies, this effect is likely due to promoter-proximal pausing induced by NELF and DSIF upon inhibition of P-TEFb activity by DRB (,). Consistent with the previously published results (), NELF was unable to induce such pausing in the absence of DSIF, even in the presence of DRB (B, lanes 5, 6, 11 and 12). These results indicated that the positive activity of DSIF is dependent on P-TEFb and independent of NELF. To further confirm the above results, we depleted P-TEFb from P1.0 fraction (C) and added it back prior to preincubation (D) or during elongation (E), The latter experiment was carried out in order to exclude the possibility that the presence of P-TEFb during preincubation and initiation steps may cause phosphorylation of RNAPII or DSIF and thus prevent the effect of NELF on elongation. As expected, P-TEFb was critical for the stimulatory activity of DSIF (D, lanes 4 to 6). When P-TEFb was absent, NELF cooperated with DSIF to repress elongation (E, lanes 3 to 6), and the addition of P-TEFb after the start of transcription alleviated this repression (E, lanes 7 to 10). Importantly, with P-TEFb added back, the full-length transcripts were synthesized to similar extent regardless of the presence of NELF, suggesting that the occurrence of NELF- and DSIF-induced promoter-proximal pausing has no influence on the subsequent DSIF- and P-TEFb-induced processive elongation. To investigate the role of DSIF in transcriptional activation , we used a reporter gene assay in HeLa cells, in which the expression of the reporter gene is controlled by Gal4-VP16. In this system, we predicted that if Gal4-VP16-mediated transcriptional activation involves the positive activity of DSIF, depletion of the large subunit of DSIF, hSpt5, would reduce expression of the reporter gene. We transfected HeLa cells with a plasmid expressing a short hairpin RNA (shRNA) targeting , together with Gal4-VP16 or Gal4DBD expression plasmids, and a luciferase reporter plasmid with Gal4-binding sites upstream of the promoter. Western blot analysis showed an 80% decrease in the amount of hSpt5 in cells transfected with shRNA No. 1, compared to cells transfected with a control plasmid, 72 h after transfection (A). A second shRNA targeting a different sequence in similarly knocked down hSpt5 protein levels in cells, indicating that the effect of shRNAs was gene-specific. We used shRNA No. 1 in the following experiments. Knockdown of caused a significant reduction in Gal4-VP16-activated expression of the reporter gene 72 h after transfection (B, upper panel), consistent with results that DSIF cooperates with Gal4-VP16 to achieve high-level activation. In contrast, when cells were transfected with a plasmid encoding Gal4DBD instead of Gal4-VP16, knockdown of had little effect on the low-level expression of the reporter gene (B, lower panel). The above results indicated that Spt5 is essential under conditions where transcriptional activity is induced, but not for transcription under basal conditions. To examine the requirement for DSIF more precisely, we adjusted the expression levels of the luciferase reporter gene by co-expressing varying amounts of Gal4DBD with Gal4-VP16. The Gal4DBD(1–94) is able to bind DNA templates, and is believed to interfere with Gal4-VP16-mediated activation in a competitive manner. As shown in C, expression of increasing amounts of Gal4DBD diminished expression of the reporter gene in both knockdown and control cells, suggesting that Gal4-VP16-mediated activation was competitively suppressed by Gal4DBD. Note that the expression level of the luciferase reporter gene in knockdown cells approached that of control cells with increasing amounts of Gal4DBD. D shows the ratio of luciferase activity in knockdown cells relative to control cells, illustrating quantitatively the results in C. These results indicated that different levels of transcription have different requirements for DSIF. Together, the results of both and assays showed that transcription at a higher level imposes a stronger requirement for DSIF. #text

Nucleotide excision repair (NER) is an evolutionarily conserved DNA repair pathway that deals with severely distorting DNA lesions including intrastrand crosslinks such as UV-induced pyrimidine dimers [reviewed in (,)]. Within NER two damage-sensing pathways are recognized: one for the entire genome, global genome repair (GGR), and one for the transcribed strand of active genes, transcription-coupled repair (TCR). In yeast, GGR requires Rad7, a protein carrying leucine-rich repeats, and Rad16, a member of the SWI2/SNF2 subfamily of putative helicases (). These proteins presumably act in a complex that might be required in chromatin remodeling to facilitate damage recognition by Rad4/Rad23 [reviewed in (,)]. As ongoing transcription is required for TCR, damage recognition is likely done by the elongating RNA polymerase (RNAP) itself. RNAP arrests at injuries in the template strand triggering, likely via additional specific factors, the recruitment of the DNA repair machinery [reviewed in ()]. Interestingly, TCR appears to be functional once a low and basal rate of transcription is achieved, beyond which there is no simple correlation between transcription and repair rates (). In , the stalled RNAP leads to the recruitment of the transcription-repair coupling factor (TCRF) Mfd, allowing for the release of RNAP and further recruitment of the repair factors (,). In eukaryotes the precise mechanism of TCR remains poorly understood. Mutations in proteins required for NER lead to severe disorders known as Xeroderma pigmentosum and Cockayne's syndrome [for review see ()]. One of these proteins, Cockayne syndrome B protein (CSB), and its yeast ortholog Rad26, share conserved functions (,) and represent putative eukaryotic TCRF candidates. CSB and Rad26 belong to the SWI2/SNF2 helicase superfamily. Although CSB has been shown to have DNA-dependent ATPase activity, an ATPase-deficient mutant partially restores CSB activity (). The putative function of CSB as a TCRF has been substantiated by reconstitution of the TCR initiation steps, in which an elongating RNAPII arrested at a DNA lesion was shown to mediate an ATP-dependent incision of the damaged DNA only in the presence of CSB (). XPG, one of the structure-specific DNA endonuclease responsible for the removal of an oligonucleotide containing the DNA lesion in NER, is another protein involved in TCR. Recent results imply a coordinated recognition of stalled RNAPII by XPG and CSB in TCR initiation in mammalian cells and suggest that TFIIH-dependent remodeling of stalled RNAPII without release may be sufficient to allow repair (). In yeast, the Rpb9 subunit of RNAPII has also been shown to contribute to TCR (,). Alternatively, and analogous to the mRNA-dependent loading of termination factors in (,), it is also conceivable that the nascent mRNA, or proteins bound to it, may be required to load repair enzymes at stalled polymerases. On the other hand, RNAPII is subject to ubiquitylation and proteasome-mediated degradation in response to UV-generated DNA damage (). It has been proposed that degradation of damage-stalled RNAPII complexes might be an alternative to TCR (). Indeed, recent studies have shown that arrested RNAPII elongation complexes are the preferred substrate for ubiquitylation, which is dependent on the C-terminal repeat domain (CTD) of RNAPII and on the Def1 protein in yeast (,). In eukaryotic cells, export of nuclear mRNA to the cytoplasm requires correct RNA-processing and the association of a number of RNA-binding proteins to form export-competent ribonucleoprotein particles (mRNP) [for review see ()]. Although there is growing evidence for transcription-coupled mRNA export, the physical nature of this coupling is not known. A connection between mRNP biogenesis and transcription is provided by THO, a conserved four-protein complex composed of stoichiometric amounts of Tho2, Hpr1, Mft1 and Thp2 () that is recruited to active chromatin (,). Null mutation of any component of THO leads to similar phenotypes of transcription impairment and mRNA export defects, as well as to a strong transcription-associated hyper-recombination phenotype (,,). Together with the mRNA export proteins Sub2/UAP56 and Yra1/Aly, THO forms a larger complex termed TREX in yeast and humans (). However, even though deletion of any of the THO genes leads to complete depletion of the complex (), THO remains stable in mutants (), indicating that it forms a core complex independently of Sub2. Strikingly, mutants of the Mex67-Mtr2 mRNA export factor—a heterodimer that mediates the interaction of the mRNP with the nuclear pore complex (NPC)—show THO-like gene expression and recombination phenotypes (). The idea of THO being functionally involved in mRNP biogenesis and export is further strengthened by the observation that mutants of the Thp1–Sac3 complex, which has been shown to function in mRNA export by docking the mRNP to specific nucleoporins at the nuclear pore entry (,) confer the same transcription, mRNA export, and hyper-recombination phenotypes as do THO/TREX mutations (,). Previously, we showed that null mutations of the and genes confer defects in NER (). With this precedent and considering that RNA-binding proteins are concomitantly assembled on the nascent mRNA to generate a stable and export-competent mRNP [reviewed in (,,)], we studied whether TCR might be connected to mRNP biogenesis and export. We found that THO, Sub2-Yra1, Mex67-Mtr2 and Thp1-Sac3 are required for efficient TCR in yeast, thus linking mRNP biogenesis and export to TCR. Using a construct in which a self-cleaving Hammerhead ribozyme was cloned between two hot spots for UV damage, we demonstrate that TCR does not depend on the nascent mRNA, neither in wild-type nor in THO and Thp1-Sac3 deficient strains. Chromatin immunoprecipitation (ChIP) analyses revealed that, beside a severe UV damage-dependent loss in processivity, RNAPII is found to be bound to chromatin upon UV irradiation in THO mutants. Interestingly, Def1, a factor responsible for the removal of stalled RNAPII from a DNA template, is essential for the viability of THO mutants subjected to DNA damage. Our results support a model in which mRNP biogenesis and export is required for efficient TCR by preventing the occurrence of defective RNAPII complexes, which may remain stalled at a DNA lesion. We used strains W303-1A, (W839-5D, R. Rothstein), (MGSC97, 3), (MGSC102, 13), (U674-1C), (RK2-6D, 45), (WH101-1A, 37), pCM185- (, DLY33), (Ura- segregant from DLY23, 46), (WMC1-1A, 36), (WFBE046, F. Fabre), which have been reported previously, and isogenic derivatives obtained by genetic crosses. The strains were obtained by replacement of the gene in diploids and subsequent tetrad dissection. For ChIP analyses, the promoter fused to the 5′-most 300 bp of the open-reading frame was integrated at the locus (). Plasmids pHG001-Rib and pHG001-rib were obtained by insertion of a fragment containing 39 bp yeast T-tracts ( promoter), 52 bp Hammerhead ribozyme sequence (), the 47III-HII fragment, and 39 bp yeast T-tracts ( promoter) into the Eco47III-BssHII sites of plasmid pRWE005 (). Yeast cells were grown in YPD-rich medium to an OD of 0.6. Plating, UV irradiation, and quantification were performed as described (). All survival curves shown represent the average of at least three independent experiments. Irradiation and repair was carried out as described () with minor modifications. Yeast cells were grown in 400 ml YPD-rich medium, or SG supplemented with the appropriate amino acids in the case of cells harboring the plasmid pHG001, to an OD of 0.8, harvested, and resuspended in SD or SG to an OD of 1.2. A 200 ml aliquot was irradiated with 230 J/m UV light using germicidal lamps (Philips T UV 15 W). The medium was supplemented to YPD-rich or with the appropriate amino acids and the cells incubated at 30°C in the dark for recovery. Fifty milliliter aliquots were taken at the indicated repair times, chilled on ice and DNA purified using the described CTAB protocol (). All steps from UV irradiation to DNA extraction were carried out in red light (Philips T LD 18 W RED). CPDs were mapped by indirect end-labeling and quantified as described (). DNA was cut with appropriate restriction enzymes and aliquots were cut at CPDs with T4-endonuclease V (T4endoV, Epicentre) or mock treated. The DNA was electrophoresed in 1.3% alkaline agarose gels, blotted to Nylon membranes and hybridized with radioactively labeled strand-specific DNA probes. Strand-specific probes were generated by primer extension. Primers to generate DNA templates and probes that hybridize to the TS- and NTS-strand, respectively, were: Rpb2-A: 5′-TCTTGGAATAATAACTTCGCGGC-3′; Rpb2-B: 5′-GGTGGATGACAAGATACATGCC-3′; pHG001-A: 5′-ATTTTTGACACCAGACCAACTG-3′; pHG001-B: 5′-TCTGCCATTGTCAGACATGTAT-3′. Membranes were analyzed and quantified with a PhosphorImager (Fuji FLA3000). The CPD content was calculated using the Poisson distribution, -ln(RF/RF), where RF and RF represent the signal intensities of the intact restriction fragment of the T4endoV- and mock-treated DNA, respectively. Region-specific damage was calculated as the signal of that region in the T4endoV-treated DNA divided by the signal of the whole lane. The corresponding signal of the mock-treated DNA was subtracted as background. The average of the initial damage generated with 230 J/m was 0.3 CPD/kb. To allow direct comparison between different strains, repair curves were calculated as the fraction of CPDs removed versus repair time. The initial damage was set to 0% repair. RNA was extracted and northern analyses performed according to standard procedures. For RNA synthesis recovery analyses, filters were hybridized with a 324-bp long fragment obtained by PCR using primers RPB2 A, RPB2 B. Northern blots were quantified using a Fuji FLA 3000 and normalized to the rRNA levels of each samples. For pHG001 ribozyme cleavage analyses, filters were hybridized with a 314-bp long fragment obtained by PCR using primers HG001-A and HG001-B. Cells were grown and irradiated as described above. Forty milliliter aliquots were taken at the indicated repair times and cross-linked with a 1% formaldehyde solution for 15 min at RT. Glycine was added to a final concentration of 125 mM, and the cell pellets frozen in liquid nitrogen and kept at −80°C. ChIP assays were performed as described (). Monoclonal 8WG16 antibody (COVANCE) and protein A-Sepharose were used to immunoprecipitate RNAPII. The GFX purification system (GE Healthcare) was used for the last purification step. All samples were treated with 200 ng photolyase (TREVIGEN) for 30 min under photoreactivating light (Sylvania F15T8 BLB) prior to real-time quantitative PCR analysis. We used 20–30-bp oligonucleotides for the PCR amplification of two fragments of (3–43 and 7621–7674) and the 9716–9863 intergenic region of chromosome V, which was used as non-transcribed control. Real-time quantitative PCR was performed using SYBR green dye in the 7500 Real Time PCR system (Applied Biosystems). Standard curves for all three pairs of primers were performed for each PCR analysis, all PCR reactions being performed in triplicate. The enrichment of each PCR amplification of interest was calculated as the ratio between the region-specific signal and the intergenic signal of the precipitated fractions normalized with respect to the corresponding ratios of the input fractions. At least three independent experiments were performed for each condition. Primer sequences are available upon request. To analyze whether defects in mRNP biogenesis and export result in impaired TCR, we studied mutants defective in both GGR and mRNA export. Isogenic mutants were generated and survival after UV irradiation was determined (, upper panel). Isogenic repair-proficient W303-1A, repair-deficient and TCR-deficient strains were used as controls. The and single mutants show no increased UV sensitivity as compared to wild-type cells. However, upon UV irradiation viability of the and double mutants dropped below the levels of the single mutant. Survival of the strain was similar to survival of the strain, whereas was less affected. Next, we analyzed mutants of the Thp1-Sac3 complex, which acts downstream of Mex67-Mtr2 on the mRNP biogenesis and export route. Isogenic , and mutants were generated and UV survival was determined (, middle panel). The and single mutants show no increased UV sensitivity as compared with wild-type cells. However, viability of the and double mutants was below the level of the single mutant upon UV irradiation. Since THO mutants have been shown to be sensitive to UV irradiation in the absence of GGR (), we performed UV survival curves of and strains for comparison of phenotype strength (, lower panel). As expected, and survival were reduced below the levels of the single mutant upon UV irradiation. UV sensitivity of cells was stronger than cells, and weaker than cells, whereas and showed similar UV sensitivity, consistent with the individual phenotype of the single mutant in other assays (,,,). We have recently described the mutant allele, which exhibits severe transcription defects but weak hyper-recombination (). We tested the allele for UV sensitivity in the absence of GGR to check whether the UV sensitivity of THO mutants was rather linked to their transcription deficiencies or to the formation of recombinogenic structures. Survival of and cells were similar (, lower panel), indicating that the observed UV sensitivity was linked to the transcription defects of THO mutants. Next, we analyzed removal of UV photoproducts in and cells at the molecular level. Isogenic repair-proficient wild-type, TCR-deficient and GGR-deficient were used as controls. Repair after UV irradiation was determined in a 4.4-kb restriction fragment containing the constitutively expressed gene by alkaline electrophoresis and indirect end-labeling (). As previously reported (), cells showed wild-type repair levels in the transcribed strand (TS) while repair of the non-transcribed strand (NTS) was strongly reduced. In cells, repair of the TS was significantly reduced while repair of the NTS almost reached wild-type levels (). In and cells, repair of the TS was severely impaired while repair of the NTS did not exhibit significant repair defects (B). Thus, our results indicate that and show defects in TCR, but not in GGR. In a previous report, molecular analysis of DNA repair in and cells indicated general defects in NER (). In these studies, the UV doses used had produced extensive DNA damage, in contrast to the conditions used in this study, in which one repair event is sufficient to restore the intact DNA (about 1 CPD per restriction fragment). The observations that mutants of the Sub2-Yra1 and Thp1-Sac3 complexes, which act downstream of THO in mRNP export, are specifically affected in TS repair lead us to examine CPD removal in and cells, using our UV irradiation conditions (A and C). Repair of the TS was significantly reduced in both strains. As observed by UV sensitivity assays in the absence of GGR (), cells were more strongly affected in TS repair than cells, reaching levels similar to cells. Likewise, repair of and cells were equally affected in DNA repair (data not shown). In the NTS, the repair levels of and cells were similar, in contrast to previous results obtained with extensive DNA damage (), indicating that GGR is not significantly affected in and cells in our conditions. In yeast, RNA synthesis is inhibited shortly after UV irradiation, probably due to the presence of CPDs in the TS of active genes. The ability of wild-type and mutant yeast cells to recover RNAPII synthesis in individual genes has been shown to mirror their strand-specific repair capacity (). To gain additional information on the connection between the recovery of mRNA levels and TCR, kinetics of the transcript levels were determined by northern analysis in , and wild-type cells (). A direct correlation between the expression levels prior to UV irradiation and repair rates was not apparent, in agreement with previous work showing the absence of simple correlation between transcription and repair rates (). However, upon UV irradiation, transcript recovery appeared to be most efficient in cells, while transcript recovery was clearly affected in all other mutant strains. This result points to THO, Sub2-Yra1 and Thp1-Sac3 behaving differently from Rad26 in response to UV damage, since they appear to undergo severe transcription impairment upon UV irradiation, in addition to their TCR deficiencies. The molecular basis underlying the requirement of functional mRNP biogenesis and export factors for TCR might rely on the proper packaging of the nascent transcript or on their effect on transcription. In repair-proficient cells, the nascent mRNA could mediate the TCR reaction in response to the transcriptional stalling occurring at DNA lesions. To test this possibility, we designed a construct containing two 39-bp long T-tract sequences inserted at different sites within the promoter-driven ORF. Between the two T-tract sequences, 52 bp encoding either an active self-cleaving Hammerhead ribozyme or an inactive mutated form () were inserted (A). Northern analysis confirmed a complete disappearance of the full length mRNA in the construct carrying the active ribozyme (B), indicating that the nascent mRNA was efficiently cleaved between the T-tracts. We first assessed whether TCR efficiency depends on the integrity of the nascent mRNA in yeast wild-type cells by comparing repair rates in the T-tracts situated upstream (T1) and downstream (T2) of the ribozyme (C and D). The initial damage was higher in T2 than in T1, likely reflecting local differences in chromatin structure. Importantly, our results show no significant difference between repair of T1 and T2, neither in the active (Rib) nor in the mutated (rib) ribozyme constructs, indicating that an intact and 5′-capped nascent mRNA is not required for efficient TCR in wild-type cells. Nevertheless, in contrast to wild-type cells, the occurrence of sub-optimal mRNP might impede the process of TCR in mutants of THO/TREX and Thp1-Sac3. To test this possibility, we used the ribozyme system to assess whether wild-type TCR can be restored in T2 after ribozyme self-cleavage of the nascent mRNA in and cells (). Repair in the full length fragment was clearly below repair levels achieved in wild-type cells, confirming the TCR deficiency of cells. Comparison of repair efficiencies in T1 and T2 did not show any significant difference, neither in the active (Rib) nor in the mutated (rib) ribozyme constructs. Similar results were obtained in cells (data not shown). Thus, our results indicate that cleavage of the nascent mRNA does not restore TCR in THO and Thp1 mutants. Consequently, we assume that the key player in the organization of TCR is the RNAPII itself, or some associated factors, rather than the nascent mRNA. Since we can rule out an active role of the nascent mRNA in TCR, it is conceivable that an intact RNAPII complex is sufficient to mediate proficient TCR. Recently, a protein called Def1 was shown to trigger ubiquitylation and degradation of RNAPII in response to UV damage as an alternative pathway to DNA repair (,,). To study the possible requirement for Def1 in the removal of trapped RNAPII presumably present in THO mutants, Δ Δ double mutants were generated and analyzed. Δ Δ double mutants were viable, but very slow growing. Since all THO/TREX mutants grow poorly at 37°C, we first investigated whether deletion of might increase the temperature sensitivity (ts) phenotype of Δ cells (A). An additive effect that made cells inviable at 37°C was observed in the double mutant as compared to the single mutant. Next, we performed UV survival curves in isogenic Δ, Δ and Δ Δ mutants (B). Viability of the Δ Δ double mutants was reduced below the levels of the Δ and Δ single mutants upon UV irradiation, indicating a synergistic effect of the two mutations on UV sensitivity. In the absence of GGR, Δ has been shown to be highly sensitive to UV irradiation (). Nevertheless, the UV sensitivity of Δ Δ Δ cells was increased as compared to Δ Δ cells (B and C). These genetic interactions between and indicate that Def1 is important for the viability of THO mutants subjected to stress and DNA damage. Given the exacerbated ts and UV sensitivity phenotypes of double mutant () and the reduced transcription processivity of THO mutants (), we determined the kinetics of RNAPII distribution after UV damage in and wild-type cells. For this purpose, the levels of RNAPII at distal positions within the large (8 kb) gene driven by the promoter were analyzed by ChIPs with antibody 8WG16 directed against RNAPII (). In wild-type cells, we observed a drop in the overall amount of RNAPII after UV irradiation that was accompanied by a significant loss of processivity along the transcribed unit (A). Recovery of both the amount of RNAPII and processivity occurred at similar rates and were nearly complete 90 min after UV irradiation. In cells, a drop in the overall amount of RNAPII was observed after induction of UV lesions (B). Both the drop in RNAPII density and its recovery were comparable to wild-type cells. However, the loss of processivity observed in cells was much stronger than in wild-type cells and its recovery very slow, 50% of the polymerases being lost between the 5′- and the 3′-end of the gene 90 min after UV irradiation. Therefore, the amount of RNAPII loaded on a transcribed unit is not the limiting factor for TCR in THO mutants. Importantly, the recovery of RNAPII association toward the end of the coding region (3′-end) with increasing repair time is consistent with the repair rates observed previously (). While the amount of RNAPII loaded on the gene was strongly decreased after UV irradiation in wild-type and cells, only a weak drop in the amount of RNAPII was observed in cells (C), probably reflecting the defects in RNAPII degradation of this mutant strain. Recovery of RNAPII on the 3′-end of the gene was slow as compared to wild-type, with kinetics similar to those of , indicating a DNA damage-dependent processivity defect. In cells (D), a weak drop in the amount of RNAPII loaded on the gene was observed after UV irradiation. The amount of RNAPII localized on the 5′-end of the gene was similar to the levels found in cells, suggesting that the mutant did not significantly alter the accumulation of RNAPII observed in the mutant. Noteworthy, the amount of RNAPII associated with the 5′-end of the gene was repeatedly higher at 60 min than at 90 min after UV irradiation in both strains. This observation was also made in strains but never in wild-type cells. Thus, in mutant strains in which TCR and/or RNAPII degradation is impeded, RNAPII tends to accumulate on the template with a peak 60 min after UV irradiation, this effect being stronger in mutants which fail to ubiquitylate and degrade the stalled RNAPII. In cells, recovery of RNAPII at the 3′-end of the gene after UV irradiation was slow, reflecting a DNA damage-dependent loss of processivity. Taken together, our data indicate that RNAPII is indeed associated with transcribed genes after UV irradiation in THO mutants, even if repair of the lesions and transcription recovery are impeded. However, no accumulation of RNAPII was detected at the 5′ of the gene in THO mutants, in contrast to and cells, in which RNAPII accumulated at the 5′-end after UV irradiation as a result of the lack of RNAPII degradation (). These results imply that TCR-deficient RNAPII complexes remain stalled at DNA lesions in THO mutant and have to be removed from the template in order to provide access for DNA repair machineries, a process that depends on Def1. The inability of human Cockayne's syndrome cells to recover RNA synthesis after DNA damage was recently found not to be due solely to the failure of these mutants to remove lesions in the transcribed strand of active genes, but to defects in the re-initiation of the transcriptional program after UV irradiation (). Although TCR of the constitutively expressed gene is strongly affected in cells depleted of Rad26—the yeast CSB homologue—we found that, in mutants, mRNA recovery is much faster as compared to THO, Sub2-Yra1 and Thp1-Sac3 mutants (). The mRNA export-dependent TCR-deficient phenotype could reflect a substantial lag in transcription initiation. This is not the case, as ChIP analysis indicated that, in contrast to CSB, RNAPII loading after UV irradiation was not affected in THO mutants (). However, the amount of polymerases reaching the 3′-end of the gene was substantially lower in . Thus, the RNAPII processivity defects of appear to be strengthened upon UV-irradiation, presumably leading to severe delays in mRNA recovery. Even though it is conceivable that poor transcription rate and/or low RNAPII processivity alone might lead to defective TCR, several lines of evidence argue against this possibility. First, no simple correlation could be found between repair and transcription rates neither in different mutants () nor at different (). Second, transcription-elongation mutants like , in which transcription elongation is impaired and RNAPII processivity significantly reduced (,), do not show defects in TCR (). On the contrary, suppresses the TCR defects of , indicating that reduced RNAPII processivity can even act positively on TCR. There is growing evidence for transcription-coupled mRNA export [for review see ()], though the physical nature of this coupling is not known. The possible functional and physical connection between a subset of transcribed genes and the NPC () suggest that deficient mRNA export could negatively affect RNAPII transcription. Our results, providing evidence for a functional link between TCR and mRNA export, support the existence of such a feedback mechanism, which would alter transcription so that it is no longer proficient for TCR. A major player in the coupling of transcription and mRNA processing is the CTD of the largest subunit of RNAPII, which acts as a loading platform for pre-mRNA processing factors [reviewed in (,,)]. Several lines of evidence indicate that CTD phosphorylation occurs upon UV irradiation and might be involved in the signaling of stalled RNAPII for TCR (,,). These features place the CTD tail as a potential target for feedback modifications of RNAPII in response to mRNP export defects. Given the association between mRNA export and efficient TCR, the nascent mRNA could act as a mediator for TCR. However, we discard the possibility that the nascent mRNA actively supports the TCR process by monitoring the repair of UV-lesions encountered by a RNAPII-complex containing either the full-length nascent mRNA or about 70 bp left over after ribozyme-mediated cleavage (). It is worth noticing that the available nascent mRNA might be even shorter than 70 bp considering that 15 to 20 bp are covered by the RNAPII holoenzyme (,). Furthermore, our results indicate that RNA 5′-capping is dispensable for TCR. Thus, our results support the idea of the RNAPII machinery being the main mediator for an active DNA damage response. Different evidence led to the proposal that the hyper-recombination phenotype of THO mutants is mediated by the formation of R-loops (RNA:DNA hybrids) behind the elongating RNAPII, as well as impaired replication fork progression (,). Here, we show that ribozyme-mediated cleavage of the nascent mRNA does not suppress the TCR deficiency of (), suggesting that the hyper-recombination and TCR phenotypes might be mediated by different intermediates. This conclusion is further supported by the observation that the non-hyper-recombinant allele shares the transcription, mRNA export and TCR phenotypes of cells () (). Recently, RNAPII processivity has been shown to be significantly reduced in THO mutants (). Our ChIP analysis of UV-irradiated cells revealed that RNAPII processivity was further impaired while RNAPII loading remained unaffected in cells (). A conclusion of these findings is that the TCR defect of cells is a consequence of damage-trapped RNAPII. Since transcription is a one-track copying process, a trapped RNAPII can only be resolved by repairing the damage so that transcription can resume or by physically removing the stalled RNAPII [reviewed in ()]. Thus, it is understandable that RNAPII degradation activities, such as Def1, might be important in backgrounds in which aberrant stalled RNAPII are formed. Indeed, we observed that the mutant exhibited profound growth defects and was highly temperature- as well as UV-sensitive (). In contrast to wild-type and cells, RNAPII accumulated at the 5′-end of upon UV-irradiation in mutant strains (). Hence, Def1-mediated RNAPII degradation appears to be important for the removal of stalled RNAPII after UV damage, in agreement with previous works showing that UV-dependent RNAPII ubiquitylation depends on Def1 (,,). Worthy of note, we find that the amount of RNAPII tends to decrease 90 min after UV irradiation in backgrounds in which TCR or RNAPII degradation is compromised. This decrease might reflect the physiological turnover of stalled RNAPII in yeast cells. Such an interpretation implies that endogenous dissociation of stalled RNAPII would be much faster than , where human RNAPII stalled at a CPD remains stable for days (). Alternatively, the reduction in the amount of RNAPII at 90 min could be due to a down-regulation of the promoter activity or to the action of alternative RNAPII degradation activities. Given that RNAPII distribution obtained at the 5′-end of was similar in and mutants, it is conceivable that RNAPII removal by Def1 serves as an essential backup mechanism to allow repair of active genes in mutants. Our results suggest that the low processivity and mRNA export defects of mutants lead to stalled RNAPII-complexes in adverse transcription conditions, which are subjected to Def1 mediated degradation if transcription elongation cannot be rescued. Further investigation of the mutants will help us to understand the mechanisms of the different phenotypes of THO mutants, from hyper-recombination to DNA polymerase stalling, and how they might depend on each other. fig #text

Ribosome biogenesis is a fundamental multistep process that, in eukaryotes, takes place largely within the nucleolus (). Late steps in both 40S and 60S ribosomal subunit (r-subunit) synthesis occur in the nucleoplasm and after nuclear export of precursor particles in the cytoplasm (,). Ribosome synthesis is evolutionarily conserved throughout eukaryotes (,), and so far most of our understanding of this process has been obtained from studies with (,). In the yeast nucleolus, three of the four rRNAs (18S, 5.8S and 25S) are transcribed as a single large primary transcript by RNA polymerase I and processed to the first detectable rRNA precursor (pre-rRNA), the so-called 35S pre-rRNA. The fourth rRNA (5S) is independently transcribed as a pre-rRNA (pre-5S) by RNA polymerase III. In the 35S pre-rRNA, the mature rRNA sequences are separated by two internal transcribed spacers (ITS1 and ITS2) and flanked by two external transcribed spacers (5′ ETS and 3′ ETS), which must be precisely and efficiently processed to ensure correct formation of mature rRNAs (). Maturation of rRNAs is a well-defined pathway () and involves numerous -acting factors that are required for the processing and covalent rRNA modification reactions, such as small nucleolar RNA–protein (snoRNP) complexes, endonucleases and exonucleases, and different base methylases (,). Concomitantly to rRNA maturation, the pre-rRNAs assemble in an ordered manner with the 79 ribosomal proteins (r-proteins) and a large number of -acting factors that are generally referred to as r-subunit assembly factors (,) (for examples of -acting factors see ). The process of r-subunit assembly is still poorly understood. An outline of this process was provided by sucrose density gradient analyses in the 1970s, which identified 90S, 66S and 43S pre-ribosomal particles (,,). Recent advances employing proteomic approaches have revealed several distinct, successive pre-ribosomal particles and refined the model for the maturation of both 40S and 60S r-subunits [for a review (,,)]. These proteomic approaches have also led to the identification of novel non-ribosomal proteins, increasing the number of -acting factors involved in ribosome biogenesis to over 180. Evidence towards an understanding of the function of many of these -acting factors has been obtained by using a complete repertoire of techniques, thus, addressing their temporal association with pre-ribosomal particles and revealing the pre-rRNA processing and nucleocytoplasmic export defects caused by their mutational inactivation or depletion [for a review, see ()]. In contrast to the non-ribosomal proteins, the precise role of the r-proteins in ribosome biogenesis is still largely unexplored and most studies have been focused on their function during translation [for examples, see () and for a review, see ()]. Moreover, and paradoxically, the specific presence of r-proteins in pre-ribosomal particles is difficult to properly assign since r-proteins are common contaminants in purified complexes (). A very recent report has systematically approached the role of individual 40S r-proteins in ribosome synthesis (). This study revealed that most of the 33 r-proteins of the 40S r-subunit play distinct and essential roles in ribosome maturation and nucleocytoplasmic transport (). However, an equivalent analysis of the 60S r-subunit proteins has not yet been reported. There are some examples indicating that mutation in or depletion of many 60S r-proteins cause deficits in 60S r-subunits (), however, the contribution to ribosome biogenesis of only few 60S r-proteins has been analysed. So far, there is only detailed functional data available for Rpl5p, which is required for binding and stability of 5S rRNA (,), Rpl25p, which is required for efficient pre-rRNA processing at site C () and Rpl10p, which is involved in recycling of Nmd3p and subsequent subunit joining (,,). Rpl3p is required for 60S r-subunit accumulation (,) and participates in the formation and proper activity of the peptidyltransferase centre (PTC) (,). In order to learn more about Rpl3p, we have investigated the effect of genetic depletion of Rpl3p on ribosome maturation and export from the nucleus to the cytoplasm. Our results indicate that Rpl3p is required for the normal accumulation of 60S r-subunits due to defects in pre-rRNA processing of 27SA and 27SB and export of pre-60S r-particles. This suggests that Rpl3p has an essential role in the assembly of early pre-60S r-particles and that aberrant pre-ribosomal particles deficient in Rpl3p are retained in the nucleus. Recently, it has been shown that depletion of human Rpl3p alters proper chromosome segregation during mitosis (), a hallmark of most cancer cells (). In this study, we show that depletion of Rpl3p leads to a G1 arrest, however, it does not seem to interfere with proper chromosome segregation, measured as percentage of plasmid loss. All yeast strains used in this study are derivatives of strain W303 (α ). JDY511 (αHIS3MX6) is a haploid disruptant that requires a plasmid-borne copy of for cell viability (). Growth and handling of yeast and standard media were performed by established procedures (,). Tetrad dissections were performed using a Singer MS micromanipulator. DH5α strain was used for common cloning and propagation of plasmids (). The strain was obtained after transformation of JDY511 [YCplac111-RPL3] with plasmid pZGA196 (a generous gift from G. Adam), which allows expression of under the control of the promoter (), and subsequent segregation of YCplac111-RPL3. Growth behaviour on YPGal and YPD plates was further studied to test the faithful complementation and the shut-off of the construct under permissive and non-permissive conditions, respectively. For unknown reasons, the strain grows better in liquid YPGal medium containing 1 M sorbitol (YPGalS). For depletion, the strain was grown in YPGalS medium at 30°C until mid-exponential phase (OD of 0.8). Cells were harvested, washed and used to inoculate cultures in YPD medium containing 1 M Sorbitol (YPDS). Cell growth was monitored over a period of 48 h, during which the cultures were regularly diluted into fresh YPDS medium to maintain exponential growth. As a control, the wild-type JDY511 [YCplac33-RPL3] strain was used. At different time points, cells were harvested and subsequently used for preparation of total protein and RNA and of cell extracts for polysome analysis. Polysome preparation and analyses were performed as previously described () using an ISCO UA-6 system with continuous monitoring at A. Total yeast protein extracts were prepared and analysed by western blotting according to the standard procedures (). The monoclonal anti-Rpl3p antibody was a gift from J.R. Warner (). RNA extraction, northern hybridization and primer extension analysis were carried out according to the standard procedures (). In all experiments, RNA was extracted from samples corresponding to 10 OD units of exponentially grown cells and RNA corresponding to equal amounts of cells (ca. 5 μg for the wild-type strain) was loaded on gels or used for primer extension reactions. Sequences of oligonucleotides used for RNA hybridization and primer extension analyses have been described previously (). To test pre-ribosomal particle export, the appropriate strains (see ‘Results’ section) were co-transformed with a pRS315 plasmid harbouring a RPL25-eGFP reporter () or a RPS2-eGFP reporter () and a pRS314 plasmid expressing the nucleolar marker DsRedNOP1 (). Then, several transformants were grown to mid-log phase in selective liquid medium, washed, and resuspended in water. Acquisition was done in a Leica DMR microscope equipped with a DC camera following the instructions of the manufacturer. Digital images were processed with Adobe Photoshop 7.0. Cells grown in logarithmic phase to an OD of 0.1 to 0.3 were harvested, fixed with 70% ethanol and DNA was stained with propidium iodide as previously described (). Stained cells were analysed using a Becton Dickinson FACScan flow cytometer using CELL QUEST software packages to collect and analyse the data (BDIS, San José, CA). Rpl3p is an essential r-protein in yeast () that, in the 1980s, was shown to be required for normal accumulation of 60S r-subunits (,). Since then, the reports of Fried and co-workers remained the only information available on the role of this protein in ribosome biogenesis (,). To study in detail the function of Rpl3p in ribosome biogenesis, we first assessed steady-state levels of ribosomes upon its depletion. To this end, JDY511 [pZGA196] ( strain) and JDY511 [YCplac33-RPL3] ( strain) were grown in liquid YPGalS and shifted to liquid YPDS for different time points. In YPGalS, the growth rate of the strain was slightly slower than that of the wild-type control strain (doubling time of about 3 and 2 h, respectively), but it even decreased after transfer to YPDS (doubling times of 5.3, 7.6, 9.2 and more than 15 h after 6, 12, 18 and 24 h in YPDS, respectively). Western blot analysis revealed a marked reduction of Rpl3p in cells that coincided with the decrease in the growth rate in YPDS (data not shown). Polysome extracts were prepared from cells harvested from the YPGalS cultures and at different times after the shift to YPDS. The strain showed profiles very similar to those of the strain when grown in YPGalS, although a slight deficit of free 60S versus 40S was reproducibly observed (A and C). However, clear alterations in the profiles appeared after 6 h and became more pronounced at longer times in YPDS (D and data not shown). The Rpl3p-depleted strain showed a strong decrease in the levels of free 60S r-subunits versus the levels of free 40S r-subunits and an overall decrease in the 80S peak and in polysomes (D). In addition, there was an accumulation of half-mer polysomes (D). Wild-type cells showed no alteration in the polysome profile when transferred to YPDS (B). These results indicate that depletion of Rpl3p leads to a strong deficit in 60S r-subunits relative to 40S r-subunits. To characterize the basis of the net deficit in 60S r-subunits of the strain, we then analysed the effect of depletion of Rpl3p on pre-rRNA processing. Total RNA was isolated from and strains at various time points after transfer from liquid YPGalS to liquid YPDS, and steady-state levels of pre- and mature rRNA species were determined by northern blot and primer extension analyses. Different oligonucleotides hybridizing to defined regions of the 35S pre-RNA were used to monitor-specific processing intermediates (A). As shown in A, depletion of Rpl3p resulted in a marked decrease in 25S rRNA steady-state levels. This is likely due to an almost complete loss of the 27SB pre-rRNA species, which already becomes apparent at the shortest shift time point to YPDS. In addition, ongoing depletion of Rpl3p led to an accumulation of 35S pre-rRNA and aberrant 23S, 22S and 21S pre-rRNAs. These aberrant species extend from the 5′ end of the 35S pre-rRNA, site A and site A to site A, respectively. Hybridizations also identified another fragment, which extends from the 5′ end of the 35S pre-rRNA to site D (A). Depletion of Rpl3p also resulted in a mild reduction in the levels of the 20S pre-rRNA and slight reduction in the levels of 27SA compared to those from the grown in YPGalS (A). The steady-state levels of low-molecular-weight rRNAs were also studied. As shown in B, depletion of Rpl3p caused a strong decrease in the 7S pre-rRNA levels and a very slight reduction in the levels of mature 5.8S and 5S rRNAs, which was clearly noticeable only at the latest shift time points to YPDS. No differences were observed in the ratios of the long and short forms of the 5.8S rRNA. To determine the levels of 27SA and distinguish between the 27SB and 27SB precursors, we assessed the levels of these precursors and of the 27SA pre-rRNA by primer extension using a probe that hybridizes to both the 27SB and 7S pre-rRNAs (probe f, see B). As seen in , the data are consistent with those from the northern hybridizations. Rpl3p depletion led to a slight decrease in 27SA pre-rRNA and a more drastic reduction in 27SA pre-rRNA. Interestingly, the intensity of the stop corresponding to site B decreased substantially upon depletion of Rpl3p, whereas that of the stop at site B decreased 6 h after transfer to glucose medium but increased to roughly normal levels at later time points. Similar results were observed when using a probe that hybridizes only to the 27SB pre-rRNAs (probe g; data not shown). Thus, these results indicate that 27SB was primarily affected upon depletion of Rpl3p. Altogether, our results indicate that depletion of Rpl3p has a major impact on processing of 27SA, which then also leads to underaccumulation of the 27SB pre-rRNA. In addition, Rpl3p depletion affects negatively ITS2 processing events; the low levels of 27SB may account for the diminution of 7S pre-rRNAs upon depletion, however, the alternative form 27SB continues to be synthesized but fails to be converted to 7S pre-rRNA. Finally, depletion of Rpl3p inhibits or delays processing at sites A–A, which leads to the accumulation of normal 35S pre-rRNA and aberrant 23S, 22S and 21S pre-rRNAs. The appearance of these aberrant rRNAs and the accumulation of the 5′-D fragment indicate that processing events in the 5′-ETS and ITS1 do not occur in the normal order following depletion of Rpl3p. To determine whether the depletion of Rpl3p impairs nuclear export of pre-60S r-particles, we first analysed the localization of the 60S reporter construct Rpl25p-eGFP () in wild-type and strains. Under permissive conditions (YPGalS medium), Rpl25p-eGFP was found predominantly in the cytoplasm in both strains. However, following a shift to non-permissive conditions (YPDS medium) for as short as 6 h, Rpl25p-eGFP accumulated in the nucleus in about 30% of the cells. This phenotype was more evident after 12 h in YPDS since around 70–90% of the cells showed a nuclear accumulation of Rpl25-eGFP (). In many cells, we observed a very bright fluorescence signal for Rpl25-eGFP that was not restricted to the nucleolus, which was detected with the nucleolar marker DsRed-Nop1p. We did not observe nuclear accumulation of the Rpl25p-eGFP reporter in the wild-type control strain grown in YPDS (data not shown). We conclude that normal and/or aberrant pre-60S r-particles accumulate in the nucleus upon depletion of Rpl3p. This phenomenon is specific for the large r-subunit, since no accumulation of pre-40S r-particles was observed when we studied the localization of the 40S r-subunit reporter Rps2p-eGFP () (data not shown) Our previous results () and unpublished observations (I.V.R., unpublished results) as well as results from Wozniak and co-workers () indicate that Rpl3p interacts functionally and physically with the WD-repeat Rrb1p protein, which has been suggested to act as the Rpl3p assembler onto pre-60S r-particles (,). Rrb1p has also been shown to be required for the metaphase/anaphase transition during the cell cycle and proper chromosome segregation (). Rrb1p also functionally interacts with the origin recognition complex component Orc6p, involved in the initiation of DNA replication (,), and the yeast Pescadillo complex, which consists in yeast of Nop7p, Erb1p and Ytm1p and is required for both ribosome biogenesis and normal progression through the S phase of the cell cycle (,). Moreover, inactivation of human orthologues of Rrb1p, Nop7p, Erb1p/Bop1p, Orc6p and Rpl3p alters proper chromosome segregation (). To study whether yeast Rpl3p is required for optimal progression through the cell cycle, we first examined cellular morphology of cells by light microscopy. A normal morphology was observed for most cells when grown in galactose medium (A). However, 6 h after transfer to glucose medium, cells increased in size and a significant percentage (about 5%) showed an elongated shape and contained enlarged buds with pronounced apical growth. Apparently, these elongated cells contained duplicated, separated nuclei as shown by DAPI staining (A). Then, we performed fluorescence-activated cell sorting (FACS) analyses with yeast cultures of the and strains in early logarithmic phase. In asynchronous wild-type cells, we detected two peaks corresponding to cells with unreplicated (1C) and duplicated (2C) genomes, with the 1C peak being slightly higher than the one of 2C. A similar pattern was observed for the cells grown in galactose medium. However, 6 h after transfer to glucose medium, more cells remained 1C compared to the strain, most likely due to an arrest or delay in the transition through the G1 phase of the cell cycle (B). In this work, we report the functional characterization of Rpl3p in ribosome biogenesis. Rpl3p, which is the largest r-protein (387 residues in ; 43.7 kDa), is evolutionarily conserved in both sequence and structure in eukaryotes, eubacteria and archea (,). Examination of the structure of Rpl3p reveals that it contains two tightly packed globular domains and two extensions (). As discussed by Dinman and co-workers (,,), the globular domains are located on the solvent side of the 60S r-subunit and bind to the domain VIA of yeast 25S rRNA or bacterial 23S rRNA (), very close to the site where the ribosome interacts with the elongation factors eEF1 and eEF2. The extensions anchor Rpl3p to the central core of the 25S rRNA. One of the extensions is at the N-terminus of the protein (residues 10 to 24) and the other is internal to the protein (residues 217 to 278). The latter extension comes very close to the PTC where it can stabilize the surrounding rRNA (). Rpl3p has been extensively studied with respect to its role in translation, more specifically as an important functional component of the PTC (,). In bacteria, Rpl3p is one of the few proteins essential for the PTC activity (). In yeast, the first identified mutations conferred resistance to the PTC inhibitors trichodermin and anisomycin () and inability to maintain the M killer virus (). More recently, it has been shown that these mutations also affect the efficiency of programmed -1 ribosomal frameshifting due to a decrease in the PTC activity (,). In contrast to its characterization in translation, very little is known about the role of Rpl3p in ribosome biogenesis. To our knowledge, pre-rRNA processing has not been studied upon loss-of-function of bacterial Rpl3p. In yeast, the depletion of Rpl3p leads to a net deficit of 60S r-subunits [(,) and our results of ], and although pulse-chase experiments have been performed, they were not of enough resolution to assess the kinetics of pre-rRNA processing (). In this paper, we describe the role of yeast Rpl3p in pre-rRNA processing. Northern blot and primer extension analyses indicate that there is a drastic reduction in the steady-state levels of almost all 27S pre-rRNAs and both 7S pre-rRNAs upon depletion of Rpl3p. As a consequence, there is an underaccumulation of mature 25S rRNAs. However, the levels of 5.8S and 5S rRNA were only mildly affected at late time points of depletion. This is in agreement with the common observation that 5.8S rRNA behaves more stable than 25S rRNA upon depletion of many -acting factors required for 60S r-subunit biogenesis [for examples, see (,)] and the published data indicating that 5S rRNA forms a stable RNP with the 60S r-protein Rpl5p (). Intriguingly, primer extension analysis shows that the levels of 27SB pre-rRNA do not change significantly upon depletion of Rpl3p. Since it is likely that Rpl3p may not have a direct role in pre-rRNA processing reactions, we assume that the depletion of Rpl3p leads to abortive assembly of early pre-60S r-particles, which entails destabilization and degradation of the 27SA pre-rRNA and its immediate products 27SB and 7S pre-rRNAs. The almost constant steady-state levels of 27SB pre-rRNA upon Rpl3p depletion might reflect changes in the relative degradation rate of aberrant pre-60S r-particles containing this precursor but deprived of Rpl3p. These aberrant pre-60S r-particles, which should contain the 60S r-protein Rpl25p properly assembled, might be defective for nucleocytoplasmic transport, as suggested by the retention of the Rpl25p-GFP in the nuclei of Rpl3p-depleted cells. This defect is apparently specific as export of small r-subunits was unaffected. Furthermore, our northern analyses clearly indicate that depletion of Rpl3p causes a decrease in the efficiency of processing at the early cleavage sites A, A and A, thereby slightly affecting the levels of mature 18S rRNA and its 20S precursor. As a consequence, a 22S, a 21S and more abundantly a 23S aberrant pre-rRNA accumulated. This type of defect in 18S rRNA synthesis is a general feature of mutations that interfere with the synthesis of mature 25S and 5.8S rRNA (). It has been proposed that these pre-rRNA processing defects arise from inefficient recycling of -acting factors that improperly disassemble from defective pre-60S r-particles (,). Finally, there is an accumulation of a 5′ ETS-D fragment, suggesting that the aberrant 23S pre-rRNA can be processed to site D upon depletion of Rpl3p. In general, point mutations in Rpl3p cause similar but much weaker pre-rRNA processing defects than its depletion (I.V.R., unpublished result). This finding suggests that the inability to incorporate Rpl3p has a more dramatic effect on the fate of pre-60S r-particles than the incorporation of functionally hampered Rpl3p mutant variants. When and how is Rpl3p assembled? The ribosome assembly process is very difficult to assess experimentally and is not very well understood. In bacteria, it has been possible to reconstitute functional r-subunits from isolated mature rRNAs and purified r-proteins (,). These studies indicate that Rpl3p is amongst one of the first r-proteins that initiate assembly. In clear agreement with this fact, Rpl3p is present on the so-called p50S precursor particles [for a review, see ()]. In eukaryotes, since there is no ribosome self-assembly system from their components, the order of assembly of r-proteins into pre-ribosomal particles has not been characterized. , pulse-chase studies have suggested that yeast Rpl3p associates at a relatively early stage of the ribosomal maturation process (). The purification of Rpl3p within early 66S pre-ribosomal particles E, E and E is in agreement with these results (). Rpl3p is one of several extension-containing r-proteins (). Steric considerations require that these proteins bind rRNA at a stage prior to the formation of significant ternary structure. Steitz and co-workers have hypothesized that during assembly, the globular domains of bacterial Rpl3p bind first to sequences of domain VI of 23S rRNA, which adopt a structure similar to the final one present in the mature 50S r-subunit (). This binding is strong and stabilizes the protein on the rRNA. Then, the extensions of Rpl3p, which depend on interactions with the surrounding rRNA to properly fold [for a real example, see ()], bind sequentially to regions of internal rRNA domains to be accommodated inside the r-subunit and gain a stable structure (,). Since the overall structure of Rpl3p and its location in the large r-subunit is highly conserved between eubacteria, archaea and eukaryotes (), we can imagine a similar mode of assembly for the yeast Rpl3p in early pre-60S r-particles. We have recently isolated nine independent recessive mutations, which are synthetically lethal with a subset of -acting factors involved in early steps in the synthesis of 60S r-subunits including the putative RNA helicase Dbp6p, the nucleolar protein Rsa3p and the putative Rpl3p assembler Rrb1p [(,,) and I.V.R., unpublished results]. Further studies, using these above mutants as well as directed mutants where the extensions and the globular domains have been specifically altered, should help to dissect the contribution of the different Rpl3p domains to early 60S ribosome biogenesis events and to get insight into the mode of assembly of Rpl3p in pre-60S r-particles. Finally, we herein describe that depletion of yeast Rpl3p leads to a G1 delay or arrest of the cell cycle, which is accompanied by a percentage of cells with abnormal cell morphology. In yeast, cell-cycle defects have been previously described for mutant or depleted strains in other ribosome biogenesis factors. In these cases, cell-cycle progression is impaired not only at the G1 phase but throughout the different stages of the cell cycle (,,). The possible involvement of Rpl3p in cell cycle has been studied in other organisms; in zebrafish, while haplo-insufficiency in many r-proteins genes predispose to cancer, that in gives rise to similar tumour incidence as for a control line (). On the other hand, transient depletion of human Rpl3p increases the percentage of abnormal mitosis and alters proper chromosome segregation (). Interestingly, mutation in yeast Rrb1p arrests cell cycle at the G2/M phase by blocking mitosis and inducing chromosome instability, and transient depletion of GRWD, the human orthologue of Rrb1p, results in an increase of abnormal mitosis and an alteration in chromosome segregation (). However, our initial experiments, using a centromeric-plasmid loss assay (), suggest that mutation, at least the alleles we have tested, does not lead to defects in plasmid replication and maintenance (I.V.R., unpublished results). Further work is clearly needed to better understand the role of Rpl3p in ribosome biogenesis and clarify its putative link to cell-cycle progression.

The structure of biological macromolecules like polypeptides and nucleic acids comprises linear polymers of a limited number of small building blocks such as amino acids or nucleotides. Even so, because of their large size and a certain degree of flexibility of the polymer chains, biopolymers are usually folded, forming a distinct 3D structure which is generally essential for the biological function of the molecules. Studies of the 3D structure of nucleic acids go back to the seminal work of Watson and Crick, who discovered the DNA double helix. Since then, a large number of 3D structures of single- and double- stranded nucleic acids have been solved by X-ray crystallography and nuclear magnetic resonance spectroscopy, which allow a deep insight into the structure–function relationships of nucleic acids. In spite of the broad availability of structural data, investigations into the relationship between the 3D structure of nucleic acids and their interaction with stationary phases in chromatographic separation systems have been quite limited (,), mainly because of the highly dynamic nature of the process, and the involvement of both a liquid and a solid phase in the phase transfer. Since DNA and RNA molecules are employed in many biochemical applications, such as polymerase chain reaction, genotyping (), DNA sequencing (), hybridization assays and gene therapy (), understanding these phase transfer processes would not only help to improve our selection and optimization of chromatographic separation systems in order to purify and analyze them, but also to gain valuable information about the behavior of nucleic acid molecules in multi-phase systems and at interphases. Commonly used models describing the behavior of solute molecules at the liquid–solid interface, such as linear free enthalpy relationships (), are based on the fundamental Hammet equation and knowledge of principal structural descriptors, such as intrinsic molar volume, Hildebrand-solubility parameters, polarizability and proton-donor/acceptor parameters. However, these structural descriptors are difficult and tedious to determine experimentally, especially for biological macromolecules. Other models rely on learning structure–retention relationships from a set of standard analytes. By doing so, Gilar () have developed a model for the retention of oligonucleotides in ion-pair reversed-phase chromatography by adding up the retention contributions of the individual nucleotides, which have been determined experimentally from analyzing homo-oligonucleotides. Since the input structural parameters for the model are only oligonucleotide length and base composition, it is only applicable at relatively high temperatures of 60–80° C where secondary structures are usually suppressed because of thermal denaturing. A similar model was used by Otvos () for peptide nucleic acids. The model was based upon retention contribution of monomer building units, which were studied by Meek () and Guo (). We propose a new model for retention time prediction which differs in two ways from the Gilar approach. First, we incorporate several additional features of the oligonucleotides into the model, the most important being predicted secondary structure information. The temperature dependency of the secondary structure is captured by adding information similar to a melting curve. The second difference lies in model creation. Support vector regression (SVR) is used instead of simple linear or logarithmic models. SVR can model non-linear relationships while optimizing both the model performance and the model complexity. These improvements lead to a reliable model with a significantly increased prediction performance. Acetonitrile (far UV HPLC grade) and acetic acid (analytical reagent grade) were obtained from Riedel-de-Haën (Seelze, Germany), triethylamine (analytical reagent grade) and ethylenediamine tetraacetate (EDTA, analytical reagent grade) were purchased from Fluka (Buchs, Switzerland). Triethylammonium acetate was prepared by mixing equal amounts of triethylamine and acetic acid. Water was purified by means of a purification system (Purelab Ultra Genetic, Elga, Siershahn, Germany). Oligonucleotides were synthesized by MWG-Biotech AG (Ebersberg, Germany) or Biospring (Frankfurt, Germany). The sequences of the 72 different oligonucleotides that were used to create training and evaluation datasets for the retention model are collected in Table S1 of the Supplementry Data. Ion-pair reversed-phase high-performance liquid chromatography (IP-RP-HPLC) was performed with a fully automated capillary/nano HPLC system (Model Ultimate 3000, Dionex, Amsterdam, The Netherlands) equipped with a low-pressure gradient micropump (Model LPG-3600) with a vacuum degasser (Model SRD-3600), an integrated column oven, a microcolumn switching unit and flow-splitting device (Model FLM-3100), a micro-autosampler (Model WPS-3000) and a UV-detector (Model UVD 3000) with a 3-nL Z-shaped detector cell (Model Ultimate). The system was controlled by a personal computer with Chromeleon Version 6.60 SP2 (Dionex). The 60 × 0.20 mm i.d. poly-(styrene/divinylbenzene) monolithic column was prepared according to the published protocol () and is available from Dionex. One to five nanograms of oligonucleotides, dissolved in 1 μl water, were injected and chromatographed with a 30-min linear gradient from 0–16% acetonitrile in 100 mmol/l triethylammonium acetate, 0.5 mmol/l EDTA, pH 7.0. The flow rate was adjusted to 2 μl/min. The gradient delay volume of the used LC system was 5.94 μl. Although retention times of oligonucleotides were highly reproducible (0.26% relative SD in absolute retention time from three repetitive injections), and () homo-oligonucleotides were coinjected as internal standards to normalize retention. Normalization was performed using Equation () ( and representing the retention time and average retention time, respectively, of the oligonucleotides) A representative separation of an oligonucleotide including the two standards is illustrated in . The 72 different oligonucleotides were chromatographed at column temperatures of 30, 40, 50, 60 and 80° C and the measured normalized retention times are summarized in Table S1 of the Supplementary Data. A number of methods can be used to derive models from training data. In this study, SVR is used for model generation and predicting retention times. Squared Pearson correlation coefficients were used to evaluate the quality of the models (R) and the predictions (Q). SVR is a machine-learning technique that uses a training dataset to derive a model for the prediction of a quantitative response. by minimizing the objective function In this equation, the first addend minimizes the model complexity while the second minimizes the ɛ-insensitive training error — i.e. training errors below a fixed ɛ are not penalized. C is a constant which defines the trade-off between these two objectives. In 2001, Schölkopf () proposed ν-SVR, a modified version of ɛ-SVR which minimizes ɛ along with the model complexity and the training error. This simplifies the use of SVR, as ɛ no longer has to be chosen . SVR, as so far described here, cannot model non-linear relationships between input features and the quantitative response. Non-linearity is achieved by transforming the input features into a higher-dimensional space using so-called kernel functions. A linear model in the transformed feature space corresponds to a non-linear model in the original feature space. Details about kernel functions can be found in (). In this study, the libSVM implementation () of ν-SVR was used with radial basis function kernel. The kernel parameter gamma and the trade-off C have been optimized using grid search and 3-fold cross-validation. Our dataset consists of retention times measured for 72 oligonucleotides ranging from 15 to 48 bases. As one focus of this study is the influence of secondary structure on the retention time, the dataset consists of oligonucleotides that contain little to no secondary structure and others where nearly all bases form a hairpin. Four sequences that form stable hairpin structures even at elevated temperatures were included. shows the average fraction of paired bases plotted against the measurement temperature. xref #text The selection of input features for performing SVR is a critical step, as the features determine the performance of the prediction. Leaving out essential features or adding unnecessary features both lead to a drop in prediction accuracy. We thus tested several different models, i.e. input feature sets. Each model consists of several model components which group closely related features together. Besides these model components, each model implicitly contains the temperature and the length of the oligonucleotide. The model components can be grouped into composition components, structural components and energy components. contains an overview of all components. Sequence components are calculated from the oligonucleotide sequence and describe the base composition or sequence details of the oligonucleotide. COUNT reflects the base composition, while CONTACT and SCONTACT contain dinucleotide frequencies, which contain information on the adjacency of bases. The key idea here is that stacking bases will influence the secondary structure and the interaction with the stationary phase. Since retention time behavior strongly depends on the secondary structure, sequence-based features alone are suitable only for high temperatures. At lower temperatures, the secondary structure requires the inclusion of structure-based components in the prediction. These components can be subdivided into two groups. The first group considers only the secondary structure of the temperature at which the measurement was performed, whereas the second group contains information on secondary structures for temperatures of 30, 40, 50, 60, 70 and 80° C. shows the predicted secondary structure of the oligonucleotide GTGCTCAGTGTAGCCCAGGATGGG at 40° C. Examples of features calculated from the predicted structure are listed in . For the simple structural components, only the measurement temperature is considered. For the multi-temperature structural components, the secondary structure and the resulting features are calculated for different temperatures. The single temperature components PAIRED and UNPAIRED have four features each reflecting the fraction of A, T, C and G inside stems and outside stems, respectively. The STRUCTURE component contains more detailed information, as it considers the fraction of A, T, C and G inside stems and inside loops and unpaired nucleotides. The multi-temperature structural components reflect secondary structure information as well. In contrast to the single temperature components, they do not only consider the measurement temperature but a range of temperatures (30, 40, 50, 60, 70 and 80° C). Adding this gradient of secondary structure information for different temperatures provides information similar to a melting curve. The simplest component in this group is MULTISTRUCT, which, for each of the six temperatures, contains the fraction of all bases that are inside stems. MULTITWO contains the fraction of bases in stems and loops as well as the fraction of unpaired bases. The most detailed representation of the secondary structure is MULTIDETAIL. It contains the fraction of bases in stems, the fraction of bases in loops, and the fractions of A, C, G and T that are unpaired. Energy components describe the ring stacking effects between adjacent bases. These energies might influence retention time as they have to be overcome in order for the aromatic ring system of the base to interact with the column packing. The EMBOSS suite () was the source of the stacking energies of base pairs. The thermodynamic parameters used for the STACKING component were taken from (). The standard approach in predicting retention time is to create temperature-specific models, i.e. each model is based on data for a single temperature. Thus, if predictions for several temperatures are needed, one model has to be created for each temperature. The model components that consider multiple temperatures allow for a more general approach. The data points of one dataset may differ in temperature, which leads to a model that can predict retention times over the whole temperature range of the training data. The advantage of this approach is that all available data can be integrated into a single, more general model. This reduces the amount of data needed for the individual temperatures. Combining datasets of all temperatures leads to a dataset with 432 data points (72 data points from six temperatures). To avoid overfitting, we included data from only one randomly chosen temperature for each oligonucleotide. The resulting model was then used to predict the retention times of the remaining measurements. From the 11 model components presented in , we created models containing one or more of these components. As the number of models that can be built out of 11 components is huge, we focused on models built out of two or three components, which mostly contain one composition component and one structural component. These models will subsequently be referred to by the names of the components they contain (e.g. ‘count_sesum’ for a model that consists of the COUNT component and the SESUM component). For each model, the prediction correlation Q in 3-fold cross-validation was determined for all temperatures. shows the prediction performance for a selection of models: In 2002, Gilar () proposed a simple mathematical model for predicting the retention time of oligonucleotides. The model takes only the overall length and the base composition of oligonucleotides into account. As Equation () = + + + ). There are two coefficients and for each base type which have to be determined experimentally. Therefore, the retention times of homo-oligonucleotides of different lengths were measured for each base type. The individual contributions of each base can then be fitted to the measured times in order to determine and . We compared the Gilar model to our model ‘count_multistruct_stacking’. The determination of the Gilar model coefficients for our dataset will be described elsewhere (manuscript in preparation). The Gilar model fits very well at 80° C but drops to R = 0.58 at 30° C (). In contrast, the ‘count_multistruct_stacking’ model maintains a performance of R ≥ 0.95 for all temperatures. shows the retention times predicted by the Gilar model at 30 and 80° C. At 80° C, the prediction performance is equal to the performance of our models. However, at 30° , the hairpin structures (marked with circles) cannot be predicted correctly any more and generally, prediction errors increase significantly (). In order to demonstrate the effect of adding secondary structure information to a sequence-based model, we compared the model error of the models ‘count’ and ‘count_multistruct_stacking’. shows the average relative model error of oligonucleotides with a similar fraction of bases involved in their secondary structure. The ‘count_multistruct_stacking’ model can compensate for the influence of the secondary structure. However, the relative model error of the ‘count’ model rises up to nearly 38 and 23% for several oligonucleotides at 30 and 50° C, respectively. It is simply impossible for the ‘count’ model to derive a reasonable model from the training data at lower temperatures as, it does not incorporate secondary structure information. At 80° C, the ‘count_multistruct_stacking’ model still has lower relative model errors than the ‘count’ model. For practical use, the number of training data points that are required for a reasonable model is very important. So we determined an average prediction performance for the range of 10–60 training data points. We used the following procedure: shows the results for the ‘count_multistruct_stacking’ model. From the figure, one can see that even for a temperature of 30° C, 40 training sequences (or more) are sufficient to construct a model describing oligonucleotide retention with acceptable accuracy. No significant improvement is observed for more than 50 data points. The methods, experiments and results presented here clearly demonstrate that predicting oligonucleotide retention time can be significantly improved using secondary structure information and SVR. The use of secondary structure information improves the prediction performance, especially at low temperatures and for oligonucleotides that form highly stable secondary structures. The second point of interest in this study was the effect that the base sequence had on the retention time. The fact that our best model contains no explicit sequence information shows that it plays only a subordinate role in this process. Rather, it is the influence of the base sequence on the secondary structure that seems to be the base sequences' main contribution. Explicitly modeling the base sequence is therefore not necessary. However, a closer examination of the influence of the base sequence remains to be done. The second performance boost came from the use of SVR, probably because of its ability to model non-linear relations and because it handles outliers better. The essential step, and limiting factor, when working with a SVR model is the selection of the training data. A good prediction performance for oligonucleotides that lie far outside the training feature space is very unlikely. Thus, it is crucial to ensure a broad coverage of feature space, both in terms of secondary structure and base composition. We tested 11 model components that cover base composition, base sequence, secondary structure and base stacking. Although our model performed very well, there is still room for improvements. New features that model additional chemical properties can be easily added to our model by appending them to the feature vector. We expect that this kind of model extensibility can further help to improve the oligonucleotide retention time prediction model in the future. Supplementary Data are available at NAR Online. . None declared.

Ribosomes are universally conserved ribonucleoproteins that translate genetic information contained in mRNAs into proteins. In conjunction with ribosomal proteins (r-proteins), rRNAs form the basic structure of the ribosome, and have a crucial role in the fundamental process of protein biosynthesis. Recent structural studies of ribosomal subunits, and the 70S ribosomes, revealed that the functional cores for decoding and peptide-bond formation consist entirely of rRNAs (). The architecture of rRNA diversifies among organisms. In mammalian mitochondria, the lengths of the rRNAs are approximately half that of prokaryotic rRNAs (,). Large segments of missing rRNA are replaced by enlarged r-proteins and other mitochondria-specific proteins (). In contrast, eukaryotic ribosomes contain elongated rRNAs, with several insertion sequences, and increased numbers of r-proteins (). While the architecture of rRNAs might have co-evolved with r-proteins, the fundamental structure and function of ribosomes is preserved in all domains of life. Variation in the RNA-to-protein ratio in ribosomes from various organisms indicates that some structural/functional flexibility between RNA and protein is permissible in the architecture of the ribonucleoproteins. However, the functional domains in rRNAs and several r-proteins are extensively conserved in all living organisms and organella. Structural analysis of the ribosomes revealed that most of the r-proteins consist of globular domains with unstructured extensions that weave into the interior of the rRNA structure, such that they fill the gaps among the rRNA helices, and play important roles in maintaining the ribosome as a functional molecular machine (). The roles of r-proteins in the assembly of the ribosomal subunits have been extensively studied using bacterial ribosomes (,). The r-proteins, such as L2, L3 and L24, are primary RNA-binding proteins that directly recognize naked 23S rRNAs (,). These proteins are thought to bind to their binding sites during transcription of rRNAs, and play important roles in ribosome assembly. The rRNA sequences that mediate the interaction with essential r-proteins may have evolved in different organisms without the loss of binding capability. L2 is a highly conserved r-protein, and the largest in the 50S subunit. Because of its functional importance and degree of conservation, L2 is considered to be one of the most evolutionarily ancient r-proteins (). Structural analysis of the RNA-binding domain (RBD) of L2 revealed that it has two motifs that are homologous to the oligonucleotide–oligosaccharide binding (OB) fold, and the Src homology 3 (SH3)-like barrel, structures that are often found in RNA- or DNA-binding proteins (). The primary L2-binding site has been biochemically characterized by footprinting and cross-linking experiments (,), and maps to a highly conserved stem-loop structure of H66 [nucleotide positions (np.) 1792–1827] in domain IV of 23S rRNA. In crystal structures of ribosomes, the globular RBD of L2 binds an internal bulge structure (np. 1799–1800 and 1817–1820) of H66, and is located in the intersubunit space of the 50S subunit (,). The solvent-exposed surface of the RBD makes molecular contacts with helices 23 and 24 in the 16S rRNA to form bridge B7b (). The amino (N)- and carboxy (C)-terminal extensions of L2 are proximal to the peptidyl-transferase center in the 23S rRNA (,,). Although it is established that the peptidyl-transferase center consists of rRNA, L2 together with L3 are required for efficient peptide-bond formation (). The multi-functional nature of L2 was revealed in experiments by Diedrich . (), using reconstituted 50S subunits lacking L2, in which it was demonstrated that L2 is involved in the association of the ribosomal subunits, tRNA binding to the A- and P-sites, and peptide-bond formation. An replacement study showed that eukaryotic L2 (L8e), or archaeal L2 counterparts, can replace L2 and form hybrid ribosomes, and that the hybrid ribosomes are translationally active and incorporated into polysomes (). According to phylogenetic analysis of rRNAs from all living organisms, there are two distinct classes of L2-binding sites in H66 (see A and B) ()(). In the crystal structures of bacterial ribosomes, a characteristic base-triple composed of C1800, G1817 and A1819 is visible in the L2-binding site of H66 (see C) (,). In addition, a G1799·U1818 pair also stacks to this base-triple, and participates in a hydrogen (H)-bond network to form a core structure in this region. This characteristic core structure of the L2-binding site appears to induce a kinked conformation of H66 (see A). This type of base-triple, designated as class I binding site, is conserved in bacteria, archea, mitochondria and chloroplasts. In eukaryotic rRNAs, the corresponding bases are replaced by A1800, U1817 and A1819 ( numbering). Assuming that eukaryotic L2/L8e-binding sites also have a similar structure as the class I binding site, these three bases are likely to form a characteristic base-triple (B), which is designated as class II binding site (see B). The class II binding site is highly conserved in eukaryotic rRNAs, while all bacteria-type ribosomes have the class I binding site. Since L8e can replace bacterial L2 in cell, these distinct classes of sequences appear to provide functionally similar binding sites for L2/L8e proteins. To investigate the functional relationship between the two classes of L2/L8e-binding sites in H66, we employed a genetic system which we have termed systematic selection of functional sequences by enforced replacement (SSER) (). This method allowed us to identify residues absolutely essential for ribosome function in cells from a randomized rRNA library. Previously, we used this approach to analyze the peptidyl-transferase center () and the conserved loop sequence of H69 (). For the current analysis, a library was constructed by complete randomization of the internal bulge sequence in H66 (B), and then subjected to SSER. The selected variants contained naturally occurring rRNA sequences from other organisms, as well as unnatural but nonetheless functional sequences. The results of this study revealed the architecture of the L2-binding site, and provided insight into the evolution of the L2–H66 interaction. In addition, genetic analysis of a ribosome variant indicated that functional interactions between intersubunit bridges could compensate for defective ribosomal function. Δ7p strain TA542 (Δ Δ Δ Δ Δ:: Δ:: Δ:: Δ/pTRNA66 pHKrrnC) () was kindly provided by Dr Catharine L. Squires (Tufts University). The rescue plasmid pRB101 was constructed by introducing the gene and the operon into pMW118 (Amp, pSC101 ori) (Nippon gene). The plasmid pRB102 was constructed from pMW218 (Km, pSC101 ori) (Nippon gene) by insertion of the operon. The plasmid pHKrrnC in strain TA542 was replaced by pRB101 to generate strain NT101, which was used as the host cell for SSER. Cells were grown at 37°C in 2× Luria–Bertani medium (2× LB, 2% tryptone, 1% yeast extract and 1% NaCl); for solid medium, 1.5% agar was added. Antibiotics were added at the following concentrations, when required: 40 μg/ml spectinomycin (Spc), 100 μg/ml ampicillin (Amp) and 50 μg/ml kanamycin (Km). To induce plasmid replacement in strain NT101, 5% sucrose was added to the LB medium. Doubling time of NT102 variants were determined by measuring the optical density at 600 nm every 15 min using a plate reader (Molecular Devices, Inc.) The template plasmid pRB102 was hyper-methylated by the DNA methyltransferases M- I, M- III and M- II (NEB), as previously described (), and then subjected to Quik-Change site-directed mutagenesis (Stratagene), according to manufacturer's instructions, using the following sets of primers: 5′-cacagcactgtgcaaacacgagtggacgtatacggtgtg-3′ (forward) and 5′-cacaccgtatacgtccactcgtgtttgcacagtgctgtg-3′ (reverse) for H66d2; 5′-aacacagcactgtgcaaacacgtggacgtatacggtgtgacgcctgc-3′ (forward) and 5′-cgtcacaccgtatacgtccacgtgtttgcacagtgctgtgtttttaataaac-3′ (reverse) for H66d4; 5′-aaaacacagcactgtgcaaaccgaaagggacgtatacggtgtgacgcctgc-3′ (forward) and 5′-gcgtcacaccgtatacgtccctttcggtttgcacagtgctgtgtttttaataaac-3′ (reverse) for H66d1bp; and 5′-gtttattaaaaacacagcactgtgcaaacgaaaggacgtatacggtgtgacgcc-3′ (forward) and 5′-ggcgtcacaccgtatacgtcctttcgtttgcacagtgctgtgtttttaataaac-3′ (reverse) for H66d2bp. PCR was carried out, and then the reaction mixture was treated with I (New England Biolabs) to digest the template plasmid, and subjected to purification using QIA-Quick column (Qiagen). NT101 was then transformed with the indicated plasmids. Two plasmids with same origin (pRB101 and pRB102 derivative) transiently coexisted in NT101 cells. To eliminate pRB101, transformed colonies were spotted onto LB-plates containing spectinomycin, kanamycin and sucrose to force plasmid replacement. The B gene of pRB101 is a counter-selectable marker, which is lethal when expressed in the presence of sucrose; thus, cells carrying only pRB102 (NT102 derivatives) are obtained. The pRB102 variants were confirmed by sequencing. A non-functional variant of pRB102 (pRB102-H66d4) amplified in DH5α was used as a template for generation of the randomized library. pRB102-H66d4 was hyper-methylated by three DNA methyltransferases, as described above. In the first round of PCR, a set of primers [5′-gtggacgtatacggtgtg-3′ (gap-F) and 5′-gtgtttgcacagtgctgtg (gap-R)] complementary to H66 was employed to generate a gapped template. This process was required to enhance the efficiency of PCR randomization, and to reduce the background for SSER. Gapped pRB102-H66d4 was then gel-purified and used as the template for the randomized library. To construct the N10 library, a set of primers, N10-F (5′-aacacgaaag tggaNNNNNN cggtgtgacg cctgcccggt gccggaaggt taattg-3′) and N10-R (5′-tccactttcg tgttNNNNca gtgctgtgtt tttaataaac agttgcag-3′), was employed to randomize 10 bases in the internal bulge region of H66. For the N6 library, a set of primers, N6-F (5′-aacacgaaag tggacNNNNa cggtgtgacg cctgcccggt gccggaaggt taattg-3′) and N6-R (5′-tccactttcg tgtttNNaca gtgctgtgtt tttaataaac agttgcag-3′), was used. The conditions for large-scale PCR to construct the randomized libraries was described previously (,). After PCR, the reaction mixture was treated with I (New England Biolabs) and λ exonuclease (New England Biolabs) to digest the template plasmid, then subjected to purification using a QIA-Quick column (Qiagen). The products were then verified by agarose gel electrophoresis. Unbiased distribution of four bases at each position of the randomized region was confirmed by direct sequencing of the plasmid library prior to the selection. Details of SSER have been described previously (,). NT101 cells were transformed using the randomized library and plated on LB-plates containing Spc/Km. Transformants were picked and suspended in LB broth, then spotted onto LB-plates containing Spc/Km/sucrose to force plasmid replacement. For variants showing a slow growth phenotype, the cell suspension was first spotted and cultivated on LB-plates containing Spc/Km, before transfer to LB-sucrose plates. Sucrose-resistant colonies (NT102 variants) were then cultured in 2× LB broth in preparation for plasmid purification and sequencing. To verify plasmid replacement, each transformant was spotted onto two LB-plates containing Spc/Amp and Spc/Km/sucrose. No growth of the spotted cells on the Amp-plate confirmed complete plasmid replacement. Functional variants were sequenced using an ABI Prism 3100 Genetic Analyzer (Applied biosystems). To determine the doubling times of NT102 variants, 2 μl of an overnight pre-culture was inoculated into 1 ml of 2× LB medium, then separated into 5 aliquots and plated in a flat-bottomed 96-well microplate (IWAKI). The microplate was incubated at 37°C with vigorous agitation in a Spectramax 190 plate reader (Molecular Device, inc.), and absorbance at 600 nm (OD) was monitored every 15 min. This analysis was based on the procedure described previously (,). Briefly, NT102 cells were grown in 50 ml of 2× LB broth containing Spc, Km and sucrose, in a 500 ml flask with vigorous shaking at 37°C. Cells were harvested by centrifugation when the OD reached 0.4–0.5. The cell pellet was resuspended in 1 ml of cold RBS buffer (20 mM HEPES–KOH pH 7.6, 6 mM Mg(OAc), 30 mM NHCl, 6 mM 2-mercaptoethenol). Cell lysates were prepared by the lysozyme–freeze–thaw method, as described previously (), and cleared by centrifugation for 15 min at 15 000 r.p.m. at 4°C. The concentration of total RNA in the lysate was estimated by measuring OD. Fifteen OD units of the lysate was layered on the top of a sucrose gradient (10–40%) prepared in RBS buffer, then separated by ultracentrifugation in a Beckman SW-28 rotor at 20 000 r.p.m. for 14 h at 4°C. Ribosomal subunits in the gradient were fractionated by Piston Gradient Fractionator (BIOCOMP) and monitored at 260 nm using a UV monitor (ATTO AC-5200). The ‘assembly ratio’ of 50S subunit is defined by calculating molar ratio of total 23S rRNA (and 5S rRNA) against total 16S rRNA in the SDG profiles. The ‘association ratio’ of 50S to 30S is defined by calculating the portion (%) of 23S rRNA (and 5S rRNA) which is incorporated into the 70S peak. Glycerol stock of variant N-16 was pre-cultured in a test tube containing 3 ml of 2× LB medium until full growth (defined as the 0 generation), and then 0.1% (3 μl) of the culture was transferred into three different aliquots of 3 ml of new medium to create three different lineages. After overnight cultivation, each lineage was transferred to a new test tube. Serial passages were repeated for 16 days. After cultivation of 80 and 160 generations (one passage corresponds to approximately 10 generations), doubling time for each lineage was determined. pRB102 derivatives were extracted from each lineage of 80 and 160 generations, and the entire 23S rRNA and most of the 16S rRNA genes were sequenced to identify secondary mutations. To confirm whether the secondary mutations in each lineage functioned as compensatory suppressors for the deleterious effect of the N-16 mutation, each of lineage-specific mutation was independently introduced into pRB102 of N-16 (0 generation), using the following sets of primers: N-16 + C1790U-F (5′-ctgtttattaaaaacatagcactgtgaaaacac-3′) and N-16 + C1790U-R (5′-gtgttttcacagtgctatgtttttaataaacag-3′) for the C1790U mutation in N-16; C1790U-F (5′-ttaaaaacatagcactgtgcaaacacgaaagtgg-3′) and C1790U-R (5′-gcacagtgctatgtttttaataaacagttgcag-3′) for the C1790U mutation in wild-type pRB102; and G1989A-F (5′-atggccagactgtctccacccgagactcagtg-3′) and G1989A-R (5′-ggtggagacagtctggccatcattacgccattcg-3′) for the C1790U mutation in both N6-16 and wild-type pRB102. According to a phylogenetic analysis of rRNAs from various organisms and organelles () (), the internal bulge region of H66 (np. 1798–1800 and 1817–1820) is highly conserved. However, the top half of H66 (1801–1816) is not conserved, even in bacteria. In mitochondrial rRNAs, for example, the top half of H66 is nearly truncated. To define the functional L2-binding site, we first examined the effect of deleting bases in the top half of H66. We used the NT101 strain, a derivative of the Δ7 strain developed in the Squire's laboratory to test the function of rRNAs with deletions. NT101 contains the rescue plasmid pRB101 (Amp) carrying both the operon and the gene, which produces toxic products in the presence of sucrose. We introduced a second plasmid pRB102 (Km) carrying the with deletion in H66. A two-bases deletion (H66-d2) in the H66 loop was introduced into pRB102 (B), which then transformed strain NT101, and a functional transformant (H66-d2) was obtained following selection on sucrose plates. Similarly, a one-basepair deletion (ΔA1805·U1812) also yielded a functional variant (H66-d1bp). However, we failed to obtain variants carrying a four-bases deletion (ΔA1808-G1811) in the loop, or a two-basepairs deletion (ΔA1805·U1812/ΔC1804·G1813), indicating that large deletions in H66 are not tolerated. Thus, we did not test additional variants using this approach. The growth rate of the functional variants (H66-d2 and H66-d1bp) was measured as an estimate of ribosomal activity (), and we found that small deletions in the top half of H66 resulted in a slight reduction of growth rate. To investigate the functional importance of the highly conserved bulge region of H66 in the 23S rRNA, we employed a novel genetic method developed in our lab, SSER (). This system allows the rapid identification of functional sequences from a library of randomized sequences. Based on crystal structure of the 70S ribosome (), ten conserved nucleotides (1798–1801 and 1816–1821) in the bulge region of H66 were chosen for randomization, as they represented the potential binding site for L2 (B). The ten bases in H66 were completely randomized using a PCR-based method, generating a pRB102 library with 1 048 576 sequence variations (N library) (B). We employed a non-functional plasmid variant, H66-d4, amplified in DH5α as the template for PCR-randomization, rather than wild-type pRB102, to exclude the possibility of wild-type sequences being generated and isolated during the selection process. We then carried out a large-scale transformation of strain NT101 with the randomized library, and transformants were selected based on kanamycin resistance. If the sequence of the incorporated plasmid was toxic, showing a dominant lethal phenotype, the transformant was eliminated in this step. As L2 is an essential primary rRNA-binding protein, it is likely that most of the sequences in the library were excluded during this step. The kanamycin-resistant transformants showed different colony sizes (data not shown), indicating that the cells contained a sequence that resulted in altered ribosomal activity. At this stage, two incompatible plasmids, pRB101 (Amp) and pRB102 (Km), transiently coexisted in the transformants. To drive plasmid replacement, we took advantage of the counter selection system in which gene on pRB101 produces toxic products in the presence of sucrose. Each cell was picked and spotted onto selection plates containing kanamycin and sucrose. If the incorporated pRB102 plasmid had a functional sequence for ribosomal activity, it rapidly eliminated the pRB101 rescue plasmid, thus yielding sucrose-resistant cells (NT102 derivatives). Besides, if the incorporated plasmid had a non-functional or very weak functional sequence for ribosomal activity, the rescue plasmid (pRB101) could not be replaced by the introduced plasmid and the transformant became sensitive to sucrose due to the gene. Most likely due to the vast number of sequence variations in the N library and the functional importance in the randomized region, we obtained only 15 sucrose-resistant clones, from approximately 22 500 kanamycin-resistant colonies. Complete plasmid replacement of the selected variants was verified by their inability to grow in the presence of ampicillin. Functional plasmids were sequenced, resulting in the identification of 14 viable sequences from the N library (). No wild-type sequences were isolated. The variants had one-, two- or three-nucleotide changes in the randomized region, as compared to the wild-type sequence. While the sequence data provided only limited information, the results indicated that the peripheral four nucleotides (U1798, A1801, C1816 and A1821) are variable, and the internal six nucleotides (1799–1800 and 1817–1820) are less variable. To more precisely define the molecular features of the functional L2-binding site, we focused on the internal six nucleotides for another round of SSER. The six internal nucleotides (1799–1800 and 1817–1820) of H66 were completely randomized to construct a pRB102 library with 4096 sequence variations (N library) (B). In this round of selection, in order to rescue less than fully functional sequences, primary kanamycin-resistant transformants were cultivated for an additional period of time before counter selection on sucrose plates. We obtained 25 sucrose-resistant colonies, and sequence analysis of the randomized region revealed 21 distinct functional clones, including one with the wild-type sequence (N-0) (). Among them, two variants (N-1 and N-10) were present twice, and one variant (N-2) was present three times. The selected variants had changes of one to four nucleotides compared to the wild-type sequence. These results demonstrated that SSER of the N library successfully identified sufficient number of functional sequences. It is noteworthy that replacement of C1800 by A1800 was functional, in light of the fact that C1800 is completely conserved in bacteria, archaea and chloroplasts. G1799 and U1820 were preferentially selected. Although most of the selected clones contained non-naturally occurring sequences, N-0 is a conserved sequence found in bacteria and archaea, and N-1 (G1799A) is a natural sequence found in some archaea. Intriguingly, N-16 is also a natural sequence found in eukaryotic rRNA () (). According to their predicted secondary structures, 20 of the selected variants could be clearly divided into two distinct classes, based on the characteristic base-triple in the L2-binding site of H66 (A and B). The two classes were characterized by either a C1800- or an A1800-centered base-triple. Eleven variants (N-1 to N-11) carrying C1800 were categorized as class I, and nine variants (N-12 to N-20) carrying A1800 were categorized as class II. In class I variants, G1817 in the binding site tended to be co-selected with C1800 (9 variants). Similarly, in the class II binding site, U1817 appeared to be co-selected with A1800 (7 variants). Hence, all variants carried at least two of the three bases of the base-triple. These results indicated that co-variation in the selected variants is required for proper organization of the L2-binding site. To examine the growth phenotype of each variant selected from the N library, we generated new constructs of each of the variants, to avoid the generation of secondary mutations at unexpected positions or compensatory mutations during selection and sub-culturing. The growth rate of each variant was measured as an estimate of ribosomal activity (). We observed a weak relationship between the number of mutations and doubling times. Of the class I variants, N-4, carrying a U1817 substitution, showed a reduced growth phenotype. In this variant, U1817 is unable to form a base-triple with C1800 and A1819. N-11, carrying C1818 and A1820 substitutions, also showed a growth defect, whereas N-7, carrying a single A1820 substitution, did not. Thus, the slow growth of N-11 likely originated from the combination of C1818 and A1820. Compared to the class I variants, the average growth rate of the class II variants was decreased. N-16, which contained a naturally occurring eukaryotic L2-binding site, exhibited the most severe phenotype. There were three class II variants, N-13, N-15 and N-16, which had a complete class II base-triple, but different bases at position 1818. N-13, carrying a U at position 1818, showed the highest growth rate of the class II variants, indicating that the growth of these variants was modulated by the specificity of position 1818. The growth defects of each variant indicated that they had decreased functionality of their ribosomes. Since L2 is known to participate in subunit association through bridge B7b, we next examined the efficiency of subunit association in eight of the rRNA variants (N-0, N-4, N-9, N-11, N-13, N-16, N-19 and N-20), using sucrose density gradient (SDG) centrifugation (A). The association ratio [(50S incorporated into 70S)/(total 50S)] and the assembly ratio [(total 23S rRNA)/(total 16S rRNA)] of the 50S subunit for each variant were calculated from the SDG profile (). We observed little influence of sequence variation on the assembly ratio of the 50S subunit (A and ). In wild-type ribosomes, 71.6% of intact 50S subunits were incorporated into tightly coupled 70S ribosomes (TC) in the presence of 6 mM magnesium (Mg). Similar SDG profiles, indicating normal subunit association, were observed in ribosome variants N-9, N-13 and N-19, variants that also showed a healthy growth phenotype. In contrast, the SDG profiles of ribosome variants with slow growth phenotypes revealed apparently lower TC ratios (A). In N-4 and N-20 ribosomes, 45.1 and 47.4% of the 50S subunits were incorporated into TC, respectively. In addition, two variants with lower growth rates, N-11 and N-16, showed a marked reduction in TC formation, with association ratios of 24.6 and 23.1%, respectively. Thus, the slow growth phenotype of H66 variants correlated well with weak subunit association. To identify genetic interactions between the L2-binding site and other sites in the rRNA, we attempted to isolate revertants of the variant with the most severe phenotype, N-16, whose doubling time was 97.7 min (). N-16 cells were divided into three lineages and cultivated for 16 days, with repeated serial passage into new medium. We succeeded in isolating three independent populations in which the growth rate was significantly improved: revertant 1 (doubling time 45.7 min), revertant 2 (doubling time 56.0 min) and revertant 3 (doubling time 50.3 min). The doubling time of revertant 1 was faster than that of wild type (50.4 min). We observed an increased amount of pRB102 plasmid extracted from this population (data not shown), indicating that increased plasmid copy number and general adaptation to the medium during cultivation accounted for the growth phenotype. We sequenced each pRB102 revertant and identified three different mutations that appeared to be responsible for the compensatory effect on the N-16 L2-binding site mutant. In revertant 1, we identified a C1790U mutation. This position is approximately ten bases upstream of the N region. Revertant 2 had a G1818U mutation, which is within the N region, changing the N-16 sequence to that of N-13, which had a normal growth phenotype. Revertant 3 had a G1989A mutation, at the position located approximately 170 bases downstream of the N region. To confirm that the C1790U and G1989A mutations functioned as suppressors for the deleterious effect of the L2-binding site mutation in N-16, each of the compensatory mutations was reconstituted into a derivative of pRB102 obtained from ancestral N-16 cells. NT101 cells were transformed with each of the compensatory mutant plasmids to obtain two NT102 derivatives: N-16 + C1790U, and N-16 + G1989A. As shown in , the growth rates of these strains were significantly increased (52.6 and 58.5 min, respectively). Thus, we confirmed that the C1790U and G1989A mutations suppress the deleterious effect of the L2-binding site mutation of N-16. Furthermore, SDG analyses revealed that both N-16 + C1790U and N-16 + G1989A had significantly improved association ratios, ranging from 23.1 to ∼65% ( and B). These results suggested that both mutations are able to suppress the defect in subunit association caused by mutation of the L2-binding site. In control experiments, to estimate the effect of the C1790U or G1989A mutation in a wild-type background, two NT102 derivatives with a single mutation (either C1790U or G1989A) were constructed. Both strains showed normal growth rates (). The C1790U mutant showed a similar subunit association ratio (64.2%) to N-16 + C1790U, while the G1989A mutant showed a slight reduction in subunit association (60.1%), compared to N-16 + G1989A ( and B). These results suggested that a C1790U or G1989A mutation alone does not exert a positive effect on the basal activity of 23S rRNA. We carried out a functional genetic selection of 23 rRNA L2-binding site variants using an advanced genetic selection system developed in our lab, termed SSER (), allowing the identification of sequences that are essential for ribosome function from a randomized rRNA library in . Using SSER, we isolated functional sequences from independent colonies, eliminating competition against other variants for survival. As long as sufficient numbers of transformants are obtained, this method can be used to test the functionality of all possible sequences, including wild-type sequences, and naturally occurring sequences found in other organisms, as well as unnatural but functional sequences that have not emerged during the process of evolution. Thus, SSER permits completely neutral genetic selection of functional sequences, as it excludes researcher's arbitrariness and bias. From the selected sequences, we can identify sequences that would not be available from a phylogenetic comparison of rRNA sequences. In addition, SSER can identify distinct classes of functional sequences with low homology to each other, as was the case for the L2-binding sites identified in this study, whereas other approaches using conventional site-directed mutagenesis may miss this distinction. Since SSER is based on the replacement of the -operon, the selection of rRNA sequences required for translating the entire proteome of is stringent. Moreover, as limited numbers of transformants are available due to practical experimental designs, the size of the randomized region for SSER was restricted. In previous studies, we successfully selected functional sequences from an N library (4096 variations) of the 2451-region of the peptidyl-transferase center (), and from an N library (16 384 variations) of the conserved loop of H69 (). In the current study, we initially selected functional sequences of a 10-base region in the L2-binding site, using a randomized plasmid library containing 1 048 576 sequence variations (N library). To date, this is the largest rRNA library used for SSER. Although we successfully obtained 14 functional sequences, from approximately 22 500 colonies in the initial screen, this experiment gave us limited information on the L2-binding site, and as such, was not practical or effective. For SSER to be effective, we had to design an appropriately sized library, taking into consideration the functional importance of the target region. From an N library of the internal six nucleotides of the L2-binding site, we successfully isolated 20 functional variations as well as the wild-type sequence. Variants could be divided into 12 class I sequences, and 9 class II sequences, even though this region is a highly conserved site in all living organisms. Since L2 is a primary binding protein in the biogenesis of the 50S subunit (,), we explored whether alterations in the binding site had a deleterious effect on 50S assembly and/or incorporation of L2 into 50S particles. We found that no variants showed decreased assembly ratio as calculated from SDG profiles (). In addition, the stoichiometry of L2 in the isolated ribosomes from the slow growth variants was examined by SDS-PAGE, and we found no evidence of decreased amounts of L2 in any of the variants (data not shown). During SSER, rRNA variants which do not have a strong binding affinity for L2 are eliminated as dominant lethal or non-functional mutants. An intriguing finding of this study is that the sequence of one of the L2-binding site variants corresponded to the eukaryotic class of L2-binding site. Among the 9 class II variants, N-16 was a naturally occurring sequence found in a eukaryotic rRNA () (). This finding demonstrates that L2 can recognize class II binding sites in ribosomes. Cytosine or adenosine at position 1800 was selected as an essential base; it acts as a central base required to form the characteristic base-triple in the L2-binding site. Interestingly, G1817 and U1817 were preferentially selected with C1800 and A1800, respectively, indicating that there is specific co-variation in the bases at positions 1800 and 1817. When C1800 was replaced with A1800 (as in variant N-12), the doubling time of the variant decreased to 63.7 min. In addition, a single G1817U mutation (N-4) also resulted in a reduced growth rate (77.0 min). However, when these mutations were introduced simultaneously (N-13), the growth rate was improved (55.4 min), demonstrating the functional interaction of these two positions, consistent with the idea that the characteristic base-triple is required for proper organization of the L2-binding site. The crystal structures of eukaryotic ribosomes at atomic-scale resolution will help illuminate the structure of the class II base-triple and the mechanism of recognition by L8e. Typically, ribosomal proteins interact indirectly with rRNAs through salt-bridges between positively charged residues on the protein and phosphate oxygen atoms on the rRNA (). According to this mechanism, proteins recognize specific surface features of regular helical segments interrupted by irregularities, such as non-canonical base pairs, base-triples, single bulged nucleotides and small internal loops (,). In the case of the L2–H66 interaction, the globular RBD of L2 mainly recognizes a backbone structure of the internal bulge region in H66. In the crystal structure of the 70S ribosome (), no base-specific interactions between the RBD of L2 and the N region were observed. This binding mechanism is consistent with a common feature of RNA–protein interactions in the ribosome, in which base-specific interactions are less important. Also, the recognition of specific structural features of the RNA by L2 can explain the finding that two distinct classes of L2-binding sites were selected from the randomized library; each one might form a similar tertiary structure that is recognized by L2/L8e proteins. From an evolutionary viewpoint, it is plausible that rRNAs originally provided two classes of aptamers for the globular RBD of L2 proteins. When the ancestral species of eukaryotes arose from prokaryotes, it is possible that the class II binding site was selected incidentally, rather than co-evolving with the L2/L8e protein. SDG analysis revealed a good correlation between the growth phenotype and subunit association of each variant. As the globular domain of L2 is located at the intersubunit surface, and forms bridge B7b with helices 23 and 24 of 16S rRNA, it is likely that functional variations in the L2-binding site affect the configuration of L2, leading to weakening of bridge B7b. The N-terminal extension of L2 directly interacts with the base region of H34 in domain II of the 23S rRNA, and the tip of H34 directly interacts with S15 in the 30S subunit to form bridge B4. It is known that bridge B4 is an important functional site for subunit association (,). Previously, we showed that even a short deletion in H34 has a great effect on subunit association (). The N-terminal extension of L2 could be affected by alterations in the L2-binding site, which modulates the relative orientation of the H34 tip and S15. In support of this, deleterious ribosome variants with altered L2-binding sites showed defective subunit association. We obtained two revertants of variant N-16, in which the weak subunit association was recovered. Each revertant contained a point mutation in domain IV of the 23S rRNA. Revertant 1 had a C1790U mutation in the base region of H66 (A). In the crystal structure of the 70S ribosome (A and B), C1790 forms a hydrogen bond with C1774, which base-pairs with G729 of H34 in domain II of the 23S rRNA. Mutation of C1790U would affect the relative configuration of H34 and H66, through the U1790–C1774 interaction. As a single mutation C1790U in the background of wild-type rRNA had little effect on the ribosome, C1790U must have a specific compensatory effect on the defective subunit association of the N-16 variant. Revertant 3 had a G1989A mutation, which is located near H64 (A). Since G1989 directly interacts with positions 1474–1476 in helix 44 of the 16S rRNA to form bridge B5 (), it is possible that the G1989A mutation modulates bridge B5, thus compensating for defective subunit association. As a single mutation in the background of wild-type rRNA, G1989A caused a slight reduction of subunit association, indicating that the G1989A mutation in N-16 had a modulatory function, perhaps in fine-tuning intersubunit association. Together, these results suggest that a variety of mechanisms of compensation of defective subunit association are possible by modulation of intersubunit bridges. The reduction in growth rate caused by antibiotic resistance can be compensated by various mutations, most of which reside in r-proteins. For example, a mutation in the S12 gene, which confers resistance to streptomycin and causes growth reduction, is compensated by second site mutations in various r-proteins, such as S4, S5, S12 and L19 (). Although these are many examples of protein–protein compensation, few examples of RNA–RNA compensation have been reported. The lack of data may be due to the lack of a suitable experimental system for this type of analysis. As described in this study, our selection system can readily provide vast numbers of functional rRNA mutants with severe growth phenotypes. In addition, we were able to successfully identify compensatory RNA mutations from revertants of an original rRNA variant. These results, and this type of analysis in general, will help elucidate the many genetic interactions between functional sites in rRNAs, unveiling the ribosomal dynamics at each step of translation.

Forward genetics, where the observation of a phenotype is followed by the identification of the responsible gene(s), provides a valuable tool to carry out functional genomics. The availability of systems to generate tagged mutations on a large scale facilitates this kind of approach, permitting the production of collections of genetically modified cells or organisms which can then be phenotypically screened (). The molecular label conferred by the tagging mutagen allows for a ready identification of the locus responsible for the observed phenotype. Due to their ability to mobilize around genomes, DNA transposons and retrotransposons have been widely used as tools to generate mutation libraries in a variety of organisms. DNA transposons are genetic elements consisting of inverted terminal DNA repeats (TRs) which in their naturally occurring configuration flank a transposase coding sequence (CDS). This transposase follows a ‘cut and paste’ mechanism to excise the transposon from its original genomic location and insert it into a new locus (). Retrotransposons, however, use an RNA intermediary molecule and retrotranscription to ‘copy and paste’ themselves in different locations of the genome. In mammals, the lack of efficient transposon systems has largely precluded the application of this type of mutagenesis. However, the awakening of the use of DNA transposons and the recent success with synthetic retrotransposons in the mouse is opening wide the door of forward genetics in this model organism (,). In 1997, Ivics . resurrected a Tc1-like DNA transposon by comparing the nucleotide sequences of a number of dormant inactive elements in salmonid fishes, predicting the active sequence and repairing the inactivating mutations. This elegant and meticulous approach resulted in an active transposon termed Sleeping Beauty (SB). The SB system was soon reported to be functional in human and murine cells (,,). Moreover, since its resurrection, both the inverted repeats and the transposase coding sequence have been optimized, which has yielded an SB element with increased mobilization activity (). This has been used in mammals for a wide range of applications, such as gene therapy, germline mutagenesis and somatic mutagenesis (,). The activity of the system needs to be regulated, because excessive and uncontrolled transposition results in genomic instability such as inversions, deletions and translocations (). These large-scale genomic rearrangements mask the more subtle and informative singular transposition events, interfering with the identification of single genes responsible for specific phenotypes. Thus, the availability of a highly active, though regulatable, transposase would be desirable. In addition to SB, (PB), a transposable element originally isolated from the genome of the cabbage looper moth has also been reported to be highly active when introduced into mammalian genomes, including human () and mouse cells (). In a direct comparison experiment involving transfection, PB has been reported to be the most active of four tested transposons, including PB, SB11 (one of the hyperactive SB transposases), Tol2 and Mos1 (). In a separate study, PB has also been shown to be more active than SB12, another hyperactive version of SB (). Although there are other hyperactive SB transposases available, which apparently show higher activity than SB11 and SB12, they have not been directly compared to PB (,). Given its good performance in mammals we have characterized and optimized the PB system for its utilization in mouse ES cells and for somatic mutagenesis . We have generated a mouse codon-optimized version of the PB transposase CDS and we have observed that it provides levels of transposition considerably higher than those of the native PB CDS. We have analysed the differential properties of the 5′ and 3′-terminal repeats of the PB transposon, uncovering the existence of promoter activity in the 5′-terminal repeat. Finally, we have created a highly active and regulatable PB transposase, by fusing the optimized PB CDS to the modified ligand-binding domain of the oestrogen receptor, ERT2. We expect that this optimized transposon system will significantly expand the utility of transposon-based mutagenesis for the genetic characterization of the mouse genome. iPB and mPB were custom synthesized. Besides the PB transposase CDS, both sequences included a Kozak element around the start ATG at the 5′-end and two consecutive stop codons at the 3′-end. EcoRI and NotI flank the 5′ and 3′-ends, respectively, and they were used to clone both CDSs into pcDNA3 or pcDNA3-KzHA. To produce the donor plasmids, the fusion gene and the bovine growth hormone polyA signal were PCR amplified from pFlexible (). En2SA was PCR amplified from T2/Onc, kindly provided by Dr L. Collier, and the EMCV IRES sequence was sub-cloned from pPRIG, which was a gift by Dr P. Martin (,). The minimal 5′ and 3′-PB terminal repeats were custom synthesized and then combined by sub-cloning to produce the desired configurations. The sequence corresponds to a HindIII/BamHI fragment from pGL4.13 (Promega). ERT2 was PCR amplified from a cDNA kindly provided by Dr P. Chambon and subcloned at the 5′ or at the 3′ of mPB. L1 linkers were created as a result of the cloning process. L2 and L3 linkers were introduced between mPB and ERT2 by sub-cloning the pairs of primers shown in Supplementary . The core aminoacidic sequence of the L3 linker was kindly provided by Dr Joseph Kaminski. The CDSs and protein sequences of mPB (GenBank accession number: EF587698) and mPB-L3-ERT2 (GenBank accession number: EF587699) are shown in Supplementary . AB2.2 ES cells were cultured on a layer of mitotically inactive SNL76/7 feeder cells and transfected by electroporation as described previously (). Puromycin selection was conducted on SNLP 76/7-4 feeders (a puromycin resistant derivative of SNL76/7) with 3 μg/ml of puromycin. Puromycin resistant ES cell colonies were stained for 15 min with 1% methylene blue in 70% EtOH, washed in distilled water over night and air-dried. This protocol produced a very low background which allowed counting colonies with diameters ⩾0.3 mm. COS-7 cells and HeLaS3 cells were grown in 6-well plates and with DMEM supplemented with 10% foetal bovine serum. Plasmids were transfected with Lipofectamine 2000 (Invitrogen), following manufacturer's instructions. Transfected COS-7 cells were washed with PBS and each well was harvested with 500 μl 2 × Laemmli buffer. Fifteen microlitres of lysate were loaded per lane in a 4–12% gradient SDS–PAGE gel (Invitrogen). The gel was transferred to a Hybond-ECL™ (Amersham Biosciences) membrane, and the membrane was dried over night and blocked with 5% non-fat dry milk dissolved in PT (0.1% Tween-20 in 1× PBS) for 1 h at room temperature. Then, it was incubated with 12CA5 anti-HA monoclonal antibody (Abcam, cat #ab16918) or with anti-actin monoclonal antibody (Sigma, cat #A5316), both diluted 1:1000 in 3% non-fat milk dissolved in PT for 1 h at room temperature. The membrane was washed 3 times for 5 min with PT and incubated with the peroxidase-labelled anti-mouse antibody from the ECL™ Western Blot Analysis System (GE Healthcare) diluted 1/10000 in 1.5% non-fat dry milk for 1 h at room temperature. The membrane was washed again as before and specific protein bands were detected using ECL™ Western Blot Analysis System (GE Healthcare). Transfected HeLaS3 cells were washed with PBS and each well was lysed for 15 min with 100 μl of 1× passive lysis buffer from the Dual Luciferase Reporter Assay System (Promega). Lysates were harvested and 20 μl of each sample were analysed for luciferase activity following the manufacture's instructions in a MicroLumatPlus LB 96V luminometer (Berthold Technologies). As the PB transposon was originally isolated from insect cells, we hypothesized that a mouse codon-optimization of the transposase CDS might produce higher levels of transposition in the murine ES cell context (). To address this possibility we synthesized the native PB transposase CDS (hereafter iPB) and a mouse codon optimized CDS version (hereafter mPB). iPB corresponds to the original wild-type coding sequence of the PB transposase, whereas mPB is a synthetic sequence coding for the same polypeptide as iPB, but where each codon has been changed for the preferred codon for translation in mouse cells. Both iPB and mPB contain a Kozak sequence built around the start methionine codon. Besides this, potential cryptic splice sites generated during the codon optimization were detected and avoided in the optimized sequence. Expression vectors were assembled by subcloning both versions of the transposase CDSs into pcDNA3. We compared their ability to promote transposition of a PB transposon carrying a promoterless puromycin resistance cassette from a transfected plasmid vector into the ES cell genome (A). iPB or mPB expression plasmids were co-electroporated with a transposon donor plasmid (5′-PTK-3′) into AB2.2 mouse ES cells and the numbers of puromycin-resistant colonies were assessed. mPB yielded considerably more puromycin-resistant colonies than iPB (B). To address saturation effects the amounts of helper or donor plasmids were varied while keeping a constant amount of the respective donor or helper plasmid (C and D). In both series of experiments, mPB provided higher levels of transposition than iPB throughout the whole range of amounts of transfected plasmids. In these experimental conditions we did not observe overproduction inhibition, a phenomenon described in certain transposon systems in which transposition rates decline when the amount of transposase exceeds certain levels (,,). Interestingly, the levels of transposition provided by iPB in these series of experiments were far below those of mPB, especially in D, where not even the highest amount of iPB plasmid is able to match the levels of transposition obtained by transfection of the lowest amount of mPB. This is likely due to the stringent conditions (5 × 10 cells per transfection and 1 μg of donor plasmid) of this experiment, aimed at analysing saturation effects. Additional experiments performed in less-restrictive conditions showed that increasing amounts of iPB provided levels of transposition in the same order of magnitude as those of mPB (Supplementary ), indicating that the differences between the performance of iPB and mPB strongly depend on the experimental settings. Despite this, we must remark that mPB showed higher levels of transposition than iPB in all the conditions tested. To ascertain if the higher transposition frequency provided by the optimized sequence was caused by higher production of transposase we subcloned iPB and mPB into pcDNA3-Kz-HA, a pcDNA3-based vector containing a Kozak sequence plus the nucleotide sequence encoding the hemagglutinin epitope (HA) (A), enabling detection of protein production from HA-iPB and HA-mPB by Western blot analysis using an anti-HA antibody. These two plasmids plus an empty pcDNA3-Kz-HA vector were transfected into COS-7 cells in triplicate. Analysis of protein lysates using an anti-HA monoclonal antibody revealed a clear band at the expected size (∼68 kDa) in the cells transfected with transposase expression plasmids, but not in the cells transfected with the empty vector (B). The amount of PB transposase in the cells transfected with HA-mPB was considerably and consistently higher than in the cells transfected with HA-iPB, confirming that higher levels of transposase production are likely responsible for the greater transposition activity provided by mPB. The original PB transposon is flanked by 13-bp terminal inverted repeats and has additional inverted repeats 19-bp long located asymmetrically with respect to the end of the element (). These inverted flanking sequences, although able to permit transposition between different plasmids in insect cells, were found to be insufficient to allow transposition from a donor plasmid to genomic DNA. In fact, additional transposon DNA sequence was found to be necessary to provide this kind of transposition (,). A 5′-terminal repeat (5′-TR) of 313 bp and a 3′-terminal repeat (3′-TR) of 235 bp have been described as the minimum PB terminal repeats (). Combinatorial experiments performed with the SB transposon have shown that a configuration containing two ‘left’ terminal repeats (or 5′-TRs) provides levels of transposition 3 times higher than the original ‘left’ + ‘right’ combination (). In order to study the properties of the 5′ and 3′-TRs of the PB element and to explore the possibility of enhancing the rates of transposition, we subcloned a promoter-less puromycin resistance cassette between different combinations of minimum 5′ and 3′-PB terminal repeats (A) (5′–5′, 5′–3′, 3′–5′ and 3′–3′) and cotransfected the resulting donor plasmids with the mPB expression cassette into AB2.2 cells. Only the cells transfected with the 5′–3′ and 3′–5′ configurations yielded significant numbers of resistant colonies (A). These results indicate that the PB transposon requires one 5′ and one 3′-terminal repeat in order to achieve transposition. The number of colonies obtained with the 5′–3′ transposon was 4.6 times higher than with the 3′–5′ configuration (A). As we had used a promoter-less selection cassette, one explanation of this observation is that the promoter sequence predicted in the 5′-TR, which is functional in insect cells, is also active in the mammalian genome (,). To address this possibility, we subcloned a luciferase expression cassette (Luc2) downstream of the 5′ and 3′-PB TRs (B) and assessed luciferase activity in transiently transfected HeLaS3 cells. The luciferase activity observed with the 5′-TR-Luc2 construct was 5 times higher than that observed in the cells transfected with the 3′-TR-Luc2 construct (B), indicating that the promoter in the 5′-TR repeat is also active in mammalian cells. The success of cancer-gene identification experiments performed with the SB transposase in mice has been strongly determined by the levels of transposase activity (,). However, an excess of transposition has been reported to create undesired genomic rearrangements in an SB-based system for mouse germline mutagenesis (). Thus, the ability to control transposition temporally would provide an additional advantage to the PB transposase. To explore the possibility of generating an inducible transposase we created fusion proteins between mPB and the modified oestrogen receptor ligand-binding domain (ERT2), at the N or C terminus of mPB (ERT2-mPB and mPB-ERT2, respectively) () (A). The ERT2 domain provides the possibility of regulating the activity of a protein by the presence or absence of 4-hydroxytamoxifen (4-OHT). In the absence of 4-OHT a protein containing the ERT2 domain is sequestered by heat-shock proteins, preventing it from functioning (). In the presence of 4-OHT, the fusion protein is released and can then play its role (). The two key requirements of inducible proteins are high activity under induced conditions and low/no activity in the absence of induction. To achieve the maximum activity and inducibility we evaluated a series of linkers between mPB and ERT2 of different lengths, charges and flexibilities (). Most of the fusion proteins appeared to have reduced activity. The L1 fusions show 100-fold reductions compared with mPB. The L2 fusions also exhibited reduced activity although one of them (ERT2-L2-mPB) was not leaky when not induced. However, the fusion proteins containing L3 linkers provided high levels of transposition (B). The L3 linker was based on a sequence which has previously been used between the GAL4-DNA-binding domain and the native PB transposase, which resulted in levels of activity of the fusion protein close to 100% with respect to the wild-type enzyme (). The number of colonies obtained in the presence of 4-OHT using mPB-L3-ERT2 was 6 times higher than those obtained with iPB, though it was less efficient than mPB. Importantly, the activity was highly 4-OHT-dependent, since in the absence of 4-OHT transposition was reduced more than 800-fold to background levels. To our knowledge, mPB-L3-ERT2 is the first PB transposase inducible by 4-OHT, it provides higher activity than iPB and it shows undetectable levels of leakiness. The recent availability of DNA transposons which are active in the mammalian genome has widened the set of tools to carry out functional genomics and transgenesis. SB, a transposon reconstructed from inactive sequences present in salmonid fishes, has already shown great utility, having being successfully used in cancer-gene identification, germline mutagenesis and human cell transgenesis experiments (,,). More recently, PB, an active transposon from the cabbage looper moth , has also been shown to work in mammalian cells, providing transposition levels higher than those displayed by certain hyperactive SB forms (). These two transposable elements have distinctive properties, for instance target sequence preference, which suggests that they will be used as complementary tools to modify and study the mammalian genome (,,,). We have characterized, optimized and further enhanced the functionality of the PB transposon/transposase system. We have synthesized a mouse-codon-optimized PB transposase CDS, which provides levels of transposition significantly higher than the native PB transposase CDS over a wide range of transposon/transposase ratios. Remarkably, the difference in performance between iPB and mPB strongly varied with the experimental conditions, although mPB always provided higher levels of transposition. Notably, overproduction inhibition, the decrease of transposition produced by excessive transposase previously observed in SB, was not detected for PB in the experimental conditions we used (,,). Intriguingly, two previous reports have tried to address the existence of this phenomenon in PB with disparate results. On one hand, Wu . () found overproduction inhibition of PB in HEK293 cells. On the other hand, Wilson . () concluded that PB lacks overproduction inhibition in HEK293 cells. Interestingly, in the same study, Wilson . transfected equivalent amounts of a pCMV-based SB12 helper and donor plasmids in HEK293 cells, and observed the previously reported overproduction inhibition. Although our results support the lack of overproduction inhibition in PB, further work is needed to clarify this issue. We have also shown that the higher transposition provided by mPB correlates with increased protein production. This result highlights the usefulness of codon optimization to increase protein levels. Besides this, the availability of two different sequences encoding PB transposase offers the opportunity to obtain different levels of protein and transposition activity. We have analysed the differential properties of the 5′ and 3′-PB terminal repeats. In contrast with the previous results obtained with different combinations of the SB terminal repeats we have only detected transposition activity when the naturally occurring combination of PB 5′-TR and 3′-TR was used (). Interestingly, we obtained higher numbers of colonies when the resistance cassette was oriented from the 5′-TR to the 3′-TR than in the opposite orientation. In these experiments the transposon contained a gene-trap selection cassette. As the 5′-TR has been described to contain a promoter active in insect cells, we hypothesized that the observed differences in colony numbers could derive from transcriptional activity from the 5′-TR activating the selection cassette (). In a promoter analysis experiment using luciferase as a reporter, we observed that the 5′-TR does act as a promoter which is 5-fold stronger than the 3′-TR. This observation should be considered in the design of PB-based promoter-less transposons, as the expression of the promoter-less cassette from the 5′-TR repeat may produce unwanted results. In our view, the simplest solution would be to rearrange the transposon so that the 3′-TR is upstream of the 5′-end of the promoter-less sequence. Remarkably, recent work has detected enhancer activity in the 3′-TR of the piggyBac transposon (). In our opinion, the presence of promoter and enhancer sequences in the piggyBac 5′-TR and 3′-TR, respectively should also be considered in gene-therapy applications, so as to avoid the negative experiences of generating activating mutations previously obtained with retroviruses (). Finally, we have engineered an inducible, optimized PB transposase. We have made and assessed six different combinations of mPB with ERT2 (). The six versions have two different arrangements of ERT2 either at the C-terminus or at the N-terminus of the PB transposase coupled with three different linkers (L1, L2 or L3) between ERT2 and mPB. Four of these fusion proteins show extremely reduced levels of transposition when compared to the parental mPB, even in the presence of the inducer, 4-OHT. However, the other two fusion proteins, ERT2-L3-mPB and mPB-L3-ERT2, produce high levels of induced transposition. Both of them contain a linker which was based on an aminoacidic sequence previously used to fuse the native PB transposase to the GAL4-DNA-binding domain (). The length and the net positive charge of this linker compared with the ones which were less successful are likely to underlie the huge differences in activity. Remarkably, mPB-L3-ERT2 yields levels of transposition 6 times higher than iPB in the presence of 4-OHT, whereas no detectable transposition was observed without induction. In the absence of 4-OHT the transposition activity provided by mPB-L3-ERT2 is reduced 800-fold, returning back to the low experimental noise levels. Thus mPB-L3-ERT2 is not only an improved version of the PB transposase, but it can also be temporally regulated. This newly engineered enzyme will be useful for diverse applications, ranging from gene therapy, where the possibility of switching on and off the enzyme would contribute to making it safer, to forward genetics cancer screens in mice, where the temporal regulation of the transposition activity would facilitate the adjustment of the dynamics of the tumourigenic processes to prevent undesired genomic rearrangements. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

DNA and RNA oligomers that contain consecutive guanine (G) nucleotides are capable of folding into stable secondary structures such as G-quadruplexes, wherein four Gs are hydrogen bonded together into a roughly square planar array (). G-quadruplexes are of remarkable stability and have been proposed to be involved in regulation of gene expression. For example, a DNA G-quadruplex structure formed in the promoter region functions as a transcriptional repressor element and an RNA G-quadruplex is believed to regulate alternative splicing of the pre-mRNA coding for hTERT, the reverse transcriptase component of the enzyme telomerase (,). Human immunodeficiency virus type-1 (HIV-1) contains the 5′AGAGA polypurine tract sequence (PPT) that is conserved in all HIV-1 strains and is present in the coding region of integrase (IN) and nef messenger RNAs. G-quadruplexes have been implicated in HIV-1 RNA dimerization () and recently were shown to occur in a reverse transcription intermediate, namely between the overlapping strands of the HIV-1 central DNA flap (). It has been shown that HIV-1 nucleocapsid (NCp) and gp 120 envelope protein exhibit a high affinity for several tetramolecular quadruplexes (,). RNA quadruplexes are more stable than their DNA counterparts and in most cases no dissociation is experimentally observed for G tracts involving five guanine quartets (). In the present study, we have used peptide nucleic acid (PNA) targeted to the folded PPT sequence of HIV-1 messenger RNA. PNAs are DNA analogues in which the -(2-aminoethyl) glycine units replace the deoxyribose phosphate backbone (). PNAs are capable of sequence specific recognition of DNA and RNA, obeying the Watson–Crick hydrogen-bonding or/and Hoogsteen schemes (). The neutral amide backbone of PNAs increases their binding affinity to DNA and RNA and the hybrid complexes exhibit high thermal stability (). Short PNA probes were shown to be able to disturb and finally to bind folded RNA structures used as target sequences (). Here, we show that pyrimidine PNAs overcome kinetic and thermodynamic obstacles and succeed to hybridize to folded PPT sequence and finally to unfold it. We have examined whether complexes formed with PNAs on the PPT sequence, likely triplex and duplex structures, affect RNA translation elongation . The cellular antisense activity of the best inhibitors was tested in streptolysin-O (SLO) permeabilized cells stably transformed with two reporter genes, the firefly luciferase () and GFP that contain upstream of the reporter genes either the wild-type HIV-1 PPT target sequence, or a mutated HIV-2 PPT sequence, respectively. The DNA, RNA and PNA oligonucleotides presented here were synthesized by Eurogentec (Seraing, Belgium). 13-mer-Acr PNA was synthesized as previously described (). Sequences and names are given in . The oligoribonucleotides (RNA) were 5′ end-labelled with [γP] ATP (3000 Ci/mmol) and 10 units of T4 polynucleotide kinase. Association and dissociation of the different PNA–RNA complexes were estimated by cooling/heating experiments, recording the UV absorbance as a function of temperature on Uvikon XL spectrophotometer (BioTek) with 1 cm optical pathlength quartz cuvettes. The temperature of cell holder was regulated by a Peltier thermosystem driven by LifePower software (DuSoTec GmbH) for the control and data acquisition. Samples were first heated to 95°C, then cooled down to 20°C and heated to 85°C at the rate of 0.5°C/min or 0.1°C/min with absorbance readings taken every 1°C at 260 and 400 nm. Samples were prepared in a buffer containing 100 mM KCl and 10 mM sodium cacodylate, pH 7. For cooling-melting temperature analysis, the baseline drift was corrected by subtracting absorbance at 400 nm from that at 260 nm. The cooling–melting curves were obtained by plotting the corrected absorbance at 260 nm versus temperature. The maximum of the first derivative of the cooling–melting curves (∂/∂) was taken as an estimation of or values. The and were estimated within, ±1°C accuracy. Unless otherwise indicated, the standard buffer (10 μl) for the PNA-binding assay contained 2 nM of 5′-end labelled RNAs in 50 mM Tris, pH 8. Samples were incubated with increasing concentrations of PNA at 25°C for 10 min. Gel electrophoresis was run at 25°C on a 15% polyacrylamide/bisacrylamide (19/1) non-denaturing gel or 12% polyacrylamide/7M urea denaturing gel. The gel and buffer contained TBE (100 mM Tris/90 mM Boric acid/1 mM EDTA) at pH 8.3. Following autoradiography a PhosphorImager was used for quantitation. The plasmid pRP159 (gift from Dr D. van Gent, Netherlands Cancer Institute, Amsterdam) was constructed by insertion of HIV-1 integrase gene (∼0.9 kb), into the host vector (Promega) behind the SP6 promoter. transcription was performed on Nde-I linearized pRP159 using standard conditions (RiboScribe SP6 RNA Probe Synthesis Kit, Epicentre Technologies). The transcripts were translated in rabbit reticulocyte lysate purchased from Promega. Lysate (17 µl) was added to the reaction mixture (25 μl) supplemented with amino acid mix (1 μl, without methionine), methionine (1 μl, 15 μCi/μl), PNA and transcript. The reaction mixture was incubated for 20 min at 37°C and 8 μl of the mixture was analysed on a 14% (w/v) Tris–Glycine, SDS–PAGE, NOVEX ™ pre-cast gel (Invitrogen). Translation kinetics experiments were carried out in the same reaction mixture. Aliquots (10 µl) were removed at the indicated times and quenched by addition of an equal volume of 2-fold concentrated loading buffer supplemented with β mercaptoethanol. Protein was quantitated using a Molecular Dynamics PhosphorImager. The CMV (+) PPT/HeLa cells stably contain two reporter genes, the firefly luciferase () gene () and the GFP gene, under the control of a bi-directional doxycycline-inducible CMV promoter. These cells contain the PPT sequence (5′AAAAGAAAAGGGGGGA) or a mutated sequence (5′ AAAAGAAGGGGA) upstream of the AUG translation start site of luciferase and GFP genes respectively (). The CMV-luc/HeLa cells contain the reporter gene () under the control of a doxycycline-inducible CMV promoter. There is no insertion of the PPT containing fragment in these cells. Both cell lines were grown in DMEM (Invitrogen) supplemented with 10% of fetal bovine serum, 2 mM glutamine, 50 U/ml of penicillin and 50 μg/ml of streptomycin. Cell culture medium was supplemented with G418 (500 μg/ml) and puromicin (2.5 μg/ml) to maintain the integrated target sequences. SLO (Institute of Medical Microbiology and Hygiene, Mainz, Germany) was used to reversibly permeabilize CMV/(+)PPT/HeLa or CMV-luc/HeLa cell lines toward PNAs according to a recently revised protocol (). SLO was conserved in PBS buffer supplemented with 0.1% BSA. Cells were washed twice and re-suspended in HBSS (Hanks’ balanced salt solution with calcium and magnesium, 10 mM HEPES, and 1% fetal bovine serum). For each experiment, in a 48-well dishes, 1.3 × 10 cells in 100 µl were permeabilized by addition of an optimized amount of SLO (110 ng) and then incubated at 37°C for 15 min, in the presence of PNAs. Resealing was achieved by addition of 800 µl of DMEM supplemented with 10% fetal calf serum and further incubation at 37°C for 20 min. Cells were then transferred in 96-well dishes (4 × 10 cells/well) and cultured at 37°C for 4 h and then induced by addition of doxycycline (Sigma) and further cultured 40 h before quantifying firefly luciferase activity, GFP and total proteins. In the same time, cells were examined with respect to permeabilization efficiency and viability by flow cytometry (Facsort, Beckton Dickinson). Briefly, during permeabilization PNAs were replaced with FITC (20 µM) and after resealing cells were washed twice with PBS and passed through flow cytometer in PBS supplemented with propidium iodide (10 µg/ml) which permitted to determine the fraction of permeabilized and viable cells in the original culture. At the end of experiment, cells were harvested for lysate (passive lysis buffer, Promega) and both the firefly luciferase activity [expressed in relative light units (RLU)], GFP fluorescence and protein concentration [Bradford reagent from Bio-Rad, expressed in optical density (OD)] were measured using a spectrofluorimeter (Wallac Victor 2 Multi-label Counter, Perkin Elmer). The luciferase activity and GFP expression shown in were normalized to the absorbance data, that reflect the amount of proteins, and then expressed as a percentage compared with the luciferase activity and GFP fluorescence levels that were obtained in SLO-treated cells in the absence of PNA. Each data point was averaged over two replicates of three separate experiments. shows the electrophoretic mobility of the 19-mer, RNA-III-wt oligoribonucleotide containing the PPT sequence of the HIV-1 integrase coding sequence (). Several slow migrating species are detected with wild-type RNA (lane 1), while the derived 19-mer mutated RNA, RNA-III-mut, in which two adenines break the G-tract migrates faster and as a single species (lane 2). Such observations supports the involvement of the G-run in the structure of the RNA PPT sequence that likely adopts a quadruplex form. Indeed, stretches of guanines can associate through hydrogen bonding to form four-stranded structures. G-quadruplexes are of remarkable stability and have a preferential affinity for monovalent cations (typically K or Na) that exhibit a suitable size for interaction with the electronegative carbonyl oxygen ring inside the G-quartet (). Recently, Mergny . () have shown that the RNA sequence r-UGU, forms extremely stable parallel RNA quadruplex that is much more stable than its DNA counterpart. The increased stability of RNA quadruplexes results from a faster association and a slower dissociation. Strand concentration and ionic environment plays an important role in the kinetics of quadruplex formation. In our case, the folded RNA structure was observed at very low 19-mer RNA-III-wt concentrations (2 nM) in the absence of added monovalent ions. One explanation is that this 19-mer RNA sequence is already folded after the synthesis and that during sample drying, setting in solution and dilution, the folded structure is maintained. This is not astonishing insofar as the lifetime of r-UG4U quadruplex at 37°C was estimated to be more than 100 years (). The multiple bands of RNA-III-wt can be assessed to folded structures isomers. We have used in our studies different batches of synthesis and always observed the same profile of migration. NaOH addition to the RNA-III-wt showed a shift of the folded structures to an unfolded structure, which migrated at the same position as mutated RNA-III-mut (A). Folded structures titration by NaOH allowed the determination of a mid-point transition from structured-RNA-III-wt to RNA-III-wt monomer at 9 mM NaOH (B). In order to determine whether short complementary PNAs can invade and bind to structured RNA targets, here RNA-III-wt, we have used electrophoretic mobility shift assays. Binding of PNAs to unfolded RNA-III-mut was also analysed. P-radiolabelled RNAs were incubated in the absence or in the presence of increasing concentrations of tridecamer PNAs, 13-mer-I and 13-mer-II. A shows that at low concentrations (<25 nM) tridecamers form complexes that migrate slower than folded RNA [complex (s)] while at higher concentrations a discrete, faster migrating complexes [complex (f)] were predominant. We postulated that retarded low mobility species result from the fixation of PNAs to the A-rich region of the target sequence not engaged in the structure formed by the G-tract while high mobility complexes result from the invasion of RNA structures by 13-mer-I and 13-mer-II PNAs.13-mer-II PNA that can form six C.G base pairs invades slightly more efficiently folded RNA than 13-mer-I PNA that can form only four C.G base pairs with the PPT RNA target. The binding of the two tridecamer PNAs was also studied on the mutated target, RNA-III-mut. This RNA sequence is not stably structured and is complementary on 11 and 8 contiguous nucleotides with 13-mer-I and 13-mer-II PNAs, respectively. Then, binding of 13-mer-PNAs can be also observed on this RNA sequence, with complexes migrating at the same position as the ones observed on the wild-type sequence after PNA-induced unfolding (A). Binding efficiencies of 13-mer-PNAs on the two RNA targets, wild type and mutated, was measured. For tridecamer-PNAs complexed to RNA-III-wt, we could not determine precisely the K values (that means the concentrations of PNAs required for the formation of 50% of complex) because of the multiplicity of the slow-mobility complexes, however these values are situated between 2 and 5 nM. On the contrary, K values can be determined on unstructured RNA-III mut; for unmodified and acridine-modified tridecamer-PNAs complexed to RNA-III-mut, K was determined around 4 nM. To further characterize PNA binding to the folded PPT RNA sequence we used truncated PNA sequences, 9-mer and 11-mer. They are complementary both to the wild-type and the mutated sequence, to the A-rich region that is not involved in the structure of the PPT RNA. The 9-mer and 11-mer sequences were also used as bis-PNAs (see for sequences), that were expected to bind efficiently to the PPT sequence. As expected, these short sequences bind to folded RNA-III-wt sequence and no unfolding was observed even at high PNA concentrations (A), consistent with the fact that the target sequence is not involved in the structure. The 9-mer and 9-mer-bis PNAs, bind very efficiently to the wild-type and mutated RNA sequences. K values are situated in the range of 2–4 nM. On the mutated RNA target, the complexes are well resolved; with the monomeric PNA 9-mer an unique complex is observed whereas with both bis-PNAs, 9-mer and 11-mer-bis PNAs, two discrete shifted complexes (A, C and C) were observed. In the case of 11-mer-bis PNA, there are two bands in complex C. Probably, 11-mer-bis PNA forms with RNA-III-wt two structurally isomeric C complexes. Similar structurally isomeric complexes formed between homopyrimidine bis-PNAs and single- and double-stranded DNA targets have been reported (). At low bis-PNA concentrations, complex C was predominant while at high concentrations complex C was predominant. The complex C formed with 9-mer bis-PNA migrates as complex formed with 9-mer PNA and corresponds to the triplex formed with the AGA purine sequence with two 9-mer PNAs. Complex C can be assigned to the triplex formation with two bis-PNAs as illustrated in A. If it was the case, this complex would contain two free PNA strands that could form a PNA–RNA–PNA triplex with another RNA target. Indeed, upon addition of an excess of 19-mer RNA super-shifted complexes have been obtained with 9-mer bis-PNA and 11-mer bis-PNA (B). B shows that in the presence of 400 nM PNAs, almost the totality of the RNA-III-wt was converted in slow migrating (s) and fast migrating (f) complexes depending on PNA sequences. Upon addition of an excess RNA-III no change in the migration of complexes were observed except for complexes formed with 11-mer-bis and 9-mer-bis PNAs that were super-shifted (B). These results support a model where each bis-PNA molecule participates in the triplex structure with only one arm; the second arm of each bis-PNA remains free (A, complex C). It is noteworthy that the lifetimes of complexes formed with RNA-III-wt are higher than those formed with RNA-III-mut as upon addition of an excess of target RNA followed by 5 min of incubation, partial dissociation of PNA–RNA-III-mut complexes was observed, whereas PNA–RNA-III-wt complexes were not dissociated (B). fig #text UV-melting curves were performed to further characterize the different complexes. For the two PNA tridecamers, 13-mer-I and 13-mer-II, the melting profiles are very different (); the heating and cooling curves are superimposed for the 13-mer-II in contrast to that is observed for the 13-mer-I. In the latter case, temperatures of dissociation and association depend on the heating rate, supporting non-equilibrium conditions during experiment which gave rise to hysteresis phenomenon. This observation is consistent with duplex formation with 13-mer-II PNA and triplex formation with 13-mer-I PNA since triplex formation is known to be slower than duplex one (). At acidic pH the melting curves of 13-mer-II PNA were similar to those obtained at neutral pH, suggesting that cytosine protonation is not involved in complex stability as expected for triplexes formed with C-rich third strand (data not shown). Altogether, these results suggest that tridecamer PNAs form at neutral pH stable complexes involving duplex (13-mer-II) or triplex (13-mer-I). To better characterize the complex formed with the 13-mer-I PNA on the PPT RNA, melting profiles were performed with the 9-mer PNA that forms a triplex on the AGA sequence. In the models proposed for triplex formation with T-rich PNAs, reflects the non-equilibrium formation of triplex while reflects in most cases the duplex dissociation (). Our experiments are consistent with this model; values were almost the same ( (9-mer) = 84°C and (13-mer-I) = 88°C) but are completely different ( (9-mer) = 58°C and (13-mer-I) = 70°C). Finally, triplex and duplex-forming tridecamer PNAs discriminate wild-type target from mutated one. An important decrease on (between 7°C and 12°C) of 13-mer-I, 13-mer-I-Acr and 13-mer-II PNAs to RNA-I-mut compared with RNA-I-wt was observed (). As expected, 13-mer-I PNA that forms a triplex with PPT sequence inhibits translation of HIV-1 integrase coding mRNA in a dose-dependent manner (A). This inhibition is accompanied by the synthesis of a truncated protein. The truncated protein size (18 kDa) is similar to the size of the polypeptide chain obtained after RNase H cleavage of the RNA duplexed to a 15-mer phosphodiester oligonucleotide targeted to the PPT region (15-mer-PPT) (A, lane 8). Interestingly, 13-mer-II PNA that forms a duplex with the PPT sequence is as efficient as 13-mer-I PNA to arrest translation elongation (A). We have determined an IC value of 0.3 μM and 0.2 μM for PNA 13-mer-I and 13-mer-II, respectively. Both tridecamer PNAs did not affect the translation of Ha- and Luciferase messenger RNAs (A for 13-mer-II and data not shown for 13-mer-I). The time course for translation elongation arrest at a fixed PNA 13-mer-I concentration was also examined (B). In the absence of PNA, translation full-length protein (p32, IN) synthesis was achieved in 4 min. In the presence of 13-mer-I PNA, the truncated p18 polypeptide was detected as early as 2 min after the initiation of translation. Truncated protein life-time is significantly higher when translation elongation was arrested with 13-mer-I PNA compared with 13-mer-II PNA: B shows that depending on the PNA concentration used, the amount of truncated protein dropped to 80, 60 and 30% after 20 min of translation in the presence of 13-mer-I PNA whereas lower quantities (60, 40 and 20%) were obtained in the presence of 13-mer-II PNA. These results may reflect the longer lifetime of PNA-RNA complexes versus the PNA–RNA complexes. These PNAs, 13-mer-I, 9-mer, 9-mer-bis, 11-mer-bis, share in common the TCT sequence. HIV integrase coding mRNA contains a AGA and AGA sequences situated respectively upstream and downstream of the PPT sequence (). C shows that high concentrations of 13-mer-I PNA (>0.4 µM) arrest translation at the PPT site and also at these two other sites, upstream (arrow a, lanes 3, 4, 8, 9) and downstream (arrow b, lanes 3, 4, 8, 9) the PPT site. In order to evaluate the sizes of truncated proteins we have targeted PO-ODNs to the PPT and secondary sites in the presence of RNase H (C, lanes 5–7). These two additional products were also observed with the shorter 9-mer, 9-mer-bis and 11-mer-bis PNAs (C). It is noteworthy that the translation arrests at these secondary sites were more pronounced for the shorter PNAs, 9-mer and 9-mer-bis, compared with the 13-mer-I PNA that appeared more selective. PNA 13-mer-II that do not contain the TCT sequence (see for sequences) induces arrest of translation only at PPT site (C, lane 2). By using the CMV (+) PPT/HeLa cell line, containing two reporter genes, the firefly luciferase gene (luc) and the GFP gene, under the control of a bi-directional doxycycline-inducible CMV promoter, we carried experiments to study the intracellular antisense activity of tridecamer PNAs that were shown as specific and efficient translation inhibitors . Two inserts containing either the wild type HIV-1 PPT sequence or a mutated sequence (HIV-2 PPT) were cloned in the 5′ transcribed region of the luciferase and GFP genes respectively (A). Therefore, it is an appropriate system to test sequence-specific cellular activity of anti-PPT molecules; PNAs targeted to the PPT sequence should induce inhibition of luciferase expression without affecting the GFP expression (A). The PNAs were introduced into the cells by permeabilization of the cellular membrane following treatment with SLO. The SLO-based protocol is among the most efficient methods, to efficiently deliver oligonucleotides of various chemistries (,,). A dose-dependent specific inhibition of luciferase expression was induced in the presence of PNA tridecamers in the SLO-permeabilized cells (B). In these experiments, an important fraction of cells (∼95%) was permeabilized with an optimized amount of SLO without inducing important cell mortality (∼20%). These parameters were measured using flow cytometry analysis and propidium iodide labelling of dead cells and fluorescein labelling of permeabilized cells. Fluorescent microscopy studies without fixation carried out after SLO-treatment using fluorescein-labelled tridecamer PNA has shown a vesicular and nuclear localization of the PNA (data not shown). Similar PNA intracellular distribution after SLO permeabilization was observed using fixed cells (). At 1 μM concentration, 13-mer-I and 13-mer-II PNA induced a strong inhibition of luciferase expression (∼70%) without affecting GFP expression (B). Treatment with the control PNA containing a scrambled sequence did not affect luciferase activity (B). Moreover, tridecamer PNAs did not inhibit luciferase activity of cells lacking the PPT insert (data not shown). xref #text

Transcription in bacteria is catalyzed by DNA-dependent RNA polymerase (RNAP). In order to form an initiation-competent promoter complex, the multi-subunit bacterial RNAPs require the participation of sigma factors. This class of initiation factors binds to the RNAP in solution prior to any DNA interaction. For historic reasons, the bacterial RNAP is referred to as the ‘core’ enzyme, while the complex of core and sigma is called the ‘holo’ enzyme. Studies with RNAP have indicated that formation of the initiation-competent, stable ‘open’ complex (RP), in which a 14 base pair (bp) region of the promoter DNA spanning positions 11 to +3 () has been melted, is a multi-step process (,). The initial, unstable, ‘closed’ complex (RP) between holo RNAP and promoter DNA isomerizes to the open complex in a reaction that includes two kinetically significant intermediates (I, I) and involves conformational changes in both the promoter DNA and the RNAP (): Here both RP and I have short half lives, of the order of seconds in transcription compatible buffers, while I and RP are much longer lived (half lives of minutes or hours) (,). Sigma factors are involved in both the recognition of specific promoter elements and the initiation of the promoter DNA strand separation, which enables the template strand to base pair with nucleoside triphosphates, the precursors of the RNA. Highly conserved aromatic amino acids, positioned on one side of an alpha helix in region 2.3 of the sigma factor () have been implicated in the melting process (). Of these, especially the Y430 and W433 ( σ numbering) have been extensively studied. Both the Y430 and the W433 have generally been assumed to interact with the highly conserved 11A residue in the 10 region. Substitutions at both positions have been found to be deleterious for promoter DNA melting (,,,), but remarkably, the Y430A substitution was also found to facilitate formation of complexes of RNAP with short model templates at low temperatures (), suggesting an inhibitory role for the Y430 residue under some circumstances. The effects of substitutions for Y430 and W433 have been found to be cumulative (,,). A substantial amount of experimental evidence supports a crucial role for the 11A base on the non-template strand in initiating the process of promoter DNA strand separation (), but not in formation of the closed complex (). It is thought that flipping the 11A out of the DNA helix and into a hydrophobic pocket on the sigma factor, nucleates DNA melting (,,). Consistent with such a role for the 11A, mismatches at 11 were found to partially compensate for the deleterious effects on open complex formation of non-template strand substitutions at 11 (,). From studies of the effects of base analog substitutions at 11, it was concluded that the N1 of the 11A is important in the nucleation of promoter DNA melting (). In much of the previous work different assay conditions and promoter DNAs were used, making some of the results difficult to compare. Thus the roles of Y430 and W433 have remained unclear. Here we revisit and extend the prior work using uniform assay conditions and template DNAs. Our results have led to the following novel insights: (i) The Y430 and W433 residues have important functions beyond any direct interaction they may have with the 11A. (ii) Substitutions for the 11A and the Y430 and W433 have similar effects in directly inhibiting the rate limiting step in the formation of an open RNAP-promoter complex. (iii) The effects of substitutions at 11A and at Y430 and W433 are cumulative. Oligodeoxynucleotides were synthesized by Invitrogen, Integrated DNA Technologies [2AP (2 amino purine) and Oligonucleotides abasic at positions 11 or 8], or TriLink (Nebularine oligos). (γ-P) ATP was purchased from Perkin Elmer, DNA modifying enzymes from either New England Biolabs or Roche and RNAP core from EpiCenter. All chemicals were from Sigma, Fisher or Amresco. Mutagenesis of the rpoD gene and sigma expression vectors are as described. σ factors were purified using the protocol of Zhi and Jin () that does not involve protein denaturation, with minor modifications: a sonicator was used to perform the cell lysis instead of using a French press; GE Healthcare HiTrap chelating HP columns were used for the Ni column chromatography. As in our hands, there was no significant difference between purified sigma factors before and after ion-exchange chromatography, we routinely omitted this step. However, buffer and salt content of the storage buffer for all purified sigma factors is similar regardless of whether this step was included. The final concentrations of the dialysis buffer components are 10% glycerol, 50 mM Tris-HCl, pH 8, 0.1 mM EDTA, 0.1 mM DTT, 0.01% Triton X-100 and 350 mM NaCl. After concentrating the protein in this buffer, an equal volume of pure glycerol was added. With the above procedure there is a detectable but insignificant contamination of the σ with core RNAP. The optimal ratio of the purified sigma factors to core enzyme for reconstitution of holo RNAP was determined by titrating varying amounts of sigma with core RNAP and monitoring formation of a stable, heparin-resistant complex between the RNAP and a strong promoter. The fold-excess of sigma over core required for maximal extents of stable complex formation differed for the various preparations of sigma factors used. Purified RNAP core (400 nM) and σ factors were incubated on ice for 1 h with an excess of sigma as determined in the assays performed above. All concentrations were adjusted using storage buffer. DNA oligonucleotides, purified as described (), were 5′ end-labeled with P by polynucleotide kinase in a reaction containing γ-P-ATP (). Unincorporated (γ-P) ATP was removed using BioRad Micro Bio-Spin 6 Chromatography Columns. Annealing of complementary DNA strands was performed in a reaction containing 25 mM Tris-HCl, pH 7.9, 50 mM NaCl, 100 nM P-labeled DNA, and 150 nM unlabeled complementary strand. Reactions were incubated at 90°C to 95°C in a heat block for 5 min, followed by slow cooling to room temperature. The concentration of the annealed DNA is expressed as the concentration of the limiting, radiolabeled strand. Each reaction (10 μl) contained, in Fork binding buffer (FBB: 30 mM Hepes, pH 7.5, 1 mM DTT, 0.1 mg/ml BSA, 100 mM NaCl, 0.1 mM EDTA, pH 8, 1% glycerol), 10 nM annealed DNA, and 50 nM of RNAP holoenzyme. Reactions were started by addition of RNAP and incubated at room temperature for 10 min. To assay for formation of stable complexes, reactions were challenged with 200 μg/ml of heparin by adding 1 μl from a 2 mg/ml stock and incubated for an additional 10 min. For reactions without a heparin challenge, 1 μl of ddHO was added and incubation was allowed to proceed for 10 additional minutes. For loading, 2 μl of non-denaturing dye solution (45% glycerol, 50 mM sucrose, 0.1% BPB and 0.1% XCFF) was added to each reaction and 9 μl was applied to a running 4% non-denaturing gel poured and run in TAE buffer (40 mM Tris-acetate and 1 mM EDTA). Gels were run at low voltage (90–100 V) for 1–2 h at room temperature. After drying, the gel was analyzed by PhosphorImaging (Molecular Dynamics) using ImageQuant 5.2 software to quantify the radiolabeled DNA that is free and RNAP-bound, in order to determine the fraction of DNA bound by RNAP. Error was determined from half of the spread of the values (two experiments), or the standard deviation (three or more experiments). : the equilibrium dissociation constant, was determined by titrating a constant annealed DNA concentration of 2 nM with RNAP (a range of 4 nM to 200 nM final concentration). Reactions were incubated, heparin challenged (10 min, 200 μg/ml), and loaded onto non-denaturing gels as described before. After PhosphorImager analysis of the radiolabeled bound and free DNA, fractions bound () were plotted against RNAP concentration using Kaleidagraph version 3.52 and fit to the following hyperbolic equation: = /(1+(/[RNAP])) + . : The first-order rate constant, , for dissociation of stable RNAP-promoter complexes was determined by mixing final DNA and RNAP concentrations of 10 nM and 50 nM, respectively, in a 65 μl volume. After a 10 min incubation at room temperature, 2.6 μl of 5 mg/ml heparin (final concentration of 200 μg/ml) was added and aliquots were removed after 0.5, 1, 2, 5, 10 and 30 min and added to 2 μl of non-denaturing dye before loading 10 μl onto a native gel as described before. After PhosphorImager analysis of the radiolabeled bound and free DNA, fractions bound in heparin-stable complexes (determined as indicated before) were plotted against time using Kaleidagraph version 3.52 and fit to the following sum of exponentials: = *(exp(−*)) + *(exp(−*)), where and are the amplitudes for the first and second decay process, with rate constants and , respectively. The first event is likely the fast dissociation of RNAP bound in non-specific and closed complexes and the second, with rate constant , (reported here as ) the dissociation of stable complexes. All errors were calculated as described for the EMSA assays. Half lives were calculated as = 0.69/. : the second-order rate constant for formation of a stable complex, , was calculated from the experimentally determined pseudo first-order rate constant () for association of RNAP and promoter DNA to form a stable complex, and , the first-order rate constant for dissociation of the stable complex. To determine , promoter DNA and RNAP at final concentrations of 10 nM and 50 nM, respectively, were mixed in a 10 μl volume and the reaction was incubated for 0.5, 1, 2, 5, 10 or 30 min before adding 1 μl of 2 mg/ml heparin (final concentration of 200 μg/ml) for 30 s (to remove all closed complexes). Addition of dye and loading of reactions is as described before. Fractions of DNA bound (determined as indicated before) were plotted against time using Kaleidagraph version 3.52 and fit to the following pseudo first order equation: = *(1 − exp(−*)). From the experimental values of , was then calculated using the following equation: = [RNAP] + . Reactions contained 10 nM annealed DNA in Tris Binding Buffer (TBB: 40 mM Tris-HCl, pH 8, 40 mM KCl, 1 mM MgCl; we found that the KMnO reactions were more efficient in this buffer () despite slightly reduced extents of heparin-resistant binding) and were started by the addition of RNAP to 50 nM (final volume 20 μl). Incubation was carried out for 10 min at room temperature. Heparin challenge was then performed for an additional 10 min by adding 1 μl of a 2 mg/ml stock to obtain a final concentration of 100 μg/ml. For experiments without heparin, 1 μl of ddHO was added instead. Then 1 μl of a freshly made 21 mM stock of KMnO was added for a final concentration of 1 mM. After 30 s, 5 μl of stop solution containing 1.5 M NaOAc, pH 8, 4 mg/ml glycogen, and 300 mM β-mercaptoethanol was added and reactions were placed on ice. After ethanol precipitation, the dried pellets were re-disolved in 70 μl of 1M piperidine, and the solutions incubated for 20 min at 90°C. Reactions were stopped by placing on dry ice. Ten microliter of 5 M LiCl was added to the thawing reactions, and the DNA was ethanol precipitated again. The dried pellets were taken up in ddHO, dried again, dissolved in 8 μl of formamide dye (10 mM NaOH, 1 mM EDTA, 80% formamide, 0.1% XCFF, 0.1% BPB) and loaded onto a 10% sequencing gel, poured and run in TBE (89 mM Tris, 89 mM boric acid, 2 mM EDTA). After electrophoresis, the gel was dried and the bands were revealed by PhosphorImaging (Molecular Dynamics). Analysis of the images was performed using ImageQuant 5.2 software. To compare extents of RNAP-induced melting of a promoter duplex, the 1T band was used to quantify strand opening, and the radioactivity of this band was divided by the radioactivity of full length Duplex DNA. A normalized background value, established by using the same quantification procedure with a lane containing labeled DNA without added RNAP, was subtracted. Finally, all results were re-normalized to the value obtained for WT RNAP with Duplex DNA. Error was determined by taking the standard deviation (three or more determinations), or half of the spread (two determinations). Our goal was to arrive at a better understanding of the roles of amino acid residues Y430 and W433 of σ, and of the 11A in promoter DNA melting. Our approach has been to combine σ containing substitutions of A, L, F, W or H for Y430 and of A, L, F, Y or H for W433 with promoter DNAs bearing substitutions for the 11A. The DNA substrates used in this work are shown in A. We have carried out most experiments using a promoter (‘Duplex’) that is truncated in the downstream direction at position +1, the start site of transcription. In the non-template strand these DNAs had the 11A or substitutions of 2AP, G, purine, or an abasic nucleotide (B). Another template (‘Mismatch’) had an A at 11 in the template strand, leading to a mismatch in the templates that contained A, 2AP, or purine at 11 in the non-template strand. Duplex templates containing a G at 11 of the non-template strand had a 11C in the template strand, and a G at this position for the Mismatch DNA. The promoter used has consensus 35 and 10 regions [the latter includes the upstream TG element, and is thus an extended 10 (,)], to ensure tight RNAP binding in the closed complex, so that effects of substitutions would mostly be on DNA melting rather than closed complex formation. Our previous work () showed that substitutions at the 10 and 35 that decreased the similarity of these regions to their consensus sequences, greatly reduced binding affinity for RNAP in the closed complex. The rationale for the use of truncated templates is that the effects of substitutions in both promoter DNA and σ are more pronounced, as compared to a longer DNA template (to position + 20) of the same sequence (data not shown). Stable complex formation at room temperature for some combinations of promoter and σ substitutions are shown in C. In this experiment, prior to loading binding reactions consisting of RNAP and labeled DNA on the gel, they were challenged for 10 min by added heparin. Heparin binds tightly to free RNAP and is a competitor with DNA for the binding to the enzyme. Due to the fact that the heparin is added in great excess, it will sequester any free RNAP that forms because of dissociation of RNAP promoter-complexes. Thus only stable complexes survive the heparin challenge. Stable complex formation between RNAP and Duplex DNA is seen to be greatly affected by the identity of the base at 11 of the non-template strand. The gel image in C shows that stable complex formation is decreased for WT RNAP-DNA complexes containing the 2AP substitution at 11 compared to the Duplex DNA (A at 11). and summarize the results collected for all experiments carried out with various combinations of DNA templates and RNAP bearing substitutions in σ. The groups of bars represent different DNA templates and the bars within each group, the various amino acid substitutions. We tested purine for its lack of substituents, 2AP as an A-analog but with the amino group at the 2—instead of at the 6 position, G, which in addition to the 2-amino group also has a 6 carbonyl group, and dSpacer which has di-deoxyribose, without an attached base (B). 2AP has found much use for its fluorescent properties (), but has been a very useful probe for RNAP promoter interactions as well (). For Duplex DNA binding to WT RNAP, among the substitutions tested, 11 purine was similar to 11A and 11 abasic was slightly worse (A and A). Here the destabilization of the 11 base pairing would (partially) compensate for any loss of contacts due to the removal of the 11A (,). Formation of a stable complex was inhibited considerably by both the 11 2AP, and the 11 G. The latter two had the greatest effect of any single change in either promoter DNA or σ tested in our experiments (A and A: compare bars within the Duplex group with the first bar of the 11 2AP and 11G groups). This is likely due to steric hindrance by substituents not present at the same positions in the canonical 11A (the NH at the 2 position of 2AP and G, and the additional carbonyl O at the 6 position of G). The absence of the 6 amino group may not be as important, as despite its absence, the promoter with 11 purine behaved similarly to that with the 11A. The 11G substitution may have been particularly detrimental due to the presence of an H at the N1, absent at the N1 of both A and 2AP (B), which had previously been shown to be important for open complex formation (). Consistent with prior observations (,), a mismatch at position 11 for some of the σ variants slightly improves stable complex formation with the 11A, 11 2AP and, to a lesser extent, 11G Duplexes (B). This effect can be interpreted to reflect the greater ease of rotation of a 11A out of the helix if it is not base paired to a T. Interestingly, the Abasic Duplex DNA with an A at 11 of the template strand (called ‘Abasic Mismatch’) was more impaired in stable complex formation than the Abasic Duplex with a T at the same position. One possible interpretation of this result is that the template strand 11T of the 11 AT base pair is recognized by the RNAP, another, that the abasic substitution has sequence-dependent structural and thermodynamic consequences (,). When RNAP mutants are used containing substitutions for Y430 of σ, small defects are seen with the Duplex or Purine Duplex DNA (A), particularly for Y430F and Y430W. These defects become much more pronounced when paired with DNA containing other substitutions at 11A, particularly with the 11 2AP. The effects of the substitutions at 430 of σ on stable complex formation by RNAP with the 11 2AP Duplex can be summarized as follows: (i) the σ with the Y430A (see also C) and Y430L substitutions tolerate the 11 2AP to a greater extent than does WT σ, (ii) the Y430H substitution has a small effect and (iii) the Y430F (see also C) and Y430W substitutions are deleterious. Unexpectedly, a very similar pattern is seen for the DNA containing the 11 abasic substitution for Duplex DNA. This indicates that the Y430 residue could not have as its sole function a role dependent on the 11 A base. Results with the same 11 substitutions for Mismatch DNA (B) are similar to the results obtained with WT RNAP, in that a slight improvement in stable complex formation is seen for most templates. All Y430 RNAP mutants, as well as WT RNAP, show large defects in stable binding to the 11G Duplex and Mismatch DNA and are therefore difficult to compare. In a similar series of experiments to those shown in A and B, the effects of substitutions at residue W433 were studied with the 11A substituted DNA templates (A and B). For stable binding to Duplex DNA, W433A and W433L are slightly defective, while W433F, Y, and H behave similarly to WT RNAP under our conditions (A). Like the Y430 mutant RNAP, these defects became more pronounced when paired with the 11 substituted DNA templates. The amino acid preferences for the 433 position are different from those at the 430 position in that an aromatic amino acid is preferred in the former, while a small to medium sized aliphatic amino acid for the latter. As with the Y430 mutant RNAP, a range of effects for substitutions at W433 was observed even when the DNA used was abasic for 11 in the non-template strand. Thus, like Y430, W433 must have a role other than that involving the 11A. The results for stable binding to the 11 Purine Duplex, the 11G Duplex (A), and the Mismatch DNA templates (B) by the W433 mutant RNAP are similar to those obtained for the Y430 RNAP (A and B). Our results clearly demonstrate RNAP's preference for certain base analogs over others at the 11 position of non-template DNA. To verify the unique properties of the 11 base, experiments similar to those shown in and for substitutions at 11 of Duplex were also carried out with Duplex containing substitutions for the 8A (data not shown). The 8 position is conserved to a lesser extent in σ promoters () than the 11 (56% versus 76%), and would therefore be expected to be more tolerant to base substitutions. Based on modeling, the 8A is also positioned in the DNA helix far enough away from the Y430 and W433 amino acids such that we would not expect an interaction to occur between the two (). In contrast to the data shown in and , it was found that for every substitution except 8G, the results were very similar to those seen with the 8A Duplex and Mismatch. The 8G substitution had deleterious consequences, but not as severe as the 11G. Thus, stable complex formation was largely insensitive to substitution at the 8 position on the non-template strand. These results confirm the important role of the 11 base on this strand. All of the above substitutions at the 11 and 8 position were also introduced in fork DNA (non-template strand to +1 and template strand to 12), but no significant differences between DNA templates or RNAP mutants were found at room temperature. The one exception was for substitution of 11G, which showed similar RNAP binding patterns as described before but just a 1.5 to 2-fold decrease in overall binding compared to 11A fork DNA. For comparison, we have also carried out experiments with different promoter DNAs. With longer DNA sequences identical to the Duplex (A) through position +1, but with an extension to +20 (), under the same conditions, no effects of substitutions at 430 and 433 of σ on stable complex formation at room temperature were observed. When a 2AP was introduced at the 11 position, the effect of the substitution was also slight, and substitutions in σ had small effects. This is in sharp contrast to the results presented in C, and . With a similar, long template containing a total of three non-consensus substitutions in regions 10 and 35, some differential effects of the substitutions were seen, although the overall binding was much reduced even for the WT RNAP (data not shown). To determine whether the presence of the extended 10 TG sequence affected our results, a version of the Duplex DNA without the TG sequence was also employed. It displayed diminished binding of wt RNAP under the standard conditions used in and , but the patterns of the substitutions tested (W433L, Y430A and Y430F), with Duplex, and 11Ab Duplex (both with CA instead of TG) was not altered, including Y430A binding better than WT to the 11Ab Duplex (data not shown). In order to try and pinpoint the step(s) in at which the substitutions in promoter DNA or σ exert their effects, we performed binding experiments in the absence of a heparin challenge to determine whether complex formation in general (i.e. both closed and open complexes) was affected. The experiments shown in as well as those shown in , and were carried out with a subset of promoter DNAs and σ. The results in A (gel image) and 4B (quantification by Phosphor Imaging) demonstrate that for both DNAs (Duplex and 11 2AP Duplex), total complex formation is similar for RNAP with WT and the three mutant σ. If the RNAP and DNA are incubated for 30 s instead of 10 min, similar amounts of total complex formation are detected (data not shown). Thus defects in stable complex formation (C, and ) cannot be explained by invoking effects of the promoter—or σ substitutions on DNA binding. Total complex formation was also not affected for RNAP binding to 11 Abasic Duplex DNA (data not shown). For the 11 G Duplex, small (less than 2-fold) differences among the various RNAP in total complex formation became apparent (data not shown), although any correlation between binding and stable complex formation is difficult to assess in view of the very low extents of stable complex formation seen with this template. The above experiments established that substitutions at 11, and at Y430 and W433 did not affect RNAP binding to promoter DNA in the absence of a heparin challenge, but rather had an effect on the fraction of RNAP promoter complexes that was heparin-resistant. To quantify the effects of the substitutions, we determined the equilibrium binding constants for formation of stable complexes (i.e. after incubation of RNAP and promoter DNA, the complexes were subjected to a 10 min challenge with heparin prior to loading of the reaction mixture on a non-denaturing gel). An example experiment is shown in A and the quantification of the data is shown in A. It is seen that the 11 2AP substitution increases the for stable complex formation to WT RNAP by almost 3-fold, indicating a significant weakening of the complex. As compared to RNAP containing WT σ, the W433L and Y430F substitutions in σ result in smaller (approximately 2-fold) increases in for the 11A Duplex, in agreement with the data shown in and . More pronounced effects of the substitutions became evident with the 11 2AP Duplex: the binding affinity of the W433L and Y430F RNAP was so weak that only a lower limit of about 25 nM could be established for . With the 11 2AP Duplex, the complex was more stable (smaller ) for the Y430A RNAP, consistent with the data shown in and . The determinations also demonstrate that the effects of the substitutions at 11 of the promoter and 430 and 433 of σ are cumulative, i.e. the mutant RNAP exacerbates the effects of the 11 2AP substitution. To gain further insight into the steps in affected by the substitutions, and in view of the possibility that the data may have been obtained near the tight-binding limit of our experiments, where discrimination between different values may not have been optimal, we measured the kinetics of stable complex formation. In these experiments we determined the RNAP concentration-dependent rate parameter, , that was then converted to values of . Studies on the rate of dissociation of the complexes were carried out to get the rate constant, and the complex half lives. The / ratio provides another estimate of the dissociation constant. Sample experiments are shown in B and C, respectively. The data are collected in B and C (values for are given in the legend to C). Within the error of our experiments, neither the 2AP substitutions in promoter DNA nor the σ substitutions in RNAP significantly affect . However, they lead to an increase in by as much as a factor of 15. The experimental conditions for the equilibrium and kinetic experiments differed: the glycerol concentration was greater in the former. Thus the values determined cannot be compared to the / ratio (and the similarities of the extents of complex formation in are fortuitous). Even so the effects of the substitutions would be expected to be comparable. However, from the kinetic results, some substitutions would be anticipated to cause an approximately 15-fold increase in , but direct measurements only show a 3-fold increase, in agreement with the supposition that we were operating near the lower limit of detectable values for . For the truncated promoter DNAs used here, the effects of the σ substitutions were most notable from the values for complexes formed with the 11 2AP Duplex: compared with the WT RNAP, the RNAP with the W433L and Y430F had a greater and shorter half life by factors of 3.5 and 2.4, respectively, while the Y430A substitution resulted in complexes with a smaller and greater half life, by a factor of 2.8. We conclude that the substitutions at Y430 and W433 of σ affect the stability of RNAP-promoter complexes. We additionally performed kinetic experiments using WT RNAP and both 11 Abasic Mismatch and 11G Duplex promoter DNA, as well as WT and W433L RNAP with 11 Abasic Duplex. We were unable to detect any significant differences in on—or off rates for the interaction of WT and W433L RNAP with 11 Abasic Duplex DNA, as compared to Duplex DNA (data not shown), despite slightly reduced levels of overall heparin resistant binding (A and A). Complexes of WT RNAP and 11 Abasic Mismatch DNA were slightly more stable than those of 11 2AP Duplex DNA (Half-life of 11.3 min for 11 Abasic Mismatch compared to 5.4 min for 11 2AP Duplex) (data not shown), while both are less stable than complexes containing the Duplex or 11 Abasic Duplex DNAs. Thus the reduced extent of stable binding seen with the 11 Abasic Mismatch DNA compared to the 11 Abasic Duplex ( and ) is due to the decreased stability of the former complex. Consistent with the slow on-rate observed in prior work (), we were not able to determine a meaningful for the interaction of WT RNAP and the 11G Duplex due to the low amounts of heparin resistant binding and for the same reason, no either. In there are two types of stable complexes: the intermediate I with, at most, a few melted base pairs at the upstream edge of the DNA region that becomes single stranded in the open complex, and the fully strand-separated open complex RP. Under most conditions the equilibrium between these complexes heavily favors the open complex. In order to determine whether the substitutions in the Duplex DNA or σ might have shifted the complex towards the I form, we determined their effects on the relative extents of promoter strand separation by subjecting RNAP-promoter complexes to KMnO probing () both with and without a prior challenge with heparin. A sample gel for the determination of the extent of strand separation of complexes surviving a 10 min heparin challenge is shown in A. As a measure of RNAP-induced promoter DNA melting, the intensity of the band generated by piperidine-induced cleavage at the oxidized 1T in single stranded regions was quantified, and expressed as percent of the amount of radioactivity in uncleaved DNA. The values are shown in B for samples that were heparin-challenged, and in 7C for samples that were not. The presence of the 11 2AP does not affect strand opening in the absence of a heparin challenge, but the extent of strand opening is reduced after incubation with heparin prior to loading the samples on the gel. The presence of the 11 2AP only affects the extent of strand separation when the equilibrium is skewed to the left by the addition of heparin, which binds to free RNAP and inhibits promoter binding. These results are similar to those obtained by EMSA with and without heparin challenge (compare left bars of the Duplex and 11 2AP Duplex in A and A, with left bars of the same DNAs in B). Just as was observed by EMSA ( and ), with the 11A Duplex the differential abilities of the RNAP with substitutions in their σ to orchestrate heparin-stable strand separation of promoter DNA are masked under our conditions (see left-hand clusters of bands and bars in A and B, respectively). Again the use of the 11 2AP Duplex allows better differentiation among the σ with substitutions at 430 and 433: For the complexes formed with the W433L mutant, a smaller fraction of the radioactivity is in cleaved DNA bands than for complexes formed with WT σ, while with the Y430A a larger fraction of the radioactivity is found in the KMnO generated bands. The Y430F is similar to the WT σ in this regard. This substitution's lack of deleterious effects here, as compared to the EMSA assays, may be partially due to the lower concentrations of heparin used (100 µg/ml in the KMnO experiments versus 200 µg/ml for the EMSA). For W433L and Y430A, the results closely parallel those obtained with the EMSA assay, which monitored formation of stable RNAP-promoter complexes. In the absence of a heparin challenge, no significant differences were seen in the KMnO patterns between W433L, Y430A and Y430F for either the 11A or the 11 2AP Duplexes. This is in agreement with the results from the EMSA experiments on the extent of formation of complexes in the absence of a heparin challenge (B). We have no data that, for any combination of substitutions in σ and Duplex DNA tested, would support the accumulation of complexes that survive a heparin challenge, but yet are not fully strand separated (e.g. I). Indeed we also have observed that stable complexes, even if containing both mutant σ and a 2AP substitution at 11, can initiate RNA synthesis. When tested on a template that is extended in the downstream direction compared to that shown in , the complexes were able to make the abortive product UpApU from UpA and UTP in proportion to their ability to form heparin resistant complexes (data not shown). The observation that the Y430F substitution is deleterious to formation of a stable complex between RNAP and promoter DNA containing 11 2AP, would suggest that the tyrosine OH group plays an unknown but important role in formation of a stable RNAP-promoter complex. With the consensus promoter that extends to +20, at low temperatures, the RNAP containing the Y430A substitution (but not Y430L) is deficient in open complex formation (data not shown): the same is the case for a similarly long promoter with non-consensus 10 and 35 regions at room temperature. Such behavior had previously been observed with the long non-consensus promoter for Y430A RNAP (). Similarly, Juang and Helmann () found that for σ, both the Y189A and Y189L (Y 189 is the ortholog of σ Y430) substitutions had detrimental effects on open complex formation with a full sized promoter. With the short Duplex promoter sequence, the RNAP containing σ substitutions Y430A and Y430L are better than the WT sequences of σ at forming a stable complex. In this respect, our results on stable complex formation at room temperature, are consistent with those from Gralla's group, who investigated total complex formation (no heparin challenge) near 0°C; under these conditions the vast majority of the complexes are expected to be unstable, closed, complexes (). The latter group concluded that for their short DNA templates, the Y430 had an inhibitory effect, which would be relieved when the Y430 residue was substituted by A (). This would also be the case for the Y430L substitution. We speculate that Y430's role is to be inserted between the bases of the DNA to initiate promoter bending. Then, for the Duplex promoter which would be difficult to bend as it lacks any sequence downstream of +1 to serve as a handle for the bending, it might be less detrimental to form a stable complex with RNAP containing σ with an A or L at position 430. As they are smaller and more flexible they would be less likely to be forced into the DNA helix, than the Y430 of WT σ. The OH of Y430 conceivably could aid in this process by engaging in H-bond formation. It has been suggested () that RNAP may compete with, and weaken, the hydrogen bonding between the 11A-T pair prior to base flipping. Because the Y430 substitutions to A or L facilitate open complex formation with Duplex DNA, it is not likely that the hydroxyl of Y430 is engaging in such a competing hydrogen bond interaction with the 11A. It had been assumed that the Y430 and W433 residues played a direct role in flipping the 11 base out of the DNA helix and into a pocket on the RNAP (,). The fact that we see a range of activities for the various substitutions at positions 430 and 433, even with a DNA that is abasic at the 11 position of the non-template strand, sets limits on models for the roles of each of the residues W433 and Y430. , there are multiple possibilities, including the following: (i) No interactions with the 11A; (ii) interaction(s) only to 11A; (iii) interaction(s) only to 11A, with other amino acids also interacting with this base; (iv) interactions to both 11A, and to another group of the DNA; (v) novel interactions with the DNA due to adaptation of the protein (,) to the DNA site that lacks the 11A. Our results with DNAs that are abasic for 11 in the non-template strand rule out (ii) and (iii), above: If the only roles of Y430 and W433 were to recognize and flip the 11A, in the absence of this base the actual residue present at positions 430 or 433 of σ would not matter, and a constant (low) level of strand separation would be expected regardless of the particular amino acid side chain at these positions. The simplest explanations for the results with the abasic DNA are possibilities (i) or (iv), but we cannot exclude (v) based on the available data. The adaptabilty of σ may in fact be of crucial importance for the recognition of a variety of non-consensus promoters by RNAP. For comparison we have also investigated the effect of an alanine substitution for R588 of σ, which is involved in recognition of the 35 region () and would be a highly unlikely interacting partner of the 11A. It is found (data from and , and data not shown) that RNAP containing WT and R588A σ, bind the following percentage of input DNA, respectively, at room temperature subsequent to a 200 μg/ml heparin challenge: Duplex, 66 ± 7 and 46 ± 1; 11 2AP Duplex, 28 ± 7 and 3 ± 0.4; 11Ab Duplex, 51 ± 2 and 22 ± 2. These numbers for R588A are very similar to those for W433L with the same DNAs, and also qualitatively similar to those for Y430F, consistent with the possibility that Y430 and W433, like R588, would not interact with the 11A either. The finding that with the short Duplex DNA at position 430 small (A) or medium sized (L) aliphatic side chains are preferred, but at position 433, aromatic amino acids [this work and also ()], must reflect the different roles these two residues play in the process of open complex formation. Perhaps the W433 is involved in stacking onto another aromatic residue. Consistent with the conclusion that amino acids 430 and 433 of σ have other (or additional) roles than interacting with the 11 base, it is found that substitutions at 430 and 433, and at 11 in the DNA (e.g. the 2AP or the G) exacerbate each others’ deleterious effects, indicating that different processes may be affected by the DNA—and σ substitutions. Also, a mismatch at 11, expected to facilitate removal of the 11 base out of the helix as shown here ( and ) and elsewhere (,) to improve open complex formation on Duplex DNA, results in only a small improvement of stable complex formation with the templates for which the 11 A has been substituted by 2AP. This suggests that there is sequence recognition subsequent to removal of the 11 base out of the helix and thus constitutes another demonstration that nucleation of strand separation is, at least, a two-step process. Our results show that substitutions at 11 in Duplex DNA, and Y430 and W433 of σ both increase the and the for the stable complex. Ten-fold effects on the off-rate were seen for the substitution of the 11A with 2AP, for the WT RNAP. However, for amino acid substitutions in σ the effects on the of complexes formed with either Duplex DNA or 11 2AP Duplex are more modest (up to 4 fold). Interestingly, the Y430A (and L) RNAP tolerates substitutions at 11A better than the WT RNAP (see also ). Both the 11 2AP substitution in Duplex DNA, and the three amino acid substitutions investigated had little, if any, effect on the for stable complex formation (B). In a previous study () using a promoter with non-consensus 10 and 35 regions and on a longer piece of DNA with extensions in both directions, we found order of magnitude effects of single substitutions in σ on the rates of formation of stable complexes, monitored just as in the experiments described here, by using EMSA subsequent to a heparin challenge. Here we did not detect such effects on , even though our experimental conditions should have readily allowed detection of both increases and decreases in its value (B). A recent study (), as well as countless prior ones [e.g. ()] also found significant effects on of base substitutions in promoter DNA. A possible explanation for the discrepancy between our data and other work is the difference in length (Duplex is truncated at +1) and the fact that A and its base analog, 2AP, both have an unsubstituted N1 that may be important in the strand separation process (). Therefore, non-consensus 35 and 10 elements, as well as substitution of 11A with another base or base analog that changes the hydrogen bonding capabilities of the N1 position could affect on-rate. With regard to an abasic site at the 11 position, our work confirms prior data suggesting that such a site causes a local disruption of the DNA helix in the surrounding area, thus aiding in RNAP-induced strand separation (,,). In any case, effects on the ‘on’ and ‘off ’ rates would both be exerted at the rate limiting step, which is the I to I conversion in either direction. As I is in rapid equilibrium with RP and I with RP (), it is also possible that the mutations in either DNA or RNAP skew these equilibria towards RP, thus decreasing the on the one hand, or towards I, thereby increasing the , on the other. We have no information on the equilibrium between RP and I (although the DNA bending step may not occur for our short Duplex promoter lacking DNA downstream of the +1 site), but our experiments did address whether the substitutions might lead to increased formation of stable complexes that are not strand separated (e.g. I). In view of the good correlation between the EMSA (determination of stable complexes; and ) and the KMnO (determination of strand opening; ) studies, there is no evidence for accumulation of I, and consequently not for inhibition of the I to RP step, due to any substitutions in σ or promoters. This is in contrast to published data, which showed that stable complexes of WT RNAP and a 11 2AP containing promoter could still form, but that no strand separation could be detected (). Likely the actual promoter sequence plays a role as our group previously showed that strand separation could occur at another 11 2AP-substituted promoter () as well. sup sub xref #text

DNA-dependent protein kinase, consisting of the catalytic subunit DNA-PKcs and the end-binding heterodimer Ku, is a core component of the nonhomologous end-joining pathway of DNA double-strand break repair. The primary function of DNA-PK appears to be regulating the end-joining process, both by sequestering DNA ends and by catalyzing serine/threonine phosphorylation of itself as well as other proteins (,). In order to assess the biological significance of specific phosphorylation events, radiosensitivity and end joining were previously examined in CHO variants harboring nonphosphorylatable S/T → A substitutions at several sites shown to be targets of DNA-PK-mediated phosphorylation, either or . Data obtained thus far suggest that whereas phosphorylation of Ku and XRCC4 is dispensable (), phosphorylation of DNA-PK itself, particularly at T2609, S2612, T2620, S2624, T2638 and T2647 (the ‘ABCDE cluster’) is critical for efficient DSB repair. For example, single S/T → A substitutions at several of these sites increase radiosensitivity of cells harboring the mutant proteins, and tandem substitutions at multiple sites in the cluster inhibit repair to a greater degree than mutations at any single site, and confer greater radiosensitivity (,). Particularly intriguing is a DNA-PKcs allele with S/T → D substitutions at all six sites in the ABCDE cluster, designed to mimic constitutive phosphorylation at these sites. In an end-joining assay with purified Ku and XRCC4/DNA ligase IV complex (X4L4), this mutant (hereafter designated DNA-PKcs-D6) promotes ligation of a cohesive-ended substrate, albeit less efficiently than the wild-type protein. Moreover, whereas wortmannin, an inhibitor of DNA-PK kinase activity, completely blocks the end joining promoted by wild-type DNA-PKcs, wortmannin has no effect on the end joining promoted by the D6 mutant (). Thus, in this assay, DNA-PKcs-D6 appears to substitute for autophosphorylated DNA-PKcs, suggesting that phosphorylation of the ABCDE cluster may be the primary means by which DNA-PK regulates the end-joining process. In DNA-PKcs-deficient CHO-V3 cells, ectopic expression of DNA-PKcs-D6 partially restores radioresistance and DSB rejoining (,), although not as robustly as might be expected from the studies. In an attempt to further elucidate the role of phosphorylation of the ABCDE cluster, end joining was examined in whole-cell extracts of DNA-PKcs-deficient M059J cells supplemented with wild-type and mutant DNA-PKcs proteins. This experimental system allows direct examination of DNA end processing and end joining, in the presence of all the proteins required to carry out joining of both cohesive and incompatible DNA ends. The results confirm the importance of the ABCDE cluster as a DNA-PK phosphorylation target, but suggest that there are other target phosphorylation sites of equal if not greater importance. Purification of DNA-PKcs from CHO-V3 cell lines harboring wild-type and mutant human DNA-PKcs alleles has been described (). Because the yield of DNA-PKcs from hamster cells is much lower than from human cells, DNA-PKcs from HeLa cells () was used for experiments that did not involve direct comparison of wild-type and mutant alleles. Nevertheless, the properties of HeLa DNA-PKcs did not appear to differ significantly from those of wild-type DNA-PKcs purified from V3 cells. To ensure reproducibility between experiments, initial frozen aliquots of DNA-PKcs were subdivided into 1 μl aliquots and flash-frozen in the presence of 1.6 mg/ml BSA. Thus, most experiments were performed with DNA-PKcs that had undergone no more than two freeze–thaw cycles. Concentrations of DNA-PKcs from V3 cells were determined by Bradford assay, and the purity and relative concentrations of the three recombinant alleles were verified by quantitation on gels stained with SYPRO orange (Supplementary Figure 1). Previous end-joining experiments employed an SV40-based shuttle plasmid (pSV56) designed to introduce a DSB in a polylinker in the intron of the T-antigen gene (). However, because the presence of the SV40 origin reduces the yield of the plasmid from bacterial cultures, the polylinker region of pSV56 was excised as a 625-bp PflMI/AvrII fragment, and cloned into pBR322 between the EcoRV and NheI sites (the PflMI cut was blunt-ended with T4 polymerase). This 5-kb plasmid, designated pRZ56, was cut at its unique MluI site, and then subjected to controlled 3′ → 5′ exonucleolytic digestion with T4 polymerase in the presence of dTTP. This procedure resects each 3′-terminal strand to the first thymine in the sequence, resulting in a 10-base 5′ overhang at one end and an 11-base 5′ overhang at the other. An unlabeled 13-mer and a 5′-P-labeled 14-mer, each complementary to one of the 5′ overhangs, were successively ligated into the overhangs, to yield site-specifically labeled substrates with partially complementary (-ACG) overhangs, as described previously (,). Control experiments showed that the oligomers were ligated onto at least 90% of the overhangs (data not shown). The labeled plasmid was purified by agarose gel electrophoresis, electroeluted, concentrated (Amicon Centricon 100) and precipitated. Alternatively, the MluI-cut plasmid was dephosphorylated and 5′-end-labeled with polynucleotide kinase and [P]ATP. DNA-PKcs-deficient M059J cells were obtained from Dr Joan Allalunis-Turner (Cross Cancer Institute, Edmonton, Alberta, Canada), and were grown in MEM alpha (Gibco) containing 10% fetal bovine serum. One day after reaching confluence, cells were trypsinized, and washed extensively with serum-containing medium and PBS. Whole-cell extracts were prepared by hypotonic swelling and Dounce homogenization as described (,), except that instead of a 3 × hypertonic extraction buffer, 0.25 volume of a 5 × buffer (83.5 mM Tris-HCl, 1.65 M KCl, 3.3 mM EDTA, 1 mM dithiothreitol) was added prior to high-speed centrifugation, to yield a slightly higher protein concentration. Extracts consistently had a protein concentration of ∼12 mg/ml (Pierce BCA protein assay), and extracts from confluent cultures showed greater end-joining efficiency than those from subconfluent cultures. Dignam nuclear extracts, prepared as described previously (,), had a concentration of ∼5 mg/ml. Reactions contained 50 mM triethanolamine–KOH pH 7.5, 1 mM ATP, 1 mM dithiothreitol, 1.3 mM magnesium acetate and dNTPs (or ddNTPs) at 100 μM each. Typically, a 16-μl reaction contained 10 μl of extract, resulting in a final concentration of 66 mM potassium acetate and 16% glycerol, and an effective Mg concentration of 1 mM (taking into account ∼0.3 mM EDTA from the extract). Buffer components were first mixed with cell extract at 22°C. DNA-PKcs and/or kinase inhibitors were then added and the solution mixed by pipeting. Finally, 10 ng substrate was added and the reaction again mixed by pipeting, and placed in a 37°C water bath, usually for 1 or 6 h. Samples were deproteinized as described () and either loaded directly onto 0.8% agarose gels, or cut with BstXI and TaqI and analyzed on 20% polyacrylamide DNA sequencing gels. Storage phosphor screens were exposed to frozen polyacrylamide or dried agarose gels, and images were analyzed with ImageQuant 3.3 software. As demonstrated previously, human whole-cell extracts contain robust activity for joining of both cohesive and incompatible DNA ends (,,). When end-joining reactions are performed in the presence of relatively low Mg concentrations, the joining is largely dependent on core end-joining factors such as DNA-PKcs, XRCC4, DNA ligase IV (,) and, in the case of incompatible ends, DNA polymerase λ (). To assess the dependence of end joining on DNA-PKcs, and specifically on DNA-PKcs kinase activity, whole-cell extracts were prepared from DNA-PKcs-deficient M059J glioma cells (). Western blotting (not shown) confirmed absence of DNA-PKcs in these extracts. End joining of an internally labeled substrate with partially complementary (-ACG) 3′ overhangs was then examined in these extracts either with or without addition of purified DNA-PKcs from HeLa cells (). When this substrate was incubated in M059J extracts, and then cut with BstXI and TaqI to release short fragments from opposite ends of the plasmid, there was some 3′ → 5′ exonucleolytic digestion, but no ligation products were detectable (B). Supplementation of the extracts with DNA-PKcs resulted in appearance of prominent 42- and 24-base labeled products. As demonstrated previously (), these products reflect ‘accurate’ head-to-tail (42-mer) and head-to-head (24-mer) end joining by a mechanism involving annealing of the overhangs, fill-in of the resulting one-base gap, and ligation (A). Although traces of 34- to 38-base products, corresponding to small deletions in the repair joints, were detected, the accurate products consistently accounted for at least 90% of the end joining. Using various batches of cell extract, 4–20% of the initial DNA ends were accurately joined. The specific DNA-PK inhibitor KU57788 (aka NU7441) () completely blocked formation of both accurate products at a concentration of 1 μM, as did the less specific DNA-PKcs inhibitor wortmannin at a concentration of 3 μM (C). A specific inhibitor of ataxia-telangiectasia-mutated (ATM) protein, KU55933 (), had little effect at concentrations expected to completely inhibit ATM kinase activity (0.2–1 μM, D and E). However, it did partially inhibit end joining at 10 μM, a concentration expected to affect DNA-PK as well (). DMSO (solvent for all these inhibitors) had only a slight inhibitory effect (<20%) on the efficiency of end joining. These results confirm the requirement for DNA-PKcs, and for its kinase activity, in the joining of incompatible ends by human whole-cell extracts. Similar DNA-PKcs dependence was seen in nuclear extracts, except that, as reported previously (), significant end joining was only seen when extracts were also supplemented with purified X4L4 (B). However, whole-cell extracts were used for all subsequent experiments because supplemented nuclear extracts gave lower yields (2–3%) of end-joined products, and showed a trace of X4L4-dependent end joining in the absence of DNA-PKcs. This latter result, which is reminiscent of studies showing accurate end joining of gapped DSB substrates in extracts of DNA-PKcs-deficient CHO strains (,), could be explained either by the presence in human whole-cell extracts of factors whose inhibitory effect is abrogated by DNA-PK, or the presence in nuclear extracts of factors that can substitute for DNA-PK in promoting end-to-end synapsis. DNA-PKcs-supplemented M059J extracts were also capable of joining a 5′-end labeled cohesive-end substrate (A). Consistent with simple cohesive-end ligation, such joining yielded exclusively 16-base (head-head), 46-base (tail-tail), 30-base and 36-base (both head-tail) labeled fragments, which comigrated with products generated by treatment with T4 ligase (not shown). For this substrate, the incubation time was reduced to 1 h, which avoided significant loss of 5′ label but did not significantly decrease the extent of joining (data not shown). Although a trace of cohesive end joining was sometimes detected in unsupplemented extracts, DNA-PKcs addition consistently increased end joining by at least 15-fold. As with the gapped substrate, KU57788 blocked end joining at concentrations expected to inhibit DNA-PKcs, with half-maximal inhibition at ∼0.02 μM and complete inhibition at 1 μM (B and C). Again, the ATM inhibitor KU55933 had no effect at concentrations expected to inhibit ATM [ ∼ 0.013 μM, ref. ()], but did block end joining at the extremely high concentrations expected to inhibit DNA-PK as well ( ∼ 2.5 μM) (C). These results indicate that joining of DSBs with cohesive as well as incompatible ends in human cell extracts is completely dependent on catalytically active DNA-PK but not ATM. To determine whether the observed requirement for DNA-PKcs kinase activity reflects a requirement for autophosphorylation of the S/T 2609–2647 cluster, end joining was examined in M059J cells supplemented with either wild-type DNA-PKcs or DNA-PKcs proteins having either S/T → A (A6) or S/T → D (D6) substitutions at all six sites in the ABCDE cluster. To ensure consistency, all three proteins were expressed in and purified from CHO-V3 cells, and previous work showed that they all have comparable kinase activity (,). shows end joining of the 1-base-gapped substrate in extracts supplemented with these wild-type and mutant DNA-PKcs proteins. DNA-PKcs-D6, designed to mimic constitutive phosphorylation of the ABCDE cluster, supported end joining, but with markedly lower efficiency than the wild-type protein (A and B). The S/T → D mutations did not affect the fidelity of end joining, as the accurate 42- and 24-base products were formed almost exclusively. DNA-PKcs-A6 also supported end joining, but even less efficiently than the D6 mutant. Nevertheless, even at the lowest concentration of DNA-PKcs-A6, joining was significantly greater than that seen with no DNA-PKcs (0.5% versus <0.05%). Moreover, in three replicate experiments, the difference between the D6 and A6 alleles was consistently more pronounced at 8 nM than at 2–4 nM (B). The formation of accurate repair products from the gapped substrate requires two end-processing steps, gap filling and ligation. To determine the effect of the phosphorylation site mutations specifically on the gap-filling step, end-joining reactions were performed in the presence of ddTTP rather than dTTP. As expected, in the presence of wild-type DNA-PKcs, ddTTP resulted in the generation of a 15-mer, corresponding to a single-base-elongated but unligated intermediate, accompanied by a decrease in intensity of the 42- and 24-base ligation products (C). The residual 42- and 24-mer products presumably reflect gap filling by residual dTTP in the cell extracts (extracts were dialyzed for only 3 h ()). With the D6 and A6 mutant forms of DNA-PKcs, there was little if any apparent gap filling. If the extent of gap filling were comparable to that of overall end joining, the 15-mer intermediate might not be detected due to its proximity to the dominant 14-mer band corresponding to the initial substrate. In any case, however, the mutant proteins are clearly much less efficient in promoting gap filling on aligned DSB ends than is wild-type DNA-PKcs. To examine the effect of phosphorylation site mutations specifically on the ligation step, similar end-joining assays were performed with a simple 5′-end-labeled substrate bearing cohesive 4-base (CGCG-) 5′ overhangs (). The results were very similar to those obtained with the substrate requiring gap filling; that is, the D6 mutant was markedly less efficient in promoting ligation than wild-type DNA-PKcs, and the A6 mutant was even less efficient than the D6 mutant (A and B). Thus, end joining of a substrate requiring only simple religation appears to be just as dependent on the availability of phosphorylation sites in the ABCDE cluster, as end joining of a substrate requiring more complex processing. Thus, gap filling, ligation and end joining overall, show roughly the same relative dependence on ABCDE phosphorylation, consistent with the model previously proposed (,) in which phosphorylation of the ABCDE cluster promotes some conformational or other change that increases accessibility of DNA ends and thereby facilitates both gap filling and ligation. The finding that the D6 mutant promotes end joining more efficiently than the A6 mutant, is consistent with a model wherein the S/T → D mutations mimic phosphorylation of the ABCDE cluster (albeit imperfectly) and promote a similar conformational change in DNA-PKcs. To assess whether this was the only essential DNA-PK-mediated phosphorylation event, joining of both cohesive and incompatible DNA ends was examined in the presence of M059J extracts, mutant DNA-PKcs proteins and the DNA-PKcs inhibitor KU57788. With the gapped DSB substrate, KU57788 completely blocked the end joining promoted by wild-type DNA-PKcs as well as both the A6 and the D6 mutants (C and D). Similarly, KU57788 reduced end joining of the cohesive-end substrate in the presence of either mutant to the extremely low level seen in the absence of DNA-PKcs (C and D). This complete inhibition by KU57788 was confirmed in at least three independent experiments with each substrate and each of the three DNA-PKcs alleles. Thus, mimicking ABCDE phosphorylation with S/T → D substitutions was not sufficient to promote end joining in the absence of DNA-PKcs activity. These results suggest that there are target sites for DNA-PKcs-mediated phosphorylation, other than the ABCDE cluster, that are important if not essential for end joining. In assays with the substrate bearing 3′ overhangs, 3′ → 5′ exonucleolytic processing in cell extracts is also apparent. Although the nuclease responsible for this digestion has not been identified, it nevertheless provides an independent measure of DNA end accessibility. Despite some variability between different batches of cell extract, exonucleolytic processing was consistently greater with the D6 than with the A6 mutant, consistent with phosphorylation of the ABCDE cluster playing an important role in controlling DNA end accessibility (C). Nevertheless, KU57788 reduced exonucleolytic processing with all three DNA-PKcs alleles, suggesting that DNA-PK-mediated phosphorylation of sites outside the ABCDE cluster contribute to regulation of end accessibility. Several lines of evidence suggest that DNA-PKcs autophosphorylation in the ABCDE cluster plays an important role in promoting nonhomologous end joining. When nonphosphorylatable S/T → A substitutions are introduced at all six sites in the cluster, DNA-PKcs no longer stimulates ligation of cohesive ends by X4L4 (). Moreover, CHO cells harboring the same DNA-PKcs-A6 allele are even more radiosensitive than cells lacking DNA-PK entirely (). Both these results are consistent with the proposed model wherein autophosphorylation of these sites induces a conformational change in DNA-PKcs that increases accessibility of DNA ends to specific end-processing enzymes such as polymerase λ and ligase IV, thus allowing end joining to proceed to completion. However, the A6 mutant does support a low level of Artemis-mediated hairpin opening and V(D)J coding joint formation, suggesting some residual activity (though perhaps only in the hairpin cleavage step) (). The coding joints formed in cells expressing this mutant are more similar to those formed in wild-type than in DNA-PKcs-deficient cells, but show even less nucleotide loss from the ends (consistent with constitutive end sequestration), and completely lack any microhomology-based joining. Conversely, a mutant with S/T → D substitutions at the same six sites promotes ligation of cohesive ends by purified X4L4, albeit less efficiently than the wild-type DNA-PKcs, even when DNA-PK kinase activity is blocked (). When expressed in DNA-PKcs-deficient CHO cells, DNA-PKcs-D6 largely restores normal V(D)J recombination, and partially rescues radioresistance (). Both these results suggest that the S/T → D substitutions mimic DNA-PKcs autophosphorylation to a considerable extent, though perhaps not perfectly. While all of these studies strongly suggest an important role for phosphorylation of the ABCDE cluster in both V(D)J recombination and DSB repair, they do not address the issue of whether such phosphorylation is sufficient to promote these processes, or whether other sites of DNA-PK-mediated phosphorylation exist that are of equal or greater importance. These results also do not indicate whether quantitative differences between DNA-PKcs alleles in restoring radioresistance reflect qualitative differences in the types of DSBs that can be effectively processed by each allele. In intact CHO cells, complete DNA-PKcs deficiency reduces DSB rejoining by only ∼40%, whereas replacement of DNA-PKcs with DNA-PKcs-A6 reduces rejoining by ∼50% (). These results suggest that some DSBs are repaired by alternative DNA-PK-independent mechanism(s) () that are partially inhibited by DNA-PKcs-A6, probably because it fails to dissociate from unrepaired DSBs (). In the whole-cell extract assay, however, the low Mg concentration restricts end joining to the ligase IV-mediated pathway (), and thus accurate joining of the substrate requiring gap filling is absolutely dependent on DNA-PKcs (). This stringency allowed the detection of a small amount of residual accurate joining promoted by DNA-PKcs-A6 (), reminiscent of the low level of restored V(D)J recombination in cells expressing this allele (). Thus, at least in cell extracts, DNA-PKcs does have some residual activity in promoting end joining, even when phosphorylation of the ABCDE cluster is completely abolished. This residual activity, like that seen with wild-type DNA-PKcs, is still blocked by KU57788, and therefore must require DNA-PK-mediated phosphorylation of sites outside the ABCDE cluster, either on DNA-PKcs itself or on other proteins. The results suggest that once these phosphorylations occur, DNA-PKcs-A6 either dissociates from some DNA ends (), or can adopt a conformation that permits some end processing. The level of end joining promoted by DNA-PKcs-D6 in cell extracts, greater than DNA-PKcs-A6 but typically 2- to 5-fold lower than wild type (), is in reasonable agreement with its 2-fold lower activity in promoting ligation by purified X4L4 (), as well as with its ability to partially restore radioresistance (∼50%) and DSB rejoining (∼20%) to DNA-PKcs-deficient V3 cells (,). However, in previous work wortmannin had no effect on the ability of the D6 allele to promote X4L4-mediated ligation (). Thus, the finding that the more specific inhibitor KU57788 completely blocks the ability of DNA-PKcs-D6 to promote end joining in cell extracts ( and ), was unexpected. The simplest explanation for these conflicting results is that the presence of other proteins at DNA ends in cell extracts (e.g. XLF, Artemis and/or the Mre11/Rad50/NBS complex) confers a requirement for additional phosphorylation events, besides DNA-PKcs phosphorylation in the ABCDE cluster—events that are apparently dispensable in the simpler reaction with DNA-PK and X4L4 only. On the other hand, the failure of the D6 mutant to fully substitute for the wild-type protein in both and assays is not particularly surprising, and is consistent with a model wherein optimum end joining requires properly timed interconversion between phosphorylated and nonphosphorylated forms of DNA-PKcs. (There is, however, the formal possibility that S/T → D substitutions do not fully mimic phosphorylated S/T residues, and that if fully ABCDE-phosphorylated DNA-PKcs could be generated, it would promote some level of end joining in the absence of other phosphorylations.) Overall, the data confirm the importance of ABCDE phosphorylation in promoting DNA-PK-mediated end joining, but suggest that there are additional DNA-PK phosphorylation targets that are as important as ABCDE cluster, if not moreso. Histone H1 may be one such target (). Because DNA-PKcs is a large and complex protein, any comparisons among the wild-type and mutant forms of the enzyme are potentially complicated by the possibility that differences in protein stability rather than in phosphorylation status could account for the observed differences in end-joining efficiency. However, we have not seen any indication of differences in stability, and all three alleles have similar levels of kinase activity. In fact, in some assays, such as DNA end protection and X4L4 recruitment, the A6 protein (which is least active in end joining) is more active than the D6 protein (), consistent with the proposal that it more strongly binds to and sequesters the DNA end. Extreme care was taken that the three alleles of DNA-PKcs were purified, stored and aliquotted identically. Moreover, although a single batch of each form of DNA-PKcs was used in all experiments in , the previously reported differences in biochemical activities were seen consistently with at least two batches of each allele (Ramsden,D.A., unpublished data). Nevertheless, the possibility that one or both of the mutant proteins tends to spontaneously degrade in a way that compromises its end-joining activity but not its kinase activity, cannot be definitively excluded. Recently, it has been reported that in irradiated cells (10 Gy) the ABCDE cluster is phosphorylated by ATM rather than DNA-PK (). However, if such ATM-mediated phosphorylation is required for DNA repair, it is difficult to reconcile the small defect in DSB rejoining seen in ATM-deficient cells () with the much larger deficit seen in cells harboring DNA-PKcs-A6 (). The distinct lack of an inhibitory effect of KU55933 on end joining at concentrations known to block ATM kinase activity ( and ) suggests that, at least in cell extracts, the ABCDE phosphorylation that is functionally relevant for repair is carried out by DNA-PK rather than ATM. At survivable levels of damage in intact cells, DNA-PK autophosphorylation might be transient and involve only an extremely small number of DNA-PKcs molecules actually bound to DSB ends, and so might not be detected in any direct phosphorylation assays. A second ‘PQR’ cluster of five phosphorylated serines has been identified in DNA-PKcs, and in contrast to ABCDE, phosphorylation of these sites appears to decrease accessibility of DNA ends (). However, even simultaneous S → A substitutions at all five PQR sites confers only slight radiosensitivity. A threonine → alanine substitution at a newly identified C-terminal site, T3950, confers radiosensitivity intermediate between normal and DNA-PKcs-deficient cells (), but while phosphorylation at this site has been demonstrated it is not known whether the phosphorylation is catalyzed by DNA-PK. Overall, the available data suggest that functionally relevant phosphorylation occurs at a large number of sites on DNA-PKcs, and that while no single site is essential, some are more important than others. Thus, while elimination of phosphorylation throughout the ABCDE cluster (as in DNA-PKcs-A6) reduces end joining to a low level, elimination of phosphorylation at all non-ABCDE sites (DNA-PKcs-D6 plus KU57788) reduces end joining to an undetectable level ( and ). Interpretation of results with the D6 mutant is complicated by the possibility that optimal end joining may require not only that these sites be phosphorylated, but that they be phosphorylated/dephosphorylated in a specific order or at a specific stage of the end-joining pathway. Although the assays with the gapped substrate can potentially discriminate between the gap-filling step and the final ligation step, mutations in the ABCDE cluster as well as DNA-PK/ATM inhibitors appear to affect both steps about equally, consistent with a model where ABCDE phosphorylation precedes both steps. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Peptides are involved in many essential processes in signal transduction and cell-to-cell communication. Several prominent examples are the peptide hormones that bind to cell surface receptors triggering a signal cascade within the cell (), peptides involved in innate immunity (), or the quorum sensing peptides found in Gram-positive bacteria (). Since methods like phage display or mRNA display facilitate the isolation of peptides that recognize protein targets, the number of synthetic peptides with novel biological functions is steadily and rapidly growing. These include enzyme and toxin inhibitors (,), peptides with antiviral activity (,) or with the ability to specifically detect tissues with histological changes associated with diseases (,). Peptides selected in this manner mostly exert their effect by binding to a specific target site. Recently, a 16-mer peptide termed Tip (Transcription inducing peptide; A) was isolated via phage display. Tip is not only able to bind specifically to its target, the bacterial transcription factor Tet repressor (TetR), but also triggers a conformational change normally induced by tetracyclines, the natural effectors of TetR (). TetR is widely used to control regulation of gene expression in prokaryotes and, as a fusion protein with an activation domain, in eukaryotes (), because it combines high specificity for its cognate DNA sequence () with extremely sensitive induction by tetracyclines (tc), especially the potent analogues doxycycline (dox) and anhydrotetracycline (atc) (,). Non-tc inducers of TetR, like Tip, may lead to alternative inducers when used as a scaffold for peptidomimetics to generate novel, small-molecule effectors. The recently published crystal structure of the TetR–Tip complex provides a molecular basis for such efforts (). Another application of Tip is to fuse it to a protein of interest and use it to analyse target protein expression by monitoring a TetR-controlled reporter (). Although a wide variety of tag-systems suitable for diverse applications exist, Tip is the only protein tag that allows quantitative analysis of protein expression. Mandatory for this application is a high efficiency of induction to ensure that low levels of Tip fusion proteins can be detected. We demonstrate here that addition of an aromatic residue between C-terminally fused Tip and TrxA along with mutations in the effector-binding pocket of TetR lead to a strong increase of reporter induction. Furthermore, the introduced TetR mutations render it insensitive to induction by tetracyclines, thereby creating exclusive specificity for Tip. Chemicals were obtained from Merck (Darmstadt), Sigma (Munich) or Roth (Karlsruhe) and were of the highest purity available. Media, buffers and solutions were prepared with Millipore water or deionized water and autoclaved. Heat labile substances were dissolved and filtered with a sterile filter (0.2 µm). Enzymes for DNA restriction and modification were obtained from New England Biolabs (Frankfurt/M.), Roche Diagnostics (Mannheim), Invitrogen (Karlsruhe) or Fermentas (St. Leon-Rot). Sequencing was carried out according to the protocol provided by PE Applied Biosystems (Weiterstadt) for cycle sequencing. All bacterial strains are derived from K12. Strains DH5α ( (r m), 1, 1, 96, 1, 44, ϕ80dΔM15, Δ()U169) () and RB791 (IN[]1, L) () were used for general cloning procedures. Strain DH5α(λ50) (Tn transcriptional fusion) (,) served as host strain for β-galactosidase assays. The plasmids pWH1200 (), pWH527- and pWH1413-derivatives for TetR variant expression (), pWH806 (), pWH2100-, pWH2200- and pWH2300-derivatives for C- and N-terminal TrxA/SbmC-XTip variant expression () were used in the studies. Class B TetR () was used in all experiments. Insertion of Met into the C-terminal TrxA-Tip fusion was done by standard 2-step PCR using pWH2101 as template, the mutagenic primer 5′-GTCGGGTGGAGCTTGGACTTGGAATG-3′ to introduce the Met codon (underlined) and the flanking oligonucleotides 5′-TGACAATTAATCATCGGCTCG-3′ and 5′-AAGGAATGGTGCATGCCTGC-3′. The product obtained was restricted with HindIII and PstI and cloned into likewise-digested pWH2101 (). The resulting plasmid was named pWH2102. Randomization of the Met at position −1 in TrxA(C)-MTip was performed by combined chain reaction (). The phosphorylated oligonucleotide 5′-GTCGGGTGGAGCT(NNK)TGGACTTGGAATG-3′ introducing the randomized codon (NNK; N = A, C, G or T; K = G or T) and the flanking oligonucleotides 5′-TGACAATTAATCATCGGCTCG-3′ and 5′-AAGGAATGGTGCATGCCTGC-3′ were used in a standard PCR reaction with pWH2102 as template. The resulting DNA was restricted with HindIII and PstI and cloned into likewise-digested pWH2102. Candidates were identified by sequencing and named pWH2102-X with X denoting the amino acid replacing the M. The construction of a plasmid-encoded SbmC(C)-Tip fusion was done by amplification of the fragment encoding the fusion from genomic DNA from BW25113(λ50)(C)-, a strain containing the linker and Tip-encoding sequence as used in the TrxA fusion fused to the 3′ end of the gene at its position in the genome (unpublished data) with the primers 5′-ATGATGAACTACGAGATTAAGCAGGAAGAGAAACG-3′ introducing a restriction site for XbaI (underlined) and 5′-GTAGTTTTACGAACCTCCACCACTAGGAGC-3′ introducing a restriction site for SphI (underlined). The product was restricted with XbaI and SphI and cloned into likewise-digested pWH610 (). The resulting plasmid was named pWH2301. Insertion of Met (underlined) into the C-terminal SbmC-Tip fusion protein was done genetically by standard 2-step PCR using pWH2301 as template, a mutagenic primer 5′-CTCGGGTGGAGCTTGGACTTGGAATG-3′ and the flanking primers 5′-ATGATGAACTACGAGATTAAGCAGGAAGAGAAACG-3′ and 5′-GTAGTTTTACGAACCTCCACCACTAGGAGC-3′, introducing restriction sites (underlined) for XbaI and SphI, respectively. The product obtained was cleaved with both enzymes and cloned into likewise-digested pWH610 (). The resulting plasmid was named pWH2301-Met. The F86A mutation was introduced into -N82A encoded by pWH527-N82A via standard 2-step PCR using a mutagenic primer 5′-CGTAATAACGCTAAAAGTAGATGTGCTTT-3′ and the two flanking primers 5′-CTCGACATCTTGGTTACCG-3′ and 5′-CGCCGTACTGCCCGCTTGG-3′. The resulting double mutant was cloned via restriction with NcoI and XbaI into pWH527 resulting in low-level constitutive expression of the TetR mutant (). Repression and inducibility of TetR was assayed in DH5α(λ50) (). The strain was transformed with pWH527-derivatives expressing TetR variants at a low level or pWH1413-derivatives expressing TetR at a higher level and with plasmids from the pWH2100/2200/2300 series encoding the different C- and N-terminal Tip fusions to TrxA or SbmC. Overnight cultures and log-phase cultures were grown at 28°C in LB medium supplemented with 100 µg/ml ampicillin and either 60 µg/ml kanamycin for pWH527-derivatives or 25 µg/ml chloramphenicol for pWH1413-derivatives. For this purpose, stationary phase cultures were diluted 1:100 in fresh medium and expression of the fusion proteins induced using 60 µM IPTG, if not indicated otherwise. The cells were then grown to an OD600 of ∼0.4 and their β-Gal activities determined as described by Miller (). Three independent clones were assayed for each combination of constructs and experiments repeated at least twice. The values obtained were normalized to the maximal β-Gal activity in the absence of TetR which was set to 100% and typically varied from 6500 to 7500 Miller units in the individual experiments. DH5α(λ50) was transformed with the derivatives listed in the respective β-Gal assays and grown under the same conditions as stated there. Cells were harvested at an OD600 of 0.4. The crude lysates were prepared by sonication and centrifugation. Ten microgram of crude lysate of each construct was loaded either on a 14% (TrxA-XTip fusions) or a 10% SDS-PAA gel (TetR variants) and electrophoresed. Proteins were transferred (120 mA, overnight at 4°C) to a PVDF membrane (BIORAD) in a Mini V8.10 blotting apparatus (Gibco-BRL) using 10 mM NaHCO, 3 mM NaCO and 20% (v/v) methanol as transfer buffer. After blocking with 0.2% I-Block™ (TROPIX) in phosphate-buffered saline (75 mM Na-phosphate; 68 mM NaCl; pH 7.8) with 0.1% Tween 20, membranes were treated either with a monoclonal anti-TrxA antibody (TrxA-XTip fusions, Anti-Thio™ antibody, Invitrogen) or a polyclonal TetR-specific antibody (SA1851; lab stock). Signals were detected with anti-mouse IgG (TrxA-XTip fusions) or anti-rabbit IgG (TetR variants) conjugated to horseradish peroxidase (Amersham Pharmacia) and the ECL+ kit (Amersham Pharmacia) following the manufacturer's instructions. A shows the relevant features of the screening strain used in this study to score the efficiency of TetR induction by TrxA fusions with Tip. This strain has been described in detail (). It contains one plasmid encoding TetR or its variants and LacI. TetR represses a chromosomally located , while LacI controls expression of the respective TrxA-Tip fusion protein. The latter is encoded on a second, compatible plasmid giving rise to IPTG-inducible β-Gal expression mediated by TrxA-Tip. For our studies, Tip was fused C- or N-terminally to TrxA, and it had been observed that the latter is a much better inducer of TetR than the former () (see B and C). A potential reason might be a sequence difference in Tip resulting from the failure of cleaving the initiator amino acid M from the N-terminus in when the following residue is W (see B) (,). This M residue is missing in the C-terminal fusion. To examine this possibility we inserted an M residue into the C-terminal TrxA(C)-Tip fusion at the corresponding position termed −1 (see B). This insertion leads to a nearly 2-fold increase in β-Gal activity, but still reaches only half of the β-Gal activity obtained with the N-terminal TrxA(C)-Tip fusion (see C). The steady-state protein levels of all three TrxA fusion proteins are similar as concluded from Western blots (see D) confirming that the induction levels are intrinsic properties of the fusions. We next asked if M is the optimal amino acid at position −1 for induction of TetR by randomizing this residue. Western blot analyses were carried out with the resulting 19 TrxA(C)-XTip variants to make sure their expression levels are the same. Only TrxA(C)-PTip showed a reduced steady-state level while all other variants were present in similar amounts as TrxA(C)-MTip and TrxA(C)-Tip (data not shown). The TrxA(C)-XTip variants were then scored for their induction efficiencies in the screening strain (see A). The results are summarized in . All activities of the TrxA(C)-XTip fusions are presented relative to 100% β-Gal activity defined by a strain lacking TetR. TrxA(C)-XTip variants with P, charged residues like D, E, K or R or with aliphatic, hydrophobic residues like V, L or I display either no or only marginal induction of TetR. All other amino acids at this position yield fusion proteins that induce TetR. They are either less active (Y), as active (S, T, C, F) or more active (G, A, N, Q, H, W) than the one with M. X-ray crystallography of TetR complexed with a 16-mer Tip oligopeptide acetylated at the N-terminus revealed details of the location of Tip inside the tc-binding pocket. The C atom of the acetyl moiety is in close proximity to residues N82 and F86 of TetR (). shows an excerpt of the structure highlighting this proximity. We assumed that a large residue at this position, as found in several active C-terminal TrxA(C)-XTip variants, might lead to sterical hindrance. Therefore, we replaced the bulky N82 and F86 residues with a smaller A residue and scored the induction properties of these mutants with TrxA(C)-MTip. The results are shown in . Even in the absence of IPTG, in which only basal expression of TrxA(C)-MTip from the leaky promoter occurs, we observed induction of TetR-N82A and TetR-F86A. However, TrxA(N)-Tip is still the more efficient inducer under these conditions. At the higher, IPTG-induced, expression level of TrxA(C)-MTip, induction of TetR-N82A reaches the same level as with TrxA(N)-Tip and wt TetR. A slightly smaller maximal induction by TrxA(C)-MTip is observed for TetR-F86A. As expected, altering the size of the binding pocket in TetR-N82A and TetR-F86A does not lead to increased induction by TrxA(C)-Tip lacking the additional M residue. The induction properties of TrxA-Tip could depend in part on the TrxA portion of the fusion protein. To address this question, we constructed Tip and MTip fusions to the C-terminus of the protein SbmC () and replaced the TrxA-Tip fusion in our screening strain by them. SbmC is a small, soluble protein of 157 residues with solvent-exposed and flexible N- and C-termini (). It is part of the SOS regulon and involved in reducing DNA damage (). Induction of TetR by SbmC(C)-MTip is slightly more efficient than induction by SbmC(C)-Tip. In contrast, TetR-N82A is induced to a 5-fold higher level of β-Gal activity by SbmC(C)-MTip compared to SbmC(C)-Tip as shown in . This result establishes an only marginal influence of the carrier protein on Tip activity. Hence, the improved induction of the TetR variants by the elongated Tip results solely from their effects on Tip–TetR interaction. We wondered if the induction efficiency profiles of the TrxA(C)-XTip variants depend on the shape of the inducer-binding pocket. In addition to the TetR-N82A and -F86A mutants, we constructed the double mutant TetR-N82A-F86A and introduced that also into the screening strain. The Western blot shown as insert in indicates that TetR and TetR-F86A are expressed to about the same level, while TetR-N82A and TetR-N82A-F86A are present in roughly half the amount of the former variants. The induction activities were scored without IPTG at the basal expression levels of the TrxA(C)-XTip variants to increase the sensitivity and are shown in . TrxA(C)-Tip shows about the same induction for TetR-N82A and TetR-F86A (∼2% β-Gal activity), but the double mutant appears slightly de-repressed under these conditions (∼4%). No other TrxA(C)-XTip variant induces TetR-N82A to a significantly higher degree than TrxA-MTip, and only TrxA(C)-FTip and TrxA(C)-WTip reach about the same induction efficiency (see the white columns in ). In contrast, TetR-F86A is very efficiently induced by TrxA-XTip variants with an aromatic residue. While TrxA-MTip shows only ∼6% β-Gal activity, TrxA(C)-HTip and TrxA(C)-YTip lead to ∼35% β-Gal activity, TrxA(C)-WTip gives rise to 53% and TrxA-FTip even to 60% β-Gal activity. The latter variant is hence the most efficient inducer of TetR-F86A. Similar to TetR-F86A, we observed the highest induction activities for TetR-N82A-F86A by TrxA-XTip derivatives with aromatic residues, albeit in a slightly changed order: TrxA-YTip exhibits the same induction efficiency for both TetR variants, while TrxA-HTip, TrxA-FTip and TrxA-WTip are better inducers for the double than for the single TetR mutant. Taken together, it appears that TrxA(C)-FTip and TrxA(C)-WTip are the best inducers for all three TetR variants, but their efficiencies increase from TetR-N82A via TetR-F86A to TetR-N82A-F86A. It is also noteworthy to mention that these induction profiles do not correlate generally to the differences in expression levels established for these three TetR variants. The TetR residues N82 and F86 contact tc as derived from the crystal structure of the TetR-tc complex (). Hence, we tested whether the three TetR mutants in which one or both of these residues are replaced by A are inducible by tc, atc or dox, three widely used inducers of TetR. The results are shown in . TetR-F86A is not induced by tc, but atc and dox are very efficient inducers. In contrast, TetR-N82A and TetR-N82A-F86A are not induced by any of these compounds. Since these two are best induced by Tip we have created functionally novel TetR mutants, which will not respond to the most commonly used inducers atc and dox anymore, but instead are highly sensitive to Tip. TrxA(N)-Tip exhibits a much higher activity as inducer of TetR than TrxA(C)-Tip (). We show here that part but not all of this activity difference is due to the fact that an additional M residue is present when Tip is N-terminal, owing to the fact that removal of the initiator M residue in is very inefficient before a W residue (). The crystal structure of the TetR–Tip complex () reveals that residues W1 to N4 of Tip are located in the tc-binding pocket, thereby piercing the TetR core. We assume that an N-terminally fused Tip may enter the inducer-binding pocket more easily because the interacting residues can enter in a head first orientation. This is not possible for a C-terminal fusion of Tip, and hence, may contribute to the lower induction efficiency. The M residue in position −1 of Tip is not the only, nor the optimal residue for increased induction of TetR. Indeed, randomizing the residue at this position in TrxA(C)-XTip revealed that several other amino acids also lead to an increase in TetR induction. The better induction seen with small residues like G, A or S could simply be interpreted as an extension of the linker sequence separating Tip from the TrxA scaffold, thereby increasing the flexibility of Tip. This simple explanation does not hold true for the increased induction seen when aromatic residues occupy position −1, as TrxA(C)-FTip is as active as TrxA(C)-MTip, and TrxA(C)-WTip and TrxA(C)-HTip are even slightly more active. Furthermore, TrxA(C)-DTip, TrxA(C)-ETip, TrxA(C)-KTip or TrxA(C)-RTip carrying charged residues exhibit no induction of TetR, while the uncharged, but similarly sized residues TrxA(C)-NTip and TrxA(C)-QTip are very good inducers of TetR. According to the crystal structure of the Tip–TetR complex, the N-terminus of Tip is located in a predominantly hydrophobic environment formed by residues L60, F67, F86, L90 and L142 of the tc-binding pocket. Placing a charged residue in the hydrophobic core of a protein should be very unfavourable (,). Hence, we conclude that the residue at position −1 must exert its function via contacts with the inducer-binding pocket of TetR. There is a clear distinction among the hydrophobic amino acids between aromatic and aliphatic residues. C-terminal TrxA(C)-XTip variants with V, L or I show a decrease in activity compared to M, while variants with H, F, Y or W induce TetR more efficiently. Comparison of the contacts in the TetR inducer-binding pocket to tc and the first five residues of Tip () shows that most of the key interactions for tc (H64, N82, F86, H100, P105) are not formed with Tip (). In particular, the space occupied by the A-ring of tc appears to be empty in the Tip–TetR complex. It is thus conceivable that the aromatic residues mimic the A-ring structure of tc best, followed by the flexible M residue with the easily polarizable sulphur, while the large hydrophobic residues fit least well. and display the proximity of the N-terminus of Tip to the residues N82 and F86 in the inducer-binding pocket of TetR. The replacement of these bulky residues by the much smaller A in single and double exchange mutants of TetR results in a dramatic increase of the induction activities of several TrxA(C)-XTip variants. These gain of function mutants reinforce the hypothesis that the residue at position −1 in TrxA(C)-XTip is located in the inducer-binding pocket facing the altered residues of the TetR variants. This conclusion is highlighted by the observation that TrxA(C)-XTip variants show different activity profiles for induction of the TetR-N82A, -F86A and -N82A-F86A mutants. We assume that this reflects the slightly different interactions of the various residues with the mutated tc-binding pockets. The increase of induction efficiency of the TrxA(C)-XTip TetR variant pairs is indeed dramatic because nearly complete induction is obtained without having to relieve the LacI-mediated repression of TrxA(C)-XTip expression. Previous studies of the expression levels of TrxA(C)-Tip have revealed that the fusion protein is barely detectable in Western blots in the absence of IPTG (). Since we also demonstrate that the induction efficiencies are not confined to TrxA fusions, we have created much more sensitive tools for monitoring the expression levels of proteins in general as compared to the one used previously (). Furthermore, two of the TetR mutants are not inducible by tc, atc or dox anymore, allowing their use in conjunction with tc inducible expression systems (). Taken together, the data presented here suggest that proteins carrying a modified Tip as a C-terminal protein tag in combination with a reporter system under control of a TetR mutant allows highly sensitive detection of their expression levels. Inserting such a tag at the C-terminus of a protein is of great advantage over the N-terminus as one can expect less interference with translation initiation ().

Human immunodeficiency virus type 1 (HIV-1) nucleocapsid protein (NC) is a small, basic, nucleic-acid-binding protein having two zinc fingers connected by a short, basic amino acid linker. Each finger contains the invariant CCHC metal-ion-binding motif (). NC binds non-specifically to the phosphodiester backbone of nucleic acids (), but also exhibits sequence-specific binding at sites with runs of Gs or T/UGs (). In addition, NC is a nucleic acid chaperone and is able to catalyze nucleic acid conformational rearrangements that lead to formation of the most thermodynamically stable structures () (reviewed in 1–3,). The chaperone function has two independent activities (): aggregation of nucleic acids, localized primarily to the N-terminal basic residues (); and weak destabilization of duplex molecules, associated with the zinc fingers (). The nucleic acid chaperone activity of NC is required for efficient and highly specific DNA synthesis. Indeed, NC plays an important role in almost every step in reverse transcription including the minus-strand (,,) and plus-strand (,) transfer events that are mandatory for synthesis of full-length minus- and plus-strand DNAs and formation of the long-terminal repeats present at the ends of proviral DNA. In minus-strand transfer, the initial DNA product of reverse transcription, known as (−) strong-stop DNA [(−) SSDNA], is translocated to the 3′ end of the viral genome (acceptor RNA) in a reaction mediated by base pairing of the complementary repeat (R) regions at the 3′ ends of the DNA and RNA molecules (,). For HIV-1, R consists of 97 nucleotides (nt). The first 59 nt in acceptor RNA and (−) SSDNA form highly stable, complementary stem-loop structures, which are referred to as transactivation response elements (TAR) RNA and TAR DNA, respectively. NC stimulates HIV-1 minus-strand transfer (3 and references therein) by transiently destabilizing the TAR structures (,,,). Destabilization of these structures promotes annealing of TAR RNA to TAR DNA (,,,,) and inhibits a competing self-priming reaction at the 3′ end of (−) SSDNA (). Loop–loop interactions have been shown to be critical for dimerization and packaging of retroviral RNA () as well as for the formation of kissing complexes containing TAR RNA (). Indeed, nucleation of the NC-catalyzed annealing step in minus-strand transfer was proposed to occur through interaction between the apical loops of TAR RNA and TAR DNA (,). However, an alternative proposal suggested that nucleation proceeds through destabilization of the 3′ and 5′ stem termini (). Based on single-molecule FRET experiments () and a detailed kinetic study of NC-promoted annealing of mini-TAR constructs (), it was also proposed that multiple pathways might be involved. More recent studies with full-length TAR suggest that a zipper mechanism involving the lower stems and bulges is the major nucleation pathway for TAR annealing in the presence of NC () (Vo, M.-N., Rouzina, I. and Musier-Forsyth, K., in preparation). The structure and thermostability of the nucleic acid intermediates are major determinants for NC-facilitated minus-strand transfer. A number of studies have emphasized the importance of maintaining the bulges in the TAR DNA structure of (−) SSDNA (,) as well as the critical role of acceptor RNA structure (,,). Interestingly, minus-strand transfer is more sensitive to the thermostability of acceptor RNA than to the stability and structure of (−) SSDNA (,). These findings are consistent with NC's weak destabilizing activity (see above) and led to the conclusion that efficient minus-strand transfer requires a delicate thermodynamic balance between the structures of (−) SSDNA and acceptor RNA and the stability of the annealed RNA–DNA hybrid (). In the present study, we have elucidated the paradoxical activity of two acceptor RNAs (RNA 70 and RNA 50) in minus-strand transfer: Despite the fact that RNA 70 has a higher predicted overall free energy of folding (Δ) than RNA 50, more transfer product is synthesized with the RNA 70 acceptor than with RNA 50 in assays with the same (−) SSDNA (). Based on extensive mutational analysis, we demonstrate for the first time that the local structure of acceptor RNA at potential nucleation sites, rather than overall thermodynamic stability, is a crucial determinant for NC chaperone activity during the minus-strand transfer step of reverse transcription. We also show that in our reconstituted system, NC-mediated annealing is more efficient than strand transfer (i.e. annealing plus reverse transcriptase (RT)-catalyzed elongation of minus-strand DNA). Since RT activity (but not annealing) requires Mg (), it seemed likely that the lower values for strand transfer might be due to the presence of a high concentration of Mg in strand transfer reactions. In fact, experiments presented below strongly suggest that Mg competes with NC for binding to the negatively charged phosphodiester backbone in (−) SSDNA and acceptor RNA. Finally, we present data indicating that for our system, destabilization of a secondary structure formed by the 5′ TAR RNA sequence and to a lesser extent, loop–loop interactions between TAR RNA and TAR DNA contribute to efficient minus-strand transfer. T4 polynucleotide kinase and proteinase K were obtained from Ambion Inc. (Austin, TX, USA). [γ-P]ATP (3000 Ci/mmol) was purchased from GE Health Sciences. HIV-1 RT was obtained from Worthington. HIV-1 NC was a generous gift from Dr Robert Gorelick (SAIC Frederick, Inc., NCI-Frederick, Frederick, MD). xref sup table-wrap fig #text xref italic #text xref sup italic #text sup xref #text In a previous study of NC chaperone activity and minus-strand transfer (), a series of acceptor RNAs truncated in U3, the 3′ region of R, and TAR, were assayed with (−) SSDNAs having comparable truncations in complementary sequences, except that a portion of U5 rather than U3 was deleted. In general, in assays with the same DNA, acceptor RNAs with low predicted free energies of folding had more strand transfer activity than more highly structured acceptors. However, we found one striking exception when two acceptor RNAs, RNA 70 and RNA 50, were assayed with the DNA 50 (−) SSDNA (see below). To determine whether this finding could lead to new insights regarding the mechanism of nucleic acid chaperone activity in minus-strand transfer, we used the model system illustrated in . The figure shows annealing of RNA 70 (A) or RNA 50 (B) to DNA 50 as well as the 20-nt U3 RNA sequence, which serves as the template for RT-catalyzed extension of DNA 50 to a 70-nt product. We have now confirmed the original observations and show the data in , for ease in following the experiments presented below. The predicted overall thermodynamic stability of RNA 70 (Δ22.9 kcal/mol) is significantly higher than the predicted stability of RNA 50 (Δ14.9 kcal/mol) (A) (,). Yet when these RNAs were each assayed with DNA 50 (B), RNA 70 exhibited a much higher level of strand transfer activity than RNA 50 (C). In addition, NC had very little effect on RNA 70 activity (C). In an attempt to understand this result, we note that the DNA–RNA hybrid formed by RNA 70 has 50 bases complementary to DNA 50 and is therefore more stable than the RNA 50–DNA hybrid, which has only 30 complementary bases (). In addition, RNA 70 and RNA 50 are folded differently, since RNA 70 has almost the entire TAR stem loop (nine bases from the 3′ end are missing), while RNA 50 has only the 5′ half of the TAR structure. Consequently, the local structures formed by their 5′ sequences (A, boxed residues) also differ. Examination of the DNA 50 structure (B) shows that there is an 11-nt single-stranded region at the 3′ end. This region is likely to be the nucleation site for the annealing reaction, since it has no secondary structure that might interfere with annealing to the complementary bases at the 5′ ends of the TAR sequence in each acceptor RNA (A). The predicted thermostabilities of the RNA 70 and RNA 50 local structures (A, boxed residues) are very similar (,). However, destabilizing the respective structures to allow initial formation of the RNA–DNA hybrid has different consequences for each RNA: With opening of the local structure in RNA 70, the full complement of 11 bases becomes available for annealing to the 11-nt sequence in DNA 50. In contrast, only seven bases become available with destabilization of the RNA 50 local structure. The remaining 4 nt are base paired and melting this additional stem could reduce the nucleation efficiency. This situation would account for the fact that the RNA 50 reaction is NC dependent, whereas the RNA 70 reaction is not (C). To test this hypothesis, our strategy was to make mutations that either stabilize or destabilize the local structure of the acceptor RNAs and then to measure the effect of the mutations on minus-strand transfer. The mutations we chose were ones that retain the original RNA 50 fold (A), thereby allowing a direct comparison of mutant and wild-type activities. A illustrates these changes and also gives the predicted overall Δ values for each mutant. The RNA 50 structure was destabilized in two ways: (i) by creating a mismatch (G46U); and (ii) by changing a G-C bp to a G-U wobble pair (C49U). We also made a more stable mutant of RNA 50 by changing the G-G mismatch to a G-U wobble pair (G48U). Since the RNA 50 mutants have the same fold as the wild-type (WT) RNA (data not shown), an increase or decrease in predicted Δ values can be attributed to increased or decreased thermostability induced by the changes in local structure. Thus, mutants RNA 50G46U and RNA 50C49U (destabilizing mutations) have lower predicted Δ values than WT RNA, while RNA 50G48U (stabilizing mutation) has a higher predicted Δ value than WT (A). Note that we did not make mutations in the upper stem containing the last 4 bases (nt 28–31) that are annealed to the 11 base sequence in DNA 50, since FOLD analysis (,) revealed that most of the potential mutations in this region would result in structures that differed from RNA 50 in their overall fold. To evaluate the behavior of the three mutated RNAs, we assayed RNA 50 and each of the RNA 50 mutants for strand transfer activity with DNA 50 (B), as a function of increasing concentrations of HIV-1 NC (B and C). [Note that the lane numbers for the gel (B) are given below the individual bars in the bar graph (C).] In contrast to RNA 70 (C; also, see below), RNA 50 strand transfer activity was stimulated by NC in a dose-dependent manner (B and C). At the highest concentration of NC (0.88nt/NC [1.2 µM]), stimulation was 3.5-fold (compare lanes 1 and 5). Both of the destabilizing mutants, RNA 50G46U and RNA 50C49U, showed significantly increased transfer activity in the absence of NC (lanes 6 and 11, respectively), but addition of increasing concentrations of NC resulted in only negligible stimulation of this activity (lanes 6–10 and lanes 11–15). In reactions with the stable mutant, RNA 50G48U, strand transfer activity in the absence of NC was reduced by 2.5-fold compared with that of RNA 50 (compare lane 1 with lane 16). However, NC stimulated the strand transfer activity of this mutant by 5-fold at the highest NC concentration (compare lanes 16 and 20). To further investigate whether local structure is a key determinant of NC chaperone activity, we also made three stabilizing mutations in RNA 70 (A). It was of interest to determine the effect of these mutations on minus-strand transfer efficiency and in particular, to see whether mutant reactions would become dependent on NC concentration. Here too, our strategy was to test only mutants that have the WT RNA 70-fold. Two point mutations were constructed by changing two G-U wobble pairs to G-C base pairs (bp) (RNA 70U28C or RNA 70U30C). We also made a double mutant by changing both U28 and U30 to C (RNA 70U28,30C), thereby creating two new G-C bp to give a total of four. As expected, the RNA 70 mutants, which have increased stability of local structure, also have higher predicted overall Δ values than WT (, see inserts in each panel). The time course of minus-strand transfer with acceptor RNA 70 and the three mutants was measured with DNA 50 (B) in the presence or absence of NC (). Examination of the end point value showed that strand transfer activity was very efficient with RNA 70 (C and A). In fact, ∼70% of the (−) SSDNA was converted to the transfer product at 60 min. Moreover, the extent of the reaction was independent of NC concentration (A). However, a change of a single G-U wobble pair to a G-C bp at positions U28 or U30 (RNA 70U28C (B) and RNA 70U30C (C), respectively) resulted in a reduced level of activity. In the absence of NC, synthesis of the transfer product at 60 min was decreased by 3.4 (U28C)- and 2.4 (U30C)-fold compared with the end point value for RNA 70 (compare B and C with A). Interestingly, in both cases, NC enhanced strand transfer (by 2- and 1.5-fold, respectively, with 0.88 nt/NC [1.4 µM]) (B and C). With the double mutant, strand transfer activity was reduced to almost background level in the absence of NC (D). In this case, where four G-C bp need to be destabilized, NC had a relatively strong stimulatory effect, producing a 4.6-fold enhancement of the end point value with 0.88 nt/NC (D). Despite NC stimulation of strand transfer with mutant RNA 70 acceptors, in each case the highest level of activity at 60 min was still lower than that observed with WT RNA. The rates of minus-strand transfer in the absence or presence of NC were also determined in reactions with each of the RNA 70 acceptors (). With WT RNA 70, the rates were essentially the same regardless of whether NC was present (, line 1). The rates for the two RNA 70 point mutants were dramatically reduced in the absence of NC, compared with RNA 70: 8-fold for RNA 70U28C and 7-fold for RNA 70U30C (, first column, compare line 1 with lines 2 and 3). However, a modest stimulation of the rates (2.6 (U28C)- and 3 (U30C)-fold) was observed with increasing concentrations of NC (, lines 2 and 3). The rate of minus-strand transfer with the double mutant was reduced to almost background level in the absence of NC (, first column, compare lines 1 and 4). With addition of NC, there appeared to be no detectable enhancement of the initial rate (, line 4). This could be due to the comparatively stable stem loop in the mutant structure, which NC initially might have had difficulty melting. Interestingly, when we compared the data for the early experimental time points, we found that at the highest NC concentration, the ratio of the values with and without NC increased significantly over a 10-min time interval (2.1 at 1 min; 3.3 at 3 min; ∼5 at 5 min; and ∼6 at 10 min) (D). This stimulation of strand transfer was higher than that observed with the single mutants (constant ratios ∼3 or 4 over 10 min). Collectively, the data presented in and strongly support our hypothesis that local structure at the nucleation site determines the efficiency of NC chaperone activity during minus-strand transfer. Thus, stabilizing mutations reduce strand transfer activity and increase NC dependence. Moreover, the larger the number of bp in the duplex that must be destabilized, the greater the effect of NC. In contrast, destabilizing mutations increase activity and are not greatly affected by addition of NC. It was also of interest to determine whether changes in the loop residues of RNA 70 would affect strand transfer. Although RNA 70 does not have the complete 3′ stem sequence of TAR, Fold analysis (data not shown) indicated that the TAR loop structure is retained (A). We made a loop mutant by changing G53 and G54 to two A residues (see highlighted residues and arrows in A), thereby replacing two G-C bp normally formed between bases in the RNA and DNA loops with two A-C mismatches. This mutation did not have a significant effect on the overall folding of the RNA (data not shown) and Fold analysis (,) showed that the predicted Δ value for the mutant (−23.4 kcal/mol) is very close to the predicted Δ value for RNA 70 (−22.9 kcal/mol; A). Interestingly, like WT RNA 70, both the rate and extent of strand transfer observed with the loop mutant were not influenced by the addition of NC (, lines 1 and 5; compare A with A). In reactions containing DNA 50, the rate of strand transfer with the loop mutant was the same as that with WT RNA (, compare lines 1 and 5). Nevertheless, the end point value of the mutant reaction was reduced by almost 3-fold (compare A with A), presumably due to the two mismatches in the final product. The data on the extent of strand transfer indicate that interaction of residues in the apical TAR loops in acceptor RNA and (−) SSDNA contribute to the stability of the transfer product. However, the lack of an effect on the rate suggests that the TAR loop is not the critical nucleation site in this system. When the assay was performed with the loop mutant and a DNA 50 compensatory mutant, DNA 50C17,18T (see highlighted bases and arrows in B), the rate was not significantly changed (, compare lines 5 and 6). Moreover, the extent of strand transfer showed only a modest increase (∼1.4-fold) compared with the value obtained with WT DNA 50 (compare B with A). This shows that creation of two A-T bp in place of the A-C mismatches had only a small effect on increasing the stability of the final product. To gain a further understanding of the mechanism of NC chaperone activity during minus-strand transfer, it is important to know how changes in local structure affect the kinetics of annealing. The annealing reaction leads to formation of an RNA-DNA hybrid containing complementary R sequences in acceptor RNA and (-) SSDNA (). In this study, the experiments were performed with WT RNA 70 and RNA 70 mutants (). The results are expressed as the percentage of P-labeled DNA 50 annealed in the reaction. A shows the kinetics of annealing with WT RNA 70. In contrast to the results obtained for strand transfer (A), NC had a small stimulatory effect on annealing. Thus, the rate was increased by 3-fold with the highest concentration of NC (, line 1), although the extent of annealing at 30 min was only minimally increased relative to the minus NC value. When the loop mutant was assayed (B), NC significantly stimulated the rate of annealing, i.e. by 3-fold with 3.5 nt/NC (0.3 µM) and by almost 6-fold with 0.88 nt/NC (1.4 µM) (, line 4). The end point values were 1.4- to 2-fold higher than the value for the minus NC control (B). Thus, the annealing data for the loop mutant also differ from the strand transfer results with respect to an NC effect (A). However, despite the stimulatory effect of NC, both the rate and extent of annealing of the loop mutant were lower than the corresponding WT values (, compare lines 1 and 4; A and B), most likely as a result of the two mismatches. In other experiments, we found that in reactions with the loop mutant and the DNA 50 compensatory mutant, the rate and extent of annealing were similar to the values obtained with WT DNA 50 (data not shown). The annealing kinetics were also analyzed with the 5′ stem-loop mutants: RNA 70U28C and the double mutant RNA 70U28,30C. In the case of the single mutant, the rate of annealing was lower than that of WT, both in the absence and presence of NC (10-fold, minus NC; ∼3-fold, plus NC) (, compare lines 1 and 2). As might be expected for a more stable mutant, NC stimulated the rate of annealing to DNA 50: by 5-fold (3.5 nt/NC, 0.3 µM) and 11-fold (0.88 nt/NC,1.4 µM). In accord with our observations on the rates of annealing, the end point values were also lower than the WT values (5-fold, minus NC; ∼1.4-fold, plus 3.5 nt/NC). Surprisingly, with 0.88 nt/NC, the extent of annealing was virtually the same as that of WT (compare A and C). With the double mutant, annealing was extremely inefficient in the absence of NC and both the rate and extent of annealing were barely detectable (D). In the presence of NC, there was a modest increase in the rate of annealing, but the actual values at each NC concentration were ∼5-fold lower than those measured for WT RNA (, compare lines 1 and 3). Interestingly, the extent of annealing was approximately the same as that achieved by the single mutant at the highest NC concentration and here too, the value was very close to that of WT (compare A with D). Taken together, the data in and show that for the WT and the mutants, there was a stimulatory effect of NC on annealing. This was true to a lesser extent, even for RNA 70 and the loop mutant, which did not exhibit such an effect in the strand transfer reactions ( and ). In addition, in the case of the stem-loop mutants whose local structures are significantly more stable than that of WT RNA 70 (), annealing was more efficient than strand transfer under our experimental conditions. The apparent increased efficiency of the annealing reaction with the 5′ stem-loop mutants compared with their activity in strand transfer was unexpected. We wondered whether the absence of RT, dNTPs, and 7 mM MgCl in the annealing reaction mixtures was responsible for this result. Addition of RT or dNTPs, either singly or in combination, to annealing reactions with RNA 70, RNA 70G53,54A, or RNA 70U28,30 did not change the level of annealing (data not shown). In contrast, when increasing concentrations of Mg were added, there was an effect on RNA–DNA hybrid formation (). With RNA 70, addition of 7 mM Mg in the absence of NC (, lane 2), NC without Mg (lane 3) and NC plus Mg (lanes 4–8), resulted in succeeding small, but reproducible increases in the efficiency of annealing compared with that of the minus NC reaction (lane 1). Interestingly, the stimulation observed with NC and Mg was independent of the actual Mg concentration (lanes 4–8). A similar pattern was observed with the RNA 70 loop mutant, except that the absolute values were reduced as a consequence of the mutation (lanes 9–16). In contrast, the double stem mutant (lanes 17–24) behaved quite differently from the other two RNA 70 acceptor RNAs. In this case, annealing activity was extremely low in the absence of NC and Mg (lane 17) as well as in the presence of 7 mM Mg without NC (lane 18). NC markedly stimulated annealing in the absence of Mg (lane 19; D), but as increasing concentrations of Mg were added together with NC (lanes 20–24), there was a corresponding decrease in the amount of annealed product formed. At 7 mM Mg, the reduction was ∼3-fold (lane 24). Viewed collectively, these results demonstrate that when annealing was only moderately stimulated by NC (RNA 70 and RNA 70 loop mutant), Mg was able to increase the efficiency of the reaction without loss of the NC effect. However, when the annealing reaction was strongly dependent on NC concentration (RNA 70U28,U30C), Mg inhibited the chaperone activity of NC. The implications of these findings are discussed below. The goal of the present study was to provide new insights into the mechanism of HIV-1 NC nucleic acid chaperone activity in the minus-strand transfer step of reverse transcription. Our approach was to use a reconstituted system with model substrates described in an earlier report (), since their structures are not overly complex and are therefore especially suitable for mutagenic analysis (). We initially focused on the question of whether efficient strand transfer is determined by the overall thermodynamic stability and structure of acceptor RNA or more directly, by local structure at the nucleation site. In assays with RNA 70 and RNA 50, we found that the higher strand transfer activity exhibited by RNA 70 is correlated with the local stem-loop structure at the 5′ end of the TAR sequence, which contains 11 bases complementary to an 11-nt single-stranded region in DNA 50 (), and with RNA 70's predicted overall thermodynamic stability, which is actually considerably higher than that of RNA 50 (A) (). In further support of our ideas, the data clearly demonstrated that stabilizing mutations in the relevant local structure dramatically reduce the rate and extent of strand transfer and increase dependence on NC ( and ), whereas destabilizing mutations lead to a marked increase in strand transfer efficiency and loss of the NC requirement (). These results indicate that local structure at the nucleation site is a critical determinant for NC chaperone activity in minus-strand transfer. This conclusion is consistent with other studies demonstrating the importance of local structure in NC-promoted annealing of to the primer-binding site in an HIV-1 RNA transcript (), RT-catalyzed extension reactions with TAR RNA mutants (), and HIV-1 recombination (,). Taken together with our findings, it would seem that the mechanism we have identified using a reconstituted system may also be relevant to the reverse transcription pathway utilized during the course of HIV-1 replication in infected cells. We considered the possibility that formation of a kissing loop between the complementary TAR RNA and TAR DNA apical loops might also contribute to efficient annealing and strand transfer in our system. Previous efforts to address this question in other TAR-based systems led to diverse results. In two studies, mutational analysis showed that nucleation via loop–loop interactions facilitates efficient NC chaperone activity in minus-strand transfer () and annealing (), whereas in other work, destabilization of the central double-stranded segment of the TAR stems was reported to be the major pathway for annealing, with only a minor role for a kissing-loop complex (). Evidence has also been presented suggesting that multiple pathways may be involved in NC-dependent nucleation of annealing (,), although more recent studies with NC and full-length TAR favor a zipper mechanism involving the lower stems and bulges () (Vo,M.-N., Rouzina,I. and Musier-Forsyth,K., in preparation.) Here we report that the major effect of mutating two of the three contiguous G residues (G53,G54) in the RNA 70 apical loop to A (A) is an almost 3-fold reduction in the extent of strand transfer (A and A). This is presumably due to the two mismatches that are created by the mutation, which results in a less stable product. A substantial effect was also observed when all three Gs were changed to A (). Like WT RNA 70, the activity of our loop mutant is independent of NC (A), in all likelihood because the RNA fold is unchanged by the mutation (data not shown). Making the compensatory changes in DNA 50 yields only a small improvement in the extent of strand transfer, so that the overall efficiency remains lower than that achieved with the WT substrates (A and B). This finding implies that the loop–loop interaction must involve more than one G-C bp, since substitution of two A-T bp is not sufficient for optimal activity. Conservation of Gs in the apical loop of TAR RNA may be related to NC's preference for binding to unpaired Gs (3 and references therein) and to the unusual stability of a kissing complex, even with only two G-C bp (,). It is of interest that mutations in the RNA 70 local structure at the 5′ end of TAR and in the apical loops both affect the extent of strand transfer ( and ). However, ‘only’ stabilizing mutations in the RNA 70 5′ stem loop have striking effects on ‘both’ the rate and extent of strand transfer and are NC dependent (). Thus, it appears that destabilization of the RNA 70 5′ local structure is likely to be the dominant nucleation pathway for minus-strand transfer in the RNA 70 system. Another issue that we address concerns a comparison between annealing and strand transfer activities in the presence or absence of NC. The RNA 70 acceptors that we tested fall into two groups, i.e. having activity that is either independent (group 1) or dependent (group 2) on NC. With the RNAs in the first group (WT RNA 70 and the RNA 70 loop mutant), strand transfer is unaffected by addition of NC (A and A), although NC slightly stimulates the extent of annealing (A and B). The lower minus NC values for annealing compared with the corresponding values for strand transfer might reflect the fact that during strand transfer, RT-catalyzed elongation of the annealed DNA drives the equilibrium towards formation of a more thermodynamically stable product with a greater number of base pairs than the number contained in the RNA–DNA hybrid. When NC is present, the rate of annealing is increased (, lines 1 and 4) so that even in the absence of RT, the end point values for annealing and strand transfer are in close agreement. The RT effect observed here is reminiscent of a somewhat similar scenario that occurs during plus-strand transfer (): annealing of the complementary primer-binding site sequences in (+) SSDNA and minus-strand acceptor DNA is ultimately favored over hybrid formation between the 3′ terminus of tRNA and (+) SSDNA, since RT extends each strand of the DNA duplex to yield the more stable double-stranded DNA transfer product (). RNA 70 acceptors with stabilizing changes in the 5′ local stem-loop structure, e.g. RNA 70U28C and RNA 70U28,30C (group 2), show a clear dependence on NC for efficient annealing and strand transfer ( and ) and a striking reduction in the rates of these reactions relative to the rates for WT RNA 70 (Table 2). In the absence of NC, the annealing and strand transfer activities of both the single and double mutants are quite low. However, in reactions with NC, the annealing activities of both mutants are greatly increased and reach a value of almost 60% at the highest NC concentration (C and D), whereas the comparable values for strand transfer are significantly lower (B and D). These results indicate that optimal experimental conditions for annealing and strand transfer may differ when activity is dependent on NC's chaperone function. We have reported similar findings in studies with NC mutants having changes in the CCHC motif (,). To investigate this discrepancy between the efficiency of annealing and strand transfer with structured RNA acceptors, we tested annealing under strand transfer conditions. We found that of the components present specifically in strand transfer reactions, it is only Mg that inhibits NC chaperone activity during annealing and in a dose-dependent manner (C). This would explain why a potential stabilizing effect of Mg on the nucleic acid substrates does not lead to greater stimulation of strand transfer by NC, as might be expected. In fact, since Mg is present in vast excess over NC in our reactions, the data strongly suggest that Mg successfully competes with NC for non-specific binding to the phosphodiester backbone of the nucleic acid substrates and partially displaces NC. Not surprisingly, since Mg does not have nucleic acid chaperone activity, we observe that formation of the annealed RNA–DNA hybrid becomes very inefficient in the presence of high Mg concentrations. This conclusion is consistent with a similar observation made by Musier-Forsyth and colleagues () (Vo,M.-N., Rouzina, I. and Musier-Forsyth,K., in preparation). The Mg concentration in virions and in infected cells is not known. However, in some instances the concentration in uninfected cells has been reported to be much lower (∼0.2–0.25 mM) than the amounts typically used in reverse transcription reactions (6–8 mM) (75 and references therein). If the intracellular Mg concentration in infected cells is also low, it would suggest that RT-catalyzed extension can proceed at a lower Mg concentration than and that, under such conditions, Mg should not interfere with the efficiency of NC-dependent annealing during HIV-1 infection. Interestingly, in reactions containing 0.25 mM Mg and NC, we find only an ∼8% reduction in annealing (C). In reactions with WT RNA 70 and the loop mutant, which have a fairly minor requirement for NC, annealing is stimulated by Mg to a small, but constant extent, over a wide range of Mg concentrations (A and B). In this case, when the ionic strength is increased, the increase in positive charge leads to greater attraction between the nucleic acid strands and results in more efficient annealing, in accord with studies on the aggregation properties of NC (,,,). In conclusion, we have demonstrated that the local structure in acceptor RNA at the nucleation site is a critical determinant for nucleation of annealing that occurs during HIV-1 minus-strand transfer. Moreover, destabilization of a short secondary structure at the 5′ end of the TAR sequence in RNA 70 appears to be the dominant nucleation pathway. Our data also point to different consequences for annealing versus strand transfer when reactions are strongly or weakly dependent on NC. Where NC has little or no effect, annealing and strand transfer occur with similar efficiencies. However, when NC-catalyzed destabilization of acceptor structure is required, annealing appears to be more efficient than strand transfer. We attribute this result to the presence of a high Mg concentration in reconstituted strand transfer reactions, which leads to a competition between Mg and NC for binding to the negatively charged phosphate groups in the nucleic acid strands. Taken together, these findings contribute significantly to a greater understanding of the mechanism of NC nucleic acid chaperone activity during minus-strand transfer.

Genetic information analyses have placed a strong demand for advanced biomolecular recognition probes with high sensitivity and excellent specificity. One such promising probe is the molecular beacon (MB) (), a short hair-pin oligonucleotide probe (A) that binds to a specific oligonucleotide sequence and produces fluorescence signal. With an inherent signal generation mechanism, MBs are able to detect nucleic acid targets without separation or the addition of extra reagents. This property makes MBs especially useful for real-time detection of DNA and RNA, which is of great significance to the study of gene expression in real-time and at the single cell level (). However, when used for intracellular analyses, MBs tend to generate dramatic false-positive signals due to nuclease degradation, protein binding (,) and thermodynamic fluctuations (). For example, it has been reported that unmodified phosphodiester oligonucleotides may possess a half-life as short as 15–20 min in living cells (). In addition to nuclease degradation, DNA-MBs are subject to a myriad of nucleic acid binding proteins in cells. The interaction of these nucleic acid binding proteins with MBs can disrupt the stem-loop structure and cause nonspecific fluorescence signal (). To overcome the bioinstability problem, MBs have been synthesized with nuclease-resistant backbones such as phosphorothioate and 2′--methyl RNA bases (). More drastically, various groups have considered neutral peptide nucleic acids (PNAs) as a scaffold for MBs (). Backbone modifications have both advantages and disadvantages. Those that retain the repeating charge continue to behave like natural nucleic acids in their hybridizing properties, balancing certain disadvantages, such as the toxicity occasionally associated with phosphorothioate-containing oligonucleotides (). To the extent that MBs incorporating such modifications resemble natural nucleic acids, however, they still may be opened by intracellular DNA- and RNA-binding proteins, many of which are believed to recognize a repeating backbone negative charge as one pharmacophore. This can lead to a signal in the absence of analyte. For example, MBs with 2′--methyl RNA bases possess a better nuclease resistance, higher affinity, increased specificity, faster hybridization kinetics and a superior ability to bind to structured targets compared to their DNA counterparts. However, 2′--methyl modified MBs open up nonspecifically in cells, possibly due to protein binding (,). Lacking repeating charges, PNAs are not degraded by nucleases and their hybridization products with RNA are thermally more stable compared with DNA–RNA and RNA–RNA duplexes. The neutral charged PNAs in a MB probe are not likely to be recognized by endogenous RNA- or DNA-binding proteins. Xi . () reported that using PNA-MBs instead of DNA-MBs for traditional fluorescent hybridization probes would benefit cell detection under a wide range of conditions. The neutral backbone, however, creates other undesirable properties. For instance, PNAs have a well documented propensity to self-aggregate () and fold in a way that interferes with duplex formation (). Furthermore, PNAs change their physical properties substantially (and unpredictably) with small changes in sequence (,), although adding charged appendages helps (). Since the environmental conditions inside of a living cell can not be optimized for the solubility of PNA, intracellular applications of PNA are limited. Considering these facts, we reasoned that a scaffold that differs as much as possible in the geometric and steric properties from ribose, but retains the repeating charge, might be the most likely to retain the desirable solubility and rule-based molecular recognition features of natural DNA, while avoiding binding to intracellular DNA- and RNA-binding proteins. Locked nucleic acids (LNAs) () offer one possible implementation of this reasoning. LNA is a conformationally restricted nucleic acid analogue, in which the ribose ring is locked into a rigid C3′- (or Northern-type) conformation by a simple 2′-, 4′- methylene bridge (,) (B). LNA has many attractive properties (,), such as high binding affinity, excellent base mismatch discrimination capability, and decreased susceptibility to nuclease digestion. Duplexes involving LNA (hybridized to either DNA or RNA) display a large increase in melting temperatures ranging from +3.0°C to +9.6°C per LNA modification, compared to corresponding unmodified reference duplexes. Furthermore, LNA oligonucleotides can be synthesized using conventional phosphoramidite chemistry, allowing automated synthesis of both fully modified LNA and chimeric oligonucleotides such as DNA/LNA and LNA/RNA. Other advantages of LNA include its close structural resemblance to native nucleic acids, which leads to very good solubility in physiological conditions and easy handling. In addition, owing to its charged phosphate backbone, LNA is nontoxic and can be delivered into cells using standard protocols that employ cationic lipid (). All these properties are highly advantageous for a molecular tool for diagnostic applications. Several reports have revealed LNA as a most promising molecule for the development of oligonucleotide-based therapeutics. We have communicated the preparation of a fully modified LNA-MB and the investigation of its properties (). experiments showed that the LNA-MB not only exhibits excellent thermal stability and single base discrimination capability, but also resists nuclease digestion and binding of single stranded DNA binding proteins. All of these properties are ideal for intracellular applications and studies. Unfortunately, the hybridization rate of this LNA-MB was relatively slow. In this study, we evaluated different LNA-MB designs on their thermodynamics, binding kinetics and enzymatic resistance. Specifically, we have designed and synthesized MBs with different DNA/LNA ratios and stem lengths. Systematic studies on the thermostability and hybridization kinetics of the MBs were performed. The binding of these LNA-MBs to single stranded DNA binding protein (SSB) was investigated. In addition, the effects of incorporating LNA bases in a MB on the activity of DNase and RNase were explored. The effect of shortening stem length on the hybridization rate of LNA-MBs was studied. The design guidelines of LNA-MBs for intracellular applications were recommended. MBs prepared are listed in and . DNA and LNA synthesis reagents were purchased from Glen Research (Sterling,VA,USA). Deoxynuclease I, Ribonuclease H and SSB were purchased from Fisher. An ABI3400 DNA/RNA synthesizer (Applied Biosystems, Foster City, CA, USA) was used for target DNA synthesis and MB probe preparation. Probe purification was performed with a ProStar HPLC (Varian, Walnut Creek, CA, USA), where a C18 column (Econosil, 5u, 250 × 4.6 mm) from Alltech (Deerfield, IL, USA) was used. UV-Vis measurements were performed with a Cary Bio-300 UV spectrometer (Varian, Walnut Creek, CA, USA) for probe quantitation. Fluorescence measurements were performed on a Fluorolog-Tau-3 spectrofluorometer (Jobin Yvon, Inc., Edison, NJ, USA). MBs possessing locked nucleic acid bases were synthesized using LNA phorsphoramidites. The controlled-pore glass columns used for these syntheses introduced a DABCYL (4-(4-(dimethylamino) phenylazo) benzoic acid) molecule at the 3′ ends of the oligonucleotides. FAM (6-carboxyfluorescein) phorsphoramidite was used to couple FAM to the 5′ ends of the sequences. The complete MB sequences were then deprotected in concentrated ammonia overnight at 65°C and purified by high-pressure liquid chromatography. The collection from the first HPLC separation was then vacuum dried, incubated in 200 µl 80% acetic acid for 15 min, incubated with 200 µl ethanol and vacuum dried before the second round of HPLC. Unless otherwise indicated, hybridization experiments were conducted with 100 nM MBs, 500 nM complimentary target sequences in a total volume of 200 µl. All experiments were conducted in 20 mM Tris-HCl (pH 7.5) buffer containing 5 mM MgCl and 50 mM NaCl. To test the nuclease sensitivity of MBs, the fluorescence of a 200 μl solution containing 20 mM Tris-HCl (pH 7.5), 5 mM MgCl, 50 mM NaCl and 100 nM MBs was measured as a function of time at room temperature. One unit of ribonuclease-free DNase I was added, and the subsequent change in fluorescence was recorded. To test the susceptibility of MB-RNA hybrids to ribonuclease H digestion, 100 nM of MBs were incubated with the same concentration of RNA target in the aforementioned buffer. The fluorescence intensity of the solution was monitored. After the hybridization reached equilibrium, 12 units of ribonuclease H were added, and the subsequent change in fluorescence was recorded as a function of time. The thermal stabilities of the MB samples were determined by a BioRad RT-PCR thermal cycler. The fluorescence intensity of 100 μl MBs (100 nM) in 20 mM Tris-HCl (pH 7.5) buffer containing 5 mM MgCl, 50 mM NaCl was monitored as a function of temperature. The temperature was brought to 10°C and increased at 1°C increments to 95°C, with each step lasting for 2 min and fluorescence monitored during each step. The melting temperatures (T values) were obtained as the maxima of the first derivative of the melting curves. Gel electrophoresis was performed to study the interaction between SSB and MBs. In 20 mM Tris-HCl buffer (pH 7.5, 5 mM MgCl, 50 mM NaCl), 5 μM MB was incubated with the same concentration of SSB. After 1 h, the solution was analyzed in a 3% agarose gel in TBE buffer (100 V) for 50 min. The gel was then stained with Gelcode blue stain solution (Pierce) for 1 h and washed with deionized water for 30 min. The image of the resulting gel was obtained by scanning on a regular scanner. Compared to DNA-MBs, MBs fully modified with LNA respond slowly to complementary target DNA or RNA sequences. shows the hybridization of a DNA-MB and a fully modified LNA-MB (LNA-MB-E0) with the same loop target DNA. The fluorescence signal of DNA-MB reached equilibrium within minutes. In contrast, the fluorescence intensity of the fully modified LNA-MB-E0 increased slowly over time, indicating its slow hybridization kinetics. The reaction did not reach equilibrium even after 20 h under the same conditions. The slow hybridization kinetics would compromise temporal resolution when obtaining the dynamic information of RNA in living cells. In order to take advantage of LNA-MBs for intracellular analysis, it is necessary to expedite LNA-MB hybridization kinetics. It is believed that two factors could lead to the slow hybridization kinetics: slow stem dehybridization rate and tendency of hybridized LNAMBs to form sticky-end pairs. First of all, hybridization of MBs with target sequences competes with the dehybridization of the MB stem. The strong binding affinity of the LNA bases in the stem is unfavorable for stem dehybridization. It has been reported that replacing DNA with LNA bases in one of the oligonucleotide sequences in an octamer duplex had no evident effect on the association rate (), but significantly decreased the dissociation rate of the octamer sequence from its complementary sequence. The dissociation rate dropped as much as 2-fold for a single base replacement and 5-fold for two base replacements. More than 30-fold decrease in dissociation rate was observed after replacing three DNA/DNA base pairs with DNA/LNA base pairs (). Therefore, inserting several LNA bases in a MB stem greatly enhances the energy barrier of opening the LNA-MB stem and significantly slows down its hybridization kinetics. Secondly, because of the excellent binding affinity of LNA, hybridized MB sequences are more likely to form sticky-end pairs (). DNA sticky-end pairing (SEP) plays an important role in cellular processes and has been well used in various biotechnological applications. For MBs, SEP is defined by the intermolecular interaction between the stems of two or more MBs when loop target DNA is present. When hybridized to its loop target DNA, a MB opens up, exposing two complimentary stem sequences. The two complementary sticky ends from two MB/target hybrids can pair to form a short double helix, leading to the association of two hybrids at one end. These two MB hybrids can form a closed structure, ([MB]), by pairing the other two sticky ends or polymerize into a multimolecular structure, [MB] ( > 3), by pairing with more hybrids. With sticky-end pairing, the fluorophore and quencher are brought together again (but from different sequences), causing the quenching of the fluorescence. Any parameters that stabilize the short double helix of the sticky-end pairs will lead to a more severe loss of the fluorescence intensity. For example, the higher the [Mg] in the solution, the more signal is lost (). The direct result of SEP is the quenching of the fluorophore in opened MBs. Therefore, when the opening of MB via target hybridization is immediately followed by SEP, only slow signal intensity increase is observed. Considering the two factors discussed earlier, we believe that the hybridization kinetics of LNA-MBs could be improved by preventing SEP, or by reducing the LNA percentage in MB sequences, especially in the stem, to lower the reaction energy barrier and speed up the opening rate of MB hair-pin structure. A series of DNA/LNA chimeric MB with a 6-mer stem were prepared as listed in . DNA-MB is a probe constructed solely from DNA bases. The LNA-MB probes were prepared by gradually replacing every base, every other base, every second base, every third base, every fourth base and every fifth base of the DNA-MB with LNA bases to produce LNA-MB-E0, LNA-MB-E1, LNA-MB-E2, LNA-MB-E3, LNA-MB-E4 and LNA-MB-E5, respectively, as shown in . Stem melting temperatures () of LNA-MBs were measured to investigate the effect of LNA percentage on the stability of MB hair-pin structure. lists the melting temperature of the MBs. It has been reported that introduction of LNA raises the of the oligonucleotide and a complementary DNA or RNA by as much as 9.6°C per LNA modification (). The incorporation of LNA bases in a MB sequence dramatically stabilized the stem duplex of the probe. When one DNA/DNA base pair was replaced by one LNA/LNA pair in the stem, as in LNA-MB-E5, the melting temperature of the MB stem increased as high as 20°C. The stem increased about 27°C when the number of LNA/LNA base pair increased to two (as in LNA-MB-E2, LNA-MB-E4). The melting temperature change was found weakly dependent on the nature of base pair change. For example, of LNA-MB-E4 was about 1°C higher than that of LNA-MB-E2. The difference between the stems of these two sequences was that the former had two G: C LNA pairs, while the latter contained one G: C and one A: T LNA pairs. Insertion of three LNA: DNA pairs in the MB stem had similar effects on the stability of MB stem as two LNA: LNA pairs did, as indicated by the of LNA-MB-E4 and LNA-MB-E3. No fluorescence signal change was observed for LNA-MB-E0, suggesting an extremely high of the probe. The hybridization of the 6-mer-stem MBs to their target DNA sequence that is complementary to the MBs’ loop sequence was investigated. A very slow increase of fluorescence signal was observed, when target DNA was introduced into the LNA-MB-E0 solution. Lowering the percentage of LNA in the probe sequences significantly improved the initial hybridization rate. For LNA-MB-E2, E3 and E4, interesting fluorescence traces were observed. The fluorescence of the solution initially had a sharp increase and then suddenly began to decay. The blue trace in shows a typical response of LNA-MB-E3 to the addition of target DNA. Similar trace was observed for DNA-MBs at higher [Mg] in previous studies and was attributed to SEP of MBs in the presence of target (). Incorporation of LNA bases in MBs significantly enhanced the affinity of the stem, as indicated by the elevated melting temperature of the LNA-MBs. Consequently, intermolecular hybridization of LNA-MBs was evident for LNA-MB-E2, LNA-MB-E3 and LNA-MB-E4. The traces of LNA-MB-E1 and LNA-MB-E0, however, were somewhat different, with no clear drop in fluorescence after the initial signal increase. The signal of these two beacon solutions increased much slower because of more LNAs in the stem, resulting in a higher tendency to form sticky-end pairs at the initial hybridization state. Several strategies could be exploited to prevent the formation of sticky-end pairs (). For example, lowering Mg concentration is proven to be an effective way to prevent SEP for DNA-MBs. However, this approach is not practical for intracellular applications. Another approach is to use a target complementary to the loop plus one of the stem sequences. The hybridization of the so-called shared-stem targets to MBs results in a MB/target hybrid with only one sticky end, making SEP impossible. The red trace in shows the hybridization of a shared-stem target (CCT AGC GCG ACC ATA GTG ATT TAG A) to LNA-MB-E3. Faster hybridization kinetics and greater signal enhancement were observed. The results (A, blue trace) demonstrated that at lower temperatures the MB was in a closed state, the fluorophore and the quencher were held in close proximity to each other by the stem, and the MB did not fluoresce. However, at high temperature the helical order of the stem gave way to a random-coil configuration, separating the fluorophore from the quencher, restoring fluorescence (B1) (). This transition, defined as the melting temperature of the stem, occurred at 84°C. In the presence of 5-fold excess of single-stranded loop target, the MB fluoresced weakly (A, green trace) at room temperature, but the fluorescence intensity increased significantly as the temperature was slowly raised. The signal peaked at around 50°C and then diminished significantly followed by an increase in fluorescence at high temperature. B2 summarizes the phase transitions occurred. At low temperatures, MBs hybridized spontaneously to the target sequences that were complementary to the loop sequences. The formation of MB/target hybrids exposed the sticky ends of MBs, leading to the formation of [MB] complex through the SEP process. As a result, the fluorescence intensity of the MB remained low even though the probe had hybridized with its target. Because the structure stability of the sticky-end pairs was weaker than that of the probe-target duplex and the hair-pin structure, the [MB] complexes were disrupted and separated into individual MB/target duplexes as the temperature increased. Consequently, the fluorophores were unquenched and the fluorescence intensity increased accordingly. Continuing to raise the solution temperature destabilized probe-target duplexes, thus the MBs were released. The free MBs returned to their closed conformation and the fluorescence decreased. As the temperature was raised further, the closed MBs melted into random coils (B2), restoring their fluorescence. This transition occurred at the same temperature (84°C) as in the MB solution free of targets. When the MB was incubated with 5-fold excess of shared-stem target DNA that was complementary to the loop plus one of the stem sequences, different changes in fluorescence were observed (A, red trace). The results showed that the MBs fluoresced brightly at low temperatures, but as the temperature is slowly raised fluorescence diminished significantly, followed by an increase in fluorescence at high temperatures. B3 summarizes the phase transitions occurred. The difference between B2 and B3 is the formation of the [MB] complexes in the case of using loop target sequences. Clearly, the SEP process decreased the overall signal change even when the target sequences were in excess and most MBs were in the open state. The use of shared-stem target sequences effectively blocked SEP formation, leading to a higher signal enhancement and significantly faster apparent hybridization kinetics. When incubated with shared-stem target DNA, the fluorescence signal of the MB was three times the intensity of the same MB with loop target DNA (A). The hybridizations of all 6-mer-stem LNA-MBs with shared-stem targets were much faster and gave much higher fluorescence signal as compared to the hybridizations with loop target DNA. The hybridization results of these probes also indicated that the less the LNA bases in a beacon sequence, the faster the hybridization rate the beacon had. compares the response of the DNA-MB, LNA-MB-E0, LNA-MB-E1 and LNA-MB-E3 to the addition of target DNA. With less DNA bases being replaced by LNA bases, the hybridization rates of MBs were significantly increased. Normal MBs are subject to nonspecific binding of proteins, such as the ubiquitous SSB. Such a binding causes normal DNA-MBs to open and give false-positive signal. In addition, protein binding lowers the accessibility of MBs for hybridization to their target sequences. A nucleic acid probe immune to protein binding will perform its analysis with higher accuracy and better targeting efficiency , where such proteins are highly abundant. The hybridization results of 6-mer-stem MBs showed that the hybridization rate of a MB was tunable by adjusting the density of LNA bases in the probe sequence. We have reported that MBs with fully modified LNA bases had no response to the addition of excess SSB. With the decrease of DNA/LNA ratio in the probe, it became a concern that the probe might be subject to DNA binding proteins. To find out how the ratio of DNA/LNA in a probe sequence would affect the binding affinity of the probe to DNA binding proteins, fluorescence measurements and gel electrophoresis experiments were carried out to study the binding of SSB to LNA-MBs. Agarose gel (3%) was used to visualize the SSB/MB complex. Due to the high molecular weight of SSB (74 kD), SSB migrates slowly in the gel. The binding of the MB to SSB would significantly expedite the migration rate of the protein because of the multiple negative charges of nucleic acids. As shown in , only one band was observed for the sample containing only SSB. The SSB barely migrated in the gel under the experimental conditions. When the DNA-MB was added to the SSB solution, an extra protein band with a faster migration rate appeared, while the intensity of the first band decreased. This clearly indicated the binding of SSB to DNA-MB. The SSB/MB complex band was observed for LNA-MB-E5, LNA-MB-E4, LNA-MB-E3 and LNA-MB-E2. Under the same experimental condition, no visible SSB/MB band was seen for LNA-MB-E1 and LNA-MB-E0. The binding of SSB to MBs was further confirmed by using fluorescence measurements. The disruption of MB secondary structure due to SSB binding displaces the fluorophore from the proximity of the quencher molecule, resulting in fluorescence increase. All the MB sequences except LNA-MB-E1 and LNA-MB-E0 had higher fluorescence intensity when excess SSB was added. For example, an approximate 4-fold increase in fluorescence signal was observed for LNA-MB-E3. Unfortunately, the signal enhancement was also affected by the purity and the stem stability of MBs; thus, the binding affinity could not be directly evaluated with the fluorescence intensity change. Further measurements of the signal changes at different concentrations of an individual MB need to be performed in order to measure the K of the corresponding probe to SSB. Nonetheless, the fluorescence measurements were consistent with the gel imaging results. Results from both assays suggested no observable binding of SSB to MBs with all LNA or alternating DNA/LNA bases. Ideal probes for targets in living cells should be stable inside the cell, should not induce the destruction or perturbation of their target, and should signal only at the presence of their target (). MBs constructed from natural deoxyribonucleotides are not suitable for this purpose, because cellular nucleases can degrade them and cellular ribonuclease H can digest the target mRNA in the region where the probe is bound. The sensitivity of the 6-mer-stem LNA-MBs to nuclease digestion were tested using DNase I. As shown in , immediate increase in fluorescence signal was observed, when DNase I was added to a solution containing DNA-MB. This result excludes the viability of using MBs constructed from natural deoxyribonucleotides for intracellular applications. With the introduction of LNA in a MB sequence, the digestion process slowed down, as seen in the case of LNA-MB-E5, LNA-MB-E4, LNA-MB-E3 and LNA-MB-E2. In contrast, MBs with DNA/LNA alternating bases (LNA-MB-E1) in the sequence or with all LNA (MB-LNA-E0) gave no observable signal change, indicating their resistance to DNase I after more than 2 h incubation with the nuclease. To test the susceptibility of LNA-MBs to nucleases found in cells, the stability of LNA-MB-E1 and LNA-MB-E0 were further tested inside living cells containing no target sequences. After injection into cells, both MBs remained close during the observation period (more than 1 h), as indicated by no observable fluorescence signal change over time. RNase H (Ribonuclease H) is an endoribonuclease that specifically hydrolyzes the phosphodiester bonds of RNA when it is hybridized to DNA. This enzyme does not digest single or double-stranded RNA. RNase H is found in both the nucleus and the cytoplasm of all cells and its normal function is to remove RNA primers from Okazaki fragments during DNA replication. Antisense RNase H activation has proved not only to be a powerful weapon in assessing gene function but is emerging as the method of choice for antisense therapeutics as well. While desirable in antisense technology, RNase H activation is not favored in RNA monitoring. The action of RNase H on MB: RNA duplexes will lead to destruction of the target RNA and loss of signal due to reformation of MB hair-pin structure (, left). Use of MBs does allow real-time monitoring of RNase H activity. right shows the response of a DNA-MB to the sequential addition of target RNA, RNase H and target DNA, respectively. The probe lit up with the introduction of RNA in the solution. After reaching the hybridization equilibrium, RNase H was added. An immediate decrease of the fluorescence signal from the probe was observed, which indicated the degradation of RNA on the MB: RNA duplex through enzyme cleavage and the reformation of MB hair-pin structure. After enzyme digestion, the DNA-MB remained intact, as evidenced by its response to the addition of target DNA. Thus, the MB will act as a catalyst inside the cell to destroy all target RNA sequences through RNase H degradation. The series of DNA/LNA chimeric MBs enabled the systematic study of the effect of LNA on the RNase H activity using fluorescence measurements. Insertion of LNA bases in DNA MB sequence slowed down the degradation of RNA in the probe: target duplex as compared to the DNA-MB. As seen in , as the DNA gap between LNA bases in a MB was shortened, the activity of RNase was inhibited more. For example, about 30% of RNA target bound to LNA-MB-E4 was immediately destroyed after the introduction of RNase H (). No significance signal decrease was observed, when the enzyme was added to the duplexes of RNA with LAN-MB-E3, LNA-MB-E2, LNA-MB-E1 and LNA-MB-E0. Ion exchange HPLC analysis of a solution containing LNA-MB-E3, RNA, and RNase H further confirmed the protection of RNA after overnight incubation of LNA-MB-E3 with RNA. There were no small pieces of RNA cleavage product observed in HPLC, as compared to the appearance of different lengths of RNA pieces for the mixture of DNA-MB, RNA and RNAse H (). Our results suggested that the number of DNA bases in a stretch for an LNA-MB should be less than three, in order to prevent significant cleavage of the target RNA. Changing DNA/LNA ratio can affect the stem stability of a MB sequence, the length of the MB stem can likewise be tailored to such a degree that it is stable enough to maintain a hair-pin structure, while presenting a low energy barrier in order for a fast hybridization kinetics. We have reported that shortening the stem length can significantly improve the LNA-MB hybridization rate. In the current study, DNA/LNA alternating MBs with 5-mer, 4-mer and 3-mer stems, designated as LNA-MB-E1-5S, LNA-MB-E1-4S, LNA-MB-E1-3S as listed in , respectively, were prepared to investigate the optimal stem length of DNA/LNA MB. As expected, the melting temperature of MB stem decreased with the decrease of probe stem length. With a 5-mer stem, LNA-MB-E1-5S had a T of 68.6°C, which was at least 28°C lower than that of a 6-mer stem LNA-MB-E1. The 4-mer-stem probe LNA-MB-E1-4S had a melting temperature of about 52.3°C. Further decrease of stem length to three base pairs did not render any stable hair-pin structure, as indicated by the negative response of MB-LNA-E1-3S to its complementary DNA and heating. As a result of decreasing stem stability, MBs with shorter stems hybridize to their target DNA much faster. For example, in the presence of 5-fold excess of shared-stem target DNA, it took about 4 h for LNA-MB -E1 to reach its hybridization equilibrium. Under the same condition, the hybribrization of LNA-MB-E1-5S reached equilibrium after about 1 h, while the 4-mer-stem probe LNA-MB-E1-4S completely opened up in less than 15 min. As shown in , the hybridization rate of LNA-MB-E1-4S was comparable to a regular DNA-MB. More than 80% of the 4-mer LNA-MB hybridized to the target DNA within 3 min, which makes it an excellent probe for introceullar imaging applications. Developing stable, sensitive and selective molecular probes is of great significance for the deciphering of life processes inside cells. LNA has showed its interesting properties, such as higher affinity, greater selectivity and better stability. To take full advantage of LNA for intracellular applications and gain a better understanding of the LNA effect on the behavior of MBs, we have synthesized and investigated a series of DNA/LNA chimeric MBs. These series of MBs allow us to systematically study the effect of the DNA/LNA ratio in molecular beacons on their thermal stability, hybridization kinetics, protein binding affinity and nuclease resistance. The number of LNA bases in a MB stem sequence has a significant effect on the stability of the hair-pin structure. The MB stem melting temperature was elevated by as high as 20°C by simply replacing one DNA pair to an LNA/LNA pair. A MB with 6-mer LNA stem had a melting temperature higher than 95°C. The melting temperature of MBs was found to decrease with lower number of LNA bases in the stem. Due to the high affinity of LNA, LNA-MBs tend to form multiple-probe complexes through a SEP process, when hybridized to loop target DNA, resulting in a very low signal response and slow apparent hybridization kinetics. SEP can be prevented by using shared-stem target DNA that is complementary not only to the loop section of the probe, but also to one of the stem sequences, as evidenced by the higher signal enhancements and faster hybridization rates. The hybridization rate of LNA-MBs could also be significantly improved by lowering the DNA/LNA ratio in the probe. The binding of LNA-MBs to SSB was also studied. Under the experiment condition, it was found that only MB sequences with DNA/LNA alternating bases and all LNA bases were able to resist SSB binding. Further detailed studies need to be conducted on how the LNA modification changes the affinity of SSB binding to nucleic acids. The prepared LNA-MBs with different numbers of DNA bases between two LNA bases enable a systematic study on the effect of LNA modification on the susceptibility of the probe to nuclease and the activity of RNase H function. It was found that the only MB sequences with DNA/LNA alternating bases, or fully modified with LNA were not subject to DNase I digestion and stable inside living cells. Our results showed that a sequence consisting of a DNA stretch less than three bases between two LNA bases were able to block RNase H function. Hybridization results from DNA/LNA-alternating MBs with different stem lengths indicated that satisfactory hybridization kinetics could be achieved from an alternating DNA/LNA MB with a 4-mer stem. The results from this study suggest a guideline for designing MBs for intracellular applications. A shared-stem MB with a 4-mer stem and alternating DNA/LNA bases ensures reasonable hybridization rates, reduced protein binding, and resistance to nuclease degradation for both target and probes. These findings will also have implications on the design of other LNA molecular probes for intracellular diagnosis, therapeutic and basic biological studies. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Fixation is a series of complex chemical modifications of macromolecules present in tissues and cells, to preserve structural and functional components as closely as possible to the living state while inhibiting autolysis, bacterial and fungal decay (). Short DNA and RNA sequences can be retrieved from conventionally fixed pathology material, but good, long-term preservation of intact nucleic acids and of protein integrity is necessary to meet the increasing number of molecular diagnostic and research techniques which are becoming available. The type and length of fixation determine the degree of preservation of intact nucleic acids in tissue (). Cross-linking fixatives such as formalin and glutaraldehyde bind amino groups and create methylene bridges (). Precipitant fixatives, including methanol, ethanol, acetone and acetic acid, denature proteins by breaking the hydrophobic bonds that make up the tertiary structure of protein molecules yet preserve secondary structure for immunohistochemistry (IHC). Other compounds include the commercially available HOPE (HEPES-Glutamic acid buffer mediated Organic solvent Protection Effect) which preserves DNA and RNA suitable for polymerase chain reaction (PCR) and reverse-transcription (RT)–PCR (,) and the reversible cross-linker dithio-bis[succinimidyl propionate] (DSP) for immunostaining, microdissection and expression profiling (). The potential value of a new universal molecular fixative (UMFIX) for preservation of macromolecules in paraffin-embedded tissue has been tested which can preserve morphology and macromolecules in paraffin-embedded tissue (). Despite the number of fixatives available, however, problems still remain for many of them including toxicity, expense, the need for rapid fixation systems and the need to employ denaturants and low melting point wax for embedding. Recently, a zinc-based fixative (zinc acetate, zinc chloride and calcium chloride in Tris buffer) originally described in 1994 () was reported to be superior for DNA and protein expression analysis in a broad spectrum of tissues which do not then require heat pre-treatment for antigen retrieval (). In other studies, zinc-fixed, paraffin-embedded tissues provided superior morphology and improved immunostaining (). Zinc compounds are non-toxic and inexpensive, non-carcinogenic and are not temperature sensitive. We evaluated a series of novel fixative recipes for immersion fixation and processing to paraffin in order to improve DNA, RNA and protein yield whilst maintaining optimal tissue morphology. A range of zinc-based salt solutions, as well as other metal-based salt solutions, was tested for potential fixation properties in comparison with standard fixation procedures. All fixatives were evaluated for morphology using haematoxylin and eosin (H&E) and IHC for actin, a widely distributed antigen not requiring antigen retrieval in formalin fixed material, and for cytokeratin, an epithelial marker and CD3, a T-lymphocyte marker, both of which require pre-treatment when in formalin fixed tissue. Preservation of nucleic acids was tested by PCR and RT–PCR. Additional chemicals were tested with one of the zinc-based fixatives, Z2: dimethyl sulphoxide (DMSO), diethyl pyrocarbonate (DEPC) and ethylenediaminetetraacetic acid (EDTA) at various concentrations. We describe a reliable, cost-effective and non-toxic fixative, Z7, which demonstrates excellent protein preservation, and which is particularly effective at preserving DNA and RNA integrity in comparison with standard fixation procedures, and allows for detailed molecular analysis procedures on fixed paraffin-embedded samples even after at least a year in storage. In all experiments, tissue samples were fixed on a shaking rotor at room temperature (RT) for 24 h. A range of chemicals was examined for their effect on morphology and nucleic acid preservation when used together with the zinc-based fixative Z2. These were: Three types of tissue from C57BL/6 male/female mice 8–12 weeks of age were used in this study. Colon was used for its variety of tissue types including epithelial, endothelial, nervous and muscular tissues and its susceptibility to autolytic degradation. Spleen was used as a solid reticuloendothelial organ and liver as a solid parenchymal organ. This selection provided a range of tissue density and structure to assess physical properties such as penetration of fixatives in different tissue types. Colon, liver and spleen tissues were dissected into pieces up to 5–6-mm thick and immediately immersed in the fixing solutions for 24 h at room temperature (RT). Tissues were also fresh-frozen in liquid N as control samples for DNA and RNA quality. For every experiment, three mice were used and experiments from each mouse were repeated once. Fixation in each of the solutions described above was followed by processing of tissues. Tissue processing was performed overnight using a vacuum infiltrating processor. Briefly, this comprised: Tissue sections were cut using a rotary microtome, and floated on distilled HO at 37°C for 5 min. Sections were picked up on charged slides and dried at 39°C overnight. The blade was wiped with 70% ethanol between blocks to avoid cross-contamination. All sections were stained with H&E and evaluated independently by three observers. All images of IHC and H&E were taken using a digital camera. Tissue sections were immunostained with a variety of primary antibodies. No antigen retrieval was performed in any of the experiments and a reagent only (no antigen) negative control was included for every experiment. Antibodies and other reagents are shown in . The two protocols for immunostaining are summarized below: DNA was extracted from 25-mg fixed paraffin-embedded tissues using a commercially available kit (Qiagen DNA Tissue Kit, Qiagen Ltd, West Sussex, UK). Three to four 20-μm sections were cut using a rotary microtome and placed in a 1.5-ml microfuge tube. One millilitre of xylene was added and the sample centrifuged for 1 min at 12 000 rpm. DNA was extracted according to manufacturer's protocol and eluted with 100 μl Tris/EDTA pH 8.0 buffer. Fresh-frozen tissue was used as a control for DNA integrity. For every experiment, a tissue sample was removed and immediately placed in liquid N followed by storage at −80°C. DNA was extracted from fresh-frozen tissue according to the manufacturer's instructions and eluted with 100 μl Tris/EDTA pH 8.0 buffer. RNA from fresh-frozen tissue was extracted using the RNeasy Tissue kit (Qiagen) according to manufacturer's protocol and the RNA was eluted in 50-μl RNAse-free water. RNA from fixed tissue was extracted using the RecoverAll Total Nucleic Acid Isolation extraction kit (Ambion Ltd, Cambridgeshire, UK). Fifteen to twenty milligrams of tissue was used as starting material. Two to three 20-μm sections (according to the size of the tissue block) were taken and placed in a 1.5-ml microfuge tube. Wax was removed with xylene and then the tissue was cleared with 100% alcohol twice. RNA was extracted according to manufacturer's protocol and eluted in 60-μl RNAse-free water. PCR and RT–PCR were performed on genomic DNA to amplify two universally expressed genes, glyceraldehyde-3-phosphate dehydrogenase (GAPDH), β-actin and Trefoil factor 2 gene (large fragment), ( has the primer sequences). In all PCR experiments, extracted DNA was diluted in RNAse-free water and 100 ng of DNA was used from each sample. Experiments were duplicated in order to verify the results. For RNA amplification, cDNA was prepared from 1 μg total RNA using a RT system (Promega Corp., Southampton, UK), and a commercial cDNA mouse β-actin primer was used (R&D Systems, GenBank Accession Number: X03672) which contains a positive control. Another primer also used for RNA amplification was S15 (small ribosomal unit). Mouse β-actin transcripts were amplified and the conditions used are summarized in . RT–PCR products were electrophoresed in 1% agarose gel, stained with ethidium bromide and visualized using an ultra violet gel imager. RNA samples were kept on ice and their concentrations measured using a Nanodrop spectrophotometer. RNA samples were prepared according to the Agilent 2100 Bioanalyser protocol and were loaded into the NanoChip or PicoChip and processed for 30 min. An equal amount of RNA was used for each experiment. The 18S and 28S ribosomal peaks were used to quantify RNA. Real-Time PCR was performed using the SYBR Green Ready Taq Mix (Sigma-Aldrich Ltd, Dorset, UK), using the same primers as for PCR and RT–PCR. A series of seven 2-fold dilutions was prepared in each run starting from 100 ng of DNA and gradually reducing to 1.5 ng of DNA. A reagent-only (no DNA) negative control sample was included in each run. DNA extracted from fresh-frozen samples was used as positive control for each experiment. Complementary DNA was prepared from 1 μg of RNA using the RT system (Promega) described above and Real-Time PCR was performed with S15 primers. A sample that contained no RT was used as negative control in each experiment. Experiments were carried out in triplicate using three sets of tissue (nine tests per tissue in total) to ensure reliability. Liver samples fixed by different methods were subjected to 2-D polyacrylamide gel electrophoresis (PAGE) analysis. For fresh-frozen samples, 2–3 sections of 10 µm with an area size of 1 cm were placed in a 1.5 ml microfuge tube with 100 µl Extraction Buffer II (Bio-Rad Laboratories, Hercules, CA). Samples were vortexed for 30 s, transferred in dry ice for 5 min and then thawed at RT. The freeze-thaw-vortex step was repeated three times and samples were centrifuged at 14 000 rpm at 4°C for 8 min. Supernatant (50 μl) from each sample was transferred to a fresh tube. For paraffin embedded samples, 30 10 µm sections were dissolved into 250 µl Extraction Buffer II (Bio-Rad Laboratories, Hercules, CA) and vigorously vortexed at RT. Samples were heated at 55°C for 30 min, 1 ml of xylene was added and samples were centrifuged at 12 000  for 10 min. The supernatant was transferred to a fresh microfuge tube and centrifuged for 8 min at 12 000 . The final supernatant of both fresh-frozen and paraffin embedded tissue was combined with a re-hydration buffer mixture containing re-hydration buffer (Bio-Rad Laboratories), Immobilized pH gradient (IPG) buffer (Amersham Biosciences, Piscataway, NJ), and bromophenol blue and subsequently re-hydrated overnight with Immobiline Drystrips (pH 4/7, 11 cm; Amersham Biosciences, Amersham UK). The isometric focusing for the first dimensional electrophoresis was performed with a Multiphore II Electrophoresis System (Amersham Biosciences). The strips were subjected to high voltage at 3500 V. IPG strips were equilibrated with Equilibration Buffer I and Buffer II (Bio-Rad Laboratories) for 15 min each. Precast ExcelGel SDS gels (Amersham Biosciences) were used for the second dimension of protein separation by a Multiphore II Flated System (Amersham Biosciences) under a constant voltage of 700 V for 3 h. A silver staining kit (Amersham Biosciences) was used according to the manufacturer's instructions to detect protein spots. All samples were run in duplicate to guarantee over 90% identity. Paraffin blocks prepared as described above were stored for 14 months at RT. DNA was extracted from these and compared with DNA extracted from paraffin blocks stored at RT for less than a week. PCR of the GAPDH and the β-actin genes, and Real-Time PCR on the GAPDH gene were performed to assess DNA stability over time. Tissues used during this study were assessed for their morphological structure and the integrity of their proteins and nucleic acids. Three independent and experienced histopathologists graded H&E and IHC results. Grades ranged from 1 (lowest) to 10. For the H&E, assessment was based on criteria such as shrinkage, cell morphology, nuclear structure and fixative penetration. For the IHC, criteria for assessment were the intensity and localization of immunostaining. The PCR and RT–PCR products were scored for their intensity in the agarose gel as a means of quantifying amount of product. A score was also given for the RNA Agilent 2100 Bioanalyser graphs compared to the fresh-frozen control samples. For all the data produced, scores were given in the following manner: 1–3 = poor, 4–6 = adequate, 7–9 = good and 10 = excellent. Statistical analysis of results comprised the Student's -test which was carried out between pairs of fixatives for each of the three organs (liver, spleen and colon). The 2-tailed paired Student's -test (assuming equal and unequal variances) was used for comparison of the mean CT values between fixatives for each tissue. For each of the tests a level of = 0.05 was taken to show significant difference and = 0.01 to show highly significant difference. Mouse colon, spleen and liver sections were each fixed using the four standard fixation techniques described above, i.e. NBF, Z2, HOPE and fresh-freezing in liquid N. PCR and RT-PCR carried out to assess DNA and RNA integrity following the fixation processes were assessed by three independent observers (data not shown). The intensity of the PCR product bands in agarose gels indicated that DNA was poorly preserved by NBF, while the best results were obtained from fixation in Z2 and HOPE. In addition, assessment of morphology following staining with H&E and immunostaining for actin, cytokeratin and CD3 (data not shown) suggested that Z2 was better for structural and protein preservation. Therefore, the zinc-based compound Z2 was chosen for continuing investigation over HOPE, in view of the high cost of HOPE. Mouse spleen, liver and colon sections were each fixed in each of the zinc acetate replacement solutions described above, i.e. Z7, Z8, Z16, Z17, Z18 and Z19. Examination of morphology and DNA and RNA quality (data not shown) indicated that Z7 gave the best results, and all other combinations were abandoned. Mouse spleen, liver and colon tissues each were fixed with each of the zinc replacement solutions, Mn2, Mg2, Ga2 and V2, and compared with fixation with NBF and Z7 as described in Materials and Methods above. Morphological assessment by three independent observers indicated that morphological structure was poorly conserved with all zinc replacement solutions Mn2, Mg2, Ga2 and V2 (data not shown). RNA was extracted from liver, spleen and colon after fixation with Mn2 and Mg2, while for Ga2 and V2 fixed tissues, RNA was extracted only from mouse liver tissues. RNA concentrations from each fixed tissue was measured using a Nanodrop 1000 spectrophotometer and these together with the results from the Agilent 2100 Bioanalyser showed that the quality of extracted RNA was inferior to that extracted from the zinc-based fixative Z7 (data not shown). For this reason, none of the zinc replacement-based fixatives were pursued further. shows the scores for morphology and DNA and RNA preservation in the comparison of the zinc-based fixative Z2, the optimum zinc-based fixation recipe Z4, and the zinc acetate replacement fixative Z7 with NBF and fresh-frozen controls. The fresh-frozen samples were only assessed for DNA, RNA and morphological quality and therefore achieved an enhanced total score, since frozen tissue may contain ice crystal artefacts and is avoided for and IHC. From the scores on , NBF was inferior to the rest of the fixatives tested for DNA and RNA preservation, although it had the best score for morphology. Fresh-frozen samples showed high quality of both DNA and RNA preservation in all tests performed. Z7 was better than NBF and the remaining zinc-based fixatives for protein preservation, morphological structure and RNA preservation and equivalent to Z2 and Z4 fixatives for DNA preservation. Overall, Z7 was given the highest score from the fixatives tested. Immunocytochemistry showed that Z7 was overall the best fixative for antigen preservation without the need for antigen retrieval for all of the antibodies tested. RNA was extracted from mouse liver sections fixed in Z2, Z7 and NBF and from fresh-frozen samples and was assessed for integrity using the Agilent 2100 Bioanalyser (). Typical traces that can be obtained from RNA in various degrees of degradation are shown for purposes of comparison. It can be seen that Z7 has two peaks corresponding to 12S and 18S RNA and that it is signi-ficantly better than both Z2 and NBF at preserving RNA structure and integrity. DNA was extracted from Z7-fixed, Z2-fixed, NBF-fixed and fresh-frozen mouse tissues and was assessed for the maximum size of fragment that could be amplified by Real-Time PCR using GAPDH primers which yield a product of 599 bp. shows results obtained from comparing the mean crossing thresholds of liver, spleen and colon tissues depending on fixation process and gene sequence amplified. Fragments up to 2.4 kb in length could be amplified from the Trefoil Factor 2 gene following Z7 fixation but could not be obtained from amplification of DNA fixed in Z2 or NBF. Results are shown in . a shows the results from comparing the mean crossing thresholds (CTs) from real-time RT–PCR using S15 primers, which yield a product of 361 bp, on mouse liver, spleen and colon tissue fixed in NBF and Z7 with fresh-frozen tissue as a control for RNA quality. After RT, the samples were amplified by conventional PCR using commercial β-actin primers (b). Results showed there was no contamination with genomic DNA. The smaller band visible (302 bp) in most samples does not correspond to the size expected from genomic DNA contamination (680 bp) but the provenance of this band is not known. Proteins were extracted from liver fixed with Z7 and NBF and subjected to 2-D PAGE. Fresh-frozen liver tissue was used as positive control. Results are shown in . These results show that Z7 is better than NBF and comparable to the fresh-frozen data. For DNA and RNA quality evaluation, a 2-tailed paired Student's -test (assuming equal and unequal variances) was carried out for comparison of the mean CT values between fixatives, after real-time PCR of mouse liver, spleen, colon and the results are shown in . DNA was extracted from paraffin blocks fixed with NBF and Z7 stored at ambient temperature for 14 months and compared with DNA samples from paraffin blocks stored for less than a week. Conventional PCR results using two housekeeping genes, GAPDH and β-actin, showed that DNA quality from Z7-fixed archival and newly stored tissues was significantly better than from all NBF-fixed tissue (a and b). No significant difference was observed between archival stored Z7-fixed and newly stored Z7-fixed samples. shows results from statistical analysis of mean CTs from real-time PCR of mouse liver, spleen and colon tissue fixed in NBF and Z7 when paraffin blocks were stored for 14 months and for less than a week, with fresh-frozen tissue as a control sample for DNA quality. There was no significant difference between the quality of DNA in Z7-fixed material between archived stored paraffin blocks and blocks stored for less than a week ( = 0.497). No significant difference in DNA quality was shown between the Z7-fixed and fresh-frozen (positive control) samples ( = 0.1608). All our modifications to the zinc-based fixative recipe showed improved quality in terms of morphology and greater quality and quantity of DNA and RNA preservation compared to the standard zinc-based fixative (Z2). The addition of 5% DMSO to the Z2 fixative (our Z4 fixative) markedly improved the morphological structure and nucleic acid and protein preservation. Modifications to the zinc-based recipes included replacing zinc acetate with zinc trifluoroacetate (our Z7 fixative), zinc citrate or zinc tartrate at the same concentration. The rationale for these modifications was the fact that the different counterions were expected to lead to different degrees of binding to the zinc and different levels of hydrophobicity, effectively allowing zinc to act faster and with better penetration into the tissue. When Z4 and Z7 were compared, Z7 was found to be a better overall fixative in terms of improved nucleic acid, protein integrity and morphology, although the quality of RNA obtained from the Agilent 2100 Bioanalyser was not as good as that obtained from fresh-frozen samples. This is likely to be due to the fact that RNA is degraded throughout the processing stage. Wester . indicated that it is the processing and/or storage in paraffin that is responsible for the degradation of RNA. When zinc was replaced with the metallic ions magnesium, manganese, gallium and vanadium to investigate potential fixative properties, poor results were obtained for morphology and RNA quality and these recipes were abandoned. In the current study, the quality of RNA was assessed using the Agilent 2100 Bioanalyser and agarose gel electrophoresis. Our initial RT–PCR and Real-Time RT–PCR analyses demonstrated that amplicons up to 361 bp in size could be consistently obtained using our methodology, in direct contrast to the results obtained using NBF-fixed tissue, and comparable to fresh-frozen material. Future studies will determine the maximum size of intact RNA and DNA that can be reliably recovered using our methodology. It was shown that our Z7 fixative was overall the best fixative for DNA, RNA and proteins. DNA fragments up to 2.4 kb were successfully amplified from Z7- fixed samples, compared to 0.6 kb from Z2- and 0.6 kb from NBF-fixed samples. Total protein was analysed successfully by 2-D gel electrophoresis following Z7 fixation, and results were better than those obtained from NBF fixation. The DNA quality of Z7-fixed archival stored samples was similar to freshly fixed and processed samples, and significantly better than that found with NBF fixative. Although zinc fixation caused shrinkage in all organs tested so far, none of the zinc-based solutions penetrated the tissue as fully as NBF. This resulted in well-fixed edges but apparently not equally well-fixed interiors. In view of the fact that tissues fixed in Z7 and taken through to paraffin blocks are significantly more stable for archiving than other fixatives (except NBF), we believe that this reduction in observed penetration of the fixative does not result in any harm to the nucleic acids. The search for an ‘ideal fixative’ suitable for preserving nucleic acid, protein integrity and tissue morphology is an essential requirement for molecular biological analysis of tissues and cells. Our results show that Z7 is a cheap, easily prepared and highly effective fixative that provides significantly improved preservation of DNA, RNA and proteins and allows improved PCR, Real-Time PCR and protein analysis, which may provide an excellent alternative to NBF for contemporary molecular pathobiology research.

Regardless of catalysis strategies in various biomolecular processes, many RNA cleavage reactions undergo an intra-molecular phosphoester transfer (). This mechanism involves a nucleophilic attack by the 2′ oxygen on the adjacent phosphorus center, followed by the formation of a pentacovalent phosphate intermediate and the subsequent departure of the 5′-oxyanion group (B). Both protein enzymes, as typified by Ribonuclease A (RNase A) () and RNA enzymes, as represented by hammerhead ribozyme (), can facilitate this RNA cleavage reaction. The mechanism on how these enzymes exert their catalytic powers is of intense interest (). RNA cleaving enzymes can bring up to 10-fold rate enhancement, as compared with the uncatalyzed reaction in the aqueous solution (). This large rate acceleration can be attributed to various factors such as conformational effect and electrostatic transition state stabilization effect (). Recently, conformational effect has been invoked to understand catalytic strategies in various enzymes (), such as in chorismate mutase, where this effect alone can contribute up to 10-fold rate enhancement (). Upon the substrate binding, a tight conformation, which is structurally close to the corresponding transition state, is usually formed within the bound substrate prior to any chemical steps. Specifically, in the RNA intra-molecular phosphoester transfer reactions, such tight conformation is referred as an ‘in-line attack’ conformation (,). Here, the attacking atom O2′ is placed opposite to the leaving O5′ atom and aligned in the direction of the broken P–O5′ bond (B). This ‘in-line attack’ conformation has been observed in the structures of a hammerhead ribozyme () and two RNA cleaving enzymes bound to their RNA substrates (,) (B). These structural data are consistent with the previous speculation that ‘in-line attack’ conformation can be important for RNA cleavage (). In the present work, we wish to quantitatively evaluate how much the conformational effect of this ‘in-line attack’ structure contributes to the rate acceleration in an enzyme-catalyzed RNA cleavage reaction. We achieved this goal by taking the advantage of our recently determined co-crystal structure of a splicing endonuclease and its RNA substrate (). The RNA splicing endonuclease is responsible for the removal of the intervening sequences in nuclear tRNA and all archaeal RNAs (). It recognizes a small RNA structure, called the bulge–helix–bulge (BHB) motif. The BHB motif comprises two three-nucleotide-bulges separated by four base pairs. The endonuclease forms a homodimer and cleaves two phosphodiester bonds located after the second bulge nucleotide in these two bulges. In this co-crystal structure, a similar conformation of the RNA substrate is captured in both pre- and post-cleavage states, suggesting that the substrate conformations closely mimic that of the transition state intermediate. To our knowledge, this structure and the structure of sarcin/ricin loop RNA bound to restrictocin () [with the sarcin/ricin RNA fortuitously trapped at a minor cleavage site rather than its canonical site ()] are the only crystal structures that have trapped the ‘in-line attack’ conformation of a pre-cleaved RNA substrate in a protein-catalyzed intra-molecular phosphoester transfer reaction (B). In the present work, we employed free energy simulations based on our crystal structure in order to quantitatively dissect the contribution of the ‘in-line attack’ conformational effect to the overall catalysis. As discussed in the Results and Discussions section, two free energy simulations were comparatively performed. One is on the complex between a splicing endonuclease and a RNA; the other is on a free RNA with the same sequence. For the bound state simulation, the crystal structure with PDB code 2GJW was used as the initial input structure (). During this set-up, the protonation states of the charged residues were determined based on their local environments. Then a stochastic boundary condition () was set up with this complex structure overlapped with a 25 Å water sphere, centered at one of the scissile phosphates. During molecular dynamics simulations, the atoms within the sphere of the radius of 22 Å around the same center were treated as the dynamic region; the atoms in this region was propagated with regular Newtonian dynamics using the leapfrog integrator, and 1 fs time step was used. The atoms in the layer between the radii of 22 and 25 Å were treated as the buffer region; the heavy atoms of this biopolymer complex in this region were harmonically restrained with the force constants scaled linearly with their distances from the sphere center, and the force constants around the boundary of this 25 Å sphere were set effectively the same as crystal B factor implies. In the buffer region, Langevin dynamics was applied with the friction coefficients set also linearly scaled with their distances from the sphere center and the friction coefficients around the boundary 25 Å sphere were set as 60. The atoms beyond 25 Å sphere were fixed throughout the simulations and their charges were scaled based on the electrostatic equations. CHARMM 27 forces fields () were utilized as the potential in these simulations, and the water molecules were described using the TIP3P model. For the non-bonded interactions, an extended electrostatic treatment was applied with the electrostatic interactions within 12 Å described by the group based coulomb interaction and these interactions beyond 12 Å described by the multipole expansion (). The simulations were performed with the temperature set as 298.15 K. For the unbound state simulations, the RNA BHB motif also from 2GJW structure was used as the simulation input. Then, a periodic boundary was set with this structure overlapped with a water box sized as 61 × 37 × 37 Å and during the dynamic propagations, constant pressure condition was employed with the pressure set as 1 atm. Dependent on the distance from RNA, the effective ionic strength ranges from 1.6 M (RNA contact region) to 0.1 M (bulk), which agrees with the study by Draper (). The simulations were treated the PME algorithm (). All the other simulation details are the same as those in the bound state simulations. During the umbrella sampling simulations (), the restraint potentials were added with the reference dihedral angle Φ incremented by 1° every 1 ns molecular dynamics simulation, starting with the ‘in-line attack’ conformation captured in the co-crystal structure for both unbound and bound states. The quadratic forms of the restraint potentials were applied and the restraint force constant was set to be 30 kcal/mol/rad. Upon the completion of the dihedral space scanning, weighted histogram analysis method (WHAM) () was utilized to generate the potentials of mean forces along the reaction coordinate Φ. All the calculations were performed using the program CHARMM (). In order to analyze various effects in the enzymatic catalysis of certain reaction, the free energy profile of the corresponding reaction in the aqueous solution is usually set as the reference (the curve colored in blue in A). As discussed in our introduction, upon binding to the enzyme, the substrate may take a tighter conformation, which is closer to the transition state. If we align the free energy profiles of the enzymatic reaction and the solution reaction by setting the free energies at this tight binding conformation to be the same, as shown in A, the free energy difference between the minimum conformations of the reactants in the enzymatic reaction and in the solution reaction can be the quantitative measure of the conformational effect contributing to the catalysis. This conformational effect plays a role in bringing a conformation far from the transition state, which is the preferred structure in the aqueous solution, to a closer one. Specifically, in the RNA cleavage catalysis by the splicing endonuclease, this conformational effect represents the prepaid free energy penalty required to reach the ‘in-line attack’ conformation due to the binding to the enzyme. The reduction of this energy barrier solely due to the binding of the enzyme is a conformational entropic effect and reflects a population shift from the unbound state (free in the aqueous solution) to the bound state (in the enzymatic environment). Based on the same free energy profile alignment in A, the free energy difference between the transition states in the solution reaction and the enzymatic reaction contributes to the rest of the barrier reduction, which is mostly electrostatic effect (). Free energy simulations were carried out on the BHB motif RNA in both the unbound and the bound states. Specifically, umbrella sampling simulations () were performed with the dihedral angle (C3′–O3′–P–O5′) of the second bulge nucleotide set as the reaction coordinate Φ (B). Changes in Φ displayed strong correlation with the changes of the in-line geometry defined by angle θ (O2′–P–O5′) (B), as manifested in our simulation results (). Other dihedral angles that include (C2′–C3′–O3′–P) and (O2′–C2′–C3′–O3′) did not show any correlation with the in-line geometry angle and were not adopted as the reaction coordinates. Free energy profiles were thus generated as a function of Φ for both the bound and the unbound RNA (). As shown in (red-colored curve), the global minimum of the BHB RNA in the enzymatic environment is located around Φ = −20°, which corresponds to the region of the optimal in-line geometry (θ = ∼155°−160°) () and the dihedral angle observed in the co-crystal structure for the 5′ cleavage site (Φ = −15°) (A, lower). In contrast, the global minimum of the BHB RNA in the aqueous solution is shifted to Φ = 45°, which corresponds to a region far from the ‘in-line attack’ conformation (θ = ∼110°−120°) (, blue). According to the scheme for estimating the catalytic effect of the ‘in-line attack’ conformation (), these free energy profiles yielded a value of 1.2 kcal/mol [G(Φ = −20°) – G(Φ = 45°)] corresponding to an 8-fold reaction rate acceleration. As shown in , the free energy uncertainties in the conformation regions determining this value are very small (∼0.04 kcal/mol) compared with the determined catalytic effect 1.2 kcal/mol. It is noted that in the bound state simulations, the free energy profile in the range from −175° to 110° cannot be determined because they have very high free energy values caused by the structural clashes between the RNA substrate with the protein environments. Since this region is a high free energy one, it does not contribute to the catalysis. Clearly, the ‘in-line attack’ conformation corresponds to a range of the dihedral angle Φ (−25–10°) (or θ > 155°) (shaded green areas in and ). A more accurate estimate of the catalytic effect of the ‘in-line attack’ conformation requires taking all the ‘in-line attack’ geometries into consideration. Based on the calculated free energy profiles (), the normalized population (normalized in the range of 360°; for the bound structure, the region from −175° to 110° contributes nearly zero weight due to its high free energy values) curves can be obtained (). Then, we can get the ratio of the normalized populations in the ‘in-line attack’ conformation regions for the unbound and the bound states; here, the normalized populations are equivalent to the relative populations between the ‘in-line attack’ conformation regions and the overall conformation spaces. This ratio should give us the quantitative measure of the ‘in-line attack’ conformational effect, which led to an 11.4-fold reaction rate acceleration. The rate acceleration remained at similar values when the average in-line angle was shifted to 145° (12.2-fold) or 160° (10.8-fold), suggesting that this measure is robustly insensitive to the precise definition of the in-line geometry. Interestingly, our calculated value of the reaction rate acceleration for the splicing endonuclease based on free energy simulations quantitatively agrees with that previously obtained on the spontaneous RNA transesterification solely due to the ‘in-line attack’ conformational effect. Soukup and Breaker () determined the rate acceleration for cleaving a near perfect in-line nucleotide to be 12-fold by comparing the spontaneous cleavage rates of an evolved ATP aptamer in the presence and absence of ATP. The agreement between our computed rate acceleration in an enzyme-catalyzed reaction and that of the spontaneous RNA transestification of an unrelated RNA underlines a generally moderate contribution of the ‘in-line attack’ conformational effect on the RNA cleavage. Based on the work on ATP aptamer (), the Breaker group () speculated that the ‘in-line attack’ conformational effect (referred as α catalysis) in the enzymes is very unlikely to exceed 100-fold. Our result on splicing endonuclease strongly supports this speculation. The ∼12-fold increase from the ‘in-line attack’ conformational effect is insignificant compared to the overall ∼10-fold increase of the reaction rate. As shown in , the apparently modest contribution from the ‘in-line attack’ conformational effect may be understood from the relatively shallow energy profile and the broad conformation distribution around the ‘in-line’ geometry of the unbound state. Consequently, even though the ‘in-line’ conformation of the substrate has a narrow distribution in its bound state, achieving this conformation through the enzyme binding can only result in a relatively minor reduction in the free energy barrier. Quantitatively, an upper limit on the conformational effect in catalysis can be estimated using 1/ρ, where ρ is the probability of the tight binding conformation distribution in solution. For instance, chorismate mutase has a relatively low population of the tight conformation in solution, and correspondingly the conformational effect in chorimate mutase can provide nearly 2 × 10-fold rate enhancement (). Hence, only when the tight conformation distribution of the substrate has very small population of the tight conformation in the unbound environment, it is possible for the enzyme to lower the free energy barrier substantially through the conformational control strategy. We can conclude that the conformation distribution of the substrate prior to, rather than after the enzyme binding, determines the upper bound of the rate enhancement through the conformational strategy than previously thought. Specifically, our simulation result appears to contradict to the general belief that the ‘in-line’ conformational effect in RNA is important for the catalysis of RNA cleavage. The tight conformation populations for the substrate are typically high in the enzymatic environment, because the intrinsic driving force for the transition state stabilization can effectively bias the reactant conformation toward the transition state structure. So for an enzyme with multiple substrates, the conformational effects for the catalysis of the reactions involving these substrates tend to be controlled by their conformational distributions in solution, although their bound conformations are usually similar. Results from this study suggest that other catalytic effects, such as the electrostatic stabilization of transition state, play more important roles in the RNA cleavage. Efforts in computation are currently being made to advance our understanding of the detailed path of RNA cleavage.

, a genus of thermoacidophilic crenarchaeotes, has provided much of the information currently available on the physiology and molecular biology of archaea from geothermal environments. Seminal studies of spp. have addressed, for example, chromatin-binding proteins (,), replication (), cell cycle (), repair (), transcription (), translation (,) as well as metabolism (). The genome sequences of three species, and , have been published (). Microarrays for these organisms are commercially available and proteomic studies have been undertaken (,). On a practical level, these advances reflect the relative ease with which spp. are manipulated in the laboratory. cells can be grown aerobically and heterotrophically on a variety of complex and defined carbon sources, either in liquid media or on plates, with doubling times as short as a few hours. Since spp. are hyperthermophiles with optimal growth temperatures around 80°C, their proteins are intrinsically stable and resistant to proteolysis. As a result, enzymes expressed in mesophilic hosts can often be purified with the aid of a heat step, which removes most proteins of the host. The structural rigidity of thermophilic proteins also appears to be an advantage for crystallization, which is a prerequisite for X-ray analysis of 3D structure. Ultimately, however, the comprehensive study of molecular phenomena in any organism requires genetic analysis and manipulation . Although numerous plasmids and viruses have been reported in spp., the development of these natural genetic elements into experimentally useful tools for has lagged behind the corresponding progress made with methanogenic and halophilic archaea (). Although different plasmid and virus-based vectors have been constructed (), to our knowledge only SSV1-based viral vectors () have been successfully applied by other groups to analyze genes . We used the plasmid pRN1 as starting point to construct shuttle vectors for . This plasmid is notable for its relatively small size (5.4 kb), copy numbers ranging from 10 to 20 in mid-log phase, and the three genes ( and ) that are conserved in other plasmids. In previous work, we have analyzed the three conserved proteins (representing two DNA-binding proteins and the replication protein) and the transcriptional activity of the plasmid (). Those studies revealed that ORF56 binds upstream of its own gene and down-regulates the expression of the co-transcript (). It therefore appears that ORF56 could be involved in regulating the copy number of the plasmid. Similarly, ORF80 is a sequence-specific DNA-binding protein, but the physiological function of this protein has remained unclear. encodes the third conserved protein, a multifunctional 110-kDa replication enzyme that appears to play a central role in replication. However, neither the exact molecular mechanism of replication nor the replication origin is known for pRN1. These questions would be open to experimental study by the availability of successful shuttle vectors, but the lack of this information also makes it difficult to predict which features of pRN1 must be preserved in constructing such vectors. We therefore took an empirical approach, in which an artificial transposon was inserted at many locations around pRN1. The resulting plasmids provided a series of potential shuttle vectors differing only in the relative location and orientation of inserted genes, from which the best-performing constructs were identified. The series also provided a way to reveal experimentally which regions of pRN1 may be important for successful propagation in hosts. This study used strains PH1-16 () and PBL2025 (), strains REN1H1 () R1, R20, S1, R1S1 and HVE10/4 H1 (this study), and MR31 (). Liquid cultures were grown in Brock's basal salts medium at pH 3.5 () or the mineral base of Grogan and Gunsalus (), supplemented with different carbon and nitrogen sources as indicated. Acid-hydrolyzed casein, i.e. NZAmine AS (Sigma), or enzymatically hydrolyzed casein, i.e. tryptone (BD Biosciences), were added at 0.1%. -(+)-xylose was added at 0.2%, and -(+)-lactose in ‘lactose-only’ medium at 0.4%. For growth of untransformed mutant strains PH1-16, R20, S1R1, R1, H1 and MR31, 20 µg ml of uracil was added to the medium. Plates were solidified by addition of 0.6% Gelrite (Sigma) and 10 mM CaCl. Plates and shake flask cultures were incubated at 75°C. The Tn5-derived transposon TnPA21 () was amplified by PCR using a primer (5′-CTGTCTCTTATACACATCT) complementary to the mosaic end sequence, the terminal inverted repeat sequence found on both ends of the transposon. Native pRN1 (accession number NC 001771) was isolated from a culture of REN1H1 containing only the pRN1 plasmid () using the Nucleo Spin plasmid extraction Kit (Macherey Nagel). Plasmid preparations from ∼100 ml of culture were combined and ethanol-precipitated to obtain 0.1 µg of pRN1 for the transposition reaction. The transposition reaction was carried out using the EZ-Tn5 transposase (Epicentre) according to the instructions of the manufacturer. Transposon and plasmid were mixed at a molar ratio of 1:1. The transposition reaction products were transformed into EC100 (Epicentre). The resulting transformants were screened for correctly inserted transposons by restriction digestion with SacI and NotI. Twenty percent of the screened colonies (a total of 80 plasmids) showed two restriction bands with a combined length of 7.2 kb and were kept for further analysis. Thirteen of the 80 constructs were chosen after more precise mapping of the insertion sites by restriction digestion. From these, the replicon (R6Kγ origin of replication and gene) introduced by TnPA21 was excised using the NotI and PspOMI sites present on the ends of the transposon sequence. The resulting pRN1 fragments interrupted at different sites were then cloned into the NotI linearized vector delta2pyrEF. The plasmid delta2pyrEF is a derivative of pBluescript with the and f1 origin regions deleted. Specifically, pBluescriptSKII(+) was cut with SspI and KpnI, re-ligated, cut with SacI and SapI and re-ligated. The genes from P2 (plasmid pBSKP-pyrEF, generously provided by Christa Schleper) were cloned into the SalI and PstI sites. For constructs pA to pN the transitional region from the delta2pyrEF part to the pRN1 part was sequenced to determine the exact insertion site and the direction of the transposon insertion and the direction of the cloning into the NotI site of delta2pyrEF. In construct pG, by restriction analysis using HindIII an additional HindIII site was found to be present and confirmed by sequencing. As the pRN1 part was not PCR-amplified, but stems from the native plasmid, we conclude that pRN1 had this mutation already when isolated, or that this mutation was introduced in during propagation of pG. Except for this point mutation, pG corresponds to the expected sequence. To construct pJ we cloned the α expression cassette () into the unique SacII restriction site of pJ. For transformation into shuttle constructs were methylated at the N4-position of the inner cytosine residues of GGCC recognition sequences to circumvent restriction by the SuaI restriction enzyme (). Plasmids were methylated as previously described () by transforming the shuttle constructs into ER1821 (New England Biolabs) bearing the additional plasmid pM.BC4I (New England Biolabs). Complete methylation was confirmed by the absence of any cutting after incubation with 5 U HaeIII for 1–4 h. Constructs were electroporated either using a Gene Pulser I or Gene Pulser II (BioRad) following the protocol of Schleper . or using a Gene Pulser Xcell (BioRad) with a constant time protocol with input parameters 1500 V, 10.2 ms, 2 mm cuvettes or using the protocol described by Kurosawa and Grogan () (1250 V, 1000 Ω, 25 µF, 1 mm cuvettes). For , regeneration was done for 30–40 min in tryptone/xylose medium, water or recovery solution before plating on tryptone/xylose plates or NZAmine/xylose plates. Best results were obtained with recovery solution. Recovery solution was prepared as a 2× concentrated solution (=1% sucrose, 20 mM β-alanine/1.5 mM malate buffer, pH 4.5, 10 mM MgSO). Directly after electroporation the 50 µl cell suspension was mixed in the cuvette with 50 µl of 2× recovery solution (room temperature), transferred into a 1.5-ml tube and incubated for 30 min at 75°C in a benchtop shaker at 600 r.p.m. before plating the cells. For lactose utilization in , plating after electroporation was not feasible. Instead, electroporated cells were regenerated for 10 min in 1 ml of Millipore water (pre-warmed) at 75°C and then directly transferred into pre-heated lactose medium and cultivated in 50 ml flasks. One microliter of genomic DNA preparation or 1–5 µl of plasmid prepared from by alkaline lysis were transformed into RbCl-competent XL1-Blue cells or into the deficient strain ER2267 (New England Biolabs). Copy numbers of the different shuttle constructs were determined as already described () by qPCR and cell number determination through plating, respectively. Genomic DNA was prepared from 1 ml of culture using the Chemagic DNA Bacteria Kit (Chemagen, Baesweiler, Germany) according to the instructions of the manufacturer. After digestion with either HindIII for constructs pA–pN or SacI for pJlacS restriction fragments were resolved in 1% agarose gels, transferred to a Hybond N membrane (Amersham) by capillary transfer, fixed by UV irradiation for 5 min on a UV transilluminator and hybridized to digoxigenin-labeled probes complementary to the gene (position 9–320 from the start of the gene from P2) and pRN1 (position 4892–5048 in pRN1) for pA to pN or (position 1124–1438 from the start of the gene from REN1H1) and pRN1 for pJ. Labeling, hybridization (50% formamide, 42°C), washing (0.5× SSC, 60°C) and detection was done using the PCR DIG Probe Synthesis Kit and the Digoxigenin Labeling and Detection Kit (Roche). Colonies from plates were transferred to Hybond N membranes and subsequently incubated for 10 min on a filter paper soaked with 0.5 M NaOH, 1.5 M NaCl then for 10 min on a filter paper soaked with 1 M Tris–HCl (pH 7.5), 1.5 M NaCl, then for 5 min on a filter paper soaked with 10× SSC. Membranes were cross-linked for 5 min on the 10× SSC filter paper using a transilluminator. Hybridization and detection were done as described for Southern blots with pRN1-specific probes. For convenience, the broad-specificity β--glycosidase (,) encoded by the gene was assayed as β-galactosidase activity. Crude extracts were prepared by a freeze–thaw method () in which cells were re-suspended in 50 mM Na-phosphate buffer, pH 7, and subjected to five freeze–thaw cycles (−196°C/+50°C). After centrifugation for 30 min at 13 000 r.p.m. the supernatant was stored at −20°C, or assayed directly. All β-galactosidase assays were conducted in triplicate in a 75°C bench top shaker. The reaction mixture consisted of 1 µl of crude extract (or water for blanks), 92 µl of 50 mM Na-phosphate buffer, pH 7 and the assay was started by addition of 7 µl of 12 mg ml ortho-nitrophenyl-β--galactopyranoside (ONPG) solution. Incubation was continued for 5 min before the tubes were rapidly cooled on ice and 100 µl of 1 M NaCO solution was added to stop the reaction. Concentration of ONPG was subsequently determined in a 96-well plate in a plate reader at 410 nm using a standard curve generated with ONPG. Protein concentration of the crude extracts was determined by the method of Ehresmann (). A qualitative β-galactosidase assay was based on hydrolysis of 5-bromo-4-chloro-3-indolyl-β--galactoside (X-gal). For liquid cultures, 200 µl of culture were mixed with 20 µl of substrate solution (20 mg/ml in dimethylformamide) and incubated at 75°C until color development was observed. To score colonies, plates were sprayed with the same X-gal solution and incubated at 75°C. Small cultures (0.2-ml each) were produced under selective conditions by transferring colonies from selective plates to xylose/tryptone medium without uracil. After 2 days of incubation, the cultures ( = 2 to 4 per construct) were sampled, diluted in sterile buffer and plated on plates with and without uracil supplementation; from the resulting colony counts, the numbers of Pyr and Pyr cells in the population was determined. The process was repeated after two cycles of transfer to uracil-supplemented liquid medium (3% inoculum), each involving growth to a final density of ∼4 × 10 CFU/ml. This resulted in a total of three measurements per population and an overall numerical expansion under non-selective conditions of ∼10. In principle, shuttle vectors can be constructed from two plasmids that replicate in different hosts simply by fusing them at two points that preserve all the important functions of each plasmid. However, in the case of pRN1, it was not clear which ORFs or intergenic regions may be important for successful replication in hosts. We therefore used transposition to generate pRN1 constructs interrupted at a number of different sites without regard to the location or its sequence context. From the initial transposition mixture, 13 distinct insertion points were chosen for further development, which included addition of the genes of as selectable marker (, ). In addition to providing more chances for a successful construct, this unbiased approach allowed us to evaluate possible differences in the performance of the vector constructs in . This would provide some of the first functional data regarding which of the conserved open reading frames are important for plasmid replication and maintenance. As we expected the replication operon to be essential, only one construct interrupted within this region was chosen for analysis. The other constructs were chosen to have the interruption sites distributed as evenly as possible over the remaining part of pRN1. The open reading frames and are also interrupted in at least one construct. We have already shown that and are very unlikely to play a role in plasmid replication or maintenance in view of their very low levels of expression (). The plasmids pA–pN were electroporated into strains representing different species. As the genes provide the selectable marker, stable uracil auxotrophs were needed as recipient strains. gives an overview of the mutants tested as recipients for the various pRN1 constructs. For PH1-16 and the different mutants, the vectors seemed to be unstable, as only very low amounts of shuttle vector could be detected in some experiments. We did observe growth under selective conditions and positive PCR reactions with pRN1-specific primer pairs, but never observed positive results in Southern blots. Thus, the transformed cells seemed to lose the vector rapidly, and the continued growth observed on uracil-free medium may have been due to reversion, or recombinational conversion of the mutations to the wild-type sequence. For MR31, electroporation yielded distinct, rapidly growing colonies on uracil-deficient plates, and growth of these primary transformants was maintained in uracil free-liquid medium (tryptone/xylose or NZAmine/xylose). Sequencing of the chromosomal locus revealed that no reversion of the mutated sequence had occurred in these clones, consistent with the nature of this mutation (an 18-bp deletion). Only one shuttle construct, pM (disrupted replication operon ), consistently failed to yield Pyr transformants in MR31. Using Southern blots we examined whether the shuttle constructs pA–pN were present in the clones selected after electroporation (A). This analysis confirmed that the vector constructs had the correct size and did not integrate into the host genome, as no bands in addition to the expected ones for the episomal form of the vectors were observed. In addition, none of the vector constructs were observed to undergo large rearrangements in , with the exception of pB. For this construct, an additional band of ∼6 kb was observed in the Southern blot, which indicated a rearrangement occurring in the host. The Southern blot was repeated with a second enzyme (SacI) and again we did not have any indication for rearrangement of the plasmid constructs except pB or for integration into the host genome (data not shown). Next, we tested by retransformation experiments if the original shuttle plasmids could be recovered intact from transformants. As shown in D, 2 to 30 retransformants per construct were checked by restriction analysis and only the correct restriction pattern was observed. In the case of pA, pC, pD and pE, the shuttle plasmids were also isolated directly from transformed cultures and analyzed by restriction digestion (E). From the initial set of constructs, plasmids pC and pE were chosen to evaluate long-term stability under selective conditions. Cultures of pC and pE transformants were cultivated continuously for ∼200 generations without uracil supplementation. Then retransformation experiments and Southern blots were repeated (B), with the same results. The direction of the insertion in a given region does not influence performance of the vector. The vectors pF, pI and pG, for example, have insertion sites within 15 nt of each other. In pG, the genes are oriented clockwise, in pI and pF counter clockwise, without detectable effects on plasmid stability or growth (). In addition, the growth phenotype of transformed cells is comparable to that of the untransformed recipient strain when supplemented with uracil. To test if the selection for uracil prototrophy ensures that every cell contains a shuttle vector, cells were plated on selective NZAmine/xylose medium and on non-selective tryptone/xylose medium supplemented with uracil. In eight different experiments comparable colony numbers were obtained on selective and non-selective plates showing that no cells escaped the selection. To prove that the vast majority of cells contained a shuttle vector, cells transformed with constructs pC and pE were plated on non-selective plates and examined by colony hybridizations with pRN1 specific probes (C). The facile generation of Pyr colonies by electroporation and direct plating on selective medium indicated that all the constructs tested, except for construct pM, could replicate in under appropriate selection. In order to provide a more stringent and quantitative comparison of these constructs, we monitored their retention in populations growing in non-selective, uracil-supplemented liquid medium. Specifically, the fraction of Pyr cells in the population at three different times was determined, by dilution and plating on uracil-supplemented and unsupplemented plates. C shows the retention of 13 constructs over ∼10 generations, corresponding to ∼1000-fold numerical expansions of the host cell populations. Most constructs showed measurable loss under these conditions, resulting in ∼10% Pyr cells in the cultures. In a few cases, however, plasmid retention was much lower. The most severe instability was seen in construct pH, in which the gene is interrupted. This result provided evidence that the small DNA-binding protein encoded by has an important role in the stable maintenance of pRN1 and related plasmids. Intermediate instability was observed for construct pJ (described below). We suspect that this construct with the very strong α promoter is a burden for the cell. Both pJ and pH yielded small or heterogeneous colonies when streaked on selective plates, consistent with the observed instability under non-selective conditions. According to qPCR results, all constructs showed copy numbers within the range of 2–8 copies per cell, except for pB that showed low copy numbers around one. For pC and pE, the time course of the copy number during batch fermentation was also determined (). The copy number increased in early and mid-log phase and decreased in stationary/death phase. This behavior has also been observed for the wild-type pRN1 plasmid (). The copy number of the wild-type plasmid in its original host strain (together with pRN2) is higher, reaching 20 copies per cell when grown on rich media containing yeast extract () and 10 copies per cell when grown on tryptone media. When pRN1 alone is present in its original host () the copy number is only about two. The shuttle vectors therefore maintain a similar copy number in as the native plasmid pRN1 in the original host strain REN1H1. To test whether the vector tolerates the insertion of sequences containing expressed genes we cloned the rather strong α promoter () together with the gene into shuttle construct pJ generating the vector pJ. The stability of this construct was tested by retransformation into and Southern blotting (A and B). The construct turned out to be stably replicated in . Staining with 5-bromo-4-chloro-3-indolyl-β--galactoside (X-gal) revealed that the β-glycosidase was expressed under the control of the heat shock promoter (C). This test could be done without prior isolation of a mutant of strain MR31 because the endogenous β-glycosidase activity is very low, i.e. ∼0.01 U/mg protein, in (). The enzyme activity was also measured quantitatively (as β-galactosidase) at different ODs of a MR31 culture transformed with pJ. Copy numbers were determined simultaneously and it was found that measured β-galactosidase activities correlated well with vector copy numbers (E), as previously observed (,). It should be noted that the β-galactosidase activity of MR31 transformed with pJ (2–11 U/mg) is much higher than the wild-type β-galactosidase activity of (0.2 U/mg) () and is comparable to the β-galactosidase activity in a viral overexpression system (1.5–5 U/mg) (). Demonstration of expression in pJ transformants suggested the possibility of a selection by conferring, or restoring, the ability to catabolize lactose or other β-glycosides. The 98/2 deletion mutant PBL2025 () was therefore tested for complementation by plasmid pJ. After three rounds of selection in liquid medium (0.4% lactose), >95% of all cells contained a shuttle vector, as shown by plating on selective lactose versus non-selective tryptone plates, and X-gal staining of colonies. On both plates equal numbers of colonies were observed, on the non-selective plate in addition to ∼300 colonies also seven white colonies were observed (). Southern blot, retransformation, growth and copy number determinations for pJ in are summarized in and indicate that pJl is stably replicated in . For direct determination of transformation efficiencies is possible, because plating of the primary electroporation mixture can be done after only 30 min of regeneration. Considering all transformations performed in the current study ( = 150), the efficiencies range from 1 × 10 to 6 × 10 transformants per microgram plasmid DNA. The batch of electrocompetent cells, electroporation protocol and regeneration procedure have an influence on the transformation efficiency, as already described (). On tryptone/xylose plates, the formation of very small colonies—that did not contain a shuttle vector—was observed in addition to the colonies of normal size that were able to grow in selective liquid medium. Controls without electroporation or without addition of shuttle vector also yielded small colonies, which were not able to grow when re-streaked on selective plates or cultivated in liquid medium. Based on their phenotype and frequency in strain MR31, we hypothesize that these ‘pseudo-transformants’ contain spontaneous mutations elsewhere in the chromosome that partially suppress the phenotype. Finally, we confirmed that complete methylation of the shuttle vectors is essential for efficient transformation of . None of the constructs pA–pN yielded transformants when unmethylated or partly protected plasmids were electroporated. Based on our results with various constructs and recipient strains, we conclude that the primary obstacle to establishing stably replicating shuttle vectors derived from plasmid pRN1 is not preservation of critical plasmid functions or identification of a required host species, but creation of a suitably reliable selection. For example, point and transposon mutants of or that showed low reversion frequencies in small scale fluctuation tests (15 × 10 – <6 × 10 reversions per cell division, unpublished data) displayed for unknown reasons higher reversion frequencies after electroporation with a shuttle construct. These problems could be avoided by the use of a deletion mutant of . In contrast to and , background growth on selective plates was not observed with . Although the basis of this difference has not been established, lacks homologs of the cytosine/uracil/thiamine/allantoin permeases (SSO1905, SSO2042) present in and (ST1564) that might facilitate growth on medium with very low uracil concentrations. Under selective conditions, the vectors could be faithfully propagated in . Rearrangements occurred in only two cases, pB and pJ, and only after many generations. We do not know why pB behaves differently in this respect, although it is the only construct interrupted in between and . This region of pRN1 contains several repeats and other remarkable features like a stretch of 17 consecutive C residues (). An interruption in this region is obviously not as well tolerated as in other regions. The only interruption site that abolished shuttle vector replication completely was that of construct pM, and is situated within the co-transcribed replication operon . For pM no viable transformants could be isolated. The other conserved open reading frame, , also called (asmid egulatory), that is present on almost all sequenced genetic elements of () is interrupted in pH. Interestingly, pH shows growth comparable to the other constructs and yields the same transformation efficiencies. Therefore seems not to be essential for replication of pRN1, at least not when selective pressure is applied. However, under non-selective conditions, construct pH was lost at a much faster rate than any other construct that could be successfully established in . The relative instability of this construct provides the first experimental evidence that the DNA-binding protein ORF80 has an important role in stable maintenance A of pRN1. The instability of this construct may also have practical uses. For example, it may facilitate transfer of -marked genes to the host chromosome, by allowing such genes to be first established on an episome, and then stabilized in the population by recombinational integration at the homologous locus. Many shuttle constructs developed so far for hyperthermophilic archaea have been observed to rearrange in (), which hampers the use of these systems. Some of these problems may, in principle, be circumvented by the use of strains designed specially for dealing with unstable constructs. Additionally reducing the growth temperature to 30°C and using only 50 µg ml of ampicillin is necessary to prevent rearrangements in the pMJ vector system (). We observed rearrangements for the construct pJ in one out of 10 preparations of this plasmid in XL1-Blue cells at 37°C and 100 µg ml of ampicillin. In general, however, the constructs pA to pN seemed to be fully stable in . In particular, we never detected rearrangements in retransformed plasmids. In Southern blots, faint traces of plasmid rearrangements were visible for some preparations from but did not interfere with successful transformation of . We have developed multicopy, non-integrative, plasmid-based shuttle vectors that are very stable in both hosts, and are suitable for the use in protein expression and reporter gene studies. Transformation is rapid and simple, involving electroporation of stable mutants and plating on uracil-deficient media. The constructs are small, enabling direct cloning into unique SacII/XmaI and NotI restriction sites. The host range so far comprises and , the two most widely used and best-studied species of for which genome sequence information is available. The presence of the shuttle constructs in the cells does not cause significant growth retardation and there is limited risk of accidentally contaminating cultures because the vectors are not infectious. It should be emphasized that performance of these shuttle constructs has now been confirmed independently in three different laboratories using slightly different electroporation and cultivation protocols. In addition, the use of as recipient strain has certain practical advantages which somewhat mitigate the inconvenience of requiring specific DNA methylation. does not contain any integrated copies of pRN1 or genes homologous to pRN1 genes. This enables detailed experiments on essential regions and proteins for pRN1 replication and maintenance without interference from plasmid-gene homologs located on the host chromosome. Because of the low sequence similarity between and there is also minimal risk of undesired homologous recombination when cloning genes of into the shuttle vector, e.g. for protein expression. is the only species so far that does not contain active insertion sequences and seems to be genetically stable (). In addition, it is the species showing the highest growth rate with doubling times of around 3–4 h during exponential growth, and exhibits efficient homologous recombination (). In this context, the series of pRN1 shuttle vectors we have constructed promises to add detailed genetic analyses to the already advanced biochemical characterization of various gene products.

We describe ‘GenVar’, a comparative genomics analysis computational pipeline, whose aim is to improve existing bacterial genome annotations as well as reveal indel polymorphisms in already-annotated protein-coding genes. GenVar is based on GeneWise, a program to analyze DNA and protein sequences that helps eukaryotic gene structure analysis (). Even though GeneWise is aimed primarily at eukaryotic genomes, we have found it useful and effective as a platform upon which to develop GenVar. Manual annotation is the currently agreed upon ‘gold standard’ to provide quality genome annotation (). This gold standard, however, does not scale or keep up with the increasing pace of microbial genome sequencing. Among the main challenges in the manual annotation process are accurately identifying missed gene calls and split genes in existing genome annotations. These are the two main features addressed by GenVar. There are two possible causes of split genes: sequencing errors or mutations. Both can cause ORF truncations or over-extensions, thus creating annotation errors if left uncorrected. Comparative analysis of closely related genomes can provide important clues to help distinguish these two cases, and GenVar also provides such clues when possible. A brief description of the GeneWise program () is necessary for a better understanding of GenVar. GeneWise combines hidden Markov models for gene prediction and for alignment, thereby making it possible to compare a single protein sequence directly to genomic DNA. The models take into account known statistical properties of genes as well as the possible presence of ‘sequencing errors’ or problems in translation. GeneWise will take genomic sequence and compare it to target protein sequences (assumed to be homologous) considering all possible ‘intermediate’ predicted sequences given by the gene prediction part of the combined model. The GenVar pipeline is composed of three conceptual steps. The first two steps generate GeneWise-required inputs: a set of protein database inputs (gwpDBs) and a set of genomic DNA inputs. Each gwpDB contains orthologous proteins from a limited number of closely related species. The genomic DNA inputs are selected to represent extended regions of both predicted genes and putative intergenic regions. By breaking up the genome to be studied and the reference protein sequences into small units, we decrease the computational cost that would be incurred if GeneWise were to be used starting from the entire genome and comparing it to a general set of protein sequences. The third step comparatively analyzes missed gene calls and sequence variants among closely related species. Sequence variants are defined as genes with frameshifts, premature stop codons, insertions and deletions. Missed gene calls are DNA regions described as intergenic in the original genome annotation that can be fully aligned with gene calls in closely related or otherwise well-annotated genomes. Once the sequence variants are identified, the variants within the context of closely related genomes are classified using a simple classification scheme. This scheme facilitates the correlation of sequence variants with phenotypic differences in the species studied; it also identifies a list of frameshifts and premature stops that may be sequencing errors. We are especially interested in pathogenic bacteria. Our underlying assumption is that gene disruptions (true split genes) and indel polymorphisms play a key role in host–pathogen evolution. The literature provides several cases of this connection. For example, split genes of the major surface protein 2 (MSP2) determine antigenic variation in the tick-transmitted pathogen (). An array of variable proteins is the source of diversity in host tropism and disease causation in the obligate intracellular bacterial pathogen (). Gene inactivation, loss and acquisition are hypothesized to be the main mechanisms that contribute to fitness and promote its adaptive microevolution (). For the development and testing of GenVar, we chose the bacterial pathogen Our motivation was as follows: three out of the six recognized taxonomic species have been sequenced, including one , one and two strains, which represent the most virulent species to humans (). In addition, is one of the world's major zoonotic pathogens for which there is no human vaccine (). Although highly similar in terms of gene content (), the six species have preferential host specificity: goats (), cattle (), swine (), dogs (), sheep () and desert mice () (). Thus the choice of offers a unique opportunity to assay and improve the quality of current genome annotations and to identify unique genetic factors that may help explain the pathogen's niche as a facultative intracellular pathogen (). Finally, is a priority pathogen of the National Institute of Allergic and Infectious Diseases (). We participate in the development of a Web resource () devoted in part to (), and GenVar is being used in that project. We have analyzed four genomes: 16M (), 1330 (), 9-941 () and 2308 (). Each of these genomes has two chromosomes (2.1 and 1.2 Mbp approximately). Our results indicate that GenVar was able to improve the existing annotations of these genomes. The analysis revealed hundreds of missed gene calls and dozens of new, probable split genes. Please note, however, that the results presented are meant to demonstrate the versatility of this new tool, rather than being an exhaustive list of every possible missed gene call, split gene or polymorphic indel in the genomes studied. Obtaining such results would require a much larger input database than the one we used. Four genomes including 16M, 1330, 9-941 and 2308 were downloaded on 4 December 2005 from the National Center for Biotechnology Information (NCBI) (ftp://ftp.ncbi.nih.gov/genomes/Bacteria). The published genome assembly fold coverage for these genomes are as follows: 16M: 9X (); 9-941: 10X (); and 2308: 7X (). 1330 does not have a published fold coverage, but we assume that it is 7X or 8X based on other genomes published by The Institute for Genomic Research. The genomes of C58 (Cereon) and MAFF303099 (both are alphaproteobacteria, like ) and K12 (generally regarded as the best annotated bacterial genome) were also used by GenVar and downloaded from the same source. Also used was the Swiss-Prot protein database (UniProt Knowledgebase Release 7.3), downloaded from . Gene functional assignments reported are those obtained from the mentioned sources. GenVar is based on the GeneWise program (,). It is partitioned into three conceptual steps, as described in the Introduction section and detailed below. The GeneWise program was downloaded from . A GenVar run is entirely automated. A run of GenVar on the 1.2 Mbp chromosome takes ∼6 h on a 500 MB RAM Pentium 4 computer running Linux. Running time increases linearly with sequence length. The first step is designed to establish, for each query genome feature (QGF), a gene-specific protein database (gwpDB), the first input for GeneWise (, panel I). A QGF is either a protein-coding gene from the existing genome annotation or a DNA region between two immediately adjacent protein-coding genes on the chromosomes (intergenic DNA regions). The gwpDB is constructed from BLAST () analysis of the QGF on a species-specific protein database. The protein database consists of proteins from closely related genomes and also those that are well annotated (see above). Consequently, the gwpDB of the QGF would include a small number of proteins yet cover all its paralog and orthologs from closely related genomes as well as from well-annotated protein sequences. The second step is devised to generate meaningful searchable DNA regions (SDR), the second data input for the GeneWise program (, panel II). By ‘meaningful’ we mean that an SDR needs to cover sequence variants involved in the protein-coding genes and missed gene calls so that they can be detected. For this purpose, the SDR is defined as a DNA region of a predicted protein-coding gene with 300 bp extensions both upstream and downstream (needed to identify split genes and indels) or an intergenic DNA region between two adjacent protein-encoding genes (needed to identify missed gene calls, besides sequence variants). The next step actually runs GeneWise and parses GeneWise outputs to identify the sequence variants and missed gene calls (, panel III). In this study, a protein-coding gene or an intergenic DNA region is considered to contain a sequence variant if and only if such variant is detected when compared to its orthologs. The orthologs are determined by the best BLAST hits from each closely related genome. Furthermore, frameshifts and premature stop codons detected in the sequence variants are further mapped at specific chromosomal locations; indels are mapped on the proteins coded by the orthologs. GenVar's output is then interpreted and linked to different organismal properties such as host specificity, host–pathogen interactions and other pathogenicity-related traits found in species (, panel IV). Classification schemes were designed to determine whether the identified split genes resulted from sequencing errors or genuine mutation and to discover species-specific/selective gene disruption and their possible patterns. Such patterns can suggest the specific association between occurrences of sequence variants and pathogenicity properties. For genes with indels, the classification scheme is simple, relying on the genome association. For split genes, the classification scheme is more complex, depending on the length of the alignments of split genes with their orthologs, the status of assigned biological functions of the orthologs and genome associations. Different genomes will have different associations. For example, the split genes from 2308 were classified into six groups (). The occurrences of these sequence variants among closely related genomes in particular protein complexes were then compared to reveal possible patterns in which genes are selectively disrupted or modified. is publicly available to noncommercial users at . GenVar results are being used to reannotate all four genomes by the PATRIC project (); some of these reannotations already are publicly available through the PATRIC website (); eventually all results will be incorporated into the reannotations and deposited in GenBank. For each split gene predicted in the S19 genome, we obtained by PCR 60 bp around the predicted disruption, resequenced this fragment, and compared to the original sequence. #text italic #text S u p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

The α and β isoforms of mammalian DNA topoisomerase II (topo II) decatenate double-stranded DNA and are involved in numerous cellular processes including replication, gene transcription, chromosomal segregation, differentiation and apoptosis as well as playing an important role in chromatin structure and remodelling (). Topo IIα is found mainly in proliferating cells and its expression levels vary substantially in different phases of the cell cycle (). In contrast, topo IIβ is expressed at constant levels throughout the cell cycle and in a broad range of cell types, with higher expression levels during embryogenesis and in tumour cells (,). Topo IIβ has also been implicated in cellular maturation and differentiation in the brain (,). The two mammalian isoforms have comparable catalytic activities in that both can complement functional defects in yeast that conditionally lack the single yeast topo II isoform (). However, -/- mice die early in embryogenesis because nuclei fail to divide properly (), while -/- mice die at birth due to defects in diaphragm muscle innervation and have significant abnormalities in neurogenesis (,). Thus, topo IIα is essential for cell division, while topo IIβ serves a critical role in development. In addition to their roles in cellular proliferation and differentiation, topo II α and β are well-established targets in cancer chemotherapy (). Clinically important and widely used topo II-targetting drugs such as doxorubicin and etoposide are considered topo II poisons because they stabilize cleaved DNA/enzyme complexes, leading to the induction of apoptosis (,). Other drugs are known as catalytic inhibitors and act at different stages of the topo II catalytic cycle, such as when topo II binds to DNA or ATP (). There is a substantial body of compelling evidence indicating that changes in topo II expression, function and/or localization can play a major role in the response of cancer cells to drugs that target these enzymes (). Both topo II α and β consist of three domains: the NH-proximal ATPase domain, the central DNA cleavage and religation domain (both of which are highly conserved), and the divergent COOH-terminal domains (CTDs) (). Mutant topo IIα lacking its CTD retains activity , indicating that the CTD has more of a regulatory function rather than being essential for catalytic activity (,). The CTDs of topo II contain most of the utilized phosphorylation sites which are COOH-proximal to the enzymes’ nuclear export sequences (NES) (). The CTDs also contain nuclear localization sequences (NLS) () and mutations in these sequences (at least in topo IIα) lead to cytosolic localization of the protein and confer drug resistance (,,,). Quite recently, Linka . () reported convincing evidence that the topo II CTDs are important determinants of the isoform-specific functions of these enzymes. Previous studies have shown that human topo IIα and topo IIβ both participate in protein–protein interactions with a diverse range of nuclear proteins, including p53 (), retinoblastoma protein (), cyclic AMP-response element-binding protein and c-Jun (), histone deacetylase (HDAC) 1 and 2 (,), 14-3-3ε () and caspase-activated DNase (CAD) nuclease (,). However, there is evidence that the screens for topo II protein binding partners to date have been incomplete. In the present study, by using yeast two-hybrid analysis with the CTDs of topo II α and β as ‘bait’, we have identified phospholipid scramblase 1 (PLSCR1) as a novel binding partner of both isozymes. While the majority of studies of this primarily plasma membrane localized protein have focused on its role in phosphatidylserine externalization during apoptosis, PLSCR1 is also actively imported into the nucleus, where it appears to have a role in cellular proliferation and differentiation (,). Since topo II is involved in proliferation (α isoform) and differentiation (β isoform), and both isoforms are known to have a role in the cellular responses to cytotoxic drugs, our findings raise the intriguing possibility that interactions between topo II and PLSCR1 may influence tumour cell growth and drug responsiveness. ‘Bait’ plasmids pOBD2/topo IIα1152-1531 and pOBD2/topo IIβ1165-1621 were prepared by PCR amplification of pBS/hTOP2 (ATCC, Rockville, MD) and pYEShTOP2B (gift of Dr. I. Hickson, Oxford, UK), respectively, using PfuTurbo® DNA polymerase (Stratagene, LaJolla, CA, USA) and oligonucleotide primers to create Nco1 and Xma1 sites at the 5′ and 3′ ends of the PCR products, respectively. The PCR products were cloned into Nco1/Xma1 digested pOBD2. pACT2/HDAC1220-482 was supplied by Dr. B. Turner (University of Birmingham, UK) (). pGEX6P-1/topo IIαCTD (containing amino acids 1171-1531) and pGEX6P-1/topo IIβCTD (containing amino acids 1184-1621) were prepared by PCR amplification of pBS/hTOP2A and pYEShTOP2B, respectively, to add SmaI and XhoI restriction sites in the correct reading frame at the 5′ and 3′ ends, respectively, and to add two glycine residues as a flexible spacer between glutathione -transferase (GST) and the topo II fragment. These PCR fragments were cloned in-frame at the 3′ end of GST using the SmaI and XhoI sites of pGEX-6P-1 (Amersham Biosciences, Baie D’Urfé, QC, Canada). Site-directed mutagenesis was performed using PfuTurbo® DNA polymerase and mutagenic forward and reverse complementary primers to amplify pGEX6P-1/topo IIαCTD containing the desired mutations. In all cases, mutations were confirmed by sequencing prior to subcloning appropriate fragments into parental vectors. The SnaB1 site in pGEX6P-1/topo IIαCTD was created using a primer with the sequence 5′-CCCCAAAACTCAAAGAACTGAAACC-3′ and its complement (new SnaB1 site italicized). The new stop codons in pGEX-6P-1/topo IIα CTD, which created pGEX6P-1/topo IIα 1171-1431, 1171-1441, 1171-1451, 1171-1461 and 1171-1477, were generated using the following primers and their complements (newly introduced Xma1 sites are italicized and inserted stop codons are in boldface): 5′-CC ACT ACC GGT AG GCT GCC CCA AAA G-3′, 5′-CCA AAA GGA ACT AGA GCT TTG AAT TCT GG-3′, 5′-G AAT TCT GGT GTC CAT GAT CCT GCC-3′, 5′-CC AAA ACC AAG CGA AGG AAG CCA TCC-3′ and 5′-CT GAC TCT AAT GAT GTT TCG AAA GC-3’. Two-hybrid screening was performed using yeast strain PJ69-4A (ATCC 201540). Yeast harbouring a bait plasmid (pOBD2/topo IIα1152-1531 or pOBD2/topo IIβ1165-1621) were transformed with a human B lymphocyte library (ATCC 87003) (provided by Dr. D. LeBrun, Queen's University), and HIS3 and ADE2 reporter gene expression monitored on histidine and adenine deficient agar plates (interaction-selective plates). Plasmids were isolated, transformed into JF1754, and then grown on leucine-deficient agar plates. Plasmids isolated from JF1754 colonies were transformed into PJ69 harbouring pOBD2/topo IIα1152-1531 or pOBD2/topo IIβ1165-1621 to confirm interactions, or into PJ69 harbouring pOBD2 to identify false positives. Plasmids that encoded putative topo II α or β interacting proteins were sequenced and the partner proteins identified by comparison to nucleotide databases. BL21(DE3)-RIL cells (Stratagene) transfected with the pGEX6P-1/topo IIα or pMAL-C2/PLSCR1 constructs were used to express GST-topo IIα or maltose-binding protein (MBP)-PLSCR1 fusion proteins, respectively. Bacteria were induced with 1.0 mM isopropyl-1-thio-β--galactopyranoside (IPTG) for 3 h, harvested and then sonicated in PBS containing a protease inhibitor cocktail (Roche, Laval, QC, Canada), DTT (5 mM) and benzamidine (10 μg/ml). Lysates containing GST-topo IIα fusion proteins were pre-cleared by ultra-centrifugation and batch-bound overnight to GSH-Sepharose 4B (Amersham Biosciences, Uppsala, Sweden). GST-topo IIα fusion proteins were eluted with 50 mM Tris pH 7.5 containing 15 mM GSH, 5 mM DTT and a protease inhibitor cocktail, and dialysed overnight against PBS. Lysates containing the MBP-PLSCR1 fusion protein was treated in a similar manner, except that it was bound to amylose resin (New England Biolabs), eluted with PBS containing 10 mM maltose, 5 mM DTT and a protease inhibitor cocktail, and dialysed against PBS containing 250 mM sucrose. Purified fusion proteins were stored at −80°C. Protein samples were resolved by Tricine gel electrophoresis (binding assays involving recombinant PLSCR1) or SDS-PAGE (all other samples) and transferred to polyvinylidene fluoride membranes in 50 mM -cyclohexyl-3-aminopropanesulfonic acid, pH 11 (for topo II immunoblots) or carbonate buffer pH 9.0 containing 20% methanol (for all other immunoblots). Membranes were incubated with primary antibodies overnight at 4°C in TBS-T (50 mM Tris-HCl pH 7.4, 150 mM NaCl, 0.1% Tween-20) containing 4% (w/v) skim milk powder. The filters were washed in TBS-T, incubated for 1 h with the appropriate horseradish peroxidase conjugated anti-mouse or anti-rabbit secondary antibody, and then developed using chemiluminescence detection (Amersham Pharmacia Biotech). Topo II α and β specific mAbs, 8D2 and 3H10, respectively, were supplied by Dr A. Kikuchi (Nagoya University, Japan) (). A PLSCR1-specific polyclonal antiserum generated against the 14 COOH-terminal amino acids of PLSCR1 (CESTGSQEQKSGVW) was supplied by Dr D. Bratton (National Jewish Medical and Research Center, Denver, CO) (). A mAb against tubulin and a polyclonal antibody against calnexin were from Sigma Diagnostics; a mAb against lysosomal-associated membrane protein (LAMP) 2 was from Santa Cruz Biotechnology (Santa Cruz, CA, USA). HeLa cell nuclear extracts were prepared by lysing approximately 100 × 10 cells in hypotonic buffer (10 mM HEPES pH 7.6, 15 mM KCl, 2 mM EDTA, 0.5 mM spermidine, 0.5 mM spermine, 0.5% NP-40) on ice for 15 min, underlaying the suspension with a 30% sucrose solution in the same buffer and centrifuging at 2200 × for 20 min. The pellet was resuspended in nuclear lysis buffer (10 mM HEPES, 100 mM KCl, 0.1 mM EDTA, 10% glycerol, 3 mM MgCl), and KCl added to a final concentration of 0.55 M. After incubation for 30 min, the nuclear lysate was centrifuged at 100 000 × and the supernatant dialysed overnight against 50 mM Tris (pH 7.5)/150 mM NaCl. The dialysate was centrifuged at 15 000 × prior to overnight incubation with antibodies against topo IIα (mAb 8D2), topo IIβ (mAb 3H10), PLSCR1, or pre-immune serum at 4°C. Gamma-Bind Sepharose™ (Amersham) was then added for 1 h prior to collecting the beads by centrifugation. Bound proteins were eluted with sample buffer and analysed by immunoblotting with topo II α and β mAbs 8D2 and 3H10 as described above. HeLa cells were grown on gelatin-coated glass coverslips until approximately 80% confluent, fixed in 3.7% formaldehyde and then permeabilized in 1% TX-100. After blocking for 1 h in 1% BSA, cells were incubated with topo II α or β mAbs and PLSCR1 antiserum (ICN, Aurora, OH, USA). Antibody binding was detected with Alexa546™ goat anti-mouse and Alexa488™ goat anti-rabbit conjugated secondary antibodies (Molecular Probes, Eugene, OR, USA) and fluorescent images were captured using a Leica TCS SP2 multiphoton confocal microscope. GST-pull-down assays of GST-tagged topo II CTD fusion proteins and endogenous PLSCR1 were performed using NP-40 lysates prepared from HeLa cells. Cells were lysed in 50 mM Tris-HCl (pH 7.5), 150 mM NaCl, protease inhibitors and 0.5% NP-40 (lysis buffer) for 15 min on ice. The lysate was cleared by centrifugation and the resulting supernatant incubated with purified GST or GST-topo II CTD fusion proteins for 90 min at 20°C. In some cases, the lysate was treated with 40 mU/ml Turbo™ DNase 1 (Ambion, Austin, TX, USA) for 60 min at 20°C prior to the addition of GST or GST-topo II CTD fusion proteins. In some experiments, recombinant PLSCR1 was used in place of HeLa cell lysates. For these studies, the MBP tag was first removed by digesting the recombinant protein with Factor Xa and incubating with X-Arrest agarose (Novagen, San Diego, CA, USA). GSH-Sepharose 4B was then added and after 1 h, the beads were collected by centrifugation. Bound proteins were eluted with sample buffer and analysed by immunoblotting using PLSCR1 antiserum as described earlier. To measure the effect of PLSCR1 on topo II activity, recombinant MBP-PLSCR1 was digested with Factor Xa as above and ∼2 pmol PLSCR1 was incubated with 0.4 U purified human topo IIα for 1.5 h at 20°C according to the manufacturer's instructions (TopoGEN, Inc., Port Orange, FL, USA). The protein mixture was then incubated with kinetoplast DNA (kDNA) for 15 min at 37°C. After terminating the reactions, the samples were resolved on a 1% agarose gel containing ethidium bromide. Dodecapeptides corresponding to residues 1430-1441 of topo IIα (TGAKKRAAPKGT) or the same amino acids in random order (KRGATAKGTAPK) were purchased from Bio-Synthesis (Lewisville, TX, USA). After removing the MBP tag with Factor Xa, recombinant PLSCR1 was incubated with increasing concentrations of peptide for 1 h at 20°C. Purified GST or GST-topo IIα1258-1531 was then added and samples were treated as described earlier for the GST- pull-down assay. Vectors encoding the topo II CTDs expressed as NH-terminal fusion proteins with the DNA binding domain of the GAL4 transcription factor (pOBD2/topo IIα1152-1531 and pOBD2/topo IIβ1165-1621) were constructed, and expression in yeast strain PJ69 was confirmed by immunoblotting (data not shown). While the GAL4-topo IIβ construct was expressed as a single protein, several degradation products of the polypeptide encoded by the topo IIα construct were seen, suggesting that the product of this construct might be unstable or toxic to yeast. This may be a general problem in yeast two-hybrid screens using human topo IIα as bait, since a screen of a HeLa cell library for topo IIβ binding partners was successful (), as were two screens for yeast topo II (,), while no yeast two-hybrid screens for topo IIα binding partners have been reported. No growth was detected on interaction-selective plates when either pOBD2/topo IIα1152-1531 or pOBD2/topo IIβ1165-1621 were co-transformed with pACT/SNF4 (negative control) while growth was detected when co-transformed with pACT/HDAC220-482 (positive control) (,), suggesting the two topo II constructs were suitable candidates for yeast two-hybrid analysis (). Screening of the library, identified several clones as potential binding partners of both the topo II α and β CTDs, one of which was confirmed by sequencing as PLSCR1. Human PLSCR1 has 318 amino acids and the clone obtained, pACT/PLSCR1, encoded a polypeptide missing amino acids 1-29 and 118-143. Yeast co-transformed with pACT/PLSCR1 and pOBD2 failed to grow on selective plates, demonstrating that transcription was not induced by pACT/PLSCR1 independently of its interaction with the topo II CTDs. To confirm the interactions of topo IIα and topo IIβ with PLSCR1, the binding of endogenous PLSCR1 to GST-tagged topo IIα CTD and topo IIβ CTD fusion proteins was investigated using HeLa cell lysates. To eliminate the possibility that the interactions between PLSCR1 and the topo II fragments were DNA-mediated, the cell lysates were pre-treated with DNase1. However, as shown in , addition of DNase1 had no effect on the binding between either of the topo II CTD fusion proteins and endogenous PLSCR1, indicating that the interactions are DNA independent. The input lane in the immunoblot shown represents 5% of the lysate used in the binding assay, suggesting that the interactions may be relatively weak, at least under the conditions of this assay, although other explanations are possible. Of some interest, the addition of a FLAG tag to the NH-terminus of PLSCR1 completely inhibited the interaction of the protein with the GST-tagged topo IIα CTD (results not shown). The reason for this is unknown but may be related to the highly charged nature of the FLAG-tag, since the fusion of MBP to the NH-terminus of PLSCR1 did not affect its binding to either of the GST-topo II α or β CTD fusion proteins (data not shown). To confirm that PLSCR1 colocalizes with topo II α and β in intact cells, HeLa cell nuclei were isolated and the relative amounts of nuclear and NP-40 soluble PLSCR1 and topo II α and β in the two cellular fractions were determined by immunoblotting. Samples were prepared from equivalent numbers of cells rather than equal protein concentrations so that the relative fraction of total cellular PLSCR1 present in the nucleus could be determined. As shown in A, the nuclear fraction contains ∼10–20% of the amount of the 35 kDa PLSCR1 found in the NP-40 soluble fraction, an estimate that is supported by confocal microscopy (see subsequently). The identity of the 70 kDa band is unknown, but it was consistently present, albeit at variable levels. It was not detected by the pre-immune serum and may be an undissociated dimer. Topo II α and β were detected only in the nuclear fraction as expected. A lack of ER and cytoplasmic contamination of the nuclear lysates was demonstrated by immunoblotting for calnexin (integral ER membrane marker), tubulin (cytoskeletal marker) and LAMP2 (lysosomal marker) (B). To confirm that endogenous topo II α and β interact with endogenous PLSCR1 in intact cells, nuclear lysates were subjected to immunoprecipitation analysis using a PLSCR1 antiserum, followed by immunoblotting with topo II α and β specific mAbs. As shown in , both topo II α and β were immunoprecipitated by the PLSCR1 antiserum, but not the pre-immune serum, demonstrating that the endogenous proteins interact These results also show that the interaction is not an artifact caused by the use of recombinant topo II CTDs rather than the endogenous proteins, or the forced colocalization of proteins that do not normally share a compartment, as can occur in yeast two-hybrid analysis. The input lane represents 5% of the nuclear lysate used in each binding assay, suggesting that as much as 5% of nuclear topo II α and β is associated with PLSCR1. A high background associated with the PLSCR1 antiserum in co-immunoprecipitation experiments precluded consistent western blot detection of nuclear PLSCR1 immunoprecipitated by the topo II α and β specific mAbs (data not shown). The nuclear colocalization of PLSCR1 and topo IIα in HeLa cells was confirmed by immunocytochemistry (). We found that 1% TX-100 and incubation at 37°C was needed to permeabilize the nucleus sufficiently to detect nuclear PLSCR1. These conditions also substantially increased staining of nuclear topo IIα. This PLSCR1 staining pattern is consistent with a previous study of PLSCR1 in HeLa cells using a different primary antibody (). It is noted that nuclear localization of PLSCR1 and its induction by interferons can vary dramatically according to cell type (). Both topo IIα and PLSCR1 were localized throughout the nucleus although, as reported by other investigators (), topo IIα appeared more abundant in nucleoli of some cells (, Panels A and D). Not unexpectedly, given the differences in the relative abundance of the two proteins, the signal intensity for nuclear PLSCR1 (Panel B) was very low compared to that for topo IIα, making colocalization studies difficult. However, examination of the separate images showed a small increase in PLSCR1 staining in the nucleoli relative to the nucleoplasm. This was not caused by overlap of the red channel (topo IIα) into the green channel (PLSCR1), since cells treated with the pre-immune serum in place of the PLSCR1 antiserum did not display an increased green signal in the nucleoli (Panel E). Unfortunately, comparable colocalization experiments of PLSCR1 and topo IIβ were unsuccessful since staining by the topo IIβ-specific mAb under the fixation conditions required to detect nuclear PLSCR1 was poor. The ability of PLSCR1 to modulate topo II activity was assessed using a topo II assay that measures the decatenating activity of the enzyme. Thus purified topo IIα was preincubated with PLSCR1 for 1.5 h at 20°C prior to incubation with kDNA for 15 min at 37°C. As shown in , the catenated kDNA did not enter the gel (lane 5) as expected, while decatenated DNA entered the gel, but its electrophoretic mobility was slightly retarded compared to linear DNA (compare lanes 6 and 7). The addition of recombinant PLSCR1 to the assay mixture significantly increased the ability of topo IIα to decatenate kDNA (compare lanes 3 and 4 versus 1 and 2). To identify the region of topo IIα that mediates binding to PLSCR1, eight GST fusion proteins encoding NH– and COOH– truncated forms of the topo IIα CTD (A, upper panel) were generated and tested for their ability to bind endogenous PLSCR1 in a GST-pull-down assay. Only one of these fusion proteins (topo IIα1171-1431) failed to bind to endogenous PLSCR1, indicating that topo IIα residues 1432-1441 are required for binding to PLSCR1 (A, lower panel). To confirm that these residues were required, a dodecapeptide corresponding to topo IIα amino acids 1430-1441 (TGAKKRAAPKGT) was examined for its ability to inhibit the binding of GST-topo IIα1258-1531 to recombinant PLSCR1 in a GST-pull-down assay. A second dodecapeptide consisting of the same amino acids in random order was included as a negative control. As shown in B, the topo IIα peptide specifically blocked the interaction between topo IIα and PLSCR1 in a concentration-dependent manner, while the random peptide had no effect. These observations provide further confirmation that residues 1432-1441 are involved in the interaction of topo IIα with PLSCR1. The CTDs of topo II α and β are poorly conserved, making alignments of these regions difficult. Nevertheless, when residues 1424-1531 of topo IIα (exons 34-35) were aligned with residues 1493-1621 of topo IIβ (exons 34-36) (), a basic stretch of seven amino acids was identified between topo IIα residues 1433 and 1439 (topo IIβ residues 1501-1507) (C) that may form a positively charged patch that mediates the interaction with PLSCR1. These putative PLSCR1 interaction motifs are both approximately the same distance from the COOH–termini of their respective topo II isoforms and occur within 9–10 residues of an intron/exon boundary (). Neither basic motif overlaps with previously identified functional NLS or NES sequences (,,) and both are highly conserved in all vertebrate species for which complete topo II sequences are available (human, hamster, rat, mouse, chicken and pig). Despite the established importance of the topo II isoforms in cell division and as clinically important drug targets, the cellular processes responsible for regulating their function and activity are still not fully defined. As the identification of protein–protein interactions can often provide insight into regulatory pathways, we undertook a search for novel topo II binding partners using yeast two-hybrid analysis. The sequence divergent topo II CTDs were used as bait because there is compelling evidence that many of their regulatory elements are contained in this region. Since topo II α and β are normally localized to the nucleus, we initially found it surprising that our analysis identified PLSCR1, which is primarily localized to the endofacial plasma membrane in most cell types, as a topo II-binding protein. However, PLSCR1 can also be found in the nucleus and other organelles, albeit at much lower abundance (,). A non-classical functional NLS has been identified in PLSCR1 (,) and it has been demonstrated that PLSCR1 binds to a unique site in the nuclear import carrier protein importin α (). Thus, there is strong evidence that under certain conditions, PLSCR1 is actively imported into the nucleus making its interaction with topo II feasible. PLSCR1 at the plasma membrane is multiply palmitoylated which, like other forms of lipid modification, often induces or enhances membrane localization (,). The factors that regulate PLSCR1 palmitoylation are not known, nor is it known precisely how palmitoylation affects the non-plasma membrane localization or function of this protein. However, when palmitoylation of PLSCR1 is inhibited, either by mutation of the palmitoylated cysteines in the COOH-proximal end of the protein or by treatment of cells with 2-bromo-palmitate, PLSCR1 becomes primarily localized to the nucleus (). Since the conditions used to prepare samples for the topo II pulldown and co-immunoprecipitation assays in the present study would be expected to yield predominantly palmitoylated and unpalmitoylated PLSCR1, respectively, our observations suggest that palmitoylation does not affect the interaction of PLSCR1 with topo II α or β, at least . However, because both topo II isoforms are almost exclusively found in the nucleus, it would seem likely that they interact primarily with unpalmitoylated PLSCR1 . Protein–protein interactions involving both yeast and human topo II have been reported to result in several different functional consequences (). In some cases, the topo II–protein interactions modulate the DNA decatenating activity of the enzyme which is critical for regulated chromosomal segregation. Thus, HDAC1 and the underphosphorylated form of the retinoblastoma protein inhibit the DNA decatenating activity of human topo IIα (,) while in contrast, the cyclic AMP-response element-binding protein (), the mitotic cdc2 cyclin-dependent kinase () and CAD nuclease stimulate this activity (). Our data indicates that PLSCR1 behaves like the latter proteins in its ability to stimulate DNA decatenation by topo IIα. To better understand the molecular basis of the interaction between topo II and PLSCR1, we investigated the region of topo IIα responsible for its binding to PLSCR1. Using a series of nested truncated GST-fusion proteins in pull-down assays, a basic stretch of amino acids (residues 1432-1441) in topo IIα in a region of the enzyme that is predicted to be unstructured was shown to be critical. The importance of these amino acids was confirmed by demonstrating that a synthetic dodecapeptide corresponding to this sequence could inhibit binding of topo II and PLSCR1. However, while topo IIα residues1432-1441 are essential, it remains possible that they are not sufficient by themselves to mediate the interaction between the two proteins. Other proteins reported to bind to the topo II CTD include HDAC1 and HDAC2 () and p53 (). It will be of interest to determine the relationships, if any, among the different topo II CTD binding proteins (which now includes PLSCR1) with respect to topo II functions (). Most studies of PLSCR1 have focused on its role in phospholipid mixing at the plasma membrane. More recently it has been suggested that PLSCR1 may have a role in suppressing tumorigenesis. For example, when human ovarian tumour cells stably transfected with PLSCR1 were implanted into athymic mice, the resulting tumours were much smaller and more differentiated than those arising from untransfected cells (). Additionally, Huang . () showed that induction of PLSCR1 arrested proliferation of human myeloid leukemia cells and stimulated their differentiation. Elevated PLSCR1 mRNA levels have also been reported to be a positive prognostic factor in a clinical study of patients with acute myelogenous leukaemia (). In contrast, when transplanted into syngeneic or athymic mice, an NH-truncated form of murine Plscr1 (designated as TRA1) is leukemogenic (). Ongoing studies are aimed at determining how the tumour-suppressor effects of PLSCR1 might be related to its effects on topo II activity. The function(s) of nuclear PLSCR1 is presently not well understood. Recently, a domain within the putative NH-proximal regulatory region of PLSCR1 was shown to stimulate expression of inositol 1,4,5-triphosphate receptor type 1 (IP3R1), a protein involved in regulating Ca release from the endoplasmic reticulum, by binding directly to its promoter (). It may be that PLSCR1 participates with other nuclear proteins to regulate expression of IP3R1 and possibly other genes by recruiting topo II or alternatively, topo II recruits PLSCR1 to specific regions of DNA in order to regulate topo II activity. For the topo II-nuclear protein interactions characterized thus far, it is the interacting protein partner that appears to be recruited by topo II (,,). For example, it has been shown that chromatin remodelling occurs in response to topo II-mediated recruitment of the mitotic cdc2 cyclin-dependent kinase (). IP3R1 seems unlikely to be the only gene regulated by PLSCR1 and indeed, PLSCR1 has been reported to bind to promoter fragments from three other uncharacterized genes (). PLSCR1 did not induce transcription from these promoter fragments in a luciferase reporter assay, an observation that could be explained if PLSCR1 is acting as a negative regulator of these genes or if a necessary adaptor protein was not present in the cells used. It has also been shown that nuclear levels of PLSCR1 can be enhanced substantially by treatment of at least some cell types with both type I and II interferons (,). Whether or not the ability of PLSCR1 to modulate topo II activity is involved in the cellular activities of interferons, however, remains to be determined. In conclusion, our study shows for the first time a physical and functional interaction between topo II and PLSCR1. Further investigation of the interactions between these proteins is needed to better understand their possible role in tumour proliferation, as well as cellular differentiation and drug responsiveness.

DNase I footprinting is a powerful technique used to identify ligand–DNA interactions at specific DNA sequences (,). For the past two decades it has been the fundamental assay used to determine the sequence-selectivity for both proteins and DNA-binding compounds (,). Through manipulations of quantified data, it has become possible to further utilize DNase I footprinting to indirectly measure thermodynamic and kinetic properties of interactions (,). Typically it is possible, from one or two footprinting experiments, to locate a ligand's preferential binding sites on a desired short (100–200 bp) target DNA sequence, and characterize the ligand by calculating the binding affinities at such sites (). Unfortunately, as many researchers will testify, DNase I footprinting is a labour-intensive procedure which may take several days to prepare and carry out (,,). Furthermore, it traditionally involves the undesirable use of radioisotopes and requires tricky ethanol precipitation steps. Modifications of the DNase I footprinting protocol have been reported, and these have sought to improve the assay through the removal of radioisotope use and/or ethanol precipitation steps (). However, use of these modified assays has not become commonplace, possibly as all describe single-tube methods that ultimately still require significant labour in order to yield data. For many researchers, including us, the greatest desire is to decrease the data turnover time of DNase I footprinting to maximize productivity. In the current age of drug discovery and ‘omics’, the need for medium/high-throughput assays is evident, with the same technique performed repeatedly, in order to satisfy large sample sets (). This is achieved in most experiments by carrying-out reactions in parallel, with the 96-well microtitre plate as the industry standard. Working in the field of drug design and discovery, it was our own wish to use DNase I footprinting as a screening tool, increasing the throughput such that it could be used to assess libraries of compounds produced by combinatorial methods. Therefore, through scrutiny of every aspect of the standard protocol for DNase I footprinting, we developed a rapid assay that utilizes the 96-well format and gives increased throughput along with decreased risk, error and processing times. In the process of developing the method, it also became clear that an equally expedient analysis procedure would be required to handle the large yield of data. To this end, a semi-automated analysis protocol was also designed, tying together gel quantification software with custom-designed programs to produce a single ‘footprinting profile’ for each ligand across several hundred base pairs of sequence. The resultant protocol provides a quick and simple microtitre-based DNase I footprinting assay that is suitable for use as a quantitative screening-tool. The proximal 705 bp of the human Topoisomerase II alpha (TOPOIIα) promoter was amplified by PCR from human genomic DNA (Promega) using primers TIIA-F (CACCGCACACAGCCTAC) and TIIA-R (TGGTGACGGTCGTGAAG) (Supplementary Figure 1). The resulting fragment was ligated into the pGEM-T Easy plasmid vector using a commercial kit (Promega). After transforming XL-10 Gold strain (Stratagene), plasmid containing the TOPOIIα fragment was amplified and purified using a midiprep purification system kit (Qiagen). The generated template plasmid was verified by DNA sequencing (UCL Services, London, UK). IR700 dye (LI-COR Biosciences) end-labelled DNA was produced in 48 × 25 μl reactions by standard (Sigma) PCR amplification from the plasmid template. Labelled and non-labelled PCR primers (Thermo) directed to the T7 and SP6 promoter sequences of pGEM-T Easy plasmid were used, with the IR700 dye attached to the 5′-end of the T7 primer when producing forward-labelled target, and attached to the 5′-end of the SP6 primer when producing reverse-labelled target. Products were purified by PCR clean-up column with 12 reactions per column (Qiagen) and finally eluted into a total of 300 μl of 0.1 mM Tris-HCl pH 8.5. The IR700 end-labelled DNA concentration of both targets was estimated by measuring optical density both at 260 nm wavelength (for DNA) and at 685 nm wavelength (for IR700 dye). DNA product was then diluted with 0.1 mM Tris-HCl pH 8.5 to give a 100 nM stock and this was stored at −20°C. In both cases, the DNA fragment consisted of the 705 bp TOPOIIα promoter region plus 177 bp of flanking pGEM-T Easy sequence. In a V-bottom polypropylene 96-well plate (Eppendorf), 50 μl drug–DNA reactions were prepared against the two IR700 dye end-labelled DNA targets. In each case, six test agents were incubated with each DNA target at eight concentrations (one column) per drug: 0, 1.6, 8, 40, 200, 1000, 5000 and 25 000 nM. Serial drug dilutions were prepared beforehand in 100% DMSO at 20× final concentration allowing 2.5 μl to be added to each reaction giving a final mix of 5% DMSO, 50 mM KCl, 20 mM Tris-HCl pH 7.5, 1 mM MgCl, 0.5 mM DTT and 4 nM IR700 end-labelled DNA. Drug–DNA reactions were incubated for 17 h at room temperature before digestion for 8 min; initiated by adding 5 μl of 50 mM NaCl, 5 mM MgCl and 5 mM MnCl containing 0.02 units of DNase I (Promega). Digestion was stopped by adding 5 μl of 50 mM EDTA and digested DNA samples were purified with a 96-well vacuum filtration system (Montage SEQ96, Millipore), resuspending in 50 μl 0.1 mM Tris-HCl pH 8.5. Samples were concentrated to dryness with a Speedvac Evaporator (Thermo) and resuspended in 0.8 μl formamide containing 0.05% bromophenol blue. Samples were denatured for 4 min at 95°C, and 48 samples (the first six columns) were loaded on a 52-lane sequencing gel [41 cm length, 0.25 mm thick, containing 1× TBE, 7 M urea, 10% formamide, 5% Sequagel XR (National Diagnostics, Hull, UK)]. A GA-specific chemical sequencing ladder () produced in advance was co-loaded in two lanes of the gel to act as markers. The gel was run for 5 h at 1500 V, 35 mA, 31.5 W, 55°C using a LI-COR model 4200 DNA Sequencer with e-Seq v2.0 Software (LI-COR Biosciences). The remaining 48 samples (the reactions with the reverse-labelled target) were subsequently loaded and run on a second gel exactly as the first 48 samples. Raw intensity profiles (pixel intensity with regard to pixel position) for all 50 lanes were grouped in a CSV datasheet following 1D gel analysis of each 16-bit TIFF image using ImageQuant TL (GE Healthcare). With no background subtraction, band detection was then carried out in ImageQuant TL for the two marker lanes, ensuring that bands matched all the G and A positions within the analysed DNA sequence before exporting the lane measurement tables to a CSV datasheet. The program, ‘footprint2’, was used to convert the gel data into differential cleavage values. ‘footprint2’ is written in Perl, and the Tk toolkit is used to provide simple graphical input dialogs. The program input is CSV MS Excel datasheets for the lane intensity profiles, the band positions of one of the marker lanes (selected from the two analysed) and the analysed DNA sequence. The program proceeds in the following sequence: Using a second Perl program, ‘hitsblock2’, the output file from the ‘footprint2’ program was combined with manually assigned differential cleavage cut-off values for each drug to give a further output of a ‘score’ value assigned to each base pair in the sequence for each of the six agents tested per gel. The ‘score’ is defined as log, where is the lowest concentration for each drug where the differential cleavage at the base pair is below the value. For each drug output data were used to plot individual graphs of ‘score’ against the target DNA sequence. As a comparison, compound 2 was footprinted against P-labelled TOPOIIα promoter DNA using a previously described protocol () with minor modifications. A GA-specific chemical sequencing ladder () was used and drug was incubated with DNA at eight different concentrations; 0, 1.6, 8, 40, 200, 1000, 5000 and 25000 nM. italic fig #text A rapid 96-well DNase I footprinting screen was developed that gives readily comparable quantitative binding data for multiple test agents over long sequences of DNA. By increasing substantially the throughput of footprinting, we have demonstrated the suitability of this assay as a screening tool relevant in the current age of drug discovery. A typical radioisotope DNase I footprinting experiment might be able to assess up to a maximum of two ligands over 100 bp of DNA, whilst requiring approximately 3–4 days to proceed from an unlabelled DNA template to an analysable gel image (,,). The 96-well format screen described herein permits the assessment of 12 ligands over 500 bp of DNA (or six ligands over 1 kb) in as little as two days, along with a greatly decreased proportion of user-time spent on difficult liquid-handling steps, which could ultimately be programmed to be carried-out by a benchtop robot system. This represents up to a 30-fold increase in data yield within a significantly shortened timescale. Indeed, within our lab, the screen is routinely used to assess 12 compounds over 500 bp every 24 h, using only one automated sequencer. The described assay has the further advantage of producing quantifiable, linearly-separated band intensities collected in real-time and does not require such unreliable steps as ethanol precipitation and the transfer of an electrophoresis gel to blotting paper. As with other previously described modifications to DNase I footprinting (), useful improvements are afforded by the introduction of an automated DNA sequencing system and the omission of radioisotope use—although in a well-equipped laboratory it would be perfectly reasonable to use the 96-well procedure with radiolabelled DNA. Despite the relative expense of automated DNA sequencing systems, their versatility allows them to be used for many other applications, notably DNA sequencing (), AFLP analysis, () electrophoretic mobility shift assays () and footprinting (). Additionally, the dye labels utilized are considerably safer, have long lifetimes and are low in cost (). It is also common to find that such systems are available for communal use in many research environments, and although fluorescent sequencing systems are more common than the LI-COR infrared system described here, continual improvement in the fluorescent dyes and optics used by automated sequencing systems should no longer prevent such fluorophores being used as an alternative to radioisotopes in exactly the same method described here. Ultimately, the optimization of the practical steps for the DNase I footprinting screen were less challenging than the development of the analysis procedure. As many features of the screen were novel, no commercially available programs were suitable for the entire analysis. The final analysis system is, in effect, a hybrid between existing image analysis software and our own programs, and as such some degree of user-input is still required, in order to transfer data between steps and to make occasional informed judgements. Nevertheless, one advantage of a semi-automated procedure over a fully automated procedure is that it allows the user to monitor the analysis at several stages and in greater depth (). With the DNase I footprinting screen in particular, the use of a semi-automated procedure allowed us to overcome the difficulty of having a wide range of band intensities within each lane (due to the inherent non-uniform cleavage of DNA by DNase I). By allowing the user to monitor the differential cleavage outputs, it became possible to ensure that the desired ‘score’ data gave an accurate description of the initial gel image data without compromising on the speed of analysis. A further advantage of the semi-automated procedure is that it introduces flexibility to the analysis. With only minor modifications, our programs can be adjusted to analyse gel images of different dimensions (length, width, and number of lanes) or even to analyse electrophoresis results from alternative sequencing systems and from conventional radioisotope footprinting. Indeed, the whole design of our 96-well DNase I footprinting screen is intentionally modular, in order to permit the user to adapt the protocol to suit their needs. For example, a researcher wanting to screen a mutant library of transcription factors on several promoters could perform the DNase I footprinting in 96-well format (incubating with proteinase K before sample clean-up, in order to remove the protein content) before resolving the gel images using standard radioisotope electrophoresis over several days. The subsequent data could then be processed rapidly using the semi-automated procedure providing the user with readily-comparable profiles of how each mutation effects transcription factor binding. The 30-fold increase in throughput afforded by the 96-well DNase I footprinting assay allows, for the first time, the use of DNase I footprinting as a screening tool for assessing the DNA-binding properties of ligands, DNA-binding proteins and potentially therapeutic test agents. In our lab, we have found it possible to work at a capacity that allows as many as 42 compounds to be characterized per week against a 500 bp DNA target when using the 5 concentration range described here. The power of the assay is illustrated by the results described here, obtained using the PBD-polypyrrole series. The full series was characterized—including full analysis over 700 bp of DNA and preparation of labelled DNA—in a total time of just three days. By comparison, our previous characterization that included analysis of the series (over 200 bp of DNA) took more than four weeks to achieve (). By comparing the ‘score’ profiles of a large number of compounds, differences in binding-site size, sequence-selectivity and affinity can all be determined with ease. In research programs to design novel drugs, these data are invaluable for selecting lead compounds, directing further synthesis and in determining the nature of drug–DNA interactions. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

The flap endonucleases (FENs) are a family of structure-specific nucleases capable of cleaving a variety of branched DNA structures (,). These enzymes recognize substrates containing at least one duplex and a single-stranded 5′ overhang such as flaps and pseudo Y structures. The major activity involves cleavage of the phosphodiester backbone one nucleotide into the double-stranded region at the junction of single-stranded bifurcation with duplex DNA (,). They also display 5′–3′ exonucleolytic activities on substrates with free 5′-ends () and can attack gaps in duplex DNA and covalently closed-circular DNA (,). These latter activities appear to be very similar to the gap endonuclease activities described recently in human FEN1 (). A functional FEN seems to be essential for all organisms. For example, the FEN activity present within the DNA polymerase I N-terminal region is essential for cell viability (), haploinsufficient mice rapidly develop tumours while mutants show embryonic lethality (). The striking similarity of the structures of FEN enzymes from a diverse range of organisms has been revealed through crystallographic structure determination. For example, , T5 exonuclease (T5FEN), DNA polymerase I FEN domain and human FEN1 enzymes share a central beta sheet feature adorned by a number of helices and interconnecting loops (). All possess binding sites for two divalent metal ions composed of conserved glutamyl and aspartyl residues and a DNA-binding feature known as a helix-three-turn-helix motif (). Based on the observation of two manganese binding sites in Pol, Steitz and co-workers () proposed a two-metal-ion mechanism for FEN-mediated phosphate ester cleavage similar to that of the Klenow 3′–5′ proof-reading exonuclease (). The roles of the divalent metal ions include substrate binding by metal ions I and II () with the site I metal generating a hydroxide ion essential for catalysis (). It also appears that just one metal binding site is required for the structure-specific endonuclease activity of the T5FEN enzyme, whereas it requires both metal binding sites to be occupied for exonucleolytic cleavage (). The DNA polymerase I protein present in eubacteria possesses FEN activity carried within the N-terminal domain (). Two groups independently reported the identification of a previously uncharacterized DNA polymerase I (Pol I) paralogue by interrogation of sequences deposited in the public databases (,). The putative gene, originally designated , was identified based upon up to 60% sequence identity with part of the 5′ region of , the gene encoding Pol I. The gene appeared to have a suitable ribosomal binding site with a GUG start codon but the open reading frame consisted only of 753 nucleotides, with no homology to detectable beyond this region (). In comparison, encodes a much larger protein, consisting of the FEN domain or 5′–3′ exonuclease domain also known as the small fragment and the larger, Klenow fragment carrying the 3′–5′ proofreading and DNA-dependent DNA polymerase activity (,). Overall, the 251 amino-acid protein encoded by the open reading frame displayed 66% similarity with the DNA Pol I protein and most importantly, it appeared to contain most of the residues characteristic of the FENs (). Interestingly, these include conserved acidic residues involved in binding to one of the essential divalent metal ions usually required for catalysis (site I) but lacks three acidic residues corresponding to site II in T5FEN. shows the crystal structure of the T5FEN with the residues in common with ExoIX identified. The gene was renamed and preliminary characterization of the gene product, expressed as a GST-fusion protein has been reported. A 3′–5′ exodeoxyribonuclease activity was reported for the -encoded protein and it was renamed exonuclease IX (ExoIX) (). In addition to this principal exodeoxyribonucleolytic function the protein possessed a deoxyribophosphodiesterase (dRPase) activity on abasic sites () and was able to remove 3′ phosphoglycolate end groups from DNA (). These activities are difficult to reconcile with the strong sequence homology ExoIX shares with the FEN family of nucleases, since they are structurally dissimilar from the apurinic/pyrimidinic endonucleases, which generally perform these types of reactions (). We were intrigued by the reported 3′–5′ exonucleolytic activity of ExoIX as it appeared to display the reverse of the normal polarity of action seen in the FENs i.e. 5′–3′. In the work reported here, we over-produced native ExoIX and found that by using both one and two-dimensional activity gels, we were able to monitor the protein and nuclease activity during fractionation. Contrary to earlier work (), the ExoIX protein could be separated from a contaminating 3′–5′ exonuclease activity. We were able to identify the associated 3′–5′ exonuclease activity as being due to low levels of initially co-purifying exonuclease III (ExoIII). Once free of contaminating ExoIII, the ExoIX protein displayed no apparent exodeoxyribonuclease activity under conditions typical for FENs. We also investigated whether ExoIX was able to interact directly with ExoIII or any other proteins present in cell-free extracts. Amongst others, ExoIX bound to single-stranded DNA-binding protein (SSB). The interaction was confirmed using chemical cross-linking. The interaction took place even in the absence of nucleic acids. The SSB tetramer was able to bind up to two monomers of ExoIX. The gene was amplified by PCR from XL-1Blue genomic DNA (Stratagene) using standard procedures () with polymerase (Promega). The forward primer d(TGAATTCTTTAAGGAGATTATAGGCTGTTCA) included the underlined ATG start codon, ribosomal binding site and an EcoRI recognition sequence. The reverse primer d(TAGGGATCCGGCTCGCCGTTA) was designed to incorporate a BamHI recognition site downstream of the stop codon. After amplification, the PCR product was treated with the above restriction enzymes and ligated into similarly treated expression vector pJONEX4 (), transformed into BL21(pcI857) and transformants were selected on LB agar plates containing ampicillin at 100 μg ml. Dideoxy sequencing was used to identify a clone, designated pJONEX/, possessing the expected sequence (). A culture derived from a single colony of BL21 (pJONEX/ pcI857) was grown in 20 l 2× YT (16 g tryptone, 10 g yeast extract, 5 g NaCl/l) containing 25 µg ml kanomycin, 100 µg ml ampicillin and 100 µg ml carbenicillin in a New Brunswick Scientific BIOFLO 4500 Fermenter/Bioreactor, incubated at 30°C, until it reached an A of 1.0, at which point the temperature was raised to 42°C for 1.5 h. The vessel was then allowed to cool passively to RT, and incubated overnight. Cells were harvested by centrifugation and washed with 0.9% w/v NaCl, aliquoted into 10–15 g portions and stored at –80°C. Cell-free extracts were prepared free of nucleic acids essentially as described previously from ∼10 g cell pellets and stored as an ammonium sulphate slurry (). The slurry was re-suspended in 50 ml ice-cold 250 mM KHPO/KHPO, pH 6.5, containing 2 mM EDTA, 5 mM DTT, 5% v/v glycerol, 250 mM NaCl (HKP buffer). The suspension was dialysed against excess HKP buffer, diluted 1:10 in low salt phosphate buffer, LSPB, (25 mM KHPO/KHPO, pH 6.5, containing 1 mM EDTA, 5 mM DTT and 5% v/v glycerol), and applied to 5 ml HiTrap SP HP and heparin HP columns assembled in tandem then washed with LSPB. The SP column was removed and the heparin column eluted with a 0–1 M NaCl gradient (200 ml) collecting 5 ml fractions. The purest fractions were identified by SDS–PAGE, dialysed overnight prior to anion exchange chromatography in 25 mM Tris–HCl, pH 8.0, containing 200 mM NaCl, 1 mM EDTA, 5 mM DTT and 5% w/v glycerol. A 5 ml HiTrap Q HP column was equilibrated in excess of Q buffer (25 mM Tris–HCl, pH 8.0, containing 1 mM EDTA, 5 mM DTT and 5% w/v glycerol). The sample was diluted 1:10 with Q buffer prior to loading. The column was washed and eluted with a linear gradient of 0–0.25 M NaCl in 120 ml. Fractions (5 ml) were checked for the presence of 28 kDa protein by SDS–PAGE. A 5 ml HiTrap heparin HP column was equilibrated as described above, then loaded with the pooled Q fractions at 15%, against LSPB at 5 ml min. The column was washed extensively then eluted with 0–350 mM NaCl, over 30 ml, followed by 350–600 mM NaCl, over 75 ml, collecting 5 ml fractions. Fractions that contained pure protein, as determined by SDS–PAGE analysis, were stored at –20°C in 50% glycerol. strain BL21 (DE3) pGEX-5-x-3() was generously provided by Dr W. A. Franklin (Albert Einstein College of Medicine, New York) and induced as previously described (). Cell lysis (6 g cell pellet) and ammonium sulphate precipitation were performed, as described for native ExoIX, and the final ammonium sulphate pellet was re-suspended and dialysed into in 25 mM sodium phosphate, pH 7.3, containing 2 mM EDTA, 5 mM DTT and 150 mM NaCl. Clarified supernatant (40 ml) was applied to a pre-equilibrated 1 ml GSTrap column (GE healthcare), run at 1 ml min. The column was washed in excess of 10 ml at 0.5 ml min and the column eluted with 5 ml 50 mM Tris–HCl, pH 8.0 containing 10 mM reduced glutathione. Samples contained the purest fractions of GST-ExoIX (50 kDa) were adjusted to 25 mM Tris–HCl, pH 8.0, filtered (0.2 µm) and applied to a 5 ml HiTrap Q anion column at 25% against Q buffer. The column was washed in excess of 50 ml with Q buffer, then eluted over a 60 ml 0–1 M NaCl gradient, at 2.5 ml min, collecting 3 ml fractions. Samples containing the purest bands of the fusion protein were adjusted to 50% v/v glycerol and stored at –20°C. Radiolabelling of d(5′-GATGTCAAGCAGTCCTAACTTTGAGGCAGAGTCC) was performed in a reaction (30 µl) containing 5 pmol oligonucleotide, [1x] kinase buffer (70 mM Tris–HCl, pH 7.6, 10 mM MgCl, 5 mM DTT), 17 pmol [γ-P] ATP and 30 U polynucleotide kinase at 37°C for 1 h. Reactions were quenched by heat denaturation at 80°C for 5 min. Labelled oligonucletides were purified using Sep-Pak Plus C18 cartridges, wetted with HPLC-grade acetonirile (100% ACN) and equilibrated with 10 mM Tris–HCl, pH 7.6. The labelling reaction was applied to the cartridge, washed extensively with 3% v/v ACN and eluted with 30% v/v ACN. The eluate was evaporated to dryness and the oligonucleotide re-suspended in a buffer containing 10 mM Tris–HCl, pH 7.6. Radiolabelled nuclease assays were performed with 1.5 nM labelled oligonucleotide and varying amounts of test protein (described in figure legend) in a buffer comprising of 25 mM Tris–HCl, pH 8.0, 10 mM MgCl, 100 mM KCl and 0.3 mg ml acetylated BSA, incubated at 37°C for 10 min. Reaction products were separated by denaturing 15% PAGE, according to standard methods (). The release of acid-soluble nucleotides from high molecular weight (Type XIV; Sigma) DNA was monitored by UV spectroscopy as described, except that the assay contained DNA at 800 μg ml in 500 μl 25 mM Tris–HCl, pH 8.0, 10 mM MgCl, 50 mM NaCl, 2 mM DTT (). A zymogram assay was used to identify nuclease activity in protein bands or spots separated by 1 () or 2D electrophoresis. Zymogram gels contained ca. 30 µg ml Type XIV DNA from herring testes (Sigma). After electrophoresis, the proteins were re-natured and MgCl added to a concentration of 10 mM. After staining with ethidium bromide, exonuclease activity was visualized as a shadow against a fluorescent background when viewed on a UV transilluminator. A fraction (5 ml) adjacent to the ExoIX peak, with prominent nuclease activity and approximate molecular weight of 28 kDa, was dialysed against a 20-fold excess of wash buffer, 25 mM sodium phosphate, pH 7.4, containing 500 mM NaCl and 5% v/v. The sample was loaded onto a 1 ml Probond column (Invitrogen), washed in excess of 20 ml with wash buffer, then protein eluted with a 0–500 mM imidazole gradient (20 ml), collecting 1 ml fractions. Fractions were assayed by DNase activity gel and Coomassie staining methods. Protein in the column flow through (containing the ca. 28 kDa DNase activity) was precipitated by the addition of ammonium sulphate to a 4 M final concentration, sedimented by centrifugation at 35 000 × for 30 min and the pellet re-suspended in 5 ml 20 mM Tris–HCl, pH 8.0, containing 100 mM NaCl and 5% v/v glycerol. The protein sample was concentrated 10-fold, by centrifugation at 6000 × for 30 min in a Vivaspin concentrator device (MWCO 10000) into 25 mM Tris–HCl, pH 8.0, containing 1 mM EDTA, 5 mM DTT and 10% glycerol. The sample was then analysed by 2D zymogram and Coomassie gels. For each Immobilized pH Gradient (IPG) strip (17 cm ReadyStrip, non-linear pH 3–10, Bio-Rad Laboratories) a protein sample was made up to 300 μl with reswell buffer, RSB (7 M urea, 2 M thiourea, 4% CHAPS, 30 mM DTT, 0.2% v/v ampholytes, pH 3–10). IPG strips were actively re-hydrated at 50 V for 16 h at 20°C in a Bio-rad Protean II IEF cell. IEF was carried out at 250 V for 15 min, linear ramping from 250 to 10 000 V over 3 h followed by 10 000 V for 60 000 V h. The second dimension was carried out on a 12.5% SDS–PAGE gel as described in Allen . (). Analytical gels were stained with Biosafe Coomassie G250 stain or assayed for nuclease activity by re-naturation and ethidium bromide staining. Protein spots were excised from the gel and digested with trypsin. Identification was effected using MALDI-tof generated mass spectrometry data analysed by MS-FIT and MASCOT web-based software, allowing for peptide mass tolerance of ±150 ppm, monoisotopic masses and variable oxidation modification. ExoIX or BSA (10 mg) were immobilized covalently, using 1 ml of CNBr-activated Sepharose 4B, essentially as described by the manufacturer (GE Healthcare). cell-free lysates were prepared as described above and the protein stored as ammonium sulphate slurry. A portion (1.8 g) of ammonium sulphate precipitate was re-suspended in a 5 ml final volume of buffer ABB (25 mM Tris–HCl, pH 8.0 containing 100 mM NaCl, 5% w/v glycerol, 2 mM DTT, with and without 1 mM EDTA) and dialysed twice against a 20-fold excess of ABB (7 ml after dialysis). Preparations in the absence of EDTA were treated with 200 U ml bovine pancreatic deoxyribonuclease I (DNase I) (Sigma-Aldrich). MgCl (2.5 mM) and CaCl (0.5 mM) were added and the sample incubated for 1 h at room temperature. Lysate (1 ml) was added to either ExoIX or BSA-linked Sepharose 4B (∼40 µg) and incubated with end-over-end rotation for 1 h at room temperature. The mixture was then centrifuged (100 × for 5 min) and the supernatant removed (FT). The pelleted matrix was washed (4 × 1 ml ABB containing 200 mM NaCl). The final wash buffer was removed, an equal volume (ca. 40 µl) of SDS-loading buffer added and the sample prepared for electrophoresis on 12.5% SDS–PAGE. Proteins were visualized by staining with Coomassie blue. GST–ExoIX pull-downs were performed using protein derived from MG1655-K12. Cells (10 g) were lysed as described for ExoIX preparation and soluble proteins (50 ml) dialysed extensively against 25 mM sodium phosphate, pH 7.3, containing 150 mM NaCl, 2 mM EDTA, 5 mM DTT, 5% w/v glycerol. A final buffer change was made into binding buffer (as above without NaCl) equilibrated for 4 h at 4°C. Soluble lysate (20 ml) was mixed with 2.0 mg of GST–ExoIX, mixed by end-over-end rotation for 1 h at 4°C before loading onto a 1 ml GSTrap column at a flow rate of 1 ml min, equilibrated with 10 ml binding buffer. Flow through was re-applied to the column at the same flow rate and allowed to cycle overnight. The column was washed with 10 ml binding buffer at 0.1 ml min. Bound proteins were eluted with 5 ml elution buffer (50 mM Tris–HCl, pH 8.0, containing 10 mM reduced glutathione), collecting 1 ml fractions. Controls were repeated as described above using either 1.0 mg GST or without the addition of supplementary protein. Protein samples were prepared in 60 µl aliquots in PBS to a final concentration of 0.4 µg µl and dialysed twice against a minimum 100-fold excess of buffer. A working stock of cross-linking reagent, ethylene glycol (succinimidylsuccinate) (EGS), was freshly prepared to 50 mM concentration in DMSO, and further dilutions obtained in PBS of 5 mM and 0.5 mM. Optimization reactions (10 µl) contained 1.5 µg each protein in PBS with a final concentration of 10, 1 or 0.1 mM EGS. Reactions were allowed to proceed at room temperature for 30 min then quenched by the addition of 50 mM Tris–HCl, pH 7.5. Samples were mixed with loading buffer (5 µl) and analysed directly by SDS–PAGE and either Coomassie stained or the polyacrylamide gel was washed with 10 volumes transfer buffer (50 mM Tris, 7 mM glycine and 20% v/v methanol) for 15 min at RT and transferred to PVDF membrane using a Bio-Rad transfer cell according to the manufacturer′s instructions. The PVDF membrane was blocked using 50 ml PBS-T (PBS buffer containing 0.05% v/v TWEEN 20) and 5% w/v milk powder for 1 h at RT or overnight at 4°C. The membrane was washed with 50 ml PBS-T for 30 min, followed by incubation with 0.2 µg ml primary antibody (rabbit anti-ExoIX, custom generated by York Bioscience) prepared in 20 ml PBS-T with 1% w/v milk powder at RT for 1 h. The membrane was washed with PBS-T for 30 min at RT, followed by probing with 0.2 µg ml HRP-linked secondary antibody (donkey anti-rabbit) prepared in 20 ml PBS-T with 1% w/v milk powder, incubated at RT for 1 h. A final wash was completed with PBS-T for 30 min. Blots were developed using ECL reagents supplied by GE Healthcare, according to the manufacturer′s instructions. Resolved protein bands or spots were excised from polyacrylamide gels, digested with trypsin and analysed by MALDI-TOF mass spectrometry (Aberdeen Proteome Facility). Proteins were identified from peptide fragments by comparison to theoretical digests of the proteome , using MASCOT () or MS-fit search tools (). The gene was expressed at high levels using the heat-induced pJONEX expression system. Soluble protein was readily obtained and purified to over 98% purity as estimated by densitomety using a combination of ion-exchange resins (A). Zymogram assays (B) on ExoIX fractions from the initial ion exchange column revealed the presence of a DNase with a similar, but not co-incident electrophoretic mobility with the major protein band (C). No DNase activity was observed which co-migrated with ExoIX protein. Liquid nuclease assays were carried out on samples from each stage of the natively expressed (i.e. untagged) ExoIX purification. These assays showed that the final purified fraction lacked any significant DNase activity (below 0.02 units, ). Similarly, the GST–ExoIX fusion protein also appeared to lack any significant nuclease activity. Two positive controls, T5 D15 5′-3′ exonuclease () and exonuclease III (New England Biolabs) were also included in the assays. These proteins showed potent exonuclease activities of 614 and 227 units, respectively. A range of different divalent metal cofactors, pH and salt conditions was examined, but in no case were we able to detect any significant DNase activity in fully purified ExoIX using this assay. We then deployed a sensitive assay using radiolabelled oligonucleotides. This showed that partially purified ExoIX samples did possess a copurifying 3′-5′ exonuclease activity but it was finally separated from ExoIX by the final heparin column (E). A fraction taken from a side peak from the first heparin column was analysed on a 2D SDS–PAGE zymogram gel. The result is shown in . The spot of nuclease activity was excised from the gel and the protein identified as exonuclease III by mass spectrometry. A 20 kDa protein from cell-free extracts was shown to interact with both covalently immobilized ExoIX and GST–ExoIX fusion proteins () but was not retained by either control matrix (GST or Sepharose-BSA) demonstrating that the interaction was with ExoIX and not the matrix or fusion protein affinity tag. B also revealed bands of ∼60 and 15 kDa that appeared to stick to the covalently immobilized ExoIX matrix. They were identified as histone-like nucleoid-structuring protein H-NS, the 17 kDa protein Skp, glyceraldehyde-3′-phosphate dehydrogenase and GroEL (). The interaction between ExoIX and SSB was further investigated using chemical cross-linking of the over-expressed and recombinant proteins. SSB protein formed the expected multimer migrating close to 70 kDa when incubated alone, but when ExoIX was present two migrating species of ∼100 and 150 kDa were clearly seen (A) on SDS–PAGE analysis. Western blotting followed by detection with anti-ExoIX polyclonal antibodies confirmed that the two higher species contained ExoIX (B). The cross linking was performed on DNA-free samples indicating that this is a direct interaction between proteins. The first reported biochemical study on ExoIX was performed using a glutathione--transferase fusion protein. It suggested that ExoIX possessed 3′-5′ exonuclease activity and a 3′-phosphodiesterase activity on DNA with a 3′-incised abasic lesion. Neither activity was quantified in terms of turnover number nor was the enzyme purified to constant specific activity (). As can be seen in , stubborn nuclease activity was present during the early stages of purification, but was finally eliminated after extensive ion exchange chromatography. Here we have conclusively shown that the DNase activity associated with ExoIX is due to partially copurifying exonuclease III. The nature of the contamination was confirmed by direct identification of the spurious nuclease from a 2D substrate gel. We have previously observed contaminating nuclease activity initially copurifying with overexpressed proteins. Though weak, such contaminating activities are readily detected using sensitive radioisotopic assays such as those deployed in the previous ExoIX study. We repeated the purification of a GST–ExoIX fusion protein using a column format rather than the originally employed batch recovery method (). Extensive washing and affinity elution resulted in a protein that lacked any detectable DNase activity as was the case with the unmodified ExoIX protein. These results are intriguing given that the amino acid sequence of ExoIX possesses the ligands equivalent to those responsible for binding the catalytically essential metal ion in the FENs (). We investigated whether ExoIX interacted directly with any other proteins such as exonucelase III. Using a pull-down format or column chromatography approaches we were unable to reproducibly show any strong interaction with exonuclease III however, several proteins did appear to be able to bind to ExoIX. Both the native and the GST–ExoIX fusion proteins were able to bind to SSB. SSB forms a tetramer and is known to bind tightly to single-stranded DNA within the cell. ExoIX appears to be able to bind to the SSB tetramer as two extra species are observed in the cross-linking mixture of SSB and ExoIX compared with SSB or ExoIX alone. Both the slowest migrating species in the ExoIX-SSB reaction reacted with the anti-ExoIX polyclonal antibodies. It should be noted that it is difficult to accurately determine the stoichiometries of binding of cross-linked proteins by SDS–PAGE analysis. However, as two slower migrating SSB–ExoIX complexes can be observed, the simplest explanation is that each SSB tetramer can bind two ExoIX molecules. We were also able to detect binding of ExoIX to the H-NS protein, a molecule with a preference for binding to curved duplex DNA and G/C-rich single-stranded nucleic acids (,). These results indicate that ExoIX is involved in someway with the replication machinery, though despite our extensive efforts we have been unable to identify any substrate on which ExoIX displays hydrolytic activity. We also identified chaperones GroEL, GroES and Skp but these may have been binding to ExoIX that had become partially denatured during the immobilization process, as a common role for chaperones is to bind disordered peptides. The presence of glyceraldehyde-3-phosphorylase was almost certainly due to co-migration on SDS–PAGE of this protein with the chaperone GroEL as they share very similar molecular masses. This study reveals that ExoIX is not a 3′-5′ exonuclease as previously suggested (), yet it does seem to interact intimately with the DNA binding proteins SSB and H-NS. The homology of the protein with the FEN family of structure-specific nucleases suggests a role in processing of nucleic acids. Studies on an null mutant failed to identify any obvious defects in a range of DNA repair pathways (). However, as ExoIX is a paralogue of the FEN domain of DNA Pol I, perhaps it functions as a back-up for this enzyme, as previously suggested (). Further studies are needed to determine the biological role of this cryptic DNA Pol I paralogue.

Screening for genetic changes to unveil molecular attributes of human specimens is important for a variety of medical applications, including genotyping for inherited disorders, prediction of the pathologic behavior of malignancies, identification of cancer biomarkers and can affect treatment decisions for individual patients (). For example, mutations in genes like EGFR can profoundly influence chemotherapeutic response in lung cancer () and the response is modulated by mutations in other genes of the same signaling pathway [e.g. K-ras, HER2, ErbB-3 (,)]. Therefore there is a need for efficient and high-throughput mutation screening of multiple genes along identified signal transduction pathways in tumor samples. Because a large portion of cancer-causing genetic changes remains unknown and can occur in numerous positions along tumor suppressor genes (e.g. p53, ATM, PTEN) mutation scanning rather than detection of specific mutations is frequently required for molecular cancer profiling. Sequencing is often considered the gold standard for comprehensive mutation analysis. Multi-capillary electrophoresis, re-sequencing arrays or pyrosequencing provide platforms for highly parallel genetic analysis (). However, the expense associated with these techniques is currently high both for instrumentation and for running-costs. Since somatic mutations for most genes are relatively rare events it can be inefficient to scan for mutations using expensive approaches that in several cases provide unnecessary data (,). Another issue with direct sequencing or re-sequencing arrays is the difficulty in detecting a small fraction of mutated alleles in the presence of a high excess normal alleles, which is frequently the case with clinical cancer samples (). As a less expensive alternative, rapid pre-screening methods such as SSCP, DGGE, dHPLC, CCM, CDCE or HR-melting are widely utilized to identify DNA fragments that contain mutations prior to performing full sequencing (,). Enzymatic mutation detection based on mismatch scanning enzymes like MutY, TDG or T4 endonuclease VII for mutation pre-screening has also been employed (), albeit with modest success since these enzymes cannot detect all possible mutations and deletions () and some of them have substantial activity on homoduplex DNA (). Recently an enzymatic mutation scanning method based on the Surveyor™ (CELI/II) nuclease (,) combined with dHPLC or gel electrophoresis detection was introduced that shows satisfactory selectivity and reliability (1% mutant to wild-type alleles is detectable) while it also identifies all base substitutions and small deletions that are important to cancer (,) or to biotechnology and plant genetic applications [TILLING method ()]. While reliable, the use of dHPLC for examining Surveyor™-generated DNA fragments is a slow endpoint detection method restricted to examining a single DNA fragment at a time and the resulting DNA fragments cannot be sequenced. This limits analysis of cancer specimens when numerous samples or genetic regions need to be screened. We introduce a new approach that enables Surveyor™ to scan for mutations over one or several PCR products simultaneously and selectively amplify and isolate the mutation-containing DNA fragment(s) via linker-mediated PCR. By selectively amplifying mutation-containing DNA from wild-type fragments, the present approach de-couples enzymatic mutation scanning from the endpoint detection step. As a result, following enzymatic action on mismatches any chosen DNA detection method (real-time PCR, gel/capillary electrophoresis, microarray-based detection) can potentially be used to identify the mutated DNA fragments in a simplex or multiplex fashion. Here we utilize real-time PCR coupled with melting curve analysis (Surveyor™-mediated Real Time Melting, s-RT-MELT) to validate the new technology. We demonstrate that this approach increases the mutation scanning throughput by 1–2 orders of magnitude when several (>100) samples are to be pre-scanned for mutations, enables mutation scanning over several PCR fragments simultaneously and mutation-positive samples can be directly sequenced when somatic mutations are at a low-level (∼1–10% mutant-to-wild-type ratio) in surgical cancer specimens. Genomic DNA from cell lines with defined mutations in p53 exons, DU145 (exon 6), SW480 (exons 8 and 9), DLD1 (exon 7) and BT483 (exon 7) was extracted from cell lines purchased from the American Type Culture Collection (ATCC), or purchased as purified DNA when available. Surgical colon and lung cancer tumor samples were obtained from the Massachusetts General Hospital Tumor Bank following Internal Review Board approval. DNA from the EGFR mutation-positive cell lines A549, HCC827, H1975 and LU011 and from formalin-fixed-paraffin-embedded lung cancer samples were obtained from the Lowe Center for Thoracic Oncology, Dana Farber Cancer Institute following Internal Review Board approval. We isolated genomic DNA using DNeasy™ Tissue Kit (Qiagen). Sequences for the 5′M13 and GC-clamp portion of the primers, as well as the gene-specific portion of the primers used in this investigation are listed in Supplementary Table 1. The M13f and GC-clamp sequence was added to the 5′ end of forward and reverse gene-specific primers respectively, or . Twenty microliter PCR reactions were performed from genomic DNA with final concentrations of reagents as follows: 1X JumpStart™ buffer (Sigma), 0.2 mM each dNTP, 0.2 μM forward and reverse primer, 1X JumpStart™ Taq polymerase (Sigma). PCR cycling was done on a Perkin Elmer 9600 PCR machine. The cycling conditions were: 94°C, 90 s; (94°C, 20 s/65°C, 20 s/68°C, 1 min) × 10 cycles, with annealing temperature decreasing 1°C/cycle, touch-down PCR; (94°C, 20 s/55°C, 20 s/68°C, 1 min) × 30 cycles; 68°C, 5 min. This PCR program was linked to a program for denaturation and re-annealing of the PCR product over 10 min. Five-microliter PCR product (300–500 ng) was mixed with 0.5 μl Enhancer™ and 0.5 μl Surveyor™ (Transgenomic) and incubated at 42°C for 20 min followed by adding 0.5 μl Stop-solution, as per manufacturer's protocol. The inactivated Surveyor™-digested product was purified with PCR QiaQuick™ purification kit (Qiagen) and eluted in 35 μl water. In some experiments, the PCR product was mixed with an approximately equal amount of PCR product from wild-type DNA prior to forming cross-hybridized sequences, to facilitate detection of homozygous mutations. Following purification of the Surveyor™-treated sample, Poly-adenine ‘tail’ was added to the 3′-ends of DNA fragments. For each reaction, we added 5 μl purified surveyor-digested PCR product to a final volume of 20 μl with final concentration of 1X reaction buffer-4, 1X CoCL, 0.2 mM dATP, 4 U Terminal Transferase (New England Biolabs). The reaction was incubated at 37°C for 10 min and inactivated by heating at 75°C for 10 min. The real-time PCR amplification was performed using Titanium-Taq™ polymerase (BD-Biosciences - Clontech) in a Smart Cycler (Cepheid) real-time PCR machine. For each reaction, we added 0.5 μl polyA-tailed DNA to a final volume of 20 μl with final concentration of 1X Titanium buffer, 0.2 mM each dNTP, 0.1 × LCGreen (Idaho Technologies), 0.2 μM m13f primer, 0.2 μM oligodT-anchor mix GACCACGCGTATCGATGTCGACTTTTTTTTTTTTTTTT [ represents A, C and G each oligodT-anchor concentration is 0.067 μM, as per RACE protocol ()], 1 × Titanium™ polymerase (Clontech— BD Biosciences). The thermocycling program was as following: 1 cycle of 94°C for 2 min, 25 cycles of 94°C for 15 s, 55°C for 20 s and 68°C for 30 s for reading fluorescence. Temperature titration was performed using different denaturation temperatures, 94–82°C to experimentally determine conditions that selectively enable mutation-containing fragments to amplify. The real-time PCR step was immediately followed by real-time differential melting curve analysis using the SmartCycler™ machine. DNA melting was performed immediately following PCR on the Smart Cycler I machine. Samples were heated from 70°C to 95°C at 0.1°C/s. Differential fluorescent intensity curves (−d/d) were smoothed and used for identification of melting peak (s). The OpenArray™ high-throughput, massively parallel real-time PCR platform () (BioTrove) was tested for compatibility with s-RT-MELT. Forty-eight samples of p53 exon 8 PCR products were generated from 48 different lung adenocarcinoma samples and mutation-containing cell lines and processed via the hybridization and enzymatic steps of s-RT-MELT. Real-time PCR in the OpenArray™ platform was performed with the LightCycler FastStart™ DNA Master SYBR Green™ I (Roche) using 0.2 μM M13f and 0.2 μM oligodT-anchor-mix as primers pre-positioned on the array through-holes () and polyA-tailed DNA as template. The cycling conditions were as follows: 1 cycle at 94°C for 2 min, 25 cycles of 90°C for 15 s, 55°C for 20 s and 68°C for 30 s for reading fluorescence using a high sensitivity imaging camera (). The real-time PCR step was immediately followed by real-time differential melting curve analysis. Raw data were exported in Excel software for further analysis. The OpenArray™ experiment was repeated twice at the company's headquarters. To estimate , the PCR denaturation temperature below which PCR is not efficient it was assumed, as an initial approximation, that >95% hypochromicity must be present for PCR to work (i.e. any given sequence must be completely denatured, otherwise it re-forms immediately when temperature is lowered in the reaction and inhibits primer binding). The percent melting (hypochromicity)-versus-temperature relations for GC-clamp-containing PCR products and Surveyor™ activity-generated products were estimated using the POLAND algorithm (), and the thermodynamic parameters determined by Blake and Delcourt for 75 mM NaCl in the solution () were used. In order to force agreement at a single point, predicted and observed values for a p53 exon 8 sequence containing a short GC-clamp were normalized at 88°C. This shift accounts for the influence on of NaCl and Mg++ content in the reaction, the presence of the SYBR-GREEN/LC-GREEN dyes and the proprietary composition of PCR buffers. The of all other PCR products was then estimated using these semi-empirically determined parameters. The ‘enriched PCR’ method by Behn . () was used to sequence codon 273 mutation of p53 exon 8 from sample CT20 and wild-type samples. In addition, a second method [Amplification via Primer-Ligation At The Mutation (,)] was used to distinguish mutant and wild-type samples by virtue of the Nla-III site generated in the mutant sample by the p53 codon 273 G > A mutation. The s-RT-MELT assay converts PCR fragments generated at positions of mutations by the Surveyor™ enzyme to fully amplifiable sequences that enable selective PCR amplification in a subsequent quantitative PCR detection method. Following denaturation and re-annealing of PCR products that leads to formation of cross-hybridized sequences at the positions of mutations (A) the sample is exposed to Surveyor™ endonuclease that recognizes base pair mismatches or small loops with high specificity () and generates a break on both DNA strands 3′ to the mismatch. The resulting DNA fragments participate in a terminal transferase (TdT) reaction that leads to polynucleotide ‘tailing’ (sequential addition of adenine, poly-A-tail) at the 3′-ends. A real-time PCR reaction is subsequently performed using adjusted conditions that enable selective amplification of the mutant-only fragments, followed by real-time melting curve analysis for identification of mutations in the presence of SYBR-GREEN™ or LC-GREEN™ DNA dye. To enable selective amplification of the mutation-containing fragments in the real-time PCR step, modified primers are employed for the original amplification from genomic DNA (B). The forward primer contains a region specific to the target gene and a high melting domain (GC-clamp), while the reverse primer contains a region specific to the target gene and an M13 tail (or ). Following the TdT tailing reaction, the M13 primer is used for real-time PCR in conjunction with a primer that binds to the poly-A tail. The denaturation temperature of the real-time PCR reaction is lowered to enable PCR amplification only for fragments that do not contain GC-clamps. Because the PCR products that escape digestion by Surveyor™ contain GC-clamps (B), these fragments do not amplify efficiently during PCR, thereby enabling selective amplification of Surveyor™-selected fragments, i.e. an effective ‘purification’ of mutation-containing fragments. The subsequent closed-tube melting curve analysis enables clear separation of true mutant sequences from PCR dimers or other artifacts. To provide initial proof of principle for unknown mutation scanning using s-RT-MELT we utilized cell lines and tumor samples containing sequencing-identified mutations at several positions of p53 exon 8. A depicts dHPLC chromatograms of the products obtained using a sample containing a p53 exon 8 G > A mutation or a wild-type sample. The standard Surveyor™-dHPLC approach () was first employed to identify the mutation following PCR amplification of exon 8 from genomic DNA. The resulting dHPLC traces contain a single product for the wild-type and two products for the mutation-containing sequences (A, curves 1 and 2, respectively). Next, s-RT-MELT was used to screen the same p53 exon 8 sequence. Following PCR amplification with GC/M13-modified primers we cross-hybridized PCR products and exposed them to Surveyor™ and TdT tailing. The subsequent real-time PCR was run at different denaturation temperatures and the products were examined either via dHPLC or via real-time melting-curve analysis. At the standard denaturation temperature of 94°C the mutation-containing sample contains two peaks, corresponding to the anticipated amplification of both Surveyor™-digested and un-digested fragments (A, curve 3). However, when the PCR denaturation temperature is lowered (e.g. 86–88°C) a single PCR product is generated for the mutant sample, while the wild-type sample demonstrates no product (A, curves 4–7). In B, real-time differential melting curves for the PCR reaction run at 88°C are depicted. A peak corresponding to the PCR product from the mutant sample is again clearly evident, which is absent in the wild-type sample. Finally, C depicts sequencing of the s-RT-MELT-generated PCR fragment, as well as the direct sequencing from genomic DNA. The G > A mutation is evident in both samples. In the s-RT-MELT product the anticipated addition of the poly-A tail at the 3′-position next to the mutation is illustrated. To examine the selectivity of s-RT-MELT, dilutions of mutant to wild-type DNA were performed using DNA from SW-480 cells that harbor a p53 exon 8 14487G > A homozygous mutation. The real-time PCR reaction was again performed at 88°C and mutant-to wild-type ratios of ∼1–10% were distinguished from the wild-type using either dHPLC (D) or melting curve analysis (E). In these samples, direct di-deoxy-sequencing could not identify a mutation if the ratio of mutant-to-wild-type was <∼30–40% (data not shown). On the other hand, sequencing of s-RT-MELT products was possible including the lower dilutions (F). sRT-MELT sequencing generated traces with poly-A tails depicting the presence and the position of the mutation, although the exact nucleotide change was less clear than the one in exon 5 (i.e. the position ±1 base from the mutation might also be confused to be a mutation). The reason for this ±1 base ambiguity of the exact position of the mutation can be probably understood. The PCR performed following poly-A tail addition contains an equimolar mixture of three reverse primers (3′ ending in V = G, A or C). Depending on the exact nucleotide at the mutation, the correct primer should in theory be preferred, while the other two primers should not allow efficient polymerase extension due to the mismatched 3′-end. However, in practice this ‘allele-specific PCR’ step occasionally allows 3′-mismatched primer extension, enabling more than one version of the primer to amplify over the position of the mutation, or alternatively the incorporation of the poly-A tail may occur ±1 base from the exact position of the mutation. We conclude that in certain cases sRT-MELT indicates the position of the mutation to within 1 base, while in others (e.g. p53 exon 5) it indicates the position ‘and’ the actual nucleotide change. Next, p53 exon 8 was amplified using DNA from a group of 48 surgical lung adenocarcinoma samples and s-RT-MELT was used for the screening of unknown mutations via melting curve analysis. Mutations at different positions along exon 8 were present in several of these clinical samples, as indicated by the shift in melting profiles obtained (H) and subsequently verified via sequencing. In this set of samples, sRT-MELT-sequencing detected a low-level mutation on a colon cancer specimen (CT20) that direct sequencing failed to identify (I). As with F, sequencing of sample CT20 indicated the position of poly-A tail addition to within one base, but the actual nucleotide change was difficult to identify. To exclude the possibility for a false positive, two independent RFLP-based methods were used to verify the presence of the mutation. Thus, since the position of poly-A tail addition was known (I, codon 273 of p53 exon 8) the mismatched primer approach by Behn . () was used to introduce an MluI restriction site for the wild-type p53 sample but not for the codon 273 mutants. Subsequent restriction with MluI enzyme followed by PCR generated a product with a 14487G > A mutation for the CT20 sample but not for the wild-type sample (Supplementary Figure 1, Frame A). As an additional verification for the low-level CT20 mutation, we observed that G > A mutation generates a Nla-III site at the position of the mutation. Accordingly, we applied ‘Amplification via Primer-Ligation At The Mutation’, a method that we described previously (,) to ligate a primer at the Nla-III-digested site, and preferentially amplified the mutant fragment in a second PCR. The sequenced PCR product identified again the 14487G > A mutation (Supplementary Figure 1, Frame B). In conclusion, sRT-MELT identified correctly a p53 codon 273 low-level mutation on CT20 that was missed by regular sequencing. This is very significant as p53 exon 8 mutations at codon 273 have been associated with bad prognosis in cancer (,). The data in A–D and H indicate a lack of substantial PCR amplification at denaturation temperatures ⩽88°C for fragments containing the GC-clamp and a selective amplification of the mutation-containing fragments for several different mutation positions on p53 exon 8. To estimate the influence of the GC-clamp length on PCR efficiency versus temperature and the PCR amplification of fragments generated for mutations lying at different positions along the sequence, a calculation based on the POLAND algorithm () was performed. The predicted minimum temperatures for substantial PCR amplification were then plotted versus the experimentally observed values. Three possibilities were simulated, no GC-clamp, 26 nucleotides (nt) GC-clamp and 117-nt GC-clamp. DNA fragments corresponding to mutations at several positions along exon 8 were also simulated and compared to the experimentally observed minimum temperatures for generating a PCR product for three samples that contained mutations at different positions along p53 exon 8 (SW480, CT5 and TL50). The results (G) indicate agreement to within ∼1.0°C between theoretical prediction and experimental observation. For denaturation temperatures in the region 85–88°C in combination with a 26-nt GC-clamp all the available mutations on p53 exon 8 are predicted to result in selective amplification of the mutation-containing fragment and inhibition of the GC-clamp-containing fragment. This prediction is consistent with the experimental results obtained from PCR temperature-titration experiments (G). The developed calculation algorithm can thus be used to predict the appropriate PCR denaturation temperature for additional PCR fragment/GC-clamp combinations. As a further validation for s-RT-MELT, we utilized the method to identify mutations in additional p53 exons. A depicts the chromatographs obtained when a 1:1 mixture of DNA from SW-480 cells (homozygous mutation at p53 14686 C > T exon 9) and from wild-type cells was screened. The real-time PCR reaction was performed at different denaturation temperatures and the products were examined both via dHPLC and via melting curve analysis for comparison. As also observed for p53 exon 8, at 94°C denaturation temperature both the Surveyor™-digested and the undigested PCR products are amplified during real-time PCR (A, curves 1 and 2, mutant and wild-type, respectively). By lowering denaturation temperature to 85°C or 84°C, a single PCR product is obtained from the mutant while no product, other than primer dimer, is obtained by the wild-type sample (A, curves 3–6). B depicts the melting curves obtained following real-time PCR at 85°C denaturation temperature for the mutant and wild-type samples. s-RT-MELT was subsequently applied in the same manner to screen for p53 mutations in exons 5–7 from cell lines and surgical colon samples harboring sequencing-identified mutations including a single-base frameshift mutation in exon 7 (listed in Supplementary Table 2). The melting curves from mutant and wild-type samples in p53 exons 5–7 are depicted in C–E. The data indicate that results similar to those obtained for p53 exon 8 are also obtained for p53 exons 5, 6, 7 and 9. Detection of mutations in EGFR exons 18–21 is of particular clinical interest as these alterations can modulate response to EGFR inhibitors in lung adenocarcinoma patients (,). F, G and H depict the application of s-RT-MELT for screening DNA from lung cancer cell lines that harbor dHPLC-identified alterations in EGFR exons 19–21, including a two-codon deletion (del L747-E749, exon 19). The ability of s-RT-MELT for detecting low-level EGFR mutations was evaluated by performing DNA dilutions of a heterozygous EGFR exon 20 into a homozygous sample. A 1–10% mutant-to-wild-type ratio was detectable in this dilution experiment (F). Finally, the application of s-RT-MELT in detecting mutations in DNA from formalin-fixed paraffin-embedded (FFPE) samples was examined by screening four clinical FFPE lung adenocarcinoma specimens. Two of these samples were known to harbor EGFR exon 21 mutations (L858R), while the other two were negative for mutations when independently evaluated via dHPLC (). I demonstrates the identification of the mutational status of these samples via s-PCR-MELT. A significant potential advantage of enzymatic mutation scanning is the ability to screen several sequences simultaneously for mutations. To demonstrate that s-RT-MELT can be used for parallel scanning of mutations in several PCR products, we mixed equimolar amounts of PCR products from p53 exons 5–9 containing mutations either in exon 8 or in exon 9. We then formed ‘cross-hybridized sequences’ and screened the mixture for mutations in p53 exons 5–9 in a single tube using s-RT-MELT, as depicted in A. Following real-time PCR and melting curve analysis, the exon 8 or exon 9 mutants were clearly distinguished from the wild-type sample (A, curves 1–3). Next, the mutant exon 8 DNA sample was first diluted 10-fold into wild-type exon 8 and the equimolar mixture of p53 exons 5–9 was prepared and screened again in a single tube via s-RT-MELT. The exon 8 mutation was again distinguished from the wild-type mixture of exons (B, curves 1–3). Since >80% of p53 mutations in human tumors are encountered in exons 5–9 (), the multiplex single-tube s-RT-MELT reaction could be used to identify most p53 mutations encountered in clinical tumor samples. Combined with multiplex PCR directly from genomic DNA, this approach could result to a convenient, high-throughput method for mutation scanning. By adopting a real-time PCR platform as endpoint detection for s-RT-MELT, the throughput for mutation scanning increases drastically over other mutation pre-screening approaches that utilize dHPLC, or capillary and gel electrophoresis. To demonstrate better this point, a highly parallel nano-technology platform was adopted for the real-time PCR step of s-RT-MELT that enables an array of 3072 nl volume real-time PCR reactions (OpenArray™ system) to be carried-out simultaneously followed by differential melting curve analysis (). As a proof of principle of the compatibility of s-RT-MELT with OpenArray™, p53 exon 8 PCR products were generated from 48 different lung adenocarcinoma samples and mutation-containing cell lines and processed via the hybridization and enzymatic steps of s-RT-MELT. The 48 samples were each dispensed in 10 replicate nano-liter volume reactions on OpenArray™ plates pre-fabricated to contain the appropriate primers and amplified in 3072 real-time PCR reactions using a denaturation temperature of 90°C in the presence of SYBR-GREEN I dye. Melting curves were subsequently obtained using the OpenArray™ melting curve analysis mode. The PCR growth curves and smoothed differential melting curves obtained distinguish clearly the mutation-containing samples from wild-type samples (C and D, representative results from 3072 reactions). Furthermore, identification of mutation-containing samples is in good agreement between the conventional and the nano-technology platforms (D versus H). These data indicate that s-RT-MELT is compatible with high-throughput nano-technology detection formats and reiterates the advantage of de-coupling enzymatic selection from the detection step. Comparison of the throughput using conventional pre-screening method (dHPLC or dHPLC/Surveyor™) to s-RT-MELT () indicates that s-RT-MELT is 1–2 orders of magnitude faster when a large number of samples (>100) are screened for mutations. If the multiplex s-RT-MELT format is adopted, the throughput can increase further. The intrinsic potential of enzymatic mutation scanning for parallel identification of mutations can, in principle, be very high since the enzyme operates on numerous distinct mismatch-containing sequences on a molecule-to-molecule basis thus providing highly parallel mutation scanning. However, in the past the selectivity of the enzymes used and the endpoint detection method has limited the realization of this potential. Here we enabled Surveyor™, an endonuclease that recognizes selectively mismatches formed by mutations and small deletions following ‘cross-hybridized sequence’ formation, to generate mutation-specific DNA fragments that are amplified and screened via differential melting curve analysis. The replacement of size-separation methods (capillary/gel electrophoresis, dHPLC) by real-time PCR technology as the endpoint detection platforms and the ability to scan more than one sequences in parallel result in a highly increased throughput for s-RT-MELT while retaining the ability to detect diverse mutations at low-levels. Cel I/II endonucleases have also been known to have exonuclease activity on 5′ DNA-ends (,). For this reason, s-RT-MELT was designed to attach an oligonucleotide linker to the 3′-DNA ends via terminal transferase (TdT) instead of using the 5′-DNA ends. The exonuclease activity also tends to degrade the attached 5′-GC-clamps in s-RT-MELT, thereby eliminating their influence in reducing amplification of un-digested fragments. We found that if exposure of DNA ‘cross-hybridized sequences’ to Surveyor™ is limited to 15–20 min, the substantial degradation of 5′-GC-clamps is avoided. For multiplexing mutation detection using several PCR products simultaneously, the size of the GC-clamp on each PCR amplicon may need to be individually adjusted to ensure that mutations along all sequence positions of the PCR products included in the mixture can be screened at a single real-time PCR temperature and that undigested fragments do not amplify. The calculational tools developed in this work can be used to guide the individual design of GC-clamps. s-RT-MELT detects heterozygous SNPs as well as mutations. As with other mutation pre-screening techniques, the presence of a SNP concurrently with a mutation might be difficult to identify without performing sequencing. Because SNPs occur at fixed positions, melting peaks originating from SNPs have a reproducible pattern and melting temperatures (,) thus in many cases they should be distinguishable from mutations. Finally, it is noteworthy that s-RT-MELT is a general methodology that may also be applied to isolate mutations using mismatch-cutting enzymes other than Surveyor™ when enzymes with satisfactory properties for mutation detection become available. Detection platforms other than real-time PCR/melting (e.g. DNA microarray-based) may also be envisioned following enzymatic mutation selection. In summary, we developed a new method for rapid mutation scanning, s-RT-MELT that utilizes the Cel I/II (Surveyor™) and terminal deoxy-nucleotide transferase (TdT) enzymes to isolate and amplify mutation-containing DNA fragments without the requirement of DNA size-dependent techniques. Besides enabling highly increased throughput, multiplexed mutation screening and direct sequencing of the identified mutant DNA fragments, s-RT-MELT also retains the advantages of the Surveyor endonuclease over alternative pre-screening methods, such as reliability and identification of genetic alterations present at low (1–10%) fractions in the sample. s-RT-MELT provides a significant advancement in unknown mutation scanning in cancer research and diagnostics as well as for general medical, biological and biotechnology applications. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Modified nucleotides are useful tools to study the structures, biological functions and chemical and thermodynamic stabilities of nucleic acids (). Recently, microarray methods were introduced to study the structure of nucleic acids (). In native RNA, a majority of nucleotides form canonical pairs and single-stranded regions are typically short, roughly 5–7 nucleotides long. Detection of these single-stranded regions by RNA binding to probes on microarrays requires that the hybrid formed be thermodynamically sufficiently stable to capture the RNA. The thermodynamic stability of nucleic acid duplexes is strongly dependent on sequence, however. For example, duplexes of RNA heptamers composed of only A-U or G-C base pairs can differ in stability (ΔΔG°) by up to 15 kcal/mol, which is over 10 orders of magnitude in K (). This complicates interpretation of microarray data. Incorporation of modified nucleotides in microarray probes can increase the thermodynamic stability of hybrid duplexes and make the thermodynamic stability relatively independent of sequence. Consequently the single-stranded character of potential binding sites in target RNA becomes the dominant factor determining binding, thus simplifying interpretation to deduce target RNA secondary structure. There are many ways to adjust the stabilities of nucleic acid duplexes (,). Initial microarray experiments to deduce RNA secondary structure () used 2′--methyl RNA probes because 2′--methyl RNA/RNA duplexes are more thermodynamically stable than DNA/RNA duplexes (,,) and 2′--methyl RNA probes are also chemically stable. The thermodynamic stability of 2′--methyl RNA/RNA duplexes can be enhanced by incorporation of LNA nucleotides (), much as LNA stabilizes DNA/DNA (,,) and DNA/RNA (,) hybrids. Here, we show that 2,6-diaminopurine substitution for A in 2′--methyl RNA or LNA nucleotides can further increase thermodynamic stabilities of hybrids with RNA and thereby reduce the sequence dependence of hybrid formation. The 2,6-diaminopurine riboside (D) is an analog of adenosine containing an additional amino group at position 2 of the purine ring. The 2-amino group allows formation of a third hydrogen bond with uridine in the complementary strand. Previous studies have shown that 2,6-diaminopurine can increase the thermodynamic stability of RNA and DNA duplexes (). The data presented here demonstrate that substitution of D for A increases thermodynamic stability (ΔΔG°) of fully complementary 2′--methyl RNA/RNA duplexes by 0.4–1.2 and 1.0–2.7 kcal/mol at 37°C, respectively, for each 2′--methyl-2,6-diaminopurine riboside (D) or LNA-2,6-diaminopurine riboside (D) present in the duplex. The results for fully complementary 2′--methyl RNA/RNA duplexes fit a nearest neighbor model for predicting stability and the effects of D and LNA substitutions are additive. Measurements of duplexes with mismatches indicate that internal D-A, D-C and D-G pairs are very destabilizing relative to D-U, thus providing specificity. Mass spectra of nucleosides and oligonucleotides were obtained on an LC MS Hewlett Packard series 1100 MSD with API-ES detector or an MALDI-TOF MS, model Autoflex (Bruker). Thin-layer chromatography (TLC) purification of the oligonucleotides was carried out on Merck 60 F TLC plates with the mixture 1-propanol/aqueous ammonia/water = 55:35:10 (v/v/v). TLC analysis of reaction progress was performed on the same type of silica gel plates with various mixtures of dichloromethane and methanol (98:2 v/v, 95:5 v/v, 9:1 v/v and 8:2 v/v). The synthesis of protected 2′--methyl-2,6-diaminopurine riboside derivative was performed according to general procedures of the synthesis of 2′--methylnucleosides with some modifications (). 2,6-Diaminopurine riboside was treated with 1,3-dichlorotetraisopropyldisiloxane () and then 5′,3′--(tetraisopropyldisiloxane-1,3-diyl)-2,6-diaminopurine riboside was methylated with iodomethane in the presence of sodium hydride (). The 2′--methylated derivative was treated with isobutyryl chloride followed by triethylammonium fluoride (,). Treatment of the last derivative with dimethoxytrityl chloride followed by 2-cyanoethyl-′-tetraisopropylphosphordiamidite gave 5′--dimethoxytrityl-2′--methyl-2,6-diisobutyryl-2,6-diaminopurine riboside-3′--phosphoramidite in overall yield . 35%. Oligonucleotides were synthesized on an Applied Biosystems DNA/RNA synthesizer, using β-cyanoethyl phosphoramidite chemistry (). Commercially available A, C, G and U phosphoramidites with 2′--tertbutyldimethylsilyl or 2′--methyl groups were used for synthesis of RNA and 2′--methyl RNA, respectively (Glen Research, Azco, Proligo). The 3′--phosphoramidites of LNA nucleotides were synthesized according to published procedures (,,) with some minor modifications. The details of deprotection and purification of oligoribonucleotides were described previously (). Oligonucleotides were melted in buffer containing 100 mM NaCl, 20 mM sodium cacodylate, 0.5 mM NaEDTA, pH 7.0. The relatively low sodium chloride concentration kept melting temperatures in the reasonable range even when there were multiple substitutions and also allowed comparison with previous experiments (,). Oligonucleotide single-strand concentrations were calculated from absorbance above 80°C and single-strand extinction coefficients were approximated by a nearest-neighbor model with D approximated as A (,). Absorbance vs temperature melting curves were measured at 260 nm with a heating rate of 1°C/min from 0 to 90°C on a Beckman DU 640 spectrophotometer with a thermoprogrammer. Melting curves were analyzed and thermodynamic parameters were calculated from a two-state model with the program MeltWin 3.5 (). For most sequences, the ΔH° derived from T versus ln (C/4) plots is within 15% of that derived from averaging the fits to individual melting curves, as expected if the two-state model is reasonable. Thermodynamic parameters for predicting stabilities of 2′--methyl RNA/RNA with the Individual Nearest Neighbor Hydrogen Bonding (INN-HB) model () were obtained by multiple linear regression with the program Analyse-it v.1.71 (Analyse-It Software, Ltd., Leeds, England, ) which expands Microsoft Excel. Analyse-It was also used to obtain enhanced stability parameters for LNA-2′--methyl RNA/RNA duplexes when the LNAs are separated by at least one 2′--methyl nucleotide. Results from T vs ln (C/4) plots were used as the data for the calculations. The derivative of LNA-2,6-diaminopurine was synthesized with an approach similar to that described for synthesis of natural LNA nucleosides (,,) (). The derivative of pentafuranose () () was condensed with trimethylsilylated 2,6-diaminopurine in 1,2-dichloroethane in the presence of trimethylsilyl trifluoromethanesulfonate as catalyst (). Treatment of derivative () with lithium hydroxide resulted in the 5′--methanesulfonyl derivative (), which was converted with lithium benzoate into the 5′--benzoyl derivative (). The application of lithium benzoate instead of sodium benzoate very significantly improved solubility of the benzoate salt in -dimethylformamide. Treatment of 5′--benzoyl derivative () with aqueous ammonia resulted in formation of (). Removal of the 3′--benzyl with ammonium formate in the presence of Pd/C () resulted in formation of LNA-2,6-diaminopurine riboside (). Derivative () was treated with acetyl chloride to produce (), which was converted into LNA-2,6-diacetyl-2,6-diaminopurine riboside (), using classical Khorana's procedure (), and later into the 5′--dimethoxytrityl derivative (). The overall yield of synthesis up to this step was 18%. In reaction of LNA-5′--dimethoxytrityl-2,6-diacetyl-2,6-diaminopurine riboside () with 2-cyanoethyl-′-tetraisopropylphosphordiamidite was converted into LNA-5′--dimethoxytrityl-N2,N6-diacetyl-2,6-diaminopurine riboside-3′--phosphoramidite () in 93% yield. It was possible to use acetyl instead of isobutyryl to protect the 2,6-amino groups of LNA-2,6-diaminopurine riboside because LNA-5′--dimethoxytrityl-2,6-diacetyl-2,6-diaminopurine riboside-3′--phosphoramidite () is soluble in acetonitrile. This is in contrast to 5′--dimethoxytrityl-2′--methyl-2,6-diacetyl-2,6-diaminopurine riboside-3′--phosphoramidite. The details concerning chemical synthesis of derivatives are described in Supplementary Data. The adenosines in duplexes of the form 5′ACWAXCA/r(3′UGZUYGU) were replaced singly or completely by 2′--methyl D (D) or LNA D (D), and the thermodynamics for duplex formation were measured (). Here WZ and XY are Watson–Crick base pairs. The results can be compared with previous measurements (,) for the unsubstituted duplexes and for the As substituted by LNA A (, see also Supplementary Data for complete thermodynamic data). When only a 5′ or 3′ terminal A is substituted by D with the same type of sugar, the average enhancement in stability at 37°C is 0.37 kcal/mol. If the middle A is substituted by D with the same type of sugar, then the average enhancement is 0.94 kcal/mol. Comparisons of ΔΔG° values for the A to D replacements in to the sum of corresponding A to D and A to A replacements, which in are listed immediately below in square brackets, indicate that the effects of replacing A with D and 2′--methyl with LNA are essentially additive. The results with DCUACCA, DCUACCA, DCUDCCD, and DCUDCCD suggest that the effects of multiple substitutions are also additive. For these sequences, the enhancement in heteroduplex stability relative to ACUACCA differs from the sum of enhancements due to the individual replacements by only 0.48, 0.29, 0.27 and 0.07 kcal/mol, respectively. The results in can be combined with previous results (,) to obtain nearest neighbor parameters for 2′--methyl RNA/RNA duplexes (). The nearest neighbor parameters with D are preliminary due to the small number of occurrences for them. The thermodynamic parameters for 5′ACUAGCA/3′r(UGAUCGU) were re-measured and the values in and Supplementary Data were used for deriving the nearest neighbor parameters. The parameters for nearest neighbors without D are similar to those reported previously (). Some single mismatches in RNA/RNA duplexes are particularly stable thermodynamically due to hydrogen bonding (). Interpretation of microarray and other data must consider potential hybridization involving mismatches. Because it is important to determine the specificity of base pairing to modified nucleotides, mismatches with D, D, A and A were studied. Most mismatches were placed at an internal position within duplexes because that is statistically the most likely occurrence. Some terminal mismatches were also measured, however. The results from optical melting experiments are listed in (see also Supplementary Data for complete thermodynamic data) and the differences between free energies of duplex formation with A-U or D-U and mismatch pairing for single internal mismatches at 37°C are summarized in . Many of the duplexes had melting temperatures <20°C, which makes measurements difficult. This is one reason that some of the transitions appear non-two-state as indicated by more than a 15% difference between ΔH° values derived from fitting the shapes of the melting curves or from T vs ln (C/4) plots. Slightly, non-two-state melts were also found for four duplexes with melting temperature >31°C. For these sequences, there was at most a 19% difference between the derived ΔH° values. The ΔG° values for these sequences are still reliable because errors in ΔH° and ΔS° compensate making ΔG° values near the T reliable (). Values from non-two-state melts are listed in parentheses in and . Mismatches at 5′- or 3′-terminal positions are typically less destabilizing than internal mismatches (). To evaluate this effect, D-G and D-G mismatches were placed at 5′- or 3′-terminal positions of 2′--methyl RNA/RNA duplexes (C). For 5′XCUACCA/3′rGGAUGGU duplexes, where X is D or D, the destabilization (ΔΔG°) is 0.78 and 0.98 kcal/mol for D-G and D-G, respectively. This is similar to the destabilizations of 0.37 and 0.58 kcal/mol when X is A or A, respectively. For 5′ACUACCX/3′rUGAUGGG duplexes, where X is D or D, the destabilization is 0.67 and 0.56 kcal/mol for D-G and D-G, respectively. This is similar to the destabilization of 0.32 and 0.48 kcal/mol when X is A or A, respectively. While the differences between destabilizing by terminal A-G and D-G mismatches are within experimental error, the D-G mismatches are all more destabilizing than A-G suggesting that this is a real, albeit small, effect. As expected, the destabilization effect (ΔΔG°) is reduced compared with the same mismatches at an internal position as listed in . There are many reasons to modify the thermodynamic stabilities of nucleic acid duplexes. The application of microarrays of short oligonucleotides to probe RNA secondary structure () is one case where it is particularly useful to have sequences that base pair strongly and isoenergetically to RNA targets. Strong pairing permits the use of short oligonucleotides so that self-folding of probe is largely avoided. Moreover, short oligonucleotides provide enhanced specificity of binding (). Isoenergetic binding further simplifies interpretation of data because binding will be primarily dependent on target structure rather than probe sequence. The synthesis of oligonucleotides with 2,6-diaminopurine described here provides a way to improve recognition of U in RNA targets by enhancing both binding and specificity. Moreover, the thermodynamic results provide approximations that allow design of isoenergetic probes. The design is relatively straightforward because the effects of non-adjacent modifications are usually additive. Short modified oligonucleotides could also be applied as antisense oligonucleotides (ASO) (,,). They could also be useful to modulate binding and biological activity related to single nucleotide polymorphism (SNP) (,) and microRNAs (). Synthesis of LNA-2,6-diaminopurine riboside was reported by Rosenbohm . () and Koshkin . (). Both used 2-amino-6-chloropurine as precursor of 2,6-diaminopurine. Rosenbohm used a saturated solution of ammonia in methanol to convert derivative of 2-amino-6-chloropurine riboside into 2,6-diaminopurine riboside and this transformation was accompanied by formation of 6--methyl derivative. Efficient synthesis (65% yield) of 2,6-diaminopurine riboside required not only specific temperature but particularly control of the pressure during this reaction. Koshkin proposed to convert the derivative of 2-amino-6-chloropurine riboside into 2-amino-6-azidopurine riboside and then into 2,6-diaminopurine riboside derivative simultaneously with deprotection of 3′--benzyl. An advantage of Rosenbohm and Koshkin approaches is universal character of 2-amino-6-chloropurine riboside derivative which beside 2,6-diaminopurine riboside can be transformed into LNA-guanosine and LNA-2-aminopurine riboside. A disadvantage is the much higher price of 2-amino-6-chloropurine than 2,6-diaminopurine. Moreover, both authors propose to use benzoyl as amino protecting group and in consequence using 40% aqueous solution of methylamine at 60–65°C for 2–4 h for deprotection of oligonucleotides containing 2,6-diaminopurine riboside. The method described herein is based on standard and much cheaper substrate as well as many well established procedures and is therefore a simple and efficient method for synthesizing LNA-2,6-diaminopurine riboside. Moreover, the chemical synthesis and deprotection of many oligonucleotides carrying LNA-2,6-diaminopurine riboside demonstrate that acetyl is very suitable for protection of amino groups in 2,6-diaminopurine residue. Facile synthesis and incorporation of 2′--methyl-2,6-diaminopurine riboside and LNA-2,6-diaminopurine riboside into oligonucleotides allowed measurements of the thermodynamics for formation of 2′--methyl RNA/RNA and LNA-2′--methyl RNA/RNA duplexes containing D and D. The results show that incorporation of 2,6-diaminopurine into oligonucleotides allows modulation of duplex stability over a wide range. Replacement of adenosine by D and D always enhances the thermodynamic stability of fully complementary 2′--methyl RNA/RNA and LNA-2′--methyl RNA/RNA duplexes. The largest stabilization is observed at internal positions where enhancements range from 0.7 to 1.2 kcal/mol with an average of 0.9 kcal/mol and 1.7–2.7 kcal/mol with an average of 2.3 kcal/mol, respectively, for D and D substituting for A. The D stabilization is in the range expected for addition of a hydrogen bond in RNA (). The D stabilization is the sum of the effects of an extra hydrogen bond and of the LNA. The enhancement (ΔΔG°) for D and D relative to A and A is less at 5′- and 3′-terminal positions where it averages 0.4 kcal/mol. This difference in stabilization at terminal and internal positions is likely due to the competition between stacking and hydrogen bonding at terminal base pairs () and to the particular sequences studied. Other sequences may show larger effects for 2,6-diaminopurine substitutions at terminal positions. The stabilities of fully complementary 2′--methyl RNA/RNA and LNA-2′--methyl RNA/RNA duplexes at 37°C can be predicted reasonably well with simple models. The nearest neighbor parameters in allow prediction of stabilities for 2′--methyl RNA/RNA duplexes using the INN-HB model () and the additional enhancement, ΔΔG° (chimera/RNA), due to an LNA sugar can be predicted from: Here n is the number of 5′ terminal LNAs, n, n and n are the number of internal LNAs in A-U, D-U and G-C pairs, respectively, n and n are the number of 3′ terminal LNAs that are U or not U, respectively. The equation is similar to that suggested previously (,), but has been updated to include the new results in . The predicted values are listed in square brackets in . Mismatches formed by D and D destabilize duplexes ( and ). At the central position of duplexes with seven pairs that melt in a two-state manner, the destabilization (ΔΔG°) ranges between 2.3 and 5.2 kcal/mol at 37°C when D was only present as a mismatch. This corresponds to K's less favorable by 42 to 4600-fold at 37°C. With the possible exception of the 5′CAA/3′GGU context, mismatches of D and D with G and A destabilize 2′--methyl RNA/RNA and LNA-2′--methyl RNA/RNA duplexes more than similar mismatches formed by A and A, respectively (). The trend of destabilization is reversed for D-C mismatches. The D-C mismatches might be stabilized by a hydrogen bond between the 2-amino group of D and the O2 of C. Interestingly, the effect of a central D-G mismatch is enhanced when both terminal base pairs are D-U in the context 5′DCUDCCD/3′rUGAGGGU (). Here, the destabilization is 4.03 and 4.59 kcal/mol when each D is 2′--methyl or LNA, respectively, compared with 1.57 and 2.99 kcal/mol when the terminal nucleotides of the probe are 2′--methyl A. For mismatches at terminal positions, the destabilization ranges from 0.6 to 1.0 kcal/mol at 37°C. Mismatches with an LNA nucleotide are usually more destabilizing than those with a 2′--methyl nucleotide (). The enhanced, variable and predictable duplex stability available from 2,6-diaminopurine substitutions with either 2′--methyl or LNA sugars makes them valuable for designing isoenergetic duplexes. The large destabilizations from internal mismatches means that oligonucleotides with 2,6-diaminopurine will be highly specific for their complementary sequence. Thus they should facilitate many applications of oligonucleotides, including microarray methods for probing RNA structure () and design of nanostructures (). p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

p68 and p72/p82 are prototypic members of the DEAD box protein family (,) with the isoforms p72 and p82 arising from the gene through the use of different in-frame translation initiation codons (). The three proteins form a subfamily with similar biochemical activities, including an ATP-dependent RNA helicase (,), and possibly also similar or even redundant cellular functions. The products of both genes are important transcriptional regulators, functioning as co-activators (,) and/or co-repressors () depending on the context of the promoter and the transcriptional complex in which they occur. They also seem to be involved in alternative splicing (,) where their target genes differ, however. It is unclear, if/how the RNA-specific biochemical activities of p68 and p72/p82 are involved in these processes as well as in DNA deglycosylation (), though at least some of the functional complexes formed contain small RNAs (,). Different specific functions of isoforms p82 and p72 are not known. In , the (only) homolog of the p68 subfamily proteins, Dbp2p (), functions in the nonsense-mediated mRNA decay pathway and plays a role in ribosomal RNA (rRNA) processing (). Ribosome assembly starts in the nucleolus by formation of a 90S particle from non-ribosomal and ribosomal proteins and a single pre-rRNA primary transcript (35S in yeast cells, 47S in mammals). Within the 90S pre-ribosome the primary transcript undergoes rapid cleavage steps, which separate the precursors to the large and small subunits. Within the precursors to the large subunit (60S pre-ribosomal particles) the pre-rRNA is further maturated to the large rRNA (25S RNA in yeast, 28S RNA in mammals) and 5.8S rRNA. The early pre-60S particles are restricted to nucleoli, but later maturating forms are released into the nucleoplasm and are eventually exported to the cytoplasm (). From this, a nucleoplasmic role of Dbp2, and possibly also p68, in rRNA processing must be envisaged, since both proteins are excluded from nucleoli in interphase cells () and have recently been shown to be associated with (at least in part) nucleoplasmic forms of 60S pre-ribosomes (, for a review see also ). Only in late mitosis, are the p68 subfamily proteins detected in the reorganizing nucleoli (prenucleolar bodies) where they may additionally be involved in the structural (re)organization process (). DEAD box proteins are believed to act as ATP-dependent modulators of RNA structure, and, not surprisingly, several family members seem to assist the structural rearrangement of pre-rRNA in the course of its correct processing and folding within pre-ribosomes (,). However, an understanding of their exact roles in ribosome biogenesis is still lacking. At least some of the protein-catalyzed rRNA structural rearrangements may proceed via a branch migration type of mechanism in which portions of two competing secondary structures transiently coexist (,) and which has been shown to be used in RNA rearrangement reactions catalyzed by p68 and p72 () or by the cyanobacterial DEAD box RNA helicase CrhR (). Here, we report that either p68 or p72/p82 is essential for pre-rRNA maturation and that knock-down of both genes induces cell death. Additional data suggest a molecular mechanism by which the p68 subfamily proteins, with the participation of U8 small nucleolar RNA (snoRNA), may promote an RNA structural rearrangement within the pre-60S ribosomal subunit, indispensable to the timed endonucleolytic cleavage of pre-28S rRNA. To generate expression plasmid pCMVp68-wt, the coding sequence of pUHDp68KT () was amplified by use of primers pCIneo-fwd and His-KT3-rev (see ) and cloned into the XbaI/BamHI-restriction site of pCMV (). p68-mutant plasmids pCMVp68DQAD and pCMVp68GNT were generated from pCMVp68-wt using the QuikChange II Site-Directed Mutagenesis Kit (Stratagene) according to the manufacturer's instructions. Plasmids pGEM-U8 and pGEM-U14 were constructed by insertion of the U8- () or U14- () cDNA in 3′ to 5′-orientation into the MCS of vector pGEM7zf(−) (Promega). All DNA constructs were confirmed by direct DNA sequencing. For protein expression, cells were transfected twice at an interval of 24 h by use of FuGENE6 (Roche) as described in the user manual. siRNAs with 3′-dTdT-overhangs were produced by Qiagen and covered the sequences shown in . siRNA transfections were carried out using HiPerFect (Qiagen) as a transfection reagent with 10–20 nM siRNA following the manufacturer's protocol. Cells were cultured at 37°C and 5% CO in DMEM with 10% FCS. For colony formation assays, 5 × 10 or 10 cells were plated in 10-cm dishes the day after transfection and after 7–14 days the resulting colonies were fixed in 5% glutaraldehyde and stained with 1% crystal violet in PBS (137 mM NaCl, 2.6 mM KCl, 6.5 mM NaHPO, 1.4 mM KHPO pH 7.4). Incorporation of 5-bromo-2′-deoxy-uridine (BrdU) was measured using the respective Roche Kit I. Photographs are representative of the whole slide. For FACS analysis, HeLa cells were fixed with ice cold 70% ethanol for 30 min at −20°C 5 days after siRNA transfection followed by an incubation with 1/10 vol. RNase A (1 mg/ml) and 1/10 vol. propidium iodide (500 µg/ml) for 30 min at 37°C and analyzed with the FACScan (Becton Dickinson). Growth kinetics were determined for three independent cultures of each siRNA transfection and counted in duplicate. The resulting mean values are given. Total RNA was prepared using the RNA Kit II (Invitek), and cDNA was synthesized with RevertAid™ M-MuLV Reverse Transcriptase (Fermentas) with either oligo-dT or gene-specific 3′-primers. For relative semiquantitative PCR analyses, Taq-DNA polymerase (Fermentas) and a serial dilution of RT-generated cDNA first strand were used to produce amplified DNA products in the linear range. The amplified products were normalized to those specific for host genes β-actin or histone 2A and analyzed by agarose gel electrophoresis. Primer pairs were used as shown in . For GAPDH, β-actin and histone primers see (). Northern blot analysis was performed as described elsewhere () with digoxigenin-labeled negative strand full-length U8 or U14 snoRNA as a probe (obtained by transcription of plasmids pGEM-U8 and -U14 and detected according to the Roche DIG Northern Starter Kit manual) or with P end-labeled DNA oligonucleotides specific for human Second internal transcribed spacer (ITS2), 18S, 28S or 5,8S rRNA (). rRNA processing was monitored by pulse-chase experiments. HeLa cells were starved of methionine for 60 min and then pulse labeled with 2.22 MBq/ml -(methyl-H)-methionine (Amersham) for 60 min. Thereafter, cold methionine (15 µg/ml) was added in order to chase the label for 60 or 120 min. From all aliquots, total RNA was extracted as described above and the incorporated radioactivity measured by liquid scintillation counting. Equal amounts of radioactivity were loaded onto a 1% agarose denaturing gel. The RNA was fractionated and transferred onto a nylon membrane (Roche). The membrane was dried, sprayed with ENHANCE (Perkin Elmer) and exposed to Fuji medical X-ray films at −80°C for 2 days. For monoclonal antibody C10, see (), for rabbit polyclonal α-human p72/p82 antibodies, see () and for monoclonal PAb421 see (). Monoclonal α-fibrillarin antibody 72B9 was a gift of Prof. U. Scheer, University of Würzburg, and rabbit α-p19 antibodies were provided by Prof. M. Montenarh, University of the Saarland. α-β-actin antibodies were from SIGMA, α-B23 antibodies as well as α-p53 antibodies DO-1 from Santa Cruz, [α-PARP] antibodies from Pharmingen and α-His antibodies from Qiagen. FITC-conjugated as well as TRITC-conjugated secondary antibodies were from Molecular Probes and horseradish peroxidase-conjugated ones from SIGMA. Western blotting experiments were performed as described by () using ECL (Roche) for detection. For indirect immunofluorescence, cells were grown on coverslips, fixed in 3.7% formaldehyde in PBS for 7 min at room temperature, permeabilized with 0.5% Triton X-100 in PBS + 1% BSA for 6 min on ice and stained with the indicated antibodies. FITC- or TRITC-conjugated secondary antibodies were added for 60 min at room temperature at 1:1000 dilution. Samples were analyzed with a fluorescence microscope (Zeiss Axioscop). For isolation of wt-p68 and its mutants, COS cells (1 × 10), transfected with the respective expression plasmids for 4 days, were extracted as described by () except that the nuclear extraction buffer contained 4 mM EDTA, which was subsequently removed by dialysis before the recombinant proteins were purified by affinity chromatography on Ni NTA–cellulose and ssDNA–cellulose as described (). Purified proteins were stored at −70°C. RNA structural rearrangement reactions and preparation of the used RNA substrates were performed exactly as described previously (). ATP binding by wild-type (wt) or mutant p68 was analyzed by UV-induced photo-cross-linking as described (). Briefly, p68 or one of its mutants (300 nM) was incubated in a buffer containing 20 mM Tris-HCl (pH 7.5), 70 mM KCl, 5 mM magnesium acetate, 0.11 MBq of [α-P] ATP (110 TBq/mmol), 10% glycerol, 1.5 mM dithiothreitol (DTT) for 10 min at 37°C. Samples were placed on ice and brought under the UV cross-linker about 4 cm under the light source. The samples were irradiated for 4 min, boiled in SDS-PAGE sample buffer, and after addition of unlabeled ATP (final concentration of 4 mM) subjected to SDS-PAGE (10% polyacrylamide). After electrophoresis, the gel was stained with Coomassie blue, dried and processed for autoradiography. ATPase assays were performed as described previously (). Briefly, p68 (or one of its mutants; 30 nM) was incubated in ATPase assay buffer containing 25 mM HEPES pH 7.8, 5 mM MgCl, 100 mM NaCl, 1 mM DTT, 0.01% albumin and 10% glycerol and 20 µM [γ-P] ATP (13 Bq/pmol) for 20 min at 37°C. The unreacted ATP was precipitated by acid-washed charcoal (20% in 0.25 M HCl, 0.25 M HPO), and after centrifugation the free in the supernatant was counted in Aquasol (Dupont). ATPase activities were determined three times for each protein. The resulting mean values are given. RNA helicase assays were performed in ATPase assay buffer containing 10 U/sample RNase inhibitor, 4 mM ATP, 35 nM (in nucleotides) helicase substrate [17 bp RNA; ()] and 10 nM wt or mutant p68 in a final volume of 20 µl as previously described (). For profiling of ribosomal subunits, HeLa cells were sonicated 2 × 15 s in 40 mM Tris-HCl pH 8.5, 150 mM NaCl, 0.1% NP40. After centrifugation at 10 000 r.c.f. for 10 min, the supernatant was adjusted to 30 mM EDTA, layered onto a linear 10–30% sucrose gradient (w/v) in 20 mM Tris-HCl pH 7.6, 30 mM KCl, 1 mM MgCl, 6 mM β-mercaptoethanol and centrifuged in a Beckman L-60 ultracentrifuge for 200 min at 37 000 r.p.m. with a Beckman SW41Ti-rotor. The absorbance of each fraction was measured at 260 nm. For sedimentation analysis of U8 snoRNPs, equal absorption units (OD) of HeLa cell extracts were analyzed on sucrose gradients as described above except that the sonication buffer contained 300 mM NaCl instead of 150 mM. From each fraction RNA was prepared and analyzed for U8 snoRNA by northern blot analysis. Different siRNAs were designed to knock-down the p68 subfamily proteins by RNA interference (RNAi; ). In HeLa cells, the p68 level was reduced effectively and specifically (by almost 95% compared to control siRNA) after transfection of a siRNA (p68 siRNA) complementary to the translational start site of the p68 mRNA (A). The absence of p68 led to a remarkable increase in cellular p72 and p82 protein and mRNA levels (but not vice versa), while that of other genes, like β-actin, remained constant (A and B). Therefore, the possibility that p68 negatively controls the expression of p72 and p82, like that of its own, at the level of splicing was considered (,,). As previously reported, we found an accumulation of partially spliced p72/p82 transcripts in HeLa cells (,) that may result from this negative expression control. In fact, by p68 depletion the level of these transcripts could be reduced (B, lanes 6 and 7). In contrast to such a rather specific effect, Lin . () reported that p68 is a basic human splicing factor (essential e.g. for the splicing of β-actin and GAPDH pre-mRNA) and these authors also speculated on functional redundancy of p72 and p68 in this function. Thus, we checked for deficiencies of essential mRNAs in HeLa cells after knock-down of one or the other as well as both DEAD box genes at once. However, relative quantitative RT-PCR analyses did not reveal any reduction in the β-actin and GAPDH mRNA after siRNA transfection for 2 days (B) or longer (e.g. 3, 4 and 5 days; data not shown), when the intronless histone H2A.X mRNA, not expected to be affected by possible splicing defects, was used to normalize the RT-PCR products. Furthermore, no DNA fragments representing unspliced transcripts of β-actin or GAPDH genes were detected by the same RT-PCR analysis, though the primers span an intron in both genes (B). Those PCR signals were obtained, however, in control reactions performed with genomic DNA. Taken together, these results do not hint at an essential function of p68 and/or p72/p82 in general splicing. We note that the knock-down experiments reported above and also those described below were confirmed by use of another p68- and p72/p82 siRNA (siRNAs, see ) as well as of other cell lines (HaCaT, MCF7; data not shown). The high rates of protein suppression obtained reflect similar or even higher siRNA transfection efficiency, confirmed by immunofluorescence analysis of respective cells (C). Some sequence identity at the cDNA level of p68 and p72/p82 enabled us to find a siRNA directed to both p68 and p72/p82 (called here as subfamily siRNA), which efficiently co-suppressed all three proteins (A and B, lanes 4 and 3, respectively). On the other hand, mixing of a p68- and p72/p82 siRNA in co-transfection experiments made it possible to titrate the absolute cellular protein levels and to prevent, for example, an increase in p72 and p82 while simultaneously maximally suppressing p68 (A, lane 5). Notably, the ratio of p72 to p82, translated at roughly similar rates from one mRNA (), was unaffected by any siRNA used. Though it has been speculated that p68 is essential for cell proliferation (,), its efficient knock-down as well as that of p72/p82 (to ∼5 and 15% of the control, respectively; see A) did only slightly alter the proliferation rate of HeLa cells (A and B) and their distribution in G, S and G phases as revealed by flow cytometry (FACS) analysis (C). In contrast, after co-suppression of all three DEAD box proteins, the proliferation of HeLa cells stopped (A) and their DNA replication was blocked (data not shown). Eventually, they became abnormally flat resembling serum-starved cells, and the size of their nucleoli decreased (see A). A release of the nucleolar proteins fibrillarin, B23 and p19 into the nucleoplasm was observed (through which the staining is softened down due to the distribution of the proteins within the whole nucleus; A), and the cellular levels of fibrillarin and B23 were moderately reduced (by ∼15%) most probably due to their mislocation () whereas that of p19 remained unchanged (B). At the same time, a clear decrease of cells in G1 phase and an increase of the sub-G1 population were observed, the latter indicating cell death (C). Accordingly, co-depletion of all p68 subfamily proteins strongly reduced the clonogenic survival of HeLa, MCF7 and COS cells by >95% as compared to the control, while most cells (about 80%) with suppressed p68 or p72/p82 survived in this assay (B). Taken together, these results indicate that p68 and p72/p82 have a redundant function which is essential for cell proliferation. Notably, the proliferation of the cells was also strongly reduced by co-transfection of individual p68- and p72/p82 siRNAs (A and B), and even at a ratio that caused maximal suppression of p68, but kept a nearly constant level of p72/p82 (preventing its increase, data not shown, but see A, lane 5). Therefore, the ‘normal’ (physiological) cellular level of p72/p82 does not preserve cell proliferation in the absence of p68, whereas that of p68, vice versa, clearly does. Co-depletion of p68 and p72/p82 does not alter the level of p53 in HeLa cells (C). Thus, the function of the human papilloma virus oncogene E6, which acts as an antiapoptotic factor by inducing degradation of p53 in HeLa cells (), seems not to be affected by our cell manipulations. Accordingly, no induction of PARP cleavage was observed (C), confirming a p53-independent way of cell death in the absence of p68 and p72/p82. On the other hand, human breast cancer cells MCF7 (expressing wt p53), which displayed a similar low clonogenic survival upon co-suppression of p68 and p72/p82 (B, lane 7), showed an increase in their p53 level, apparently resulting in apoptotic death as evidenced by PARP cleavage indicative of caspase 3 activity (C). Thus, our results confirm that inhibition of ribosome biogenesis can induce cell death in a p53-dependent and -independent manner as has been shown before by depletion, e.g. of transcription factor TIF-IA () and B23 (), respectively. Mutant and wt p68 were transiently overexpressed in COS cells from expression plasmids containing a simian virus 40 origin of DNA replication, which reproducibly showed high transfection (90%, revealed by GFP cloned into the same vector; data not shown) and high protein expression efficiencies (A), the latter being essential to override endogeneous p68 subfamily proteins. A single amino acid substitution (Lys to Asn) in the ATP-binding (Walker A) motif of p68 (GNT-p68) had a drastic effect on COS cell proliferation and, given that 10% of the cells resisted the transfection, nearly completely abolished their clonogenic survival (B). This effect was specific to GNT-p68, as DQAD-p68, a mutant with a single amino acid substitution (Glu to Gln) in the DEAD box (Walker B) motif as well as exogenously expressed wt-p68 only moderately affected cell proliferation. Expression of the mutants apparently does not interfere with the negative expression control of p72/p82 by endogeneous p68 (A), and thus we can exclude that the still low amounts of p72/p82 can substitute for the p68 function in GNT-p68- or DQAD-p68-expressing cells. Notably, both mutants have no ATPase (D) and RNA helicase (E) activity, whereas ATP-binding activity is lost only in GNT-p68 (C, see also 31,39). Thus, exogenous GNT-p68 is able to compete with the endogenous protein for function resulting in a dominant negative phenotype, and it has to be tested in the absence of endogenous protein whether only ATP binding but not ATP hydrolysis or RNA unwinding of p68 is essential for cell proliferation. Nevertheless, a possible correlation of the ATP-binding activity of p68 with the capacity to catalyze rRNA rearrangement processes with an RNA branch migration complex as an intermediate product (see F) was considered. In fact, p68 and p72 have recently been shown to catalyze similar reactions (), and as it is shown in G, isolated GNT-p68 is mostly inactive in such an assay, while the same function of DQAD-p68 in comparison to wt-p68 was nearly unaffected (compare lines 5–7). In contrast to previous work (), a dependence of RNA rearrangement on ATP binding, but not hydrolysis, became obvious here most probably because p68 was prepared in the presence of high concentrations of EDTA in the extraction buffer to efficiently remove protein-bound nucleotides. Notably, RNA rearrangement is catalyzed by isolated p72 as well (). The overall rate of protein synthesis decreased in p68 and p72/p82 co-suppressed cells, apparently due to a reduction in 60S (but not in 40S) ribosomal subunits (data not shown). This and the destruction of nucleoli after subfamily siRNA transfection prompted us to look for a possible role of p68 and/or p72/p82 in ribosome biogenesis (). To determine the dynamic processing of rRNA precursors and intermediates, we carried out a pulse-chase labeling analysis of newly synthesized rRNA in HeLa cells after transfection of respective siRNAs for 70 h (B, left panel). The pulse-labeled 47S rRNA precursor was readily detected in all experiments after a 60-min pulse, indicating that transcription of rRNA genes was not blocked by any siRNA used. In the chase process, the amount of mature 28S rRNA readily increased in cells transfected with control- or p68 siRNA and, after a 120-min chase, was roughly five times that of the 32S intermediate (with a 28S:32S ratio of 5.32 and 4.66, respectively; see C). Similar results were obtained, when p72/p82 was knocked down (data not shown), indicating an effective processing of the 32S precursor in the absence of either p68 or p72/82. Cells transfected with subfamily siRNA, in contrast, showed a decreased level of labeled 28S rRNA, and the ratio of 28S:32S was reduced to 0.53 after the 120-min chase (C), whereas the maturation of 18S rRNA seemed not to be significantly affected in any case. These results were confirmed when we investigated the cellular levels of the different rRNA species by Northern blot analysis 4 days after siRNA transfection. For this, total cellular RNA was probed with sequences specific for ITS2-containing rRNA intermediates, 28S RNA, 18S RNA or 5.8S RNA (B, right panel). p68 and p72/p82 co-suppressed cells accumulated the 32S rRNA precursor whereas 12S, 5.8S and 28S rRNA levels were diminished. Thus, the simultaneous down-regulation of the p68 subfamily proteins inhibits processing of the 32S intermediate into the mature 28S rRNA. Since we also noticed an increase in the pulse-labeled 47S precursor rRNA, this may hint to a defect also in very early steps of pre-rRNA processing in the absence of p68 and p72/p82. Though U8 snoRNA also has some other function in rRNA processing, it is certainly essential for the accumulation of mature 28S and 5.8S rRNA (). Its 5′ end (outer 15 nucleotides) has the capacity to base pair with sequences at the 5′ end of 28S in pre-rRNA (). This interaction seems to transiently inhibit a premature 5.8S interaction at the same site of 28S that is essential for processing later on and still present in the mature ribosome. Thus, it has been proposed that U8 snoRNA is displaced from 28S by hybridization of the 3′ end of 5.8S rRNA. Such an RNA rearrangement may be accomplished by the protein-controlled formation of an RNA branch migration complex, which is then resolved by branch migration to dissociate the U8 snoRNA (,; see also A). U8 belongs to the class of fibrillarin-associated snoRNAs (), and a direct interaction of fibrillarin with p68 and p72 has been demonstrated (,). Therefore, we checked the cellular level of U8 snoRNA after suppression of either one or both DEAD box genes. Northern blot analysis of total RNA from HeLa cells revealed an increase of U8 (2.2-fold) 2 days after both genes were co-suppressed, whereas U14, another processing snoRNP required for pre-rRNA cleavage on the pathway of 18S rRNA synthesis (,), was not affected (B; we stress that at this time point cell proliferation was not yet diminished, see A). When p68 or p72/p82 were depleted individually, almost no increase in the level of U8 was observed, confirming a redundant function in this step of rRNA processing. Sedimentation analysis of extracts from subfamily siRNA-transfected cells showed that the increase in U8 snoRNA appeared in both free U8 snoRNPs (ribonucleoproteins; 11S) and the 60S/90S pre-ribosomal subunit fractions (C). Thus, in the absence of p68 and p72/p82, the displacement of U8 snoRNA from the pre-ribosomes seems to be hampered, and the cell tries to overcome this block by an overproduction of U8 snoRNA, which is usually limiting in ribosome biogenesis (). We have shown here that p68 and p72/p82 fulfill a redundant function in cell proliferation and viability. Our results are based on mutant studies and knock-down experiments with siRNAs, targeting either unique sequences in the individual mRNAs or a sequence common to both mRNAs and co-suppressing all three subfamily proteins at once. The diversity of the used siRNAs should exclude off-target effects due to, for example, partial sequence complementation. We point out that p72 and p82 could not functionally be discerned here, which will, however, be an interesting future task. Cells show only a moderately reduced proliferation rate when p72/p82 is suppressed on a normal p68 background and the same is true, vice versa, for p68 knocked-down cells when p72 and p82 are overexpressed due to the cessation of the p68 control. This moderate reduction in cell proliferation in the absence of only p68 or p72/p82 may be attributed to other most probably individual function(s) of these proteins not analyzed here. The negative expression control of p68 over p72/p82 seems to function, at least in part, at the level of splicing as the nuclear pool of partially spliced p72/p82 (as well as p68) pre-mRNA is nearly completely cleared at low p68 while that of the mature mRNA increased. Thus, p68, like yeast Dbp2p (), seems to regulate its own expression and, in addition, that of its closest relative most probably by an intron-mediated mechanism, though an indirect effect cannot yet be excluded. The expression of less-related DEAD box genes such as Ddx20 (Gemin3, DP103; 46) and Ddx42 () was not found to be affected (C.J. and H.S., unpublished data). Selective effects on pre-mRNA splicing, including alternative splicing (,), correspond with the previously demonstrated association of p68 with spliceosomes () but seem to contradict a recently reported essential splicing function of p68 (). Accordingly, we could not detect any impairment in the splicing of the housekeeping genes β-actin and GAPDH by efficient knock-down of p68. In addition, various pre-mRNA substrates with different -acting elements spliced equally well in mock- and p68-depleted HeLa nuclear extracts (H.U.-S., unpublished data). So far, we cannot explain this discrepancy, especially since a possible rescue of the implicated p68 splicing function by p72/p82, as discussed by (), is excluded here by the co-suppression experiments. We note, however, that our data are in agreement with very recently published results on Drosophila p68 (). Co-suppression of p68 and p72/p82 by RNAi leads to disintegration of the nucleolar structure, cell cycle arrest and cell death. Concomitantly, we observed a decrease in 60S ribosomal subunits by a partial inhibition of the processing of pre-rRNA, more precisely the cleavage of the ITS2 in the 32S rRNA precursor most probably performed by B23 (,). Such a defect may also be deduced from the demonstrated increase in the cellular U8 snoRNA level and in the fraction of U8 associated with 60S pre-ribosomal subunits, as has been reported for U14 snoRNA after depletion of the putative helicase Dbp4p (). This view is underscored by the interaction of p68 with fibrillarin (,), a known component of U8 snoRNPs (). The inhibition of rRNA processing seems to lead to a disintegration of the nucleolar structure, cell cycle arrest and eventually cell death, as has been reported for the disturbance of other steps of rRNA synthesis or maturation (,,,). In MCF7, cell destruction seems to be executed via the induction of the p53 tumor suppressor protein, and thus our observation confirms the idea that some sort of stress sensor monitors the nucleolar structure and function and regulates p53 levels (). How this monitoring, however, is translated into action in cells without functional p53 such as HeLa cells, is unknown (). For the processing of pre-28S RNA to proceed, U8 snoRNA must be displaced from 28S rRNA most probably by competitive hybridization of 5.8S rRNA via an intermediate RNA branch migration complex (), reminiscent of reactions previously shown to be catalyzed by p68 and p72 (). We have shown that p68 ATP binding but not hydrolysis is essential for the formation of similar branch migration complexes, which in case of low stability as described here, spontaneously disintegrate at 37°C (). Mutant p68, defective in ATP binding (GNT-p68), is incompetent to induce such a structural rearrangement and also drastically reduces cell proliferation in contrast to another mutant without RNA helicase activity (DQAD-p68), which can still bind ATP. The moderate reduction in the clonogenic survival of DQAD-p68 and also wt-p68 overexpressing cells may be attributable to the mislocation of p68 proteins to the nucleolus due to overexpression (see also ) or inhibition of RNA polymerase II activity as reported by Andersen . (). Under normal conditions, p68 subfamily proteins are excluded from the nucleoli in interphase cells most probably by binding partners in the nucleoplasm, such as transcription factors, splicing complexes (see above) and the A-Kinase anchoring protein AKAP95, a nuclear matrix protein not occurring in nucleoli (). AKAP95 is exposed to cytoplasmic components in mitosis and reenters the forming nucleus in telophase (). It is conceivable that it then recruits p68 and p72/p82 from prenucleolar bodies, where they are transiently localized just at that time (). Therefore, if these DEAD box proteins are involved in the displacement of U8 snoRNA by 5.8S rRNA, as implied from our experiments, this process may proceed in the nucleoplasm of interphase cells where, however, it could be difficult to detect because of potential short half-lives of respective intermediates (). The same may halt for U8 snoRNA, which after its displacement from the 32S rRNA immediately may return to the nucleoli, where it had been located exclusively so far (). In fact, it is not precisely known in which nuclear compartment ITS2 cleavage proceeds, but if the timing of ribosome biogenesis is modulated by hybridization of U8 to 32S as proposed by others (), then the exclusion of p68 and p72/p82 from the nucleolus could provide a means on how premature interaction of 5.8S and 28S rRNA sequences may be prevented. In , 18 putative RNA helicases have been implicated in ribosome biosynthesis (,), among them Dbp2p (), which all do not seem to have redundant functions. Dbp2p-depleted cells exhibit slow growth and cold-sensitive phenotypes but do not die (,) most probably because the necessary structural alterations in rRNA, catalyzed by Dbp2p, can proceed spontaneously at normal growth temperature, though less efficiently. In fact, yeast apparently does not possess a U8 snoRNA homolog, instead ITS2 elements seem to provide in some of the functions proposed above for vertebrate U8 snoRNA, thereby possibly interacting with 25S rRNA sequences less stably and allowing spontaneous formation of the 5.8S–25S hybrid structure as a prerequisite for further processing (). Nonetheless, the ribosome biogenesis (in contrast to the mRNA decay) phenotype of dbp2Δ cells is complemented by heterologous expression of human p68 (). In mammals, displacement of U8 and formation of the 5.8S–28S structure by a spontaneous rearrangement reaction may also be possible to a certain extent. This and some rest protein always left by the RNAi method may explain the residual 32S pre-rRNA processing observed under p68 subfamily knock-down conditions.

Ribonuclease P (RNase P) was originally described as an endoribonuclease that processes the 5′ leader sequence of precursor tRNA (). In bacteria, RNase P is a small ribonucleoprotein complex (), consisting of a catalytic RNA and a protein cofactor (,). The crystal structures of bacterial RNase P RNAs have recently been resolved, revealing the locations of the substrate-binding domains and the active sites in these RNA enzymes (). By contrast, when compared with their bacterial counterparts, nuclear forms of eukaryal RNase P are large ribonucleoprotein complexes (). Biochemical purification analyses have shown that a highly purified nuclear RNase P from HeLa cells has at least 10 distinct protein subunits associated with a single RNA species, H1 RNA (,). These protein subunits are termed Rpp14, Rpp20, Rpp21, Rpp25, Rpp29, Rpp30, Rpp38, Rpp40, hPop1 and hPop5 () (,). Since this purified HeLa RNase P has been enzymatically defined by virtue of its ability to cleave precursor tRNA (,), the potential existence of other forms of RNase P complexes with different subunit compositions is not excluded. Similarly, nuclear RNase P possesses nine protein subunits, designated Pop1p, Pop3p, Pop4p, Pop5p, Pop6p, Pop7p/Rpp2p, Pop8p, Rpp1p and Rpr2p (), most of which are homologous to protein subunits of human RNase P () (,). In addition, these protein subunits are shared with RNase MRP (), a mitochondrial and ribosomal RNA-processing ribonucleoprotein (). However, it is not known if these protein subunits are shared with the mitochondrial form of human RNase P, a ribonucleoprotein particle shown to have an RNA moiety that is identical to H1 RNA (). Rpp21, Rpp29, Rpp30, Rpp38 and hPop5 are highly conserved proteins that have homologs in Archaea () (). Rpp21, Rpp29 and H1 RNA are sufficient for reconstitution of RNase P activity in tRNA processing (). The archaeal Rpp21, Rpp29, Pop5, Rpp30 and Rpp38 proteins are required for efficient reconstitution of thermostable RNase P ribonucleoprotein (,). Nonetheless, recent findings reveal that protein subunits of RNase P are individually dispensable for enzyme activity (). In the case of the RNase P, it has been shown that pairs of its protein subunits are sufficient for reconstitution of enzyme activity (). Thus, archaeal Rpp21 with Rpp29 or Pop5 with Rpp30 are sufficient for RNase P RNA-based catalysis (). These pairs of archaeal proteins interact with each other in two-hybrid system (). The reconstitution studies described above underline the conserved role of archaeal and eukaryal RNase P RNAs in substrate recognition and cleavage. Recent progress in modeling the tertiary folding of eukaryal RNase P RNA uncovers that it has a conserved core structure similar to that of its bacterial counterpart (). Mutations that disrupt the predicted tertiary folding of H1 RNA abolish catalysis (). Remarkably, Kikovska et al., has shown that H1 RNA is active in tRNA processing in the absence of any protein (), a finding that is consistent with previous observation that H1 RNA alone binds to precursor tRNAs and has a conserved catalytic core (,). Since most of the protein subunits of nuclear RNase P are shared with RNase MRP, including those responsible for reconstitution of RNase P activity , it seems that the RNA subunits alone may be sufficient for diversification of function of these catalytic ribonucleoproteins as tRNA and rRNA endoribonucleases (,). Hence, the precise role of the protein subunits of RNase P and RNase MRP in hydrolyzing the phosphodiester bond in different RNA substrates remains unknown. Apparently, protein subunits may allow substrate recognition and catalysis by facilitating the proper folding of the RNA subunits of RNase P and RNase MRP. Protein subunits may also define some structural properties of RNase P and RNase MRP. For instance, Rpp20 and Rpp25 transiently associate with 12S monoparticles of RNase MRP, but these subunits dissociate from subsets of RNase MRP bound to 60–80S pre-ribosomal complexes (). Additionally, protein subunits may permit the recognition of yet unknown RNA substrates for RNase P and RNase MRP or implicate these complexes in other biological settings, such as chromatin binding and transcription (33; see below), cell cycle progression () and RNA metabolism (). Accordingly, studying the protein subunits of RNase P and RNase MRP could be useful in elucidating the diversification of functions of these two evolution-related ribonucleoprotein complexes (). #text A recent study has shown that human nuclear RNase P is required for transcription of tRNA and other small noncoding RNA genes by pol III in whole HeLa cells and cell extracts (). RNase P exerts its role on transcription through association with chromatin of tRNA and 5S rRNA genes as determined by chromatin immunoprecipitation analysis (). All the protein subunits of RNase P tested so far, i.e. Rpp14, Rpp20, Rpp21, Rpp29, Rpp30, Rpp38 and Rpp40, can be found associated with chromatin of tRNA and 5S rRNA genes in rapidly dividing cells () (33; also unpublished data), an indication that a multi-protein RNase P complex binds to the chromatin of these genes. Rpp25 has also been noted to be associated with nucleosomes (). Binding of these protein subunits to chromatin is dynamic, in the sense that they associate with tRNA and 5S rRNA genes in dividing HeLa cells and dissociate from these genes when cells cease proliferating. Furthermore, chromatin occupancy by RNase P associates with active gene transcription in extracts and in cells, and coincides with that of pol III, which could be brought down with active RNase P ribonucleoprotein in coimmunoprecipitation experiments (). Knockdown of the essential protein Rpp29 of human RNase P by RNA interference did not affect the binding of RPB8 (a core component of pol III) to tRNA and 5S rRNA genes, even though this knockdown results in severe inhibition of RNase P activity and pol III transcription (). Additionally, Rpp29 disengaged from target genes independently of RPB8 in cells that ceased proliferating (). Hence, the possibility exits that recruitment of pol III to tRNA and 5S rRNA genes is independent of that of RNase P (or at least Rpp29). The association of RNase P with chromatin of tRNA genes could be explained in terms of coordination of tRNA gene expression at transcription and processing steps. However, RNase P is also critical for transcription of 5S rRNA, 7SL RNA and U6 snRNA genes (), whose precursor transcripts are not recognized as substrates for RNase P. Hence, RNase P acts solely as a transcription factor for pol III in transcribing these latter small RNA genes. Screening of a genomic library identified RPR1 RNA, the RNA subunit of RNase P, as the specific overexpression suppressor of very slow growth at 37°C due to a small deletion of Bdp1, a subunit of the transcription factor TFIIIB complex (). Processing of the 5′ leader sequence of precursor tRNA is defective in cells producing this mutant Bdp1p, called Bdp1Δ253–269. Transcription of the gene is selectively diminished when recombinant Bdp1Δ253–269 replaced wild-type Bdp1p in an pol III transcription system. The physical interaction of RNase P with Bdp1p was demonstrated by coimmunoprecipitation and pull-down assays, implying a role for TFIIIB in 5′ end processing of precursor tRNA (). In , early processing of precursor tRNAs occurs in the nucleolus, which is enriched with RNase P RNA (). Additionally, tRNA gene families, which are dispersed in various chromosomes, colocalize with 5S rDNA genes at the nucleolus (,). This nucleolar clustering of yeast tRNA genes depends on transcription-complex formation and the existence of proficient promoters. Clustering of tRNA and 5S rRNA genes in the nucleolus forms pol III transcription sites containing concentrations of pol III and its general transcription factors, TFIIIB, TFIIIC and TFIIIA. Such nucleolar sites may also initiate nucleolar organization of the tRNA-processing pathway that includes 5′ end processing by RNase P (,). It should be noted that in contrast to the localization of pol III and RNase P in the nucleolus in , confocal immunofluorescence microscopy analyses of HeLa cells reveal that specific protein subunits of human pol III and RNase P primarily colocalize in the nucleoplasm, even though fluorescent signals are also visible in defined spots in large nucleoli (R.R. and J.N., unpublished data). #text Archaeal and eukaryal RNase P are ensembles of structurally and functionally related RNAs with highly conserved protein subunits (8–10,14,17,20,50). Reconstitution experiments of archaeal and eukaryal RNase P activities reveal that two protein subunits, e.g. Rpp21 and Rpp29, with their corresponding RNA moieties are sufficient for having endonucleolytic cleavage of tRNA substrates neutral pH7 and 5-30 mM divalent ion (,). Notably, Rpp29 but not Rpp21 can activate a bacterial RNase P RNA (,). In addition, Pop5 has an α–β sandwich structure that bears structural similarity to the bacterial RNase P protein (). Since archaeal and eucaryal RNase P RNAs are active under unphysiological conditions of high ion concentrations (,), it would be interesting to check if a single protein cofactor (Rpp29, Pop5 or other) can reconstitute some activity under physiological reaction conditions. Previous studies which revealed ambiguous properties related to RNase P or its subunits that were not consistent with its activity as a tRNA-processing enzyme could be reexamined from the perspective of its transcriptional activity. For instance, Rpp20 that exhibits ATPase activity () and chromatin-binding capability () may enable RNase P (and/or RNase MRP) to use ATP as a cofactor for binding and modulating chromatin structure and function. Rpp20, as well as Rpp25, belong to the Alba superfamily of proteins which seem to have originated as RNA-binding proteins, attaching to a variety of ribonucleoprotein complexes, including RNase P and RNase MRP (,) and then being recruited as chromatin-binding proteins (). A dual role for Rpm2p, a component of yeast mitochondrial RNase P, in tRNA processing and transcription has also been described (). The molecular mechanisms by which human RNase P binds to chromatin of noncoding RNA genes and controls transcription by pol III are not known. Future work will reveal if RNase P acts at initiation, elongation and/or termination of transcription. In addition, the physical and functional links between pol III and RNase P during the cell cycle remains to be studied. Future studies will also unveil the significance of the effect of binding of RNase P (or its subunits) to hundreds of small noncoding RNA genes and others (R.R. and N.J., unpublished data) on chromatin structure, organization and function of the human genome. Whatever the outcome of these studies, the discovery that human RNase P is a chromatin-binding complex which is critical for normal gene transcription () expands the definition of this entity as an enzyme that hydrolyzes a phosphodiester bond in precursor tRNA.

The enzyme topoisomerase II is responsible for resolving catenanes and supercoils in chromosomal DNA that are generated during DNA metabolic processes. It plays an essential role in condensation and segregation of chromosomes at mitosis (). Topoisomerase II is of considerable interest to human medicine, because it is an important target for cancer therapy (). It is also suspected that topoisomerase II can be converted into a potent DNA toxin by minor (and as such repairable) DNA lesions frequently induced by environmental factors (e.g. UV radiation, oxidative stress) (,). Moreover, certain nutritional constituents (e.g. flavonoids) are known to disturb the enzyme's normal catalytic cycle (,), which is thought to contribute to translocations within the MLL-locus that trigger infant leukemia (). These adverse properties of topoisomerase II could be the ultimate reason why vertebrates maintain two genetically distinct isoforms (denoted α and β) (), while lower eukaryotes have only one. The divergence of vertebrate topoisomerase II into α and β isoforms remains enigmatic, because the two enzymes are very similar in structure and function. They share a high degree of overall sequence homology with 68% identity and 86% similarity (,). So far, the only major difference between the two isoforms is a preferential relaxation of positive supercoils by the IIα isoform (), whereas other basic catalytic aspects are very similar (). Moreover, they have the same capacity for complementing essential topoisomerase II functions in temperature-sensitive Δtop2 yeast mutants (,). Despite these similarities, the two isozymes apparently play different biological roles in vertebrate cells (). Human cell lines lacking the α isoform encounter serious problems at mitosis because chromosome segregation is deficient (,). For similar reasons, mouse embryos lacking the TOP2α gene, fail to develop beyond the 4–8-cell stage (). In contrast, mammalian cell lines lacking topoisomerase IIβ pass normally through mitosis, and their most prominent phenotype is a decreased sensitivity towards topoisomerase II poisons (). These findings indicate that all essential topoisomerase II functions in cell-cycle-related events, such as DNA replication and sister chromatid segregation, can be performed by the α isozyme, while the β isozyme does not play an essential role in proliferating cells. And yet, TOP2β −/− mice are not viable. They suffocate shortly after birth due to developmental defects of motor and sensory neurons () and the brain (). These defects most likely reflect a requirement of topoisomerase IIβ activity in regulating the expression of genes important at later stages of neuronal differentiation (). This view has convincingly been confirmed by the recent finding that the β isoform plays an important role in the regulation of gene transcription, in as much as it introduces double-strand breaks at promoter regions of several genes, which are required for the proper signal-dependent activation of these genes (). In this respect, it is of interest that topoisomerase IIβ is constitutively expressed in all cells of the mammalian organism (), probably because expression is driven by a promoter with features characteristic of housekeeping genes (), whereas expression of topoisomerase IIα is repressed as soon as cells stop proliferating (,). Therefore, the β isoenzyme is the only type II topoisomerase available in quiescent cells. In synopsis, the data available clearly suggest that the two isoforms have different biological functions in vertebrate organisms. Here, we address the question of precisely which features render topoisomerase IIα essential for cell proliferation, and conversely, lack of which features prevents topoisomerase IIβ from adopting these functions. We have approached this problem by identifying those parts of topoisomerase IIα and IIβ that are responsible for isoform-specific functioning inside the living mammalian cell. We started from the assumption that those portions of the enzymes that are most divergent between α and β forms would be most likely to mediate isoform-specific functions. Sequence comparisons, limited proteolysis experiments and crystallographic studies suggest that topoisomerase II is composed of three major structural and functional domains (,,). With the exception of brief stretches at the N-terminal ends (first 27 or 43 amino acids of α and β, respectively), the ATPase and the central breakage/reunion domains are similar between the two isoforms, whereas the C-terminal regions differ both in size and sequence (). Thus, divergent and homologous portions of human topoisomerase IIα and IIβ were combined in a varied manner to form α/β-chimeric enzymes that were tested for unique, isoform-specific functions in human cells, most notably for localization of the enzymes at mitosis (,) and for complementation of a conditional topoisomerase IIα knockout in human cells (). We aligned amino acid sequences of topoisomerase IIα and IIβ using the default settings for ClustalW v.1.83 (WWW Service at the European Bioinformatics Institute, ) (). Exchange of defined regions between cDNAs of human topoisomerase IIα and IIβ at the positions indicated in B was accomplished by overlap-extension PCR (). Fragments encompassing the divergent C-terminal regions alone were generated by PCR. Chimeric and truncated cDNAs were inserted into a bicistronic expression vector () used previously for stable expression of biofluorescent human topoisomerase IIα and IIβ in human cells (). Here, the vector was modified to provide C-terminal fusion with enhanced yellow fluorescent protein (YFP). To facilitate simultaneous visualization of topoisomerase IIα and IIβ, a tricistronic expression plasmid was generated, in which topoisomerase IIα fused to CFP was placed in the first, and topoisomerase IIβ fused to YFP in the second cistron (). A vector expressing YFP alone served as a control. All new constructs were checked by DNA sequencing. Human embryonal kidney 293 cells (# DSMZ ACC 305, German Collection of Microorganisms and Cell Culture, Braunschweig, Germany) were transfected with these constructs. Stable transgenic cell clones were selected and maintained in medium containing 0.4 µg ml puromycin (details see:48). Epifluorescence microscopy was done with a Zeiss Axiovert 100 inverted light microscope equipped with an on-stage heating chamber (ΔTC3 from Bioptechs, Butler, PA, USA), a heated 63×/1.4 NA oil immersion objective system, a mercury lamp and appropriate filter sets. Confocal imaging was done with a Zeiss LSM 510 META inverted confocal laser-scanning microscope equipped with a 63×/1.4 NA oil immersion objective. To maintain a constant temperature of 37°C for live cell imaging, the confocal microscope was built in a ZEISS Incubator XL. Cells were cultured under the microscope in CO-independent medium (Invitrogen, Karlsruhe, Germany). To analyze complementation of topoisomerase IIα function, we used human HT-1080 cells, in which both alleles of the TOP2A gene are disrupted. The cells are rescued by transgenic expression of human topoisomerase IIα from a tetracycline repressible construct stably integrated into the genome (). Upon transfection of these cells (designated HTETOP) with the various chimeric constructs, complementation of topoisomerase IIα function was determined by comparing the number of stable cell clones obtained by selection with tetracycline (1 µg ml) versus puromycin (0.4 µg ml) (details see:26). For fluorescence activated cell sorting, cells were grown to ∼80% confluence, washed in PBS, trypsinized and resuspended in ice cold PBS at 10 cells ml. For each measurement, 20 000 cells were analyzed in a FACScan flow cytometer (BD bioscience) at a high flow rate setting with an Argon ion laser tuned to 488 nm. Isolation of cell nuclei and extraction of nuclear proteins (400 mM NaCl) followed published procedures (). Whole cell lysate was prepared from 3 × 10 cells suspended in D-PBS (Invitrogen, Karlsruhe, Germany) by addition of an equal volume of 2-fold lysis buffer (62.5 mM Tris-HCl, pH 6.8, 10% glycerol, 4% SDS, 20 mM DTT, 500 mM urea, 5 mM AEBSF, 0.04% bromophenol blue) followed by ultrasound treatment (15 s, 14 W, 20 kHz, tip diameter 2 mm). For assessment of activity by immunoband depletion, cells were cultured with 100 µM VM26 (teniposide, Bristol-Myers Squibb, Munich, Germany) for 30 min prior to harvesting. Cell lysate (5 × 10 cells/lane) was subjected to SDS PAGE and transferred to PVDF membranes (Immobilon P, Millipore, Bedford, Maryland, USA). Blots were probed with the following antibodies: (i) mouse monoclonal antibodies against GFP (clone JL8, Clontech, Heidelberg, Germany), which cross react with YFP and subsequently are referred to as ‘YFP antibodies’; (ii) rabbit peptide antibodies raised against a peptide of amino acid residues 1514–1531 of human topoisomerase IIα (CIC, Genosys Cambridge, England); (iii) rabbit polyclonal antibodies raised against a peptide of amino acid residues 1586–1621 of human topoisomerase IIβ (designated 670) (); (iv) various other antibodies against C-terminal epitopes of human topoisomerase IIβ serving as a control for results obtained with 670. These include rabbit polyclonal antibodies against amino acid residues 1341–1621 (designated H-286, Santa Cruz, Heidelberg, Germany) and 1611–1621 (designated 779) (), and mouse monoclonal antibodies against amino acid residues 1583–1601 (clone 3H10) (). Magnetic beads (Dynabeads M-280, Dynal/Invitrogen, Oslo, Norway) coupled to sheep anti-mouse IgG were loaded with YFP antibodies (Anti-GFP, mixture of two mouse monoclonal antibodies, Roche, Basel, Switzerland) according to the manufacturer's instructions (60 µg of antibodies per 10 beads). Loaded beads (4 × 10) were incubated (2 h at 4°C) with nuclear extract (200 µg total protein) in a final volume of 400 µl binding buffer (5.5 mM NaHPO, 1.2 mM NaHPO, pH 7.4, 265 mM NaCl, 5% FCS, 13.75% glycerol, 2.25 mM EDTA, 0.35 mM DTT, 10 μg ml aprotinin, 1 mM AEBSF). Subsequently, beads were washed once with three volumes of binding buffer (20 min, 4°C), followed by four washes (10 min, 4°C) with three volumes of washing buffer (5.5 mM NaHPO, 1.2 mM NaHPO, pH 7.4, 890 mM NaCl, 13.75% Glycerol, 2.25 mM EDTA, 0.35 mM DTT, 10 μg ml aprotinin, 1 mM AEBSF). These washing steps were found crucial for disrupting non-covalent interactions of YFP-fused topoisomerase II with endogenous topoisomerase species. Immunoprecipitates were finally eluted from the beads by boiling for 10 min in sample buffer (31.25 mM Tris-HCl, pH 6.8, 5% glycerol, 3% SDS, 2 mM DTT, 2 mM EDTA, 10 μg ml aprotinin, 1 mM AEBSF). Eluates were subjected to SDS PAGE and silver staining (4 × 10 beads per lane) or western blotting (2 × 10 beads per lane). Alternatively, immunoprecipitates were washed twice with decatenation buffer (50 mM Tris-HCl, pH 7.6, 100 mM KCl, 10 mM MgCl, 1 mM ATP, 0.5 mM DTT, 0.5 mM EDTA, 30 μg ml BSA) and incubated (2 h, 37°C) with 300 ng catenated kinetoplast DNA from (kDNA, TopoGen Inc., Columbus, USA), with or without 1 mM ICRF-187 (Zinecard, Pharmacia & Upjohn, Kalamazoo, MI, USA), in a final volume of 27 µl decatenation buffer. The reaction was stopped by adding 1% SDS and 0.1 mg ml proteinase K, and DNA reaction products were analyzed by agarose gel electrophoresis. Topoisomerase IIα and IIβ are composed of three functional domains, which are bordered by protease-sensitive sites. Limited proteolysis experiments revealed that the C-terminal domains begin at residues 1263 (IIα) and 1296 (IIβ), respectively (,). However, sequence alignment shows that the region of high diversity between the two isoforms extends beyond this domain border and reaches up to amino acid positions 1171–1179 (IIα) or 1185–1191 (IIβ) (,). Interestingly, type II enzymes from chlorella viruses lack this divergent C-terminal region (), and truncation of human topoisomerase IIα at amino acid 1175 produced a catalytically active variant (). An extended truncation at position 1121, however, destroyed enzyme activity () probably due to deletion of the primary dimerization region (). In summary, these findings suggested to us residues 1173–1531 and 1186–1621 of human topoisomerase IIα and IIβ, respectively, as candidate regions for the isoform-specific regulation. These regions encompass a maximum of heterogeneity (bearing only ∼32% identical amino acid residues) and can be exchanged between the two isoforms without interfering with the basic enzymatic functions. In fact, it has been reported that exchanging of the C-terminal regions of murine topoisomerase II isoforms at positions corresponding to amino acids 1173 and 1186 of human topoisomerase IIα and IIβ, respectively, gives rise to chimeric enzymes that are fully active upon heterologous expression in yeast (). Therefore, we chose to exchange the same C-terminal regions of the human enzymes. Because these regions extend beyond the C-terminal domains defined by limited proteolysis (,), they are herein referred to as C-terminal regions (CTRs). We also selected short divergent stretches at the N-terminal ends of human topoisomerase IIα and IIβ (first 27 or 43 amino acids, respectively) to be exchanged between isoforms. These regions are only ∼14% identical, whereas the rest of the N-terminal domains and the core domains (amino acids 28–1172 and 44–1185 of IIα and IIβ, respectively) are very similar with ∼81% identical amino acid residues. A synopsis of sequence homologies and sites chosen for exchanging regions between topoisomerase IIα and IIβ is shown in A. When transfected into HEK 293 cells, each of the constructs depicted in B gave rise to viable cell lines supporting stable expression of the YFP-fused proteins. Cell clones with intermediate expression levels were selected and expanded for further analysis. All clones exhibited growth rates and gross morphologies similar to cells not transfected or expressing YFP alone. To assess the integrity of the fusion proteins and to compare their relative expression levels, we subjected the cells to western blotting and probed the blots with YFP antibodies. YFP-fused full-length, non-chimeric topoisomerase IIα and IIβ and the various topoisomerase IIα/β chimeras were detected as single protein bands of expected size. There were no smaller bands detected in addition by YFP antibodies (A, top, odd numbered lanes). Thus, we could exclude rearrangements of the transgenes and safely assume that yellow fluorescence of the cells was entirely due to the desired YFP-fused protein. All constructs supported similar expression levels allowing a comparison of data between these cells. To compare YFP-fused and endogenous enzymes, blots were probed with isoform-specific antibodies against C-terminal epitopes of topoisomerase IIα or IIβ (A, middle and bottom, respectively). The YFP-fused proteins could clearly be discriminated from the corresponding endogenous enzymes as additional bands of slower migration. From comparison of lanes it became evident that endogenous levels of topoisomerase IIα and IIβ were similar in all transfected cell clones (A, middle and bottom, odd numbered lanes) and similar to those in untransfected cells (not shown), indicating that none of the YFP-fused enzymes interfered with endogenous topoisomerase II expression. It should also be noted that the desired exchanges of CTRs outlined in B were confirmed by the presence of unique C-terminal epitopes of topoisomerase IIα and IIβ in the products of the various constructs shown in middle and bottom panel of A. To compare expression levels of endogenous and YFP-fused proteins within each clone, we intended to compare signal intensity within the lanes of western blots stained with isoform-specific topoisomerase II antibodies (A, middle and bottom, odd numbered lanes). However, upon testing several antibodies directed at various unique epitopes of human topoisomerase IIβ, we found that all antibodies tested did not exhibit the same preference for the endogenous form and heterologously expressed YFP-fused variants of the enzyme. On the contrary, the antibodies preferentially recognized one or the other form of the enzyme (B). Thus, conclusions about expression levels drawn from such analyses are in our eyes unreliable. However, it should be noted that all cell clones expressing YFP-fused enzyme constructs exhibited growth rates and morphologies indistinguishable from untransfected cells, suggesting that the enzymes were at least expressed at physiologically tolerable levels. Similar analyses were carried out on cell clones expressing YFP fusion proteins of the CTRs of topoisomerase IIα and IIβ alone (C). These constructs also gave rise to single protein bands. However, the apparent molecular weight (∼100 kDa in both cases) was larger than expected from the amino acid sequence (70 kDa for α CTR-YFP and 80 kDa for β CTR-YFP). This could be due to phosphorylation, since these regions harbor the majority of phosphorylation sites (). To determine whether the exogenous enzymes were as active in the cells as the endogenous enzymes, we employed immunoband depletion. The assay is based on the stabilization of covalent complexes of topoisomerase II and genomic DNA by specific poisons (). As a consequence, signals specific for active topoisomerase II molecules are depleted from immunoblots due to retention of topoisomerase II•DNA complexes in the gel slots. As demonstrated in A, treatment with the topoisomerase II poison VM26 depleted YFP-fused non-chimeric full-length enzymes and topoisomerase IIα/β chimeras from immunoblots to a similar extent (A, top, compare lanes 1–4 with lanes 5–12). The extent of depletion was also comparable to that of endogenous topoisomerase IIα (A, middle) and IIβ (A, bottom). It should be noted that in this analysis antibody-derived biases (compare B) did not play a role since they apply in the same manner to the bands to be compared in quantitative terms. These results confirm that all the topoisomerase IIα/β chimeras were active in the cells. Moreover, it could be deduced that activity levels were comparable to those of non-chimeric enzymes (exogenous or endogenous) because bands were depleted to similar extent. Corresponding analyses of cells expressing the YFP-tagged α CTR or β CTR confirmed that these proteins were catalytically inactive, as predicted (not shown). We have previously shown by immunohistochemistry () as well as localization of biofluorescent topoisomerase IIα and IIβ () that the most obvious difference between the isoforms is their association with metaphase chromosomes. This observation is demonstrated in the first two rows of B. YFP-fused topoisomerase IIα (row 1) and IIβ (row 2) exhibit a similar distribution in the interphase nucleus (left). However, at metaphase (right) the α isoenzyme accumulates on the condensed chromosomes, whereas the β isoenzyme diffuses mostly into the cytosol and shows only a marginal association with chromosomes. The phenomenon is even more clearly seen in A showing cells co-expressing topoisomerase IIα and IIβ fused to CFP and YFP, respectively. At interphase (left panel), the two isoenzymes colocalize, whereas at mitosis (right panel) topoisomerase IIα (shown in red) is chromosome bound, and topoisomerase IIβ (shown in green) predominantly resides in the cytosol. Thus, binding to metaphase chromosomes can be used as experimental readout of isoform-specific functioning of topoisomerase IIα and IIβ in proliferating mammalian cells, and we have used it to characterize the various topoisomerase IIα/β chimeras (B, rows 3–6) and the two CTRs alone (B, rows 7 and 8). It becomes readily apparent, that the ability to bind to metaphase chromosomes is fully retained in all enzyme varieties bearing the α CTR (B, rows 1, 3 and 6) and to some extent also in this enzyme portion alone (B, row 7). Moreover, enrichment in mitotic chromosomes is lost from topoisomerase IIα, when its CTR is replaced with that of the β isoform (B, row 5). Conversely, this property is fully gained by topoisomerase IIβ, when its CTR is replaced with that of the α isoform (B, row 6). Similar effects are not apparent upon exchanging the non-conserved N-terminal stretches of the isoenzymes (B, compare rows 3 and 4). It should be noted that the β CTR alone had a diffuse distribution in the cell at metaphase, but it was not entirely excluded from chromosomes, as was YFP alone (B, row 9). In summary, these observations suggest that (i) the non-conserved CTR of topoisomerase IIα promotes binding of the enzyme to metaphase chromosomes, (ii) the corresponding CTR of topoisomerase IIβ is much less capable of performing this function, and (iii) the non-conserved N-terminal stretches of the two isozymes do not play a role in targeting the enzymes to metaphase chromosomes. Given that topoisomerase II functions as a dimer, unambiguous deductions from the data in can only be made, when dimerization of topoisomerase IIα/β chimeras with endogenous enzyme varieties can be excluded, because this would render the behavior of the YFP-linked moiety attributable to an unpredictable mixture of homo- and heterodimers. In fact, we have previously assumed dimerization between GFP-fused, wild-type topoisomerase II and corresponding endogenous enzyme molecules based on GFP-directed immunoprecipitation (IP) protocols (). We therefore wanted to clearly define the dimerization state of the topoisomerase IIα/β chimeras by YFP-directed IP followed by SDS PAGE, protein staining and immunoblotting. When we applied previously described low-salt IP conditions () to extracts from cell lines investigated here, the results were highly inconsistent, and non-reproducible; often both endogenous topo II isoforms were detected in the IPs by immunoblotting (data not shown). To address these problems we applied harsher buffer conditions with higher salt concentrations. Under these conditions, no prominent protein bands, other than the expected YFP-chimeras, were detected in silver-stained gels of any of the IPs (A, top), at least not in the size range where endogenous topoisomerase IIα or IIβ would migrate (i.e. 170–180 kDa). Similarly, endogenous topoisomerase IIα or IIβ were undetectable in immunoblots of the IPs (A, middle-bottom and bottom), the only protein species detected being the ones directly targeted by the precipitating YFP antibody. It is unlikely that this difference to our previous results was brought about by experimental dissociation of topoisomerase II dimers because they are known to be extremely salt-stable (). We find it more likely that the presence of endogenous topoisomerase II isoforms in IPs of GFP-tagged topoisomerase II, as reported previously by us (), was due to the multimerization of topoisomerase II homodimers (), which is known to decrease with increasing ionic strength (). To further demonstrate that topoisomerase II dimers were not disrupted during high salt IP, we tested whether the enzymes retained their catalytic activity (indicative of functional dimeric enzymes) in the final IPs. As demonstrated in B, IPs of all the YFP-tagged proteins exhibited strong DNA decatenation activity, thus attesting to the integrity of the enzyme dimers throughout the IP procedure. Thus, the most plausible conclusion is that all the exogenous topoisomerase II species heterologously expressed in HEK 293 cells undergo homodimer formation. Because these observations were based on IPs of YFP-tagged proteins, they do not rule out the formation of endogenous topoisomerase II heterodimers (). They do, however, allow for an unambiguous interpretation of the data obtained here with the various topoisomerase IIα/β chimeras. Thus, localization of these proteins (B) clearly indicates a decisive role of the non-conserved α CTR in targeting the core portion of topoisomerase II to metaphase chromosomes. Our observation that targeting to metaphase chromosomes is promoted by the α CTR () suggests that this region may also enable topoisomerase IIα to perform its essential functions in proliferating cells. We tested this hypothesis by complementation studies making use of human HT-1080 cells, in which both alleles of the TOP2α gene are disrupted and cell proliferation is supported by expression of topoisomerase IIα from a tetracycline repressible vector stably integrated into the genome (). These HTETOP cells, which die when topoisomerase IIα is depleted by the addition of tetracycline, were stably transfected with the YFP-fused topoisomerase II constructs investigated here (B) and the frequency with which colonies could form in the presence of tetracycline was measured. The results are summarized in . In the absence of any transfected constructs, no viable cell clones emerged in the presence of tetracycline (row 7), but the YFP-fused version of topoisomerase IIα was able to support proliferation of the cells in the presence of tetracycline, as expected (row 1). The YFP-fused version of topoisomerase IIβ, however, consistently gave rise to a much lower frequency of clones (row 2). Most interestingly, topoisomerase IIα lost most of its ability to support cell proliferation upon replacement of its CTR with that of the β isoform (row 5), whereas topoisomerase IIβ furnished with the α CTR gained this ability (row 6). A similar effect was not observed upon exchanging the non-conserved N-terminal stretches of the isozymes (rows 3 and 4). In summary, all versions of human topoisomerase II bearing the α CTR were capable of supporting cell proliferation (rows 1, 3 and 6), whereas all those furnished with the β CTR (rows 2, 4 and 5) were highly inefficient in this respect. It is, however, important to note that the latter constructs were not completely incapable of supporting cell proliferation. The average complementation index for all constructs bearing the α CTR was 1.21 (derived from , rows 1, 3 and 6). In contrast, the average complementation index for all constructs bearing the β CTR was 0.13 (derived from , rows 2, 4 and 5). Because no viable cell clones emerged from mock transfection (, row 7), these observations suggest that topoisomerase IIβ (or enzyme chimeras bearing the β CTR) can also support cell proliferation to some extent. Two explanations for this result can be considered: (i) topoisomerase IIβ could be able to substitute for the α isoform, when present at much higher levels; (ii) binding sites normally occupied by topoisomerase IIα (e.g. sites on mitotic chromosomes) could become freely accessible to the β isoform when the α isoform is absent. These sites could then be occupied by topoisomerase IIβ irrespective of its concentration in the cell. To investigate these hypotheses, we compared expression levels and localization of YFP-fused topoisomerase IIα or IIβ in clones supporting cell growth in the presence of tetracycline (). From comparison of representative cell clones subjected to western blotting and probing with YFP antibodies (A, top) it became readily apparent that complementation by YFP-fused topoisomerase IIβ (lanes 3, 4 and 5) requires much higher expression levels than complementation by YFP-fused topoisomerase IIα (lane 2). Flow cytometry confirmed this finding, showing that YFP-fluorescence was ∼10-fold brighter in various cell clones complemented by YFP-fused topoisomerase IIβ than in a reference cell clone complemented by YFP-fused topoisomerase IIα (B). Notwithstanding our hesitations about quantitative comparisons between YFP-fused and endogenous topoisomerase II species based on topoisomerase II-directed immunoblotting (see B), it should be noted that the cellular complement of YFP-fused topoisomerase IIα enabling full cell proliferation was hardly detectable by topoisomerase-directed immunoblotting (A, middle, compare lane 1 with lane 2), whereas YFP-fused topoisomerase IIβ complementing topoisomerase IIα function gave a much more intense signal than corresponding bands of endogenous topoisomerase IIβ (A, compare bands within lanes 3, 4 and 5). Thus, in our model system, the β isoform seems only able to substitute for topoisomerase IIα when it is highly (at least 10-fold) overexpressed. C further suggests that, upon repression of topoisomerase IIα expression, binding sites on mitotic chromosomes normally occupied by this enzyme do not become freely accessible to the β isoform and are not readily occupied by it. If that were the case, topoisomerase IIβ should accumulate on metaphase chromosomes upon repression of topoisomerase IIα. However, this was clearly not the case in any of the cell clones complemented by YFP-fused topoisomerase IIβ (C, rows 2–4). In all these cell clones, YFP-fused topoisomerase IIβ was mostly localized in the cytoplasm during mitosis. We failed to detect an accumulation of the enzyme on metaphase chromosomes, as seen in the same cell model with YFP-fused topoisomerase IIα (C, row 1). It should also be noted that localization of YFP-fused topoisomerase IIβ during interphase (C, left) and metaphase (C, right) was identical to that observed in HEK 293 cells where endogenous expression of topoisomerase IIβ was not silenced (A and B). In summary, these morphological data suggest that the binding equilibrium of topoisomerase IIβ at metaphase chromosomes is (i) independent of the absence or presence of topoisomerase IIα, and (ii) not significantly influenced by the total cellular level of topoisomerase IIα and IIβ. Thus, the weak but notable ability of the β isoform to complement the function of the α isoform cannot be due to an increase in its propensity to interact with mitotic chromosomes upon removal of topoisomerase IIα. It is well established that the C-terminal domains of both yeast and human topoisomerase II are dispensable for the enzyme's basic catalytic activity (,). On the other hand, a large body of evidence suggests that the C-terminal domain plays a role in regulating the cellular functioning of topoisomerase II. Most notably, it contains crucial nuclear localization signals (,,) and sites phosphorylated in a cell-cycle-related manner (). Since the C-terminal domains are the most divergent portions of the two mammalian isoforms of topoisomerase II (), it has been proposed that they determine specific functions differing between these two isoforms (). Here, we provide direct evidence in support of this view. We demonstrate that the divergent CTRs of topoisomerase IIα and IIβ govern two features in which the two isoforms characteristically differ, namely binding to mitotic chromosomes and support of cell proliferation. We show that YFP-fused topoisomerase IIα is preferentially chromosome-bound during mitosis and fully supports proliferation of cells lacking endogenous topoisomerase IIα. In contrast, the majority of YFP-fused topoisomerase IIβ is not chromosome-bound at mitosis, and clones emerged from complementation experiments at greatly reduced frequencies. The specific features of the α isoform were stringently linked to the presence of the α CTR. Replacement of the CTR in topoisomerase IIβ with the α CTR produced an enzyme chimera that behaved like topoisomerase IIα, whereas the converse experiment produced an enzyme chimera behaving like topoisomerase IIβ. The requirement of topoisomerase IIα for proper cell division has been suggested by indirect evidence showing an essential role in chromosome segregation, which is not readily adopted by the β-isoform (e.g. ). More recently, depletion of topoisomerase IIα by various experimental strategies resulted in each case in an impaired separation of chromosomes in anaphase. (,,). Here we describe a striking coincidence between the ability of all versions of topoisomerase II furnished with the α CTR to complement such a lack of endogenous topoisomerase IIα and the propensity of the complementing construct to bind to mitotic chromosomes. We even observe that the α CTR alone preferentially binds to metaphase chromosomes. It remains unclear whether chromosome binding is due to direct DNA-interactions, as suggested for various prokaryotic type II topoisomerases (,), or to interactions with other proteins, e.g. condensins () or HSP90 (). However, the strict correlation between the binding to metaphase chromosomes and the support of cell proliferation suggests a mechanistic connection between the two. It can be hypothesized (i) that efficient separation of sister chromatids and proper cell division depend on a high, local concentration of active topoisomerase II at the mitotic chromosome, (ii) that under physiological conditions only the α isoform accumulates in sufficient concentrations at the mitotic chromosome, and (iii) that this feature is promoted by an intrinsic ability of the α CTR to bind to metaphase chromosomes. These hypotheses would assign to the α CTR the function of an adaptor that shifts the binding equilibrium of the entire enzyme molecule towards the bound state and thus provides the chromosome at mitosis with sufficient topoisomerase II activity to perform extensive DNA-decatenation in the course of sister chromatid segregation. Our finding that topoisomerase IIβ can also support cell proliferation when expressed at extremely high levels supports such a hypothesis: sufficient local concentration of active topoisomerase II at the mitotic chromosome cannot only be acquired by expressing normal levels of an enzyme having a high affinity (due to the α CTR), but also by expressing highly increased levels of an enzyme having a much lower one (due to the β CTR). The above interpretation is more difficult to fit with complementation studies carried out in yeast that show that both mammalian isoforms are equally capable of rescuing temperature-sensitive Δtop2 yeast mutants (,). One explanation could be that yeast might be unable to discriminate between the two mammalian isoforms. However, this explanation is unlikely because mouse topoisomerase IIα and IIβ can be discriminated by yeast, in as much as they are distributed in a distinguishable manner in yeast cell nuclei (). Another explanation could be that yeast is more tolerant to changes in topoisomerase II expression levels than human cells, which are readily killed by overexpression of these enzymes (). Since high copy number vectors were used in the yeast studies for expression of the complementing enzymes, expression levels of the β isoform could have been high enough to enable efficient complementation of topoisomerase II functions in the same manner as seen here in a human cell line. Another possible interpretation of our data is that the topoisomerase IIα plays an essential role during replication. It has been shown in yeast that topoisomerase II is required for DNA replication, when topoisomerase I is lacking (), because movement of DNA replication complexes through the DNA double helix induces positive supercoils ahead of this machinery. Recent work in yeast demonstrates that topoisomerase II relaxes chromatin even more efficiently than topoisomerase I (). In mammals, topoisomerase IIα, but not IIβ, appears to be a key player in removal of this type of torsional stress during replication (), and it was postulated that this isoform-specificity is determined by the divergent C-terminal regions (). The residues that were suggested to play this role in replication are all within the α CTRs analyzed here. In addition, a study in chicken fibroblast showed that topoisomerase IIα, but not IIβ, co-localizes with sites of replication, and this targeting was also mediated by the α CTR (). Unfortunately, it is difficult to determine whether topoisomerase IIα plays a truly essential role during replication that cannot be complemented by other proteins (e.g. topoisomerase I or topoisomerase IIβ). Although silencing of topoisomerase IIα in human cells () and mice () causes a defect in chromosome segregation, suggesting that its essential role is during mitosis rather than S-phase, this phenotype could conceivably be caused by loss of an essential topoisomerase IIα function during the late phases of replication. Lack of such a function could still allow for a progression into metaphase followed by mitotic catastrophes due to unresolved DNA catenanes. Thus, an essential role of topoisomerase IIα in relaxation of positive supercoils generated at late stages of replication cannot be excluded, and our observation that the α CTR is required for efficient support of cell proliferation by topoisomerase II may just as well reflect a specific involvement of the α isoform in DNA replication (). Regardless of the exact role of the α CTR, we demonstrate in this article that it confers a unique, proliferation-associated functionality to the topoisomerase II core enzyme (either version of it), whereas the β CTR is much less efficient in this respect. It is therefore plausible that the two versions of the CTR cooperate in differential targeting of topoisomerase IIα and IIβ, thus providing unique functionality of the two isoforms in proliferating cells.

‘RNA repair’ is a versatile mechanism to rectify programmed breaks in tRNAs and mRNAs incurred during tRNA processing and under conditions of cellular stress. Examples include virus-mediated tRNA repair to thwart a host antiviral response to bacteriophage infection (), splicing of intron-containing tRNAs () and unconventional mRNA splicing during the unfolded protein response (,). In each of these cases, the inciting event is a site-specific endonuclease cleavage of the target RNA to generate a 2′,3′ cyclic phosphate end and a 5′-OH end. Both broken ends must be healed before they can be sealed by an RNA ligase. Healing entails hydrolysis of the 2′,3′ cyclic phosphate to form a 3′-OH and phosphorylation of the 5′-OH to form a 5′-PO. Two pathways of tRNA repair have been delineated, which proceed through distinct 3′ end-healing steps that result in different products of the RNA ligase reaction (,). The first pathway is catalyzed by the familiar T4 enzymes polynucleotide kinase-phosphatase (Pnkp) and RNA ligase 1 (Rnl1). Pnkp is a bifunctional enzyme that remodels the ends of the broken tRNA by converting the 2′,3′ cyclic phosphate to a 3′-OH, 2′-OH and by phosphorylating the 5′-OH end to form a 5′-PO. Rnl1 then joins the 3′-OH and 5′-PO RNA ends to form a standard 3′–5′ phosphodiester at the repair junction (A) (). The second pathway is catalyzed by yeast tRNA ligase (Trl1), a multifunctional enzyme composed of separable healing and sealing domains (). The healing domain, Trl1(389–827), consists of a cyclic phosphodiesterase (CPD) module that hydrolyzes the 2′,3′ cyclic phosphate of the proximal tRNA half-molecule to a 3′-OH, 2′-PO and a polynucleotide kinase module that converts the tRNA 5′-OH to a 5′-PO (A). The ligase domain, Trl1(1–388), then joins the healed ends to form a tRNA with a 2′-PO, 3′–5′ phosphodiester at the splice junction. [The junction 2′-PO is ultimately removed by a phosphotransferase, Tpt1, that is specific to the yeast tRNA repair pathway ()]. Recent studies highlight a plethora of target-specific endoribonuclease toxins in bacteria () and fungi () and their role in defending the organism against non-self species and viruses. This raises the question of whether RNA repair systems exist in bacterial cells as a means of evading programmed RNA breakage. Our identification of candidate RNA healing and sealing enzymes in several bacterial proteomes (,) suggests an RNA repair capacity, but the scant knowledge of the genetics and ecology of those bacteria provides no clues to what types of RNA damage might be subject to repair. Here, we demonstrate that an enzyme from (Pnkp) can catalyze both of the end-healing steps of tRNA splicing . The biochemical pathway of tRNA repair by the Pnkp can be reprogrammed by a mutation in the 3′ end-healing domain that yields a 2′-PO product instead of a 2′-OH. His-tagged versions of T4 Rnl1, T4 Pnkp and yeast Trl1(1–388) were produced in and purified from soluble lysates by Ni-agarose chromatography as described previously (,,). Yeast Trl1(389–827) was produced in as a His-Smt3 fusion and recovered from a soluble extract by Ni-agarose chromatography. The His-Smt3 tag was removed with the Smt3-specific protease Ulp1 and the tag-free Trl1(398–927) protein was recovered in the flow-through fraction during a second round of Ni-agarose chromatography. SDS-PAGE analysis of the enzyme preparations is shown in . Wild-type Pnkp and mutated or truncated versions of Pnkp were produced in as His-tagged fusions and purified from soluble bacterial extracts by Ni-agarose chromatography (,). The 200-mM imidazole eluates were dialyzed against 150 mM NaCl in buffer A (50 mM Tris-HCl, pH 8.0, 10% glycerol, 1 mM EDTA, 1 mM β-mercaptoethanol) and then applied to 1-ml columns of DEAE-Sephacel that had been equilibrated in buffer A. The columns were washed with 5 ml of buffer A and then eluted stepwise with 200 and 500 mM NaCl in buffer A. The Pnkp proteins were recovered in the DEAE flow-through, apparently free of contaminating nucleic acids (which were detected in the 0.5 M NaCl eluate). SDS-PAGE analysis of the Pnkp preparations is shown in . The intron-containing pre-tRNA is a chimera consisting of the mature tRNA sequence of plant pre-tRNA plus the intron and anticodon of pre-tRNA (). This pre-tRNA was generated by transcription of BstN1-cut plasmid pNtY9-T7-M1 by T7 RNA polymerase in the presence of [αP]ATP. A reaction mixture (100 µl) containing 80 mM HEPES-KOH (pH 7.5), 24 mM MgCl, 40 mM DTT, 2 mM spermidine, 2.5 mM CTP, UTP and GTP, 0.5 mM [α-P]ATP, 5 µg template DNA, 120 units RNAsin (Promega), 0.5 units yeast inorganic pyrophosphatase (Sigma) and 300 units T7 RNA polymerase (New England Biolabs) was incubated for 3 h at 37°C. The labeled pre-tRNA was purified by electrophoresis through an 8% polyacrylamide-urea gel. The pre-tRNA was eluted from an excised gel slice in 650 µl of 10 mM Tris-HCl (pH 8.0), 1 mM EDTA. This procedure yielded 380–420 pmol of pre-tRNA. pre-tRNA cleavage reaction mixtures (100 µl) containing 50 mM Tris-HCl (pH 7.5), 100 mM KCl, 10 mM MgCl, 1 mM DTT, 40 µM spermine, 40 pmol [P[AMP-labeled pre-tRNA and 1.5 µg tRNA-splicing endonuclease (produced in as a His fusion and purified by Ni-agarose chromatography) were incubated for 20 min at 65°C. The RNA products were phenol-extracted, precipitated with ethanol (28–32 pmol of cleaved tRNA recovered), resuspended in 100 µl of 10 mM Tris-HCl (pH 8.0), 1 mM EDTA and stored at −20°C. The finding that the essential healing and sealing components of the yeast tRNA-splicing system can be replaced by their bacteriophage analogs () attested to the portability of RNA repair systems from widely distant taxa. However, certain restrictions apply. T4 Pnkp can substitute for the yeast kinase-CPD domain only in tandem with T4 Rnl1 and the yeast ligase domain functions only in tandem with yeast kinase-CPD. Because the kinase modules of the phage and yeast healing enzymes are structurally homologous and the reaction products are identical (a 5′-PO RNA end), the critical factor appears to be the distinctive 3′-OH, 2′-PO end configuration generated by yeast CPD versus the 3′-OH, 2′-OH end produced by T4 Pnkp. In other words, the yeast tRNA ligase appears to require the 2′-PO terminus to seal tRNAs . To better delineate the specificity of end-healing and end-sealing during tRNA repair/splicing, we exploited an tRNA-splicing system designed by Englert and Beier (). The broken tRNA substrate was generated by treating a P-labeled intron-containing pre-tRNA with a tRNA-splicing endonuclease, which led to quantitative release of a linear 21-nt intron and the formation of two ‘half-tRNA’ molecules: a 37-nt 5′ fragment and a 39-nt 3′-fragment (B). Reaction of the broken RNA with a combination of yeast Trl1(1–388) and Trl1(389–827) resulted in a mature spliced tRNA and the circularization of the intron. Splicing of the tRNA halves and intron circularization were also seen with T4 Rnl1 plus T4 Pnkp. However, the ligase domain of yeast Trl1 was unable to splice the tRNA or circularize the intron when paired with T4 Pnkp (B). This observation echoes precisely the inability of Trl1(1–388) to function together with T4 Pnkp . This incompatibility reflects a requirement for a 3′-OH, 2′-PO terminus in order for the ligase component of Trl1 to seal the broken RNA ends . Pnkp catalyzes the phosphorylation of 5′-OH termini of DNA or RNA polynucleotides and the dephosphorylation of 2′,3′ cyclic phosphate, 2′-phosphate and 3′-phosphate ribonucleotides (,,). These characteristics are consistent with a role in end healing during RNA or DNA repair. Pnkp is an 870-aa polypeptide composed of three catalytic domains: an N-terminal module that resembles the kinase domain of T4 Pnkp, a central phosphoesterase module and a C-terminal module that resembles the adenylyltransferase domain of polynucleotide ligases. The distinctive feature of Pnkp known repair enzymes is that its 3′ end modification component belongs to the binuclear metallophosphoesterase superfamily (,). The salient question is whether Pnkp can perform a RNA repair reaction and, if so, to which of the two known repair pathways it adheres. We found that Pnkp functioned in tRNA splicing as the sole source of end-healing activity when added in tandem with T4 Rnl1, resulting in the formation of a spliced tRNA product and an intron circle (). However, splicing activity was attenuated severely when yeast Trl1(1–388) replaced phage Rnl1 as the source of the RNA sealing activity (). We surmise that the predominant outcome of Pnkp healing of a tRNA 2′,3′ cyclic phosphate end is not the 3′-OH, 2′-PO required by yeast tRNA ligase. To better address how Pnkp acts at a 2′,3′ cyclic phosphate end, we compared the release of inorganic phosphate (P) from 2′,3′ cAMP by Pnkp alone versus a duplicate sample in which the products were treated with calf intestinal phosphatase (CIP) prior to assay of P release (A). Whereas Pnkp converted 16% of the input 2′,3′ cAMP to P, treatment of the reaction product with CIP increased the yield of P to 90% of the input 2′,3′ cAMP substrate. (Treatment with CIP alone released <2% of the P from 2′,3′ cAMP.) These results indicate that a nucleoside 2′,3′ cyclic phosphate is converted by Pnkp to a phosphomonoester prior to the release of P. The transient nature of the phosphomonoester intermediate would account for why Pnkp-mediated tRNA repair proceeds down the bacteriophage-type pathway. Studies of the catalytic mechanism of the Pnkp phosphoesterase have shown it to be plastic, insofar as certain mutations in the active site alter the reaction outcome (). In particular, an H189D change abolished the phosphomonoesterase activity with -nitrophenylphosphate as a substrate, without affecting the phosphodiesterase activity with bis--nitrophenylphosphate (). Here, we find that Pnkp-H189D failed to release inorganic phosphate from 2′,3′ cAMP, while converting 88% of the input 2′,3′ cAMP to a monoester that was hydrolyzed by CIP (A). The reaction in substrate excess proceeded to near completion with apparent pseudo-first-order kinetics (B) at an initial rate of 20 s. The transformation of Pnkp into a CPD-only end-healing enzyme by the H189D mutation allowed us to probe the outcome of the CPD reaction, by performing TLC analysis of the reaction mixture as a function of reaction time in the presence of Pnkp only (no CIP). This revealed a single-step conversion of 2′,3′ cAMP to a faster moving product that comigrated with 2′ AMP (C). We infer that a 2′ phosphomonoester is the relevant intermediate in tRNA end healing by wild-type Pnkp. Testing the activity of Pnkp-H189D in tRNA splicing revealed that its RNA repair specificity had been transformed so that its healing activity was now channeled through the yeast-type repair pathway. H189D exhibited a gain of function in working with the yeast ligase to perform both tRNA splicing and intron circularization (). Pnkp-H189D had no repair activity in the absence of Trl1(1–388) (not shown). Moreover, a truncated version of H189D (aa 1–472) containing just the kinase and phosphoesterase modules retained full activity in tRNA splicing in combination with Trl1(1–388) (). Thus, the C-terminal adenylyltransferase domain of Pnkp was noncontributory to the observed tRNA repair activity. We introduced into the H189D-healing domain a second mutation in the ATP-binding motif of the polynucleotide kinase module. The K21A change abolished 5′ kinase activity, without impacting the cyclic phosphoesterase function (). The K21A-H189D double mutant was inert in tRNA repair (), proving that the bacterial enzyme is the source of the 5′ end-healing function for tRNA splicing . The present study illuminates several aspects of RNA repair: (i) the demonstration of a repair-competent RNA end-healing enzyme encoded by a bacterium (to our knowledge, the first case so documented); (ii) repair pathway choice based on the compatibility of the 3′ end-healing and sealing reaction specificities; and (iii) the ability to reprogram the choice of repair pathway by a single missense mutation in the 3′ end-healing component. These properties suggest that RNA repair enzymes and pathways can quickly evolve (or co-evolve) their specificities to suit a particular biological context, e.g. the CPD reaction chemistry and stringent specificity of yeast Trl1 for a 3′-OH, 2′-PO terminus ensures avoidance of sealing RNA ends not generated by programmed cleavage events (). Although Pnkp is clearly tRNA repair-competent , our initial attempts to replace Trl1(389–827) with Pnkp-H189D in yeast were not successful. We suspect that the non-portability of this RNA repair enzyme reflects idiosyncratic aspects of foreign protein expression or localization in budding yeast, rather than any mechanistic deficit, insofar as we were also unable to complement a strain with the Trl1 ortholog from fission yeast . Based on the detection of Pnkp-like proteins with similar size and domain organization in many other bacterial genera ( and ), we propose that bacterial RNA repair is hiding in plain sight.

The interferon regulatory transcription factor (IRF) family plays a critical role in the activation of interferon genes upon virus infection (). One of the most studied members of this family, IRF-3, is normally in a latent state but upon virus infection is activated by phosphorylation of certain Ser and Thr residues. Phosphorylated IRF-3 translocates to the nucleus and binds to its target DNA sequences as a dimer. According to structural studies, IRF-3 has two compact domains (a): the N-terminal DNA-binding domain (NTD) and the C-terminal transactivation domain (CTD), responsible for their dimerization upon virus-induced phosphorylation (). In IRF-3, these domains are connected by a 70-residue Pro-rich linker containing a nuclear export signal (NES) sequence (,). The conformational state of this linker is unclear but since it is Pro-rich, it has been predicted to be unfolded. However, our analysis of its sequence showed that the 15-residue segment of the linker that includes the NES, has a very high helix propensity (b). Moreover, we have found a definite repeat of the apolar residues which in the helical conformation would form the apolar face specific for the polypeptides forming coiled-coils, e.g. leucine-zippers. Mimicking phosphorylation of Ser/Thr residues by substitution with phosphomimetic Asp () demonstrated the essential role in activation played by these residues grouped on a short loop (residues 383–410) in the C-terminal domain: the 5D mutant (S396D, S398D, S402D, T404 and S405D) dimerizes, translocates into the nucleus and activates the target genes (). The key question is how phosphorylation of these residues activates IRF-3. Specifically, how does phosphorylation of the C-terminal domain lead to dimerization of IRF-3 and how do changes in this domain increase the ability of its N-terminal domain to bind DNA? Phosphorylation of five Ser/Thr residues substantially increases the negative charge of the C-terminal domain, which must increase repulsive electrostatic interactions between these domains and also between the complete, intact IRF and the negatively charged DNA (). It has so far been assumed that in the latent inactive state, the C-terminal domain masks the N-terminal domain from DNA and virus-induced phosphorylation relieves their association, thereby activating IRF-3 (,). The N-terminal domain is however positively charged; so increasing the negative charge of the C-terminal domain might be expected to strengthen its association with the N-terminal domain, as well as decrease C-terminal domain interactions in the dimer. The molecular mechanism of IRF activation is thus far from clear. In this study, we used the NTD of IRF-3, the full-length wild-type IRF-3 and two phosphomimetic mutants, 5D (S396D, S398D, S402D, T404 and S405) and 2D (S396D and S398D). A plasmid encoding the 113-residue NTD was kindly provided by Dr C. Escalante. It was expressed in BL21(DE3)pLysS cells and purified as described (,). The expression plasmid encoding a full-length glutathione -transferase (GST)–IRF-3 fusion was kindly provided by Dr J. Hiscott. Modified versions with substitutions S396D and S398D (2D IRF-3) and substitutions S396D, S398D, S402D, T404D and S405D (5D IRF-3) were made by mutagenesis in our group. GST–IRF-3 (1–427) was isolated from BL21 DE3 pLysS following a 3 h induction with 1 mM IPTG at 20°C. Bacterial extracts in PBS containing 1% Triton X-100 were incubated with glutathione-agarose beads for 30 min at room temperature. After three washes with PBS, the fusion proteins were stored on the beads with the addition of protease inhibitors. Resuspended beads were incubated with glutathione (50 mM Tris, 10 mM reduced glutathione) and released fusion protein was incubated with thrombin for 48 h at room temperature to cleave the GST tag. The solution was then dialyzed against a buffer containing 100 mM NaCl, 20 mM Tris, pH 7.3, 1 mM DTT, 0.1 mM EDTA, and the sample was loaded onto a DEAE column to separate IRF-3 from GST and degradation products. IRF-3 was concentrated and dialyzed against the appropriate buffer. The same procedure was used for expression and purification of the IRF-3 2D and 5D mutants. Concentrations of NTD and IRF-3 proteins were determined using extinction coefficients of: = 31 010 and 90 620 Mcm, respectively. The purity and integrity of all the proteins used were checked by PAGE and MALDI mass-spectrometry and found to be better than 95%. All oligonucleotides were purchased from Integrated DNA Technologies, Inc. and additionally purified by anion exchange FPLC on a Mono-Q column, using a linear 0.1–1.0 M NaCl gradient in 10 mM Tris-HCl buffer (pH 7.0), 1 mM EDTA, 20% acetonitrile. The DNA was precipitated with ethanol, pelleted and air dried. Concentrations of single strands and duplexes were determined from the of the nucleotides after complete digestion by phosphodiesterase I (Sigma) in 100 mM Tris-HCl (pH 8.0). To determine the concentration of labeled single strands, the additional contribution of FAM absorption at 260 nm ( = 28 000 Mcm) was taken into account. Fluorescence spectra of the proteins and the fluorescence anisotropy of DNA, corrected for the G-factor, were measured on a SPEX FluoroMax-3 spectrofluorimeter under control of DataMax software (version 2.10). The intrinsic fluorescence of the proteins was excited at 280 nm and monitored over the wavelength range from 290 to 400 nm. The instrument has a thermostated cell holder and software-controlled water bath. A 0.4-cm path-length quartz Suprasil cell was used. All measurements were conducted in 30 mM sodium phosphate buffer (pH 7.4), 100 mM NaCl, 1 mM DTT. Association constants of the monomeric WT IRF-3 and the 5D mutant with the 13-bp PRDI and 26-bp PRDIII–PRDI DNA duplexes labeled with FAM were evaluated by direct fitting of the measured binding isotherms as described (). is the concentration of free dimer; is the total concentration of protein; is the dissociation constant of the 5D dimer; is the dissociation constant of the 5D dimer–DNA complex; [DNA] is the total concentration of the DNA; is the fluorescence anisotropy of DNA observed in the course of titration, whilst and are the anisotropies of the free and bound DNA, respectively. The fluorescence of FAM was excited at 490 nm and registered at 520 nm. The association constants () were measured at 20°C and provided the Gibbs energies of association, ΔG = ln(). The free energy of unfolding by guanidinium hydrochloride (GdmCl) was determined by monitoring changes in the intrinsic fluorescence of the IRF-3 protein, as described in (,). For direct fitting of the denaturation plots, the following equation was used: The free energy of the IRF-3 5D dimer dissociation by GdmCl was obtained by measuring changes of its ellipticity at 220 nm and fitting to the following equation: Native PAGE was performed using the PhastSystem (Pharmacia). For analysis of IRF-3 monomers and dimers, PhastGel Gradient 4–12% gels were used. The concentration of proteins in loading buffer (30 mM Na-phosphate, 100 mM NaCl, 1% β-mercaptoethanol, pH 7.4) was varied from 0.2 to 0.6 mg/ml. Stained gels were scanned and the volumes of bands ( and ), corresponding to the monomer and the dimer of IRF-3 (5D), were calculated using ImageQuant TL software. Estimation of dimerization constants used the following equation: shows the results of native PAGE of IRF-3 WT and its 2D and 5D mutants. One can see that the mobilities of the 2D and 5D monomers are greater than that of the wild type, as expected for proteins differing in their net overall charges: IRF-3 WT, 9; the 2D mutant –11, and the 5D mutant –14 (). It can also be seen that only the IRF-3 5D mutant forms dimers. Analysis of the electrophoregrams obtained at several concentrations of the 5D mutant (not shown) indicated that the dissociation constant of the dimer is on the order of 10 nM, i.e. the dimerization Gibbs energy is −(45 ± 3) kJ/mol. IRF-3 contains 14 tryptophans, 5 of which are in the NTD, 9 in the CTD (114–189), while the linker has none (b). Structural changes in the NTD and CTD can thus be monitored using their intrinsic fluorescence. a and b shows fluorescence spectra of the isolated NTD and the full-length wild-type IRF-3, respectively. Increasing concentrations of GdmCl result in a decrease of fluorescence intensity and in a red shift of the maximum to 356 nm that is characteristic for fully solvent-exposed tryptophan. Plotting the wavelength of the fluorescence maximum, , versus GdmCl concentration (c) shows that the NTD unfolds in one stage at ∼2 M GdmCl, while unfolding of the full-length IRF-3 WT proceeds in two stages: one at ∼2 M, the other above 3 M GdmCl. Clearly the first stage is associated with unfolding of the NTD and the second with unfolding of the CTD. In the 2D mutant (d), the first stage of unfolding is little changed by the Ser/Asp substitutions in the CTD. That is, pseudo-phosphorylation at these sites does not affect the stability of the NTD: the position of the fluorescence maximum at low concentrations of GdmCl (<1.5 M) is invariant and similar to that of IRF-3 WT, and their mid-transitions are at the same concentration of GdmCl (2.2 M). In the case of the 5D mutant, the fluorescence spectrum in the absence of GdmCl is blue-shifted by 2 nm relative to the spectra of the IRF-3 WT and 2D monomers (d). This shift in the initial indicates an increased screening of tryptophans from the solvent, which could be caused by dimerization of the 5D monomers. Thus, the gradual rise of to the value specific for the monomeric proteins (346 nm) shows that the 5D dimers dissociate upon increasing GdmCl concentration up to 1.5 M. Unfolding of the NTD in 5D takes place at the same concentrations of GdmCl as in the 2D and WT IRF-3 proteins, but unfolding of the CTD becomes less cooperative and occurs at lower GdmCl concentrations than in the wild-type protein, indicating some reduction in stability of this domain. Another specificity of the GdmCl-induced changes in the fluorescence spectra of WT IRF-3, containing two structural domains, the NTD and CTD, is the presence of two isobestic points at ∼388 and 341 nm (b). The first, at 388 nm, appears at GdmCl concentrations <3 M. The characteristic changes in spectra in this range of GdmCl and the long-wavelength position of the isobestic point clearly demonstrate similarity with that of the isolated NTD (a) and show that unfolding of this domain proceeds in a highly cooperative manner. The shorter wavelength isobestic point, at ∼341 nm, which occurs at GdmCl concentrations above 3 M, can thus be assigned to the cooperative unfolding of the remainder of the molecule, the CTD. Plotting the fluorescence intensities at the isobestic points allows one to separate the two transitions in the WT IRF-3 (a and b), and thereby determine the GdmCl dependence of the folded fraction for the N-terminal and C-terminal domains (a). This yielded the m-factor that specifies the change in solvent-accessible surface area upon unfolding and the Gibbs energy of unfolding that characterizes the stability of the domains () (b, ). An important aspect of this analysis is that the 5D mutant is monomeric at GdmCl concentrations above 1.5 M. shows that (a) the WT CTD is twice as stable as the NTD; (b) the stability of both domains decreases with increasing negative charge on the CTD, but this effect is much more pronounced for the CTD; (c) the m-factor of the CTD in the 5D mutant is only about half that of the WT protein. a presents CD spectra of the wild-type IRF-3, 2D and 5D mutants at 20°C. The substitutions S396D and S398D (2D) increase the negative value of the ellipticity at 207 nm, showing that the two extra negative charges lead to partial unfolding of the protein. One would expect that appearance of five extra negative charges would result in further unfolding of the protein. In fact, the three additional charges in 5D protein result in an increase of its helicity: the ellipticity at 207 nm decreases with simultaneous increase of negative ellipticity at ∼220 nm. Since this 5D substitution is made in the C-terminal domain, the observed structural changes should be attributed to this domain. The increase of order in this domain with increased negative charge is surprising and can be explained only by its dimerization. According to the above fluorescent measurements, dissociation of the 5D dimer occurs at ∼1.5 M GdmCl (c). CD titration of the IRF-3 5D at 220 nm (b) shows that its negative ellipticity drops sigmoidally upon increasing the GdmCl concentration up to 1.5 M, while the ellipticity of the wild-type protein does not significantly change in this range of GdmCl concentrations. It appears that dimerization and ordering are two conjugate effects specific for the 5D mutant: the increase of negative charge in the CTD induces its restructuring resulting in the formation of the stable dimer. Analysis of the dissociation isotherm of the 5D dimer with increase of GdmCl concentration (b, inset) gives –18 kJ/(mol M) for the m-factor and for the standard Gibbs energy of dimerization in the absence of GdmCl, Δ = –(44 ± 1) kJ/mol, i.e. the dimerization constant = 8 × 10 mol. This value of the Gibbs energy of dimerization is in accord with that estimated above by native PAGE analysis, –(45 ± 3) kJ/mol. The large dimerization energy observed for the 5D mutant assumes the formation of quite extended contacts between the interacting CTDs. The extended contact surface is formed upon phosphorylation of S396, S398, S402, T404 and S405, so the appearance of negative charges might cause displacement of the ‘phosphorylation’ loop containing these residues (). This would result in the exposure of a substantial surface: ∼590 Å of polar area and 1060 Å of apolar area (). Since the larger part of the exposed surface is apolar, hydrophobic interactions would play a significant role in the dimerization of the C-terminal domains. It remains unclear, however, how much the linker might participate in this dimerization process, in particular its NES segment. The high helix propensity of this segment suggests that the observed increase of helicity upon dimer formation might be due to refolding of this segment. Moreover, the regular repeat of the apolar residues (b) suggests that two NES segments from two IRFs could form the coiled-coil, i.e. the leucine-zipper. It is tempting then to ask if this might be important for IRF-3 translocation. IRF-3 binding to DNA was studied by fluorescence anisotropy titration of 5′ fluorescein (FAM)-labeled DNA. a and b shows the results of titrating the PRDIII–PRDI 26-bp DNA and the PRDI 13-bp DNA with IRF-3 WT and its 2D and 5D mutants. It is seen () that binding of the WT protein and the 2D and 5D mutants to the 13-bp single-site DNA (PRDI) do not differ substantially, although the isolated N-terminal domain binds somewhat stronger, probably because of its greater net positive charge. The increase in negative charge in the 2D and 5D mutants decreases their affinity for DNA (), indicating that the C-terminal domain is not very remote from the DNA. If the linker region were fully unfolded, the charge increases in the C-terminal domain would not affect the affinity of the N-terminal domain for DNA. The reduced affinities of the 2D and 5D mutants for the 13-bp DNA thus suggest that the linker is not unfolded. Furthermore, since substitutions mimicking phosphorylation in the C-terminal domain do not substantially increase the DNA-binding ability of the N-terminal domain to the single-site duplex, one can conclude that phosphorylation of the C-terminal domain does not lead to unmasking of the DNA-binding site on the N-terminal domain. Monitoring the unfolding of IRF-3 by GdmCl using intrinsic fluorescence showed that full-length WT IRF-3 and the 2D and 5D mutants unfold in two distinct cooperative stages, i.e. they have two compact structural domains. There is no signal, however, from the linker because it does not contain Trp residues. Thus, we do not know if the linker is completely unfolded, has partially folded structure, or is tightly bound to one of the domains to form a single cooperative unit. To investigate this, we studied the compactness of the isolated N-terminal domain, full-length wild-type IRF-3 and the 2D and 5D mutants by dynamic light scattering. The observed hydrodynamic radii, , of these molecules are given in . One can see that of WT IRF-3 is about 1.7 times greater than that of the isolated NTD. This is the expected increase in radius, bearing in mind that the mass of WT IRF-3 is about four times that of the NTD and if one assumes that the structure of the WT IRF-3 is compact and approximately spherical. The two structural domains of the protein cannot therefore be connected by an extended flexible linker and must be in immediate contact. It should be noted that if the 75-residue segment were extended and flexible with a length of ∼24 nm, the N- and C-terminal domains would act as independent kinetic units and the average of the WT IRF-3 would not differ significantly from that of the isolated NTD. It is notable that for the 2D mutant is significantly larger than that of WT IRF-3. Since the 2D mutant remains a monomer this implies an increase in the asymmetry (/) of IRF-3 that could be due to some structural reorganization. This might be a result of relieving interactions between the C-terminal domain and a part of the linker upon ‘phosphorylation’ of S396 and S398 (,) and/or partial release of the ‘phosphorylation loop’ (residues 383–410). Surprisingly, for the 5D dimer is only about 6% greater than that of 2D, despite a doubling of the mass. This can only be explained by a reduction in the asymmetry upon forming the dimer. Dimeric IRF-3 5D can be regarded as a single kinetic unit containing two DNA-binding sites that interacts with the two binding sites on the 26-bp DNA to form a complex that is also a single kinetic unit. One could expect therefore that the Gibbs energy of binding dimeric IRF-3 5D to the 26-bp DNA will be twice that of binding to the single site 13-bp DNA, i.e. −33 × 2 kJ/mol = –66 kJ/mol. However, this value substantially exceeds the observed Gibbs energy of binding the 5D dimer to 26-bp DNA, –46 kJ/mol (). This difference cannot be explained by the difference in the number of kinetic units which are fixed in these two cases since the translational entropy factor is known to be small, TΔS = (5 ± 4) kJ/mol (). A possible reason for the DNA-binding energy of dimeric IRF-3 being less than twice the monomeric could be the work needed to distort either the double-site DNA or the IRF dimer structure upon binding. The PRDIII- and PRDI-binding sites are not on the same face of the DNA (b): the centers of two sites are separated by 13 bp and, assuming a 36° twist per base pair, the total twist between them is ∼460° – 360°= 100°. It is notable that the PRDIII- and PRDI-binding sites are separated by the spacer -GGGAG- (), which might accommodate the deformation needed to engage the two sites on the dimer. One could argue that binding of dimeric IRF-3 to its two ‘twisted’ sites would not require work because the DNA-binding domains are connected with their dimerization domains by the long linker (), and therefore both NTDs could approach the recognition sites on the twisted faces of the DNA without its distortion. However, as shown by the light scattering data, in dimerized IRF-3 the C-terminal domain keeps the DNA-binding NTD on a very short leash. Thus, the binding of dimeric IRF-3 to the 26-bp DNA containing PRDI- and PRDIII-binding sites must result either in DNA distortion or distortion of the interdomain linker. Similar situations of homo-dimeric DBDs binding to DNA containing two binding sites have been observed previously: interaction of the λ-Cro dimer with two major groove DNA sites separated by a 3 bp spacer (-GCG-) induces a DNA bend of ∼40° () and in the DNA complex of the papilloma virus E2 homo-dimer with DNA containing recognition sites separated by 4 bp, a bend of 43° is induced (). In both these complexes, a ‘recognition’ helix from each subunit binds into the major groove and the bending is towards the protein, resulting in significant compression of the minor groove, which might be energetically very expensive. A feature of these two cases is that the two DNA recognition sites are palindromic, a situation typical for homo-dimer–DNA interactions. In contrast, the two IRF-3 recognition sites could be described as inexact tandem repeats, suggesting that considerably more distortion of the DNA must occur than in the E2 and λ-Cro complexes: this might be principal basis of the free energy deficit observed in the present study. e m a i n c o n c l u s i o n o f t h i s s t u d y i s t h a t d i m e r i z a t i o n o f I R F - 3 , c a u s e d b y i t s p h o s p h o r y l a t i o n , i s t h e s o l e r e a s o n f o r t h e i n c r e a s e d a b i l i t y o f t h i s t r a n s c r i p t i o n f a c t o r t o b i n d i t s t a r g e t D N A i n t h e I F N - β e n h a n c e o s o m e t h a t c o n t a i n s t h e t a n d e m P R D I I I – P R D I - b i n d i n g s i t e s . B i n d i n g o f t h i s d i m e r i c p r o t e i n t o t h e d o u b l e - s i t e D N A m u s t i n d u c e c o n s i d e r a b l e d i s t o r t i o n o f t h e D N A a n d p o s s i b l y t h e l i n k e r r e g i o n c o n n e c t i n g t h e t w o d o m a i n s o f I R F - 3 .

DNA ligases are ubiquitous enzymes that seal DNA nicks via three sequential nucleotidyl transfer reactions (). In the first step, nucleophilic attack on the α phosphorus of ATP or NAD by ligase results in release of PP or NMN and formation of a covalent ligase-adenylate intermediate in which AMP is linked via a phosphoamide (P–N) bond to Nζ of a lysine. In the second step, the AMP is transferred to the 5′ end of the 5′ phosphate-terminated DNA strand to form a DNA-adenylate intermediate. In the third step, ligase catalyzes attack by the 3′-OH of the nick on DNA-adenylate to join the two polynucleotides and liberate AMP. DNA ligases are grouped into two families, ATP-dependent ligases and NAD-dependent ligases, according to the cofactor required for ligase-adenylate formation (). An NAD-dependent DNA ligase (LigA) is found in every bacterial species (). The structures of several LigA enzymes have been determined by X-ray crystallography (). They contain the ligase catalytic core, composed of a nucleotidyltransferase (NTase) domain and an OB-fold domain, flanked by an N-terminal domain (called domain Ia) and three C-terminal domains: a tetracysteine Zn-finger, a helix-hairpin-helix (HhH) domain and a BRCT domain (). Domain Ia is unique to NAD-dependent ligases and required for the reaction of LigA with NAD to form the ligase-adenylate intermediate (). LigA is essential in all bacteria tested to date (). The first identification and characterization of a bacterial ATP-dependent ligase was reported in 1997 for (). In the ensuing years, putative ATP-dependent ligases have been annotated in scores of bacterial species that have been subjected to genome sequencing, though only a few of them have been studied biochemically (). and have an exceptionally rich assortment of ATP-dependent ligases (named LigB, LigC and LigD) in addition to NAD-dependent LigA. Whereas mycobacterial LigA and LigB display vigorous ligase activity, LigD and LigC are relatively feeble at sealing nicked DNAs (). Genetic and biochemical analyses implicate LigD and LigC in a bacterial pathway of nonhomologous end joining (NHEJ) that is strictly dependent on a bacterial homolog of the NHEJ factor Ku (). The fact that Ku and LigD are jointly encoded by dozens of diverse bacterial genera implies that NHEJ is broadly relevant to bacterial physiology, notwithstanding that neither Ku nor LigD are essential in or under laboratory growth conditions (,). However, Ku and LigD are important in protecting spores against DNA damage incurred under conditions of environmental stress (). is a soil bacterium that parasitizes plants and causes crown gall disease. The genome (,) consists of a 2.8-Mb circular chromosome, a 2.1-Mb linear chromosome, a 543-kb plasmid (pAT) and a 214-kb virulence plasmid (pTi). Transfer of the Ti plasmid to the host and integration of T DNA into the plant genome are the critical events underlying crown gall disease. T DNA integration is believed to occur via NHEJ, yet ablation of the Ku and Lig4 proteins that comprise the NHEJ machinery of the plant host has no effect on T DNA integration (,). One of the striking features of the proteome is the presence of six putative ATP-dependent DNA ligases in addition to the canonical NAD-dependent LigA (). The predicted product of ORF 0840 on the circular chromosome (which we now designate AtuLigB) is a 541-aa polypeptide homologous to LigB. AtuLigB (like MtuLigB) consists of a core ATP-dependent ligase domain fused to an N-terminal segment that resembles the DNA-binding domain (DBD) of human DNA ligase 1 () (). AtuLigB and MtuLigB are members of a distinct ligase clade that includes mammalian DNA ligase III, archaeal ligases, poxvirus ligases and LigB homologs from diverse bacteria (e.g. and many others). In the present study, we focus on two homologs of LigD, which we designated AtuLigD1 and AtuLigD2, respectively. Like PaeLigD, AtuLigD1 (771-aa) and AtuLigD2 (840-aa) consist of a central ATP-dependent ligase domain fused to an N-terminal phosphoesterase module and a C-terminal polymerase-like domain () (). LigD is composed of the same three domains, albeit arranged in a different order (,,). AtuLigD1 is encoded by ORF 4632 on the linear chromosome, which is immediately adjacent to ORF 4631 that encodes a homolog of Ku (hereafter named AtuKu1). This operon arrangement suggests an NHEJ function for AtuLigD1 and AtuKu1. AtuLigD2 is encoded by ORF 5055 on the AT plasmid. The AtuLigD2 gene is located next to an oppositely transcribed gene cluster (a putative operon) that encodes two other Ku homologs (ORFs 5049 and 5050; designated AtuKu2 and AtuKu3) and a homolog of mycobacterial LigC (ORF 5051, designated LigC2). There are two additional LigC paralogs in : one encoded on the AT plasmid (ORF 5097; LigC1) and another on the Ti plasmid (ORF 6090; LigC3). The AtuLigC paralogs are minimal ligases (345–353-aa) consisting only of the nucleotidyltransferase and OB modules with no auxiliary flanking domains (). Genetic evidence implicates mycobacterial LigC (which is nonessential for bacterial growth) in a minor pathway of Ku-dependent NHEJ that is evident when the major LigD-dependent NHEJ pathway is ablated (). We can surmise that neither AtuLigD2, AtuKu2, AtuKu3, AtuLigC2, AtuLigC1 nor AtuLigC3 are essential for viability of , insofar as strains cured of pAT or pTi, or both plasmids, were reported to grow as well as a wild-type strain containing the pAT and pTi plasmids (). As an initial step in understanding how ligases D and C might contribute to a putative NHEJ pathway, we purified and characterized Atu ligases D1, D2, C1, C2 and C3. We find that four of the five proteins (LigC1 being the exception) are able to seal a nicked duplex DNA substrate , albeit with relatively low turnover. We focus in detail on the additional enzymatic functions associated with AtuLigD1 and AtuLigD2, especially their catalysis of 3′ end-healing reactions relevant to DNA repair, as described previously for LigD (). Whereas AtuLigD2 has an associated polymerase activity, AtuLigD1 does not. We report that bacterial Ku stimulates the repair of DNA double-strand breaks by ligases D2, C2 and C3. A DNA encoding AtuLigD1 was PCR-amplified from strain C58 genomic DNA (a gift of Dr Andrew Binns) using primers designed to introduce NheI sites at the start codon and 3′ of the stop codon. The PCR product was digested with NheI and inserted into pET28b (Novagen). DNA encoding AtuLigD2 was amplified using primers designed to introduce NdeI sites at the start site and 3′ of the stop codon. The PCR product was digested with NdeI and inserted into pET16b. DNAs encoding AtuLigC1, AtuLigC2 and AtuLigC3 were amplified by PCR with primers designed to introduce an NdeI site at the start site and a BamHI site 3′ of the stop codon. The PCR products were digested with NdeI and BamHI and inserted into pET16b. The inserts were sequenced completely to exclude the acquisition of unwanted changes during amplification and cloning. The pET-AtuLig plasmids were transformed into BL21(DE3). Cultures (1 l) of BL21(DE3)/pET-AtuLig were grown at 37°C in Luria–Bertani medium containing either 0.05 mg/ml kanamycin (for AtuLigD1) or 0.1 mg/ml ampicillin (for AtuLigD2) until the reached 0.6. The cultures were adjusted to 0.5 mM isopropyl-β--thiogalactopyranoside and then incubated at 17°C for 15 h. Cells were harvested by centrifugation and the pellets were stored at −80°C. All subsequent steps were performed at 4°C. Thawed bacteria were resuspended in 50 ml of lysis buffer (50 mM Tris-HCl, pH 7.5, 1.5 M NaCl, 10% glycerol, 15 mM imidazole). Lysozyme and Triton X-100 were added to final concentrations of 50 µg/ml and 0.1%, respectively. The lysates were sonicated to reduce viscosity and insoluble material was removed by centrifugation. The supernatants were applied to 2-ml columns of Ni-nitrilotriacetic acid-agarose (Qiagen, Chatsworth, CA, USA) that had been equilibrated with lysis buffer. The columns were washed with 50 ml of lysis buffer and then eluted stepwise with 4-ml aliquots of buffer A (50 mM Tris-HCl, pH 7.5, 0.4 M NaCl, 10% glycerol) containing 50, 100, 200 and 500 mM imidazole. The polypeptide compositions of the fractions were monitored by SDS-PAGE. The LigC and LigD proteins were recovered predominantly in the 200 and 500 mM imidazole eluate fractions, respectively. Protein concentrations were determined by using the Bio-Rad dye reagent with bovine serum albumin as the standard. The recombinant protein preparations were stored at −80°C. The yields were as follows: AtuLigC1 (32 mg), AtuLigC2 (1.3 mg), AtuLigC3 (2 mg), AtuLigD1 (14 mg) and AtuLigD2 (1 mg). Gene fragments encoding LigD1-(1–186) and LigD2-(1–193) (the PE domains) were amplified by PCR with sense strand primers that introduced an NdeI site at the start codon and antisense primers that introduced a new stop codon and a flanking BamHI site. The PCR products were digested with NdeI and BamHI and inserted into pET16b. Alanine mutations were introduced into the pET-AtuPE plasmids by overlap extension PCR. The pET-AtuPE plasmids were transformed into BL21(DE3). Induction of protein expression, preparation of soluble bacterial lysates, and purification of the recombinant PE proteins by Ni-agarose affinity chromatography were performed as described above for full-length Atu ligases. The PE proteins were recovered predominantly in the 200 and 500 mM imidazole eluates. The yields from 200-ml cultures were as follows: AtuLigD1 PE (14 mg) and AtuLigD2 PE (6 mg). Gene fragments encoding LigD1-(536–771) and LigD2-(569–884) (the POL domains) were amplified by PCR from pET-AtuLigD plasmids with sense strand primers that introduced an NheI site at the new start codon for AtuLigD1 POL and an NdeI site at the new start codon for AtuLigD2 POL. The PCR products were restricted and inserted into pET28b (for AtuLigD1 POL) or pET16b (for AtuLigD2 POL). Protein expression, preparation of soluble bacterial lysates, and purification of the recombinant PE proteins by Ni-agarose affinity chromatography were performed as described above. The POL proteins were recovered predominantly in the 200 mM imidazole eluates. The yields from 1-l cultures were as follows: AtuLigD1 POL (6 mg) and AtuLigD2 POL (15 mg). The gene encoding the 293-aa Ku protein (PA2150) was amplified by PCR from genomic DNA with primers designed to introduce an NdeI site at the start site and a BamHI site 3′ of the stop codon. The PCR product was digested with NdeI and BamHI and inserted into pET16b. Induction of Ku expression, preparation of a soluble bacterial lysate from a 1-l culture, and purification of the recombinant Ku by Ni-agarose affinity chromatography were performed as described above, except that the Ni-agarose column was eluted stepwise with buffers containing 50, 100, 200, 500 and 1300 mM imidazole. Ku was recovered predominantly in the 500 and 1300 mM imidazole eluates, which contained 2.3 and 1.2 mg of protein, respectively. Aliquots (40 µg) of the Ni-agarose preparations of AtuLigC, AtuLigD or AtuLigD PE or POL proteins, or PaeKu were mixed with catalase (30 µg), bovine serum albumin (30 µg) and cytochrome (30 µg). The mixtures were applied to 4.8-ml 15–30% glycerol gradients containing 50 mM Tris-HCl (pH 8.0), 0.2 M NaCl, 1 mM EDTA, 2.5 mM DTT, 0.1% Triton X-100. The gradients were centrifuged at 50 000 r.p.m. in a Beckman SW50 rotor for 16 h at 4°C. Fractions (∼0.2 ml) were collected from the bottoms of the tubes. The polypeptide compositions of the gradient fractions were analyzed by SDS-PAGE. Aliquots of the fractions were assayed for ligase, POL or PE activity as specified in the figure legends. A 24-bp DNA duplex containing a centrally placed 3′-OH/5′-PO nick was formed by annealing a 5′ P-labeled 12-mer DNA strand and an unlabeled 12-mer 3′-OH strand to a complementary 24-mer DNA strand as described previously (). Ligation reaction mixtures (20 µl) containing 50 mM Tris buffer as specified, 5 mM DTT, 5 mM MnCl, 1 pmol of P-labeled nicked DNA substrate, 250 µM ATP where specified, and ligases as specified were incubated for 20 min at 37°C. The reactions were quenched by adjusting the mixtures to 10 mM EDTA and 48% formamide. The products were resolved by electrophoresis through a 15-cm 18% polyacrylamide gel containing 7 M urea in TBE (90 mM Tris-borate, 2.5 mM EDTA). The products were visualized by autoradiography and quantified by scanning the gel with a Fujifilm BAS-2500 imaging apparatus. The 5′ P-labeled D10R2 and D11p primer-templates were prepared as described previously (,). Reaction mixtures (10 µl) containing 50 mM Tris-acetate (pH 6.0), 5 mM DTT, 0.5 mM MnCl, 0.5 pmol P-labeled D10R2 or D11p primer-templates, and AtuLig PE domain as specified were incubated at 37°C for 20 min. The reactions were quenched by adjusting the mixtures to 7 mM EDTA and 31% formamide. The products were resolved by electrophoresis through a 40-cm 18% polyacrylamide gel containing 7 M urea in TBE. Reaction mixtures (20 µl) containing 50 mM Tris-HCl (pH 7.5), 5 mM DTT, 5 mM MnCl, 100 µM each of ATP, GTP, CTP and UTP (rNTPs) or 100 µM each of dATP, dGTP, dCTP and dTTP (dNTPs), 1 pmol 5′ P-labeled 12-mer/24-mer primer-template, and AtuPOL proteins were incubated at 37°C for 20 min. The reactions were quenched by adjusting the mixtures to 10 mM EDTA and 48% formamide. The products were resolved by electrophoresis through a 15-cm 18% polyacrylamide gel containing 7 M urea in TBE. Ligases C1, C2 and C3 are 350-aa, 345-aa and 353-aa polypeptides composed of N-terminal nucleotidyltransferase and C-terminal OB domains (). Within the nucleotidyltransferase module is an adenylate-binding pocket composed of six motifs (I, Ia, III, IIIa, IV and V; highlighted in Supplemental Figure S1) that define the polynucleotide ligase/mRNA capping enzyme superfamily of covalent nucleotidyltransferases. Motif I (KWDGYR) includes the lysine nucleophile to which AMP becomes covalently linked in the first step of the ligase reaction. Motifs Ia, III, IIIa, IV and V contain conserved amino acids that contact AMP and play essential roles in one or more steps of the ligation pathway. Located near the C-terminus of the OB domain is motif VI, which is important for formation of the ligase-adenylate intermediate. Motif VI of the AtuLigC paralogs (RExQ) deviates from the canonical motif VI (RxDK) found in the majority of ATP-dependent ligases. To evaluate the biochemical properties of the AtuLigC proteins, we produced them in as His-AtuLigC fusions and purified them from soluble extracts by adsorption to nickel-agarose resin and elution with buffer containing imidazole. SDS-PAGE revealed that each preparation contained a predominant 40 kDa polypeptide corresponding to His-AtuLigC (). A synthetic duplex DNA substrate containing a single nick () was used to gauge the DNA-sealing activity of the AtuLigC proteins, which was evinced by the conversion of the 5′ P-labeled 12-mer strand to a 24-mer product. We observed no nick-sealing activity for two independent preparations of recombinant AtuLigC1. Therefore, all further studies of AtuLigC were conducted with the LigC2 and LigC3 isozymes. AtuLigC2 catalyzed nick sealing in the absence of added ATP (A); this activity was ascribed to preformed LigC-AMP in the enzyme preparations. The extent of ATP-independent sealing was proportional to input AtuLigC2 and there was little accumulation of the AppDNA intermediate (which migrates ∼1.5 nt steps slower than the 12-mer pDNA strand). From the slope of the titration curve, we estimated that ligase-AMP comprised 16% of the AtuLigC2 preparation. ATP-independent nick sealing by AtuLigC2 required a divalent cation cofactor, which could be manganese, magnesium or cobalt (not shown). Calcium, copper and zinc did not support nick sealing (not shown). Manganese-dependent sealing was optimal over a range of pH values from 6.0 to 8.0 in Tris-acetate or Tris-HCl buffers (not shown). Lowering the pH to 4.5 in Tris-acetate resulted in complete inhibition of sealing and the accumulation of DNA-adenylate (AppDNA) as the sole product (not shown). ATP failed to stimulate AtuLigC2 nick-sealing activity under otherwise optimal conditions because it elicited the accumulation of high levels of the DNA-adenylate intermediate (B). This effect was more pronounced when magnesium was the divalent cation cofactor. At the highest level of input AtuLigC2 tested in this experiment, in which nearly all the pDNA strand had reacted, AppDNA comprised 56% of total labeled DNA (B, Mg). This ATP-dependent trapping phenomenon, which was described previously for eukaryotic viral DNA ligases (,) and mycobacterial DNA ligases C and D (), results from dissociation of the enzyme from the DNA-adenylate and immediate reaction of the free ligase with ATP to yield ligase-AMP, which cannot rebind to AppDNA and thus cannot catalyze the phosphodiester formation step (there being only one adenylate-binding pocket on the enzyme). AtuLigC3 reacted with the nicked substrate in the absence of ATP, but the major product was the AppDNA intermediate rather than the ligated 24-mer (A). At the highest level of LigC3 tested, AppDNA comprised 41% of the labeled material while only 16% of the label was sealed 24-mer. From the slope of the titration curve, we estimated that 12% of the AtuLigC3 preparation was preformed ligase-AMP. Thus, AtuLigC3 appeared to dissociate readily from the AppDNA intermediate and was loathe to rebind and seal the adenylylated strand even in the absence of an ATP trap. ATP-independent sealing by AtuLigC3 was optimal at pH 6.0–6.5 with manganese as the cofactor (not shown). Cobalt supported nick sealing by AtuLigC3, but magnesium was less active and calcium, copper and zinc were ineffective (not shown). Inclusion of ATP stimulated formation of the AppDNA intermediate (by at least a factor of 6, calculated from the extent of total product formation as a function of input LigC3), while suppressing formation of the ligated 24-mer (compare C-Mn and 2A). This effect was evident in the presence of manganese or magnesium (C). The quaternary structures of AtuLigC2 and AtuLigC3 were examined by zonal velocity sedimentation in 15–30% glycerol gradients. Marker proteins catalase (native size 248 kDa), BSA (66 kDa) and cytochrome (12 kDa) were included as internal standards in the gradients. The ligase activity profiles for AtuLigC2 and AtuLigC3 comprised single peak components sedimenting between BSA and cytochrome (Supplementary Figure S2), which paralleled the abundance of the AtuLigC2 and AtuLigC3 polypeptides (not shown). A plot of the values of the three standards versus fraction number yielded a straight line. We calculated values of 3.4 for AtuLigC2 and 3.4 for AtuLigC3 by interpolation to the internal standard curves. The results are consistent with a monomeric quaternary structure for AtuLigC2 and AtuLigC3. In summary, these experiments show that LigC2 and LigC3 resemble LigC in that they have feeble DNA nick-sealing activity in the presence of ATP. LigD1 and LigD2 are 771-aa and 884-aa polypeptides composed of a central ligase domain flanked by putative N-terminal phosphoesterase (PE) and C-terminal polymerase (POL) domains. The ligase domains include all of the canonical nucleotidyltransferase motifs (Supplementary Figure S3). We produced AtuLigD1 and AtuLigD2 in as His-AtuLigC fusions and purified them from soluble extracts by nickel-agarose chromatography. SDS-PAGE revealed that the LigD1 and LigD2 preparations contained major polypeptides of 90 and 102 kDa, corresponding to the respective His-tagged LigD proteins (). AtuLigD1 and AtuLigD2 migrated slightly faster and slower, respectively, during SDS-PAGE than did LigD (840-aa; ∼97 kDa by SDS-PAGE). AtuLigD1 and AtuLigD2 catalyzed nick sealing in the absence of added ATP, with little accumulation of AppDNA (A). From the slopes of the titration curves, we estimated that ligase-AMP comprised 5 and 27% of the AtuLigD1 and AtuLigD2 preparations, respectively. ATP-independent nick sealing by AtuLigD1 and AtuLigD2 required a divalent cation cofactor, which could be manganese, magnesium or cobalt. Calcium, copper and zinc did not support activity (data not shown). B shows that inclusion of ATP stimulated the manganese-dependent activity of AtuLigD1 by a factor of 8 (calculated from the extent of total product formation as a function of input LigD1), with relatively little trapping of the AppDNA intermediate. In contrast, when magnesium was the divalent cation, the enzyme was 6-fold less reactive overall (as judged by consumption of substrate) and AppDNA was the majority product (B, Mg versus Mn). ATP stimulated the manganese-dependent reaction of AtuLigD2 with nicked DNA by a factor of 8, with AppDNA accumulating at lower enzyme concentrations and ligated 24-mer predominating at higher levels of input LigD2 (C). ATP trapping of the AppDNA intermediate during the LigD2 reaction was more pronounced in the presence of magnesium (C). AtuLigD1 and AtuLigD2 were sedimented in 15–30% glycerol gradients with internal markers catalase, BSA and cytochrome (Supplementary Figure S4). The nick-sealing activity profiles for AtuLigD1 and AtuLigD2 comprised single peaks that coincided with the AtuLigD1 and AtuLigD2 polypeptides. We calculated values of 4.5 for AtuLigD1 and 5.5 for AtuLigD2 by interpolation to the internal standard curves. These results are consistent with monomeric quaternary structures for AtuLigD1 and AtuLigD2. (The value of 4.5 for AtuLigD1 is slightly lower than expected for a globular protein of 90 kDa, suggesting that it might have an elongated shape.) The primary structure of the C-terminal domain of AtuLigD2 from aa 569 to 884 is similar throughout its length to the POL domain of LigD (aa 533–840) (Figure S3). In contrast, the C-terminal module of AtuLigD1 is truncated by 63-aa compared to AtuLigD2 (Figure S3). To assess what activities, if any, are associated with the AtuLigD2 and AtuLigD1 POL domains, we produced AtuLigD2-(569–884) and AtuLigD1-(536–771) in as His fusions and purified them from soluble extracts by Ni-agarose chromatography. SDS-PAGE showed that the preparations were highly enriched with respect to the recombinant D2 and D1 POL domains (A, left panel). The LigD2 POL domain was capable of template-directed DNA or RNA synthesis, as gauged by its ability to extend a primer-template composed of a 5′ P-labeled 12-mer DNA strand annealed to a complementary 24-mer strand (A, right panel). In contrast, the LigD1 POL domain displayed no polymerase activity with rNTP or dNTP substrates (A, right panel). A mixing experiment revealed that primer extension by LigD2 POL was unaffected by addition of an equal amount of LigD1 POL (not shown); thus the inactivity of the LigD1 POL protein was not caused by an inhibitor in the preparation. We surmise that the C-terminal truncation of AtuLigD1 compromises its POL function. Thus, AtuLigD2 and AtuLigD1 are not functionally equivalent. The native size of the active LigD2 POL domain was gauged by glycerol gradient sedimentation. The 38 kDa LigD2 POL polypeptide sedimented as a discrete component (peaking in fraction 19) between the BSA and cytochrome markers (B, top panel), coincident with the peak of the templated ribonucleotide addition activity (B, bottom panel). An value of 3.7 for LigD2 POL was calculated by interpolation to the internal standard curve. These results are consistent with a monomeric quaternary structure for LigD2 POL. Templated ribonucleotide addition by the POL domain required a divalent cation cofactor (C). Testing various metals at 5 mM concentration showed that activity, expressed as the percent of the input primer that had been extended, was optimal with manganese (89% extended) or cobalt (80% extended). Whereas most of the products formed in manganese were elongated by one or two nucleotides, the cobalt-directed products comprised a ladder extending to the end of the template strand (C). Magnesium was less effective (43% of primer extended) and the reaction was limited to a single step of rNMP addition (C). Cadmium and copper were weak activators of the polymerase; calcium and zinc were ineffective (C). Manganese, cobalt and magnesium displayed relative activities and product distributions similar to that seen in C over a concentration range of 0.3–10 mM (not shown). Further characterization of the POL activity is shown in . When provided with manganese and a mixture of the four dNTPs or rNTPs (100 µM of each nucleotide), increasing concentrations of input POL catalyzed progressive elongation of the 12-mer primer to yield a ladder of products 13–18 nt long (A, left panel). Very little of the primer was elongated to the end of the template strand, even under conditions of POL excess when all of the primer had been extended. The efficiency of the extension reaction was quantified as the percent of the input 12-mer primer strand that was elongated by at least one nucleotide. A plot of primer utilization versus input POL showed that ribonucleotide substrates were preferred over deoxynucleotides by a factor of 8 (A, right panel). A kinetic analysis of primer extension by excess POL in the presence of 100 µM rNTPs or dNTPs is shown in B. The first cycle of ribonucleotide addition was rapid, with nearly all of the 12-mer primer elongated to 13-mer (with some 14-mer) at 10 s, the earliest timepoint in the experiment. We estimated from the 10 s datum an apparent rate constant of ≥0.21 s for the first rNMP incorporation step. It was apparent that the second cycle of templated ribonucleotide addition was much slower, insofar as the 13-mer product formed within 10 s was gradually converted to 14-mer or longer product over the subsequent 5 min. By quantifying the flux through the 13-mer species (not shown), we estimated a rate constant of 0.016 s for the second cycle of rNMP incorporation. The 13-fold rate decrement after the first cycle of ribonucleotide addition by AtuLigD2 POL agrees well with the findings for LigD POL (), which becomes progressively less active with each ribonucleotide added at the 3′ primer terminus (). The kinetics of dNMP incorporation by AtuLigD2 POL were much slower than extension with rNTPs (B). The first cycle of dNMP addition displayed a pseudo-first order pattern with an apparent rate constant of 0.0062 s. Thus, by comparing the rate constants, LigD2 POL is at least 34-fold faster at incorporating a ribonucleotide than a deoxynucleotide during the first reaction cycle at a DNA primer end. The effect of limiting the rNTP pool is shown in C (top panel), where we see that LigD2 POL was remarkably efficient at scavenging ribonucleotide substrate and adding it to the primer-template at stoichiometric ratios of rNTP to primer. For example, virtually all of the primer was elongated at least once during the 20-min reaction at a 3:1 ratio of rNTP to primer-template (156:50 nM). At limiting rNTP concentrations—20 and 39 nM—the yields of extended primer were 19 and 33 nM, respectively. LigD2 POL was ∼8-fold less effective at scavenging dNTPs at the same limiting concentrations (C bottom panel). LigD (PaeLigD) has a 3′-ribonuclease/3′-phosphatase activity, whereby it resects a short tract of 3′-ribonucleotides on a primer-template substrate to the point at which the primer strand has a single 3′-ribonucleotide remaining (). The failure to digest beyond this point reflects a requirement for a 2′-OH group on the penultimate nucleoside of the primer strand. The ribonucleotide resection activity resides within the 187-aa N-terminal PE domain and is the result of at least two component steps: (i) the 3′-terminal nucleoside is first removed to yield a primer strand with a ribonucleoside 3′-PO terminus; (ii) the 3′-PO is hydrolyzed to a 3′-OH. The 3′-ribonuclease and 3′-phosphatase activities are both dependent on manganese. The PaeLigD PE domain also catalyzes hydrolysis of the 3′-PO of an all-DNA primer-template substrate (). The primary structures of the N-terminal segments of AtuLigD1 and AtuLigD2 resemble that of the PaeLigD PE domain (Supplementary Figure S3), which has no apparent structural or mechanistic similarity to any previously characterized nucleases or 3′-phosphatases. Extensive mutational analysis of the PaeLigD PE domain has identified an ensemble of side chain functional groups that are essential for phosphoesterase activity, which is conserved in the two LigD proteins (,). In order to probe whether the AtuLigD paralogs possess an end-healing function, we produced the N-terminal PE domains of AtuLigD1 (aa 1–186) and AtuLigD2 (aa 1–193) in as His fusions and purified them from soluble bacterial lysates by Ni-agarose chromatography. SDS-PAGE analysis verified that the preparations were enriched to the same extent with respect to the LigD1 and LigD2 PE polypeptides, which migrated anomalously at ∼30 kDa (predicted size 24 kDa) and ∼33 kDa (predicted size 25 kDa), respectively (A). This electrophoretic anomaly echoes the properties of the PE domain (). We also produced and purified two mutated versions of each of the Atu PE domains, in which the conserved counterparts of essential PaeLigD residues His42 and His84 were substituted by alanine. The purities of the AtuLigD1 H35A and H77A PE proteins and the AtuLigD2 H40A and H82A PE mutants were comparable to the respective wild-type versions (A). The PE proteins were reacted with a 5′ P-labeled D10R2 primer-template composed of a 12-mer primer strand with two terminal ribonucleotides and a 24-mer DNA template strand (B). The wild-type AtuLigD PE domains converted virtually all of the input labeled strand to a more rapidly migrating end-product, D10R1 (B). This species migrated identically to the D10R1 product generated by the LigD PE domain (not shown). A kinetic analysis of the ribonucleotide resection reaction in enzyme excess showed the transient appearance of small amounts of D10R1p at early times (D), suggesting that the Atu PE domains remove the terminal ribonucleotide in a two-step pathway similar to that described for the LigD PE domain (). The 3′-phosphomonoesterase activity of the AtuLigD PE domains was demonstrated directly by reaction with a 5′ P-labeled D11p primer-template composed of a 3′-phosphate-terminated 11-mer primer strand annealed to a 24-mer DNA strand (C). The wild-type AtuLigD PE domains converted the input D11p strand to a more slowly migrating 3′-OH end-product, D11 (C). The 3′-ribonucleotide resection and polynucleotide 3′-phosphatase activities of the AtuLigD1 and AtuLigD2 PE domains were abolished by each of the histidine-to-alanine mutations (B and C). These results verify that the 3′ ribonuclease and 3′ phosphomonoesterase activities are intrinsic to the AtuLigD PE domains and that they rely on the same essential histidines as the homolog. The quaternary structures of the PE domains were examined by zonal velocity sedimentation in a 15–30% glycerol gradient with marker proteins catalase, BSA and cytochrome included as internal standards. After centrifugation, the polypeptide compositions of the odd-numbered gradient fractions were analyzed by SDS-PAGE and aliquots of the fractions were assayed for shortening of the D10R2 primer-template. The LigD1 and LigD2 PE proteins sedimented as a discrete peaks between BSA and cytochrome , coincident with the 3′ ribonuclease activity (not shown). We infer that the isolated PE domains are monomers. The 3′ ribonucleotide resection activity of the AtuLigD PE domain was optimal at pH 5.5–8.0 in 50 mM Tris-acetate or Tris-HCl buffer (data not shown) and was strictly dependent on a divalent cation, preferably manganese or cobalt (). A manganese titration experiment showed that the 3′ ribonuclease activity was optimal at 0.2–5 mM MnCl (not shown). Magnesium and calcium were ineffective at 0.5 mM concentration. Cadmium, copper and zinc (0.5 mM) were capable of sustaining progressively reduced activity that resulted in the formation of the initial D10R1p product, little of which was converted to the D10R1 end-product in the presence of copper or zinc (). The experiments in probed the contribution of the 5′ single-strand tail of the primer-template to the ribonucleotide resection reactions. We annealed the labeled D10R2 primer to a series of incrementally shortened template strands to form substrates with 12-, 8- or 4-nt 5′ tails attached to identical 12-bp duplex segments, and a substrate consisting only of the 12-bp duplex with no single-strand tail. We examined the kinetics of the reaction of the AtuLigD1 PE domain with this set of primer-templates under conditions of enzyme excess (, top panel). As noted above, the labeled D10R2 strand of the 12-mer/24-mer substrate was converted to a 3′-OH end-product, D10R1, with low levels of the 3′-phosphorylated species, D10R1-p, appearing transiently at 0.5–1 min. The extent of substrate decay and the distribution of the reaction products were strongly affected by shortening of the 5′ tail (, top panel). The primer-template containing an 8-nt 5′ tail was consumed at 44% of the rate of the ‘standard’ DNA with the 12-mer tail (, bottom left panel) and there was a clear delay in the conversion of the D10R1-p intermediate to the D10R1 end-product, i.e. the intermediate persisted for the entire time course and exceeded the level of the end-product for the first 5 min of the reaction (, top panel). Further shortening of the tail to 4 nt slowed the rate of substrate consumption to 7% of the standard primer-template (, bottom panel) and there was scant formation of the D10R1 end-product. The reaction rate on the blunt duplex was 4% of the rate on the standard primer-template. Similar effects of tail shortening on the rate of resection and the distribution of the D10R1-p and D10R1 products were observed for the AtuLigD2 PE domain (, middle and lower panels). The DNA-binding protein Ku is critical for NHEJ-mediated repair of linear plasmid DNA in mycobacteria (). Mycobacterial Ku interacts physically and functionally with mycobacterial LigD (,). In particular, Ku stimulates the joining of linear DNA fragments by mycobacterial LigD (). In light of the genetic evidence that LigC functions in a Ku-dependent NHEJ pathway (), it was of interest to query whether Ku might affect also the strand-sealing activity of LigC . The abundance of LigC and LigD enzymes in affords a unique opportunity to address the Ku-ligase connections. However, the situation is complicated by the multiple Ku isoforms in the proteome. Mycobacteria specify a single Ku protein, which has a homodimeric quaternary structure (). Eukaryal Ku is a heterodimer of separately encoded subunits (). The homomeric and heteromeric association properties of the three Ku polypeptides are not known and the combinatorial possibilities are daunting. Our initial efforts to produce and purify recombinant His-tagged versions of the individual AtuKu proteins were not fruitful, e.g. Ku1 and Ku2 were not expressed in . Consequently, we produced and purified the single Ku protein encoded by (PaeKu; locus tag PA2150; Genbank accession NC_002516). PaeKu was chosen as a stand-in for Ku in light of the shared domain organization and biochemical activities of LigD and LigD2. Analysis of the quaternary structure of recombinant PaeKu by zonal velocity sedimentation in a glycerol gradient revealed it to be a single discrete component that overlapped the ‘heavy’ side of the BSA marker (A). We surmise that PaeKu is a homodimer of the recombinant 35 kDa Ku polypeptide. To approximate an NHEJ-like double-strand break repair scenario , we tested the ability of Atu ligases to join linear pUC19 plasmid DNA that had been digested with BamHI. The reaction mixtures contained ∼120 fmol plasmid DNA, corresponding to 240 fmol double-strand breaks. Whereas addition of a molar excess of AtuLigC2 (1.9 pmol) or AtuLigC3 (3.1 pmol) over the available DNA ends resulted in little end joining, the inclusion of Ku clearly stimulated end joining, as evinced by the appearance of a ladder of linear concatemers, comprising dimer, trimer, tetramer and pentamer products (B). The Ku-mediated stimulation of break repair by LigC2 and LigC3 was proportional to the amount of Ku added in the range of 30, 60 and 120 ng, corresponding to 0.43, 0.86 and 1.7 pmol of Ku homodimer. Ku by itself (120 ng) had no strand-joining activity (B). The absence of monomer circles among the reaction products signifies that Ku promotes intermolecular repair events by LigC2 and LigC3. In contrast, reaction of T4 DNA ligase with the same linear DNA yielded a significant amount of monomer circle along with linear concatemers (not shown). Ku also strongly stimulated plasmid end joining by AtuLigD2 (B). The amount of LigD2 added (250 ng) sufficed to join a small fraction of the input DNA, generating mostly linear dimer and trimer products. Supplementation with increasing amounts of Ku resulted in reaction of nearly all of the input DNA and its conversion to higher order concatamers (tetramer or greater) (B). In contrast, Ku had no apparent effect on plasmid end joining by AtuLigD1 (data not shown). These experiments provide the first evidence that Ku directly affects the double-strand break repair activity of a LigC-type enzyme and they implicate LigD2, LigC2 and LigC3 as likely agents of Ku-dependent NHEJ in . The menu of DNA ligases in the proteome is especially rich, owing to the proliferation of paralogs within the LigC and LigD clades, which is accompanied by an expansion of paralogs of Ku. The complexity of the bacterial NHEJ apparatus now ranges from the relatively simple state found in (which has a single Ku, a single LigD and no LigC), to progressively more complex forms in (single Ku, single LigD, single LigC), (single Ku, single LigD, two LigCs) and (three Ku paralogs, two LigDs, three LigCs). Understanding the meaning (if any) of the larger complement of putative NHEJ ligases in (and other plant-associated bacteria such as and ) requires a systematic dissection of their biochemical properties, focused on the following questions: (i) Do the putative ligases really have strand-joining activity? (ii) What other catalytic functions are associated with the ligases, particularly the LigD paralogs? (iii) How do the paralogs compare biochemically to each other? (iv) How do they compare to putative orthologs from other species? (v) Are they affected by Ku? The present study addressed these issues and thereby illuminates shared features within the LigC and LigD clades, as well as remarkable differences. The key findings are: (i) Not all putative ligases have demonstrable strand-joining activity, e.g. AtuLigC1; (ii) Atu ligases C2, C3, D1 and D2 are relatively feeble at sealing singly nicked DNA duplexes, a characteristic shared with mycobacterial LigC and LigD (); (iii) AtuLigD1 and AtuLigD2 have an intrinsic 3′ end-healing function similar to that of LigD; (iv) AtuLigD1 and AtuLigD2 are not functionally equivalent, because only LigD2 has an intrinsic polymerase activity; (v) AtuLigD2 POL prefers ribonucleotide substrates; (vi) The capacity of Atu LigC2, C3, and D2 to seal double-strand breaks (a model NHEJ-like reaction) is enhanced strongly by Ku. The Ku stimulation of plasmid end joining by AtuLigC is significant, in that it provides the first biochemical support for the genetic evidence in mycobacteria of a Ku/LigC-driven pathway of NHEJ (). It is remarkable that LigC2 and LigC3 are stimulated by Ku derived from , which does not have a LigC protein. We presume that Ku assists in approximating the duplex ends for sealing by the bacterial ligase, although the molecular details remain obscure. Clearly, the functional interaction of LigC and Ku extends across species lines and does not require ancillary structural modules flanking the core ligase domain. This differs from the LigD–Ku interaction, where the POL domain is implicated in binding to Ku (). If such an interaction is indeed critical for Ku function in the LigD-mediated NHEJ pathway, then it also appears to cross species lines, because we find that PaeKu stimulates sealing of broken plasmid ends by AtuLigD2 . The observation that PaeKu does not stimulate plasmid joining by AtuLigD1, which lacks the complete POL domain, provides additional support for the idea that Ku–LigD interactions occur via POL. However, we do not exclude an alternative scenario in which AtuLigD1 interacts specifically with one of the AtuKu paralogs (e.g. AtuKu1, which is encoded in a putative operon adjacent to AtuLigD1). Ultimately, it will be of interest to test whether LigD and Ku proteins from different bacterial species can be exchanged . The polymerase activity of LigD is a major contributor to the mutagenic character of bacterial NHEJ at blunt and 5′-overhang DNA breaks (,). Whereas the single LigD enzymes encoded by mycobacteria and have catalytically active POL domains, the situation is more complex in . The domain order of AtuLigD1 and AtuLigD2 (PE-LIG-POL) mimics that of PaeLigD, yet only AtuLigD2 has an intrinsic polymerase activity. The C-terminal POL domain of AtuLigD2 comprises a complete homolog of the PaeLigD POL domain, whereas the POL domain of AtuLigD1 is truncated at its C-terminus. The equivalent of the missing segment of AtuLigD1 POL constitutes the ‘back’ surface of PaeLigD POL, away from the active site (). As PaeLigD POL exemplifies a minimized version of the archaeal-eukaryal polymerase-primase family (), it is reasonable that the C-terminal deletion in AtuLigD1 results in loss of activity, most likely because the fold of the residual POL domain is compromised, notwithstanding that the domain contains most of the catalytic residues in the POL active site. The presence of a rump POL domain in AtuLigD1 suggests that LigD1 and LigD2 evolved by duplication of a single full-length ancestral LigD gene, followed by deletion of the C-terminal segment of the LigD1 POL module. The salient feature of AtuLigD2 POL is its much higher rate of templated ribonucleotide addition at a DNA primer terminus compared to deoxyribonucleotide addition. This property is shared with PaeLigD POL () and is likely to be relevant to NHEJ . NHEJ provides the means to repair double-strand breaks in the absence of a sister chromatid homolog, i.e. in quiescent cells that have a single copy of the bacterial genome. In the absence of ongoing DNA replication, dNTP pools are likely to be limiting because ribonucleotide reductase is cell-cycle regulated (); thus the shared capacity of LigD POL enzymes to efficiently scavenge and preferentially incorporate ribonucleotides could be advantageous for bacterial survival during quiescence, by providing a temporary ‘ribo patch’ at the 3′-OH side of the repair junction. The present study underscores that the 3′ end-healing functions of the PE domain of bacterial LigD are conserved across species (e.g. and ) and among LigD paralogs within a species (AtuLigD1 and AtuLigD2). The PE domains each catalyze two phosphoesterase reactions: (i) a phosphodiesterase-type resection of a diribonucleotide end of a primer-template to leave a single ribonucleotide and (ii) a phosphomonoesterase reaction that removes a 3′-PO from either a terminal ribo or deoxyribonucleotide of a primer-template to leave a 3′-OH. The phosphodiesterase activity of the AtuLigD PE domains is apparently limited to resecting an RNA 3′ terminus, insofar as we observe no 3′ nucleotide resection activity on an all-DNA primer-template and the resection reaction of the AtuLigD PE proteins halts once the primer has a single 3′ ribonucleoside. The ribonucleotide resection activity of the AtuLigD PE domains is strongly stimulated by the 5′ single-strand tail of the complementary DNA strand. These features are shared with the PaeLigD PE domain (). In contrast, it has been reported that LigD has an intrinsic 3′ exonuclease activity on DNA phosphodiesters (). Although the observed DNA exonuclease of full-length MtuLigD was suppressed by a mutation in the PE segment of the protein (), it was surprising that no DNA exonuclease activity was detected for the isolated PE domain (). We have purified the PE domain of MtuLigD and find that it performs a manganese-dependent 3′ ribonucleotide resection reaction on a 3′ diribonucleotide-containing primer-template and that it removes the 3′-PO from an all-DNA primer-template (Supplementary Figure S5). Thus, the PE domain of LigD has the activities ascribed to the homologous modules of and LigD. We surmise that 3′ ribonuclease and 3′ phosphatase reactions are the relevant repair functions of the PE domain that are conserved in all LigD proteins characterized to date. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Aprataxin is the causative gene product for early-onset ataxia with ocular motor apraxia and hypoalbuminemia, also called ataxia with oculomotor apraxia type 1 (EAOH/AOA1) (,), which is characterized by early-onset progressive ataxia, ocular motor apraxia, peripheral neuropathy and hypoalbuminemia (). Aprataxin interacts with the X-ray repair cross-complementing group 1 protein (XRCC1) and poly(ADP-ribose) polymerase-1 (PARP-1), which are a scaffold protein and a molecular nick sensor in DNA single-strand break repair (SSBR), respectively (), suggesting that aprataxin is associated with SSBR. Consistent with these observations, the lymphoblastoid cells from patients with EAOH/AOA1 show a high sensitivity to HO and alkylating agents, which cause DNA single-strand breaks (SSBs) (). In addition, aprataxin has the histidine triad (HIT) motif, which functions as nucleotidyl hydrolases (), and the zinc finger motif, which is a DNA- or RNA-binding motif. These findings led to the hypothesis that aprataxin has catalytic activity in SSBR. SSBs are discontinuities in the sugar–phosphate backbone of one strand of a DNA duplex. In neurons, DNA is under continuous threat of damage by endogenous reactive oxygen species (ROS) or exogenous environmental genotoxins, and more than tens of thousands of SSBs arise in each cell per day (). SSBs can arise directly from sugar damage, or indirectly from base damage via the enzymatic cleavage of the phosphodiester backbone as intermediate products of SSBR (). The formation of another type of SSB is mediated by topoisomerase I (TOP1), which cleaves one strand of a DNA duplex during transcription or DNA replication (,). If SSBs are not repaired rapidly, they can cause transcription blockage or they are converted to lethal DNA double-strand breaks (DSBs), leading to cell dysfunction and cell death (). Most SSBs are accompanied by the loss of a single nucleotide; this loss must be filled by DNA polymerase and subsequently sealed by DNA ligase. DNA polymerase β (Pol β) and DNA ligase III (Lig3) are most commonly employed for gap filling and ligation during SSBR (). SSBs arising directly from sugar damage or indirectly from base damage typically possess damaged 3′-ends: 3′-phosphate, 3′-phosphoglycolate (3′-PG), or 3′-α, β-unsaturated aldehyde ends. Because these damaged 3′-ends are not suitable for DNA polymerase or DNA ligase, enzymatic processing is required in SSBR to restore the damaged 3′-ends to suitable 3′-hydroxyl ends. 5′-Polynucleotide kinase 3′-phosphatase (PNKP) removes damaged 3′-phosphate ends, but not other damaged 3′-ends. Apurinic/apyrimidinic endonuclease (APE1) can remove 3′-phosphate, 3′-PG and 3′-α, β-unsaturated aldehyde ends (). However, the 3′-PG and 3′-phosphate removal activities of APE1 are ∼70-fold lower than the AP endonuclease activity of APE1 (). Therefore, it is assumed that other enzymes can alternatively process these damaged 3′-ends. In the present study, we hypothesized that aprataxin removes damaged 3′-ends and facilitates the repair of SSBs. Here we show that aprataxin has a novel removal activity with a unique substrate specificity toward damaged 3′-ends including 3′-PG and 3′-phosphate ends, and that disease-associated mutant forms of aprataxin lack this activity. The results indicate that aprataxin has an important and direct role in SSBR, that is, it removes blocking molecules from DNA 3′-ends, and that neurodegeneration in EAOH/AOA1 may be caused by the accumulation of unrepaired SSBs with damaged 3′-ends. Recombinant human Pol β, PNKP and Lig3 were expressed in silkworm (Katakura). The cDNA fragment of each protein was amplified by PCR using the cDNA library (Human Ovary Marathon-Ready cDNA: Clontech) as the template and then inserted into pYNGHis (Katakura). The purified plasmid vector was mixed with purified cysteine proteinase-deleted viral DNA and then cotransfected in BmN cells of silkworm larvae. The obtained recombinant virus was used for the infection of silkworm pupae. Infected pupae were incubated at 25°C, and then frozen for 144 h postinoculation. Harvested pupae were lysed, and the His-tagged protein was purified by immobilized metal affinity chromatography (MagExtractor® His-tag protein purification kit, TOYOBO). Mutant aprataxin cDNAs were produced using the GeneTailer site-directed mutagenesis system (Invitrogen). Wild-type long-form aprataxin (LA) cDNA, the forkhead-associated (FHA) domain of aprataxin (1–114 amino acids; FHA), the splicing variant of aprataxin (175–343 amino acids, GenBank accession number NP_7782411; SA) and the mutant forms of aprataxin cDNA fragments were inserted into the BamHI/XhoI site of pGEX-6P-3 (GE Healthcare Bio-Science Corp.). Glutathione S-transferase (GST) fusion proteins were overexpressed in Rosetta 2 (DE3) pLysS (Novagen) bacterial cells containing these plasmids. The bacterial cells were grown at 37°C until they reached an of 1.0 and then induced with 1 mM isopropyl-1-thio--galactopyranoside and grown for another 3 h before harvesting. Cell pellets were lysed using BugBuster HT (Novagen) with EDTA-free complete protease inhibitor cocktail (Sigma). The GST-tagged proteins were purified using a GST fusion protein purification kit (Bulk GST Purification Modules®, GE Healthcare). Purified fractions were desalinized using Slide-A-Lyzer® dialysis cassettes (Pierce Biotechnology), separated by denaturing SDS-PAGE, and stained with CBB. Western blotting was performed using an anti-GST antibody (GE Healthcare Bio-Science Corp.) and an anti-aprataxin antibody (ab31841, Abcom). The sequences of DNA substrates used were as follows: FITC-21 (5′-[FITC]CTACGTCAGATCTGCGGATGT-3′), 21-OH (5′-CTACGTCAGATCTGCGGATGT[OH]-3′), 21-P (5′-CTACGTCAGATCTGCGGATGT[P]-3′), 24 (5′-CTCTAGCACTTGAGGCTATCCATG-3′), 23 (5′-TCTAGCACTTGAGGCTATCCATG-3′), 45 (5′-CATGGATAGCCTCAAGTGCTAGAGACATCCGCAGATCTGACGTAG-3′), FITC-21-PG (5′-[FITC]CTACGTCAGATCTGCGGATGT-[PG]-3′), FITC-21-Y (5′-[FITC]CTACGTCAGATCTGCGGATGT-[Y]-3′), and 45/21(U) (5′-[FITC]CTACGTCAGATCTGCGGATGUCTCTAGCACTTGAGGCTATCCATG-3′). 5′-FITC-labeled oligonucleotides were synthesized and purified by high-performance liquid chromatography. A 5′-FITC-labeled 3′-PG 21-mer oligonucleotide (FITC-21-PG) was obtained from Thermo Electron Corporation. A 5′-FITC-labeled 3′-phosphotyrosine 21-mer oligonucleotide (FITC-21-Y) was purchased from Midland Certified Reagent Company. A 5′-FITC-labeled 21-mer oligonucleotide with 3′-phospho-α, β-unsaturated aldehyde was prepared using uracil DNA glycosylase (Invitrogen) and endonuclease III (BioLabs). To generate an apurinic/apyrimidinic (AP) site at the 21st uracil from 5′, 10 μM oligonucleotide FITC-45/21(U) was incubated with uracil DNA glycosylase (2 unit) for 1 h at 37°C in a mixture containing 20 mM Tris-HCl (pH 8.4), 10 mM MgCl and 50 mM KCl. The reaction product was annealed with 10 μM oligonucleotide 45. Annealed oligonucleotides (2 μM) were treated with endonuclease III according to the manufacturer's instructions to generate the 5′-FITC-labeled 21-mer oligonucleotide with 3′-phospho-α, β-unsaturated aldehyde. To analyze the activity of aprataxin to remove 3′-phosphate, 1 pmol of oligonucleotides phosphorylated at the 3′-end was incubated in a mixture containing 100 mM Tris-HCl (pH 6.0), 10 mM MgCl, 10 mM β-ME and 0.1 mg/ml BSA with the indicated protein (25, 50 and 100 nM aprataxin, and 50 nM PNKP) for 1 h at 37°C. To analyze the activity of aprataxin to remove 3′-PG, 3′-phospho-α, β-unsaturated aldehyde or 3′-phosphotyrosine ends, 5′-FITC-labeled 21-mer oligonucleotides with 3′-PG, 3′-phospho-α, β-unsaturated aldehyde or 3′-phosphotyrosine ends were incubated with aprataxin (25, 50 and 100 nM), APE1 and TDP1 in a mixture containing 25 mM Tris-HCl (pH 7.5), 10 mM MgCl, 0.5 mM ATP, 0.1 mg/ml BSA and 1 mM DTT for 1 h at 37°C. Reactions were stopped by adding an equal volume of gel-loading buffer (80% formamide, 10 mM EDTA and 0.1% bromophenol blue). The products were separated by electrophoresis in a denaturing 20% polyacrylamide-8 M urea gel and visualized using a Typhoon 9400 scanner (GE Healthcare). We used APE1 (TREVIGEN) and TDP1 (Abnova Corporation) as the positive controls. GMP- and AMP-lysine hydrolysis assay was performed using a method originally developed for FHIT with a fluorogenic substrate, GpppBODYPY (Molecular Probes) (). We followed the protocol of this method previously described with a slight modification. Briefly, 100 μM GpppBODYPY or ApppBODYPY was incubated with the indicated amounts of recombinant proteins (His-LA, LA, SA and FHA) for 1 h at 37°C in 20 μl of 20 mM Na HEPES (pH 7.0), 0.5 mM MnCl and 0.2 mg/ml BSA. Reactions were stopped by adding an equal volume of gel-loading buffer (80% formamide, 10 mM EDTA and 0.1% bromophenol blue). The products were separated by electrophoresis in a denaturing 20% polyacrylamide–8 M urea gel and visualized using a Typhoon 9400 scanner (GE Healthcare). We used Fhit (Upstate) as the positive control. DNA 5′-adenylate was essentially prepared as previously described (). Briefly, a 12.5 μM 5′-phosphate 3′-FITC-labeled 24-mer oligonucleotide (P-24-FITC: 5′-[P]CTCTAGCACTTGAGGCTATCCATG[FITC]-3′) was annealed with a 25 μM 3′-phosphate 21-mer oligonucleotide (21-P: 5′-CTACGTCAGATCTGCGGATGT[P]-3′) and 25 μM complementary oligonucleotide 45. The annealed oligonucleotides were treated with 100 nM T4 DNA ligase in ligation buffer (50 mM Tris-HCl (pH7.5), 10 mM MgCl, 5 mM DTT, 25 μg/ml BSA and 1 mM ATP) overnight at 37°C. Because the nick cannot be ligated owing to 3′-end blocking with phosphate, the adenylation of the 5′-end occurs as an abortive ligation intermediate. The reaction product, 5′-AMP 3′-FITC-labeled 24-mer oligonucleotide (AMP-24-FITC), was then denatured and purified by separation on a denaturing 20% polyacrylamide–8 M urea gel. Following gel extraction, AMP-24-FITC was annealed with a 3′-hydroxyl 21-mer oligonucleotide (21-OH) and oligonucleotide 45. DNA 5′-adenylate hydrolysis assay was performed as described previously with a slight modification (). In brief, the 45-mer duplex DNA harboring a nick with 5′-adenylate ends was incubated with aprataxin or PNKP in a mixture containing 50 mM Tris-HCl (pH 7.5), 1 mM EDTA, 5 mM DTT for the indicated times at 37°C. Reactions were stopped by adding an equal volume of gel-loading buffer (80% formamide, 10 mM EDTA and 0.1% bromophenol blue). The products were separated and visualized as described above. For the reconstitution of DNA repair, 5′-FITC-labeled 3′-phosphate 21-mer oligonucleotides were annealed with 45- and 23-mer oligonucleotides in a mixture containing 1 mM MgCl, 20 mM Tris-HCl (pH 8.0) and 1 mM NaCl to form SSBs with 3′-end blocking with phosphate. The 5′-FITC-labeled 3′-hydroxyl oligonucleotide (FITC-21-OH) was used as the control of SSBs. For another DNA repair assay, FITC-21-PG oligonucleotides were annealed with 45- and 23-mer oligonucleotides to form SSBs with 3′-end blocking with PG. These substrates were incubated with Pol β, Lig3 and aprataxin for 90 min at 37°C. Reaction was stopped by adding an equal volume of gel-loading buffer. The products were separated by electrophoresis in a denaturing 20% polyacrylamide–8 M urea gel and visualized using a Typhoon 9400 scanner (GE Healthcare). We constructed a histidine-tagged long-form human aprataxin (His-LA) using the baculovirus-protein-expression system (A). His-LA was purified by a standard technique, followed by further purification by immobilized metal affinity and gel filtration column chromatography to remove any contamination by other proteins. A single band, which was reactive to the anti-aprataxin antibody, was observed by CBB staining for fractions from 14 to 19 (C and E). The aprataxin-rich fraction (No. 15, B–E) was used in the following studies. To investigate the enzymatic activity of aprataxin against damaged 3′-ends, we examined the 3′-end processing activities of the recombinant aprataxin employing various 3′-end-modified oligonucleotides as substrates. When 3′-phosphate oligonucleotides were used as substrates, the 3′-phosphatase activity of aprataxin was clearly demonstrated by the mobility shift on a denaturing 20% polyacrylamide–8 M urea gel, in which 3′-phosphate oligonucleotides were electrophoresed faster than 3′-hydroxyl oligonucleotides (A). By treating 3′-phosphate oligonucleotides with aprataxin, the bands corresponding to 3′-hydroxyl oligonucleotides appeared in a concentration-dependent manner (A, lanes 3–5). Next, we determined whether aprataxin removes 3′-PG ends. Notably, as shown in B, aprataxin removed 3′-PG ends and generated 3′-hydroxyl ends (B). In contrast, aprataxin failed to remove 3′-α, β-unsaturated aldehyde or 3′-phosphotyrosine ends under the same conditions as those in which aprataxin removed 3′-phosphate or 3′-PG ends (C and D). These results indicate that aprataxin has 3′-end processing activities toward 3′-phosphate and 3′-PG ends, but not toward 3′-α, β-unsaturated aldehyde or 3′-phosphotyrosine ends. To confirm the 3′-phosphatase and 3′-PG hydrolase activities and determine the catalytic domain of aprataxin, we constructed five types of the aprataxin fusion protein including wild-type full-length aprataxin (long-form aprataxin, LA), two types of full-length mutant form of aprataxin (P206L and V263G), which are most frequently found in Japanese patients, the N-terminal FHA domain (FHA) of aprataxin, and aprataxin without the FHA domain (short-form aprataxin, SA) in the bacterial expression system (A and B) (). The GST-tagged long-form aprataxin (LA) showed 3′-phosphatase and 3′-PG hydrolase activities comparable to those of His-LA obtained from the baculovirus expression system (A and B). Although SA removed 3′-phosphate and 3′-PG (A and B), its removal activity was lower than that of LA. FHA did not show removal activity (A and B). Next, we examined whether mutant forms of aprataxin remove 3′-phosphate and 3′-PG ends. Although the mutant aprataxin proteins were unstable, we were able to produce full-length mutant forms of aprataxin (B). These purified mutant proteins contained a ∼63 kDa form, that is, full-length aprataxin and a ∼46 kDa form that retained NH-terminal GST. Neither P206L nor V263G removed 3′-phosphate or 3′-PG ends (A and B). The affinity and catalytic activity of these aprataxin proteins for these two substrates are shown in . LA showed the highest 3′-phosphatase and 3′-PG hydrolase activities, whereas SA showed the activities less than one-half of those of LA. FHA, P206L and V263G lacked the activities. These results indicate that the C-terminal region of aprataxin, which includes the HIT motif, is indispensable for these activities and the N-terminal region of 1–174 amino acids might enhance these activities. We compared these activities of LA with those of PNKP or APE1 in parallel experiments. The turnover number () of the 3′-phosphatase activity of LA was 10-fold lower than that of PNKP, whereas the of the 3′-phosphatase activity of PNKP under our conditions was lower than that previously reported () (). On the other hand, the of the 3′-PG hydrolase activity of LA was ∼20-fold higher than that of APE1. To investigate the substrate specificity of aprataxin, we compared the DNA 3′-phosphatase and 3′-PG hydrolase activities of aprataxin on single-stranded (ss) and double-stranded (ds) DNAs (). The 3′-phosphatase activity of LA on recessed, gapped or nicked ds DNA is approximately one-half of that on ss DNA (A–C). In contrast, the 3′-PG hydrolase activity of LA on both ss and gapped ds DNA with 3′-PG ends is 1.5-fold higher than that on recessed or nicked ds DNA (D–F). As aprataxin has been reported to have GMP- and AMP-lysine hydrolase activities, which are present in most proteins of the HIT superfamily (), we investigated the GMP- and AMP-lysine hydrolase activities of aprataxin employing GpppBODIPY and ApppBODIPY as substrates. The full-length aprataxin fusion proteins His-LA and LA showed very low or no GMP- or AMP-lysine hydrolase activity at the same or a higher concentration used for the 3′-end processing assay (A and B). Recently, Ahel . have shown that aprataxin removes the adenylate residues on 5′-ends of SSBs via its hydrolase activity, thus resolving the abortive DNA ligation intermediates (). Therefore, we investigated the 5′-adenylate monophosphate (AMP) removal activity of our GST-aprataxin fusion proteins. As similarly observed in their previous study (), LA removed an adenylate residue from a 5′-end of nicked ds DNA (A). In addition, SA showed a much lower 5′-AMP removal activity than LA, and FHA and disease-associated mutant forms (P206L and V263G) of aprataxin lacked the activity ()B and C. To confirm that the 3′-end processing activity of aprataxin is sufficient for SSBR by following the reactions carried out by Pol β and Lig3, we then reconstituted SSBR using gapped ds DNA with damaged 3′-ends as substrates. In this assay, aprataxin was assumed to restore damaged 3′-ends including 3′-phosphate and 3′-PG ends to 3′-hydroxyl ends that are suitable for Pol β and Lig3 (A and B, right panel). First, we incubated the gapped ds DNA consisting of 3′-phosphate 21-mer oligonucleotides and 5′-phosphate 23-mer oligonucleotides annealed to complementary 45-mer oligonucleotides with Pol β and Lig3 in the presence or absence of His-LA. Here, 45-mer products were generated only in the presence of His-LA (A, left panel). Next, we incubated gapped ds DNA consisting of 3′-PG 21-mer oligonucleotides and 5′-phosphate 23-mer oligonucleotides annealed to complementary 45-mer oligonucleotides with Pol β and Lig3 in the presence or absence of His-LA. As expected, 45-mer products were generated only in the presence of aprataxin (B, left panel). These results clearly indicate that the 3′-phosphate and 3′-PG removal activities of aprataxin are sufficient for the 3′-end processing in reconstituted SSBR with Pol β and Lig3. Here, we show that aprataxin specifically removes damaged 3′-ends including 3′-PG and 3′-phosphate ends. Through this action, aprataxin might act with DNA polymerase and ligase to repair SSBs with these damaged 3′-ends. In addition, the C-terminal region of aprataxin is responsible for the removal activity and the region of 1–174 amino acids of aprataxin might enhance the removal activity. Furthermore, the disease-associated mutant forms of aprataxin P206L and V263G lack the removal activity, strongly suggesting that the loss of the removal activity of aprataxin is closely linked to the pathogenesis of EAOH/AOA1. The removal of damaged 3′-ends is important for repairing SSBs. SSBs arising directly from sugar damage usually possess 3′-phosphate or 3′-PG ends, whereas SSBs arising indirectly from base damage, via the enzymatic cleavage of the DNA backbone, possess 3′-α, β-unsaturated aldehyde, or 3′-phosphate ends (,). The formation of another type of SSB is mediated by TOP1, which cleaves one strand of a ds DNA during transcription or DNA replication and covalently binds at the 3′-ends with the tyrosyl-DNA phosphodiester linkage () (). Several enzymes including a multiprotein complex with XRCC1 have been proposed as candidates for removing these damaged 3′-ends. PNKP might restore damaged 3′-phosphate ends, but not other damaged 3′-ends (,). APE1 is assumed to remove damaged 3′-phosphate, 3′-PG or 3′-α, β-unsaturated aldehyde ends; however, its enzymatic activity for removing 3′-phosphate or 3′-PG is extremely lower than its AP endonuclease activity (). The results using fusion proteins expressed in baculovirus and bacterial expression systems clearly showed that aprataxin restores the damaged 3′-PG and 3′-phosphate ends to 3′-hydroxyl ends, but not 3′-α, β-unsaturated aldehyde ends nor phosphotyrosine ends. Moreover, the 3′-PG removal activity of aprataxin is higher than that of APE1 and prefers the gapped ds DNA as well as ss DNA, although the 3′-phosphate removal activity of aprataxin is lower than that of PNKP and prefers the ss DNA, not ds DNA. The results suggest that aprataxin is associated with the removal of damaged 3′-ends, particularly 3′-PG ends, directly induced by sugar damage, whereas APE1 is associated with the removal of damaged 3′-α, β-unsaturated aldehyde ends induced by base damage as well as the endonucleolytic cleavage of an AP site (A and B). By contrast, although both PNKP and aprataxin might restore damaged 3′-phospohate ends, (B and C), these proteins are found in mutually exclusive complexes with XRCC1 (); therefore, they might function independently of each other in the repair of SSBs induced by sugar damage (B and C). The precise physiological roles of PNKP and aprataxin in SSBR, however, await further investigations. In DNA ligation, DNA ligase transfers AMP residues to 5′-ends of the DNA breaks and catalyzes the displacement of the adenylate residues on 5′-ends by attacking the adjacent 3′-hydroxyl group. When 3′-ends are damaged and non-ligated, the adenylate residues remain on 5′-ends of SSBs as abortive DNA-ligation intermediates. Recently, Ahel . have shown that aprataxin removes the adenylate residues on 5′-ends of SSBs (). As reported in their article (), aprataxin used in the present study also removed an adenylate residue from a 5′-end of nicked ds DNA. These results indicate that aprataxin contributes to processing the ‘dirty ends’, both 3′- and 5′-ends, of the DNA breaks. Furthermore, the disease-associated mutant forms of aprataxin lost their 3′- and 5′-end processing activities, suggesting the importance of the activities for EAOH/AOA1 pathogenesis. To determine the main physiological role of aprataxin in SSBR, we have to show the accumulation of SSBs with these dirty ends in tissues from patients with EAOH/AOA1 in future studies. The clinical phenotype of EAOH/AOA1 is specifically restricted to the nervous system, raising the possibility that the activity of aprataxin to remove 3′-phosphate or 3′-PG ends plays an important role in neurons. Tyrosyl-DNA phosphodiesterase 1 (TDP1), the causative gene product for spinocerebellar ataxia with axonal neuropathy type 1 (SCAN1), removes the TOP1 peptide from 3′-ends in SSBs induced by TOP1. In addition to its tyrosyl-DNA phosphodiesterase activity, TDP1 has been also proposed to remove 3′-PG ends and function in the repair of SSBs induced by oxidative stress (,). The clinical phenotype of SCAN1 is similar to that of EAOH/AOA1, because they include cerebellar degeneration, posterior column involvement and peripheral axonal neuropathy without predisposition to malignancy (). These similarities between EAOH/AOA1 and SCAN1 suggest that the removal of damaged 3′-ends, particularly those induced by oxidative stress, is crucial for the viability and function of neurons, in particular Purkinje cells and dorsal root ganglion cells. The unique properties of neurons including a high rate of ROS production and a high transcriptional activity may lead to the generation of more SSBs compared with non-neuronal cells. Furthermore, although mitotic cells can utilize a homologous recombination machinery to compensate for a defect in SSBR, postmitotic neurons are unable to repair damaged DNA molecules in a DNA-replication-dependent manner and they are highly dependent on SSBR for DNA integrity. Therefore, neurons might be vulnerable to a defect in SSBR. The reason why a specific neuronal subtype is vulnerable to impairment of the aprataxin-dependent or TDP1-dependent SSBR system remains to be elucidated. Further studies on the mechanism underlying neuronal death and dysfunction caused by a defect in SSBR should provide new insights into therapeutic approaches for neurodegenerative disorders. p p l e m e n t a r y D a t a i s a v a i l a b l e a t N A R o n l i n e .

A large body of evidence demonstrates that impaired responses to genotoxic insults are involved in cancer development. After a DNA insult, a coordinated cellular response takes place, including damage recognition, signal-transduction pathways (with both and proteins) and the response to the DNA damage (cell cycle checkpoints, DNA repair and apoptosis) (). These interwoven responses insure genome integrity. As a consequence, mutations in genes encoding proteins involved in the main DNA repair pathways, such as nucleotide excision repair (NER), translesion synthesis (TLS), mismatch repair (MMR), nonhomologous end-joining (NHEJ) or base excision repair (BER) may cause chromosome and genetic instability, telomere defects and cancer predisposition (). Fewer than 20 DNA repair disorders have been identified so far. Besides the well-known xeroderma pigmentosum (XP), Cockayne's syndrome (CS) and trichothiodystrophy (TTD) syndromes, which affect the two branches of the NER pathway (global genome repair transcription-coupled repair), mutations in other DNA repair pathways have also been identified. For instance, mutations in some components of the NHEJ have been observed for ligase IV (LIG4 syndrome), Artemis (RS-SCID, human radiosensitive severe combined immunodeficiency syndrome), DNA-PKcs (M059J glioblastoma cells) and Cernunnos (). Also, chromosomal instability associated with impaired RecQ helicase functions is associated with Bloom syndrome (BS protein), Werner syndrome (WS; WR protein) and Rothmund–Thompson syndrome (RTS) (). Mutations are also found in different proteins that act as proteins for DNA lesions and/or chromatin alterations, or proteins for conveying the damage signal to downstream proteins. These hereditary mutations have been found in several syndromes and diseases, such as the Li-Fraumeni syndrome (), ataxia-telangiectasia (AT; ), ATR-Seckel syndrome (ATR; ), AT-like disease (ATLD, ), Nijmegen breakage syndrome (NBS, ) and Fanconi anemia (FA). It turns out that cells arising from these patients represent the major source of DNA repair-deficient human cells (). However, the major limitation is the large genetic variation in the human population. As a consequence, to study the relationship between the main DNA repair pathways I envisaged establishing a set of isogenic human cell lines in which one specific gene is silenced by the RNA interference method. Because RNAi is an endogenous natural pathway, its power as a gene-silencing tool is greater than that of other approaches. Most studies describing RNA interference in cultured human cells are based on transient transfections of siRNA duplexes or plasmids, or transient infection of virus-expressing siRNA constructs. In the strategy based on siRNA-expressing plasmids, use is made of integrative plasmids that could lose their siRNA cassette during selection pressure. An elevated number of siRNA-based plasmids per cell or an elevated viral titer may saturate the RNAi machinery (RISC). As a consequence, overexpression of siRNA through either siRNA duplexes or integrative vectors may lead to mistaken conclusions in transient experiments. Hence, long-term maintenance of gene silencing is riddled with pitfalls. To overcome these shortcomings, I have designed episomal plasmids driving the siRNA synthesis. I have harnessed the efficiency of the siRNA approach with Epstein–Barr virus (EBV)-derived vectors in order to establish cell lines carrying a few plasmid copies per cell. For many years, pEBV-based plasmids have been used to efficiently modify human cell genotypes (,). These plasmids bear the latent replication origin of the EBV virus () and allow the expression of only one viral protein, EBV nuclear antigen-1 (EBNA-1). The main features of these vectors have been recently reviewed (). EBNA-1 tethers EBV episomes to metaphase chromosomes, providing the basis for their nuclear retention and their successful segregation at mitosis (,). Interestingly, the presence of EBNA-1 does not interfere with the differentiation of human cells (). EBV DNA replication is semi-conservative and is coordinated with genomic replication (). OriP appears to be regulated like human endogenous DNA replication. Since these vectors behave like endogenous transcription units, it was expected that they would tightly regulate siRNA expression over a long period of time. So far, I have developed and characterized >200 replicative pEBVsiRNA plasmids for their ability to impose efficient, specific and very long-term gene silencing in different human cell lines. In terms of DNA repair, my main aim was to establish a set of isogenic cell lines in which the expression of one specific DNA repair gene is abrogated. Here >40 genes were targeted, and my older clones reached >500 days in culture. These older clones maintained impressive gene silencing. This approach allows us to validate the choice of an efficient siRNA sequence that remains efficient in cells even several months following transfection. This also affords us the opportunity to compare short- and long-term effects of RNAi. Here, I present new clones that will help us to untangle the relationships between the different DNA repair pathways. The crucial step in RNAi is the unfailing recognition of the cognate mRNA by the siRNA sequence. Since Tuschl and colleagues described the initial design paradigm for efficient 21 nt siRNA (,), many rational designs have been developed, based on a better understanding of RNAi biochemistry (,). I adopted the algorithm developed by Sonnhammer and collaborators (). All my siRNA sequences targeted the open reading frame of a specific gene. Cloning was performed using my pEBVsiRNA-LacZ’ vectors carrying either a hygromycin B or puromycin resistance cassette (pBD751 and pBD899, respectively) (). These vectors carry the H1 RNA polymerase III promoter (RNA pol III) to drive the transcription of short hairpin RNA (shRNA), which gives rise to siRNA-like molecules . These cloning plasmids allow blue/white screening of bacterial colonies carrying a recombinant pEBVsiRNA plasmid with the DNA coding for a shRNA sequence. Hence, shRNA cloning reaches 100% efficiency, thus reducing the time of selection and characterization of recombinant plasmids. For each gene, 2–4 vectors were assessed for both short- and long-term silencing (several months). In each case, I selected the best vector according to three criteria: (i) highly efficient silencing of the gene of interest, (ii) maintenance of this gene silencing over time and (iii) accuracy of the expected phenotype () when this phenotype is known. HeLa (cervical adenocarcinoma) cells were maintained in Dulbecco's modified Eagle's medium (DMEM; Invitrogen, Carlsbad, CA, USA) supplemented with 10% fetal calf serum (FCS), 100 U/ml penicillin and 100 μg/ml streptomycin, under 5% CO. HeLa cells were transfected using 2 μl of LipofectAMINE2000 (Invitrogen) with 2 μg of DNA. Transfected cells were propagated in culture in the presence of hygromycin B (125 μg/ml; Invitrogen). After transfection of HeLa cells, 12 clones per gene were propagated in culture, and growing clones were analyzed for loss-of-function. Thereafter, only one clone was selected and added to my battery of knock-down clones (other clones being frozen). Among the 40 genes targeted, 20 are listed in . The other targeted genes have also been successfully silenced for a long period in culture, and the new knock-down clones have to be further characterized (data not shown). Among them, I have targeted the or genes. For acute treatments, 400 000 cells from established clones were seeded per 6-cm-diameter dish 2 days before treatments. Cells were treated with UVC 10 J/m, γ rays 6 Gy (Cs γ-ray source; dose rate of 1.9 Gy/min; IBL 637 CisBio International, Bedford, MA, USA) or etoposide (Vépéside-Sandoz; 25 μM for 1.5 h) and harvested 24 h later. Flow cytometry analysis was described elsewhere (). Briefly, adherent cells were collected by trypsinization, washed with PBS and fixed in 75% ethanol at 4°C for at least 24 h. Cells were washed twice in PBS and nuclear DNA was stained with propidium iodide (Sigma, St. Louis, MO, USA; 4 μg/ml) in the presence of RNase (Sigma; 10 μg/ml) in PBS for at least 30 min. Stained cells were analyzed on a FACScalibur (Becton Dickinson, Franklin Lakes, NJ, USA) using CellQuest software. Here, ∼10 000 cells gated as single cells using FL2A/FL2W scatter were analyzed. Procedures were described elsewhere (). For NHEJ, whole-cell extracts and DNA substrates were prepared as described elsewhere (). I targeted genes of the main pre- and post-replicative DNA repair pathways. Preliminary experiments were performed to silence the and genes in order to circumvent the transcription-coupled (TC) and global genome (GG) NER pathways. XPA is involved in both TC- and GG-NER pathways and XPC is only required for GG-NER. Numerous knock-down clones were characterized, and all of them displayed very low or nearly undetectable amounts of either XPA or XPC protein (). Several XPA and XPC clones have been recently described in detail (,). As expected, XPA and XPC clones exhibited an ‘XP’ phenotype with great sensitivity to UVC and an impaired unscheduled DNA repair synthesis (UDS) after UVC irradiation (). Strikingly, in HeLa or MCF-7 cells, XPC cells displayed major growth disadvantages in comparison with their XPA counterparts. This could be correlated with the critical interactions of XPC (C-terminal portion) with hHRad23B, centrin 2 or TFIIH (). I have now enlarged this approach by focusing on two other genes tightly associated with the XPC pathway, and . Two interpretations have been proposed: (i) hHR23A&B may stabilize the XPC protein by protecting it from 26 proteasome-dependent protein degradation (), and (ii) hHR23B may be a key event in the displacement of XPC from damaged DNA by the XPA–RPA complex (). I found that knock-down dramatically hampered HeLa cell growth, as did gene silencing. In contrast, the shutting down of the gene expression did not significantly affect cell growth (data not shown). I also focused my attention on the and genes (Cockayne's syndrome genes). shows the remaining CSB protein content observed in two CSB HeLa clones 120 days after transfection. The human CSB () belongs to a large complex that transiently interacts with the transcription machinery to check whether RNA synthesis occurs faithfully (for review ). As a SWI/SNF chromatin-remodeling factor, CSB affects the chromatin conformation of both active genes and those found throughout the genome. CSB participates in TC-NER in human cells, but presumably in other DNA repair mechanisms too, because it was found tightly associated with other DNA repair proteins, such as Ogg1 and PARP1. While CSA and CSB nullizygous mice are not viable (), I found that all isolated CSA and CSB cells maintained a residual amount of the targeted protein, suggesting that these proteins could be essential for cell survival in the absence of genotoxic injuries. The NHEJ pathway includes at least seven factors: Ku70, Ku80, DNA-PK(cs), Artemis, Xrcc4 and DNA ligase IV (Lig IV), and recently Cernunnos (). The Ku70–Ku80 heterodimer binds to DNA ends and recruits the PI-3K-like kinase DNA-PKcs (catalytic subunit of the DNA-dependent protein kinase). DNA-PKcs phosphorylates, associates with, and activates the Artemis endonuclease, which is required to resolve DNA hairpin structures. One of the targets of DNA-PKcs is the XRCC4 protein, which associates with DNA-Lig IV. The XRCC4/Lig IV complex is responsible for the final ligation step (for review ). To impair the NHEJ pathway, I targeted Ku70, XRCC4, DNA-PKcs and Lig IV. All attempts to silence the gene with different vectors (e.g. pBD699 plasmid; C) led to cell death after ∼15 days of culture (in the presence of hygromycin B), further indicating that the Ku70 protein is an essential factor in human cells. This contrasts with Ku70 knock-out mice, which are viable, but display severe defects in DSBR, growth and B cell development (for review ). In contrast, I successfully established XRCC4 (pBD694), DNA-PKcs (pBD743) and more recently Lig IV (pBD940) cells for a very long period of time. The protein content of knock-down HeLa clones was monitored constantly, as shown in and . The establishment of XRCC4 and DNA-PKcs clones demonstrates that the cells were sensitive to ionizing radiation and DSB-inducing agents. The marked reduction in either XRCC4 or DNA-PKcs protein levels triggers a failure in DSB rejoining activity, as shown by an NHEJ assay. NHEJ activities were assessed using whole-cell extracts and a wide range of substrates (). In these conditions, NHEJ activities were reduced by 70–80% in DNA-PKcs and XRCC4 cells, depending on DNA end configuration after cutting with different restriction enzymes: cohesive ends, blunt ends, 3′ end/3′ end, 5′ end/blunt end, 3′ end/blunt, 5′ end/3′ end (). As controls, KIN17 clones were used. This study indicated that the reduction in the XRCC4 protein level abrogated the re-circularization of different linear substrates, suggesting that XRCC4 is a limiting factor in NHEJ. The establishment of XRCC4 HeLa clones is interesting because no human cell line lacking this protein has been identified to date. Only Artemis, Lig IV and Cernunnos are known V(D)J/NHEJ factors deficient in human heritable diseases. In parallel to DNA-PKcs and XRCC4, I have recently isolated different Lig IV clones (). We observed that Lig IV deficiency lowered NHEJ activities to a lesser extent than did DNA-PKcs and XRCC4 (data not shown). My stable knock-down clones may allow us to answer numerous exciting questions. For instance, there are at least two NHEJ complementary sub-pathways with different kinetics, the high-speed classical D-NHEJ (strictly dependent on and protein) and a low-speed B-NHEJ (Backup; also termed microhomology-based end-joining pathway). These two sub-pathways seem to contribute to the major rescue of IR-induced DSBs in higher eukaryotes (,). It appeared that NHEJ activity was tightly dependent on both XRCC4 and DNA-PKcs proteins. Beside, XRCC4 protein may be a limiting factor in NHEJ, but not DNA-PKcs, because XRCC4 cells displayed no detectable re-circularization activity (‘OC’ and ‘CC’ products). On the other hand, most of the building blocks of B-NHEJ are unknown; DNA ligase III (Lig III) has been identified as one (). Lig III participates in the last step of the BER pathway by sealing the remaining single-strand break. Lig III is stabilized by its tight interaction with XRCC1 (for review ). I have now established different Lig III clones (two clones are depicted in A, lanes 1 and 2) that will allow us to investigate this issue. Interestingly, Lig III cells were easily obtained with a nearly undetectable protein level, as shown by western blot and immunocytochemical staining. Furthermore, a cell population continues to maintain an undetectable Lig III protein level 40 days after transfection, strongly suggesting that the loss of this protein did not alter the viability and growth of HeLa cells in the absence of genotoxic injuries (A, lane 3). I also switched off the expression of genes of the Rad52 epistasis group, such as Rad51, Rad52 and Rad54, in order to abrogate a part of HR activity. Here, I present results from Rad51 and Rad54 clones. gene silencing induced cell death and few clones with a reduced Rad51 protein level could be expanded in mass culture. Preliminary observations indicate that Rad51 cells display an expected phenotype, as discussed below ( and ). Another interesting field of investigation is the sensor/transducer pathway, which is critical for genomic stability. The MRN complex (MRE11, Rad50 and NBS1) plays an essential role in the intracellular signaling pathway activated after DNA damage, in particular in the cellular response to stalled replication forks. MRN acts as a DNA damage sensor, continuously localized at sites of unrepaired DNA damage. MRN also participates in other essential mechanisms, such as DNA replication, cell cycle checkpoints and telomere maintenance. Nbs1 binds to γH2AX in the vicinity of DSBs, and recruits Rad50 and Mre11. While Mre11 has both nuclease and DNA-unwinding activities, Rad50 adopts a ‘V-like’ conformation to act as a bridge to hold together the broken ends of DSBs and prevent extensive degradation of the DNA (for review see ,). Targeting of MRN components is essential to understand the role of the MRN complex in the NHEJ and HR pathways. and are essential genes in vertebrates, which limits experimental work. Experiments are usually performed with cells from patients displaying hypomorphic mutations in either (ATLD disease) or (NBS syndrome). Recently, one patient with two germline mutations in RAD50 has been identified (for review see ). For instance, most human NBS cells express an NH2-terminally truncated Nbs1 protein that contains an intact Mre11-binding domain (for review see ). The latter cells may maintain several biological activities associated with the MRN complex. In mouse models, the assessment of MRN functions has also been hampered by the lethal phenotype of null Mre11, Rad50 and Nbs1 mutants in cultured cells and (for review see ). Only mouse models carrying Rad50S alleles or mimicking the mutations identified in human NBS and A-TLD patients have been characterized (for review see ). I have isolated stable knock-down HeLa clones for the three components of the MRN complex. In a HeLa genetic background, all of my Rad50 clones exhibited a very low Rad50 protein level, even 6, 45, 69, 76, 120 or 149 days after transfection, with impressive stability over time (). Hence, the reduced Rad50 protein level did not significantly impair cell growth. Interestingly, in all experiments performed, the lowered Rad50 protein level in Rad50 clones triggered a marked fall in both Mre11 and Nbs1 protein levels. gene silencing induced a severe decrease in the protein level of the two other components of the MRN complex (Rad50 and Nbs1). In contrast, the nearly undetectable level of the Nbs1 protein in NBS1 cells did not induce any change in Mre11 or Rad50 protein levels. This result was observed with different pEBVsiRNA vectors per gene and in either short- or very long-term gene silencing experiments (C and D). This observation suggests that the trimeric complex MRN supports different interactions between each building block. Rad50 and Mre11 are essential for the maintenance and stability of the MRN complex, certainly in the vicinity of DSBs. The loss of either Rad50 or Mre11 triggers destabilization of the whole MRN complex and the disappearance of the other components. This was already observed in a patient exhibiting a truncated mutation in the gene associated with reduced expression of three members of the MRN (). Interestingly, the loss of Nbs1 did not interfere with the stability of either Rad50 or Mre11 proteins, as previously mentioned with mouse models (for review see ). These compelling findings show that my stable knock-down cell model has the hallmarks of MRN stability. Besides the MRN complex, I targeted other essential proteins involved in the pathway signaling the presence of DNA damage, such as the nuclear protein kinases ataxia telangiectasia mutated (ATM) and ataxia telangiectasia- and Rad3-related (ATR). ATM and ATR belong to a conserved family of proteins termed the ‘phosphatidylinositol 3-kinase-like protein kinases’ (PIKKs). PIKKs are conserved from yeast to mammals, and respond to various stresses by phosphorylating substrates in the appropriate pathways. The chief transducer of the DSB signal is ATM (for review see ). Another member of this family of kinases is DNA-PKcs, which has been described above for its gene silencing. Cells from AT patients (or ATM-nullizygous mice) are sensitive to ionizing radiation and other agents that induce DSBs. AT cells fail to activate the IR-induced G1/S or G2/M checkpoints, and exhibit radioresistant DNA synthesis, which is indicative of an impaired S-phase checkpoint (for review see ). ATR mediates responses to a broad spectrum of genotoxic stimuli, including DNA replication inhibitors (e.g. hydroxyurea), UV radiation, IR and agents such as -platinum that induce DNA interstrand cross-links. I am currently assessing loss-of-function in ATM and ATR clones. In my ATM cells (clone BD935/12; 70 days after transfection), there was no appearance of phosphorylated ATM protein after induction of DNA damage, and, consequently, p53 protein was absent (J. Bouley, personal communication). Interestingly, phosphorylation of p53 is necessary for an IR-induced G1/S arrest via transcriptional induction of the cyclin-dependent kinase inhibitor p21 (). gene silencing did not significantly affect HeLa cell growth, even after several months of culture. In contrast, ATR cells displayed an elevated rate of spontaneous cell death. This was in agreement with previous reports showing that while ATM−/− mice are viable, ATR-deficient mice die early during embryogenesis. Similarly, conditional knock-out of gene function in human cells leads to a loss of cellular viability. In an effort to reveal the interconnecting features of DNA repair pathways, I treated my stable clones with genotoxic agents that inflict major damage on DNA, triggering defects in the cell cycle progression. I postulated that defects in a specific DNA repair pathway could be easily and rapidly assessed by means of flow cytometry. I have already applied this method to XPA and XPC cells (). I used as genotoxic agents UVC, γ rays or etoposide (VP16), for the following reasons: (i) UVC-induced DNA damage blocks DNA replication and transcription, (ii) γ rays induce DSBs, which are a major threat to genomic stability and (iii) etoposide is a ‘topoisomerase II (topo II) poison’, which converts topo II into a DNA-damaging enzyme by disrupting the cleavage-religation equilibrium; VP16 induces accumulation of DSBs, activation of DNA damage sensors, cell cycle arrest, and initiation of apoptosis or repair (for review see ). All of my stable clones were cultivated, treated and analyzed using the same procedure at least three times. Briefly, cells were treated with UVC (10 J/m), γ rays (6 Gy) or etoposide (25 μM for 1.5 h in the culture medium). Twenty-four hours later, cells were trypsinized, fixed with EtOH (70%), and stored at +4°C before flow cytometry analysis. Not all doses used induced lethality in control cells at the time of the analysis. Only adherent cells were analyzed by means of flow cytometry. Here, I show some of these results. It was noteworthy that I observed no marked differences in cell cycle progression in response to genotoxic insults as a function of the age of the cells in culture. Afterwards, data from flow cytometry analyses were constantly checked with classical clonogenic cell survival assays. In this manner, I could assess the sensitivity of my clones to both low and high doses of genotoxic agents after a short or long period in culture. Long-term gene silencing opens up new areas in the field of DNA repair. Because evolution has retained highly sophisticated DNA repair processes that prevent most of the damage inflicted on the genome, it is necessary to unravel the connections between these pathways in a genetically homogeneous cell model. For the first time, I present stable clones silenced for genes acting as sensors/transducers ( and ), or genes of different DNA repair pathways ( or ). Other HeLa clones have also been established, such as PARG (in culture for 113 days) and Ogg1 (in culture for 180 days), which display a spectacular loss-of-function (G. de Murcia and A. Campalans, personal communications). I have also constructed and characterized new efficient pEBVsiRNA plasmids directed against , XRCC1, and new clones are now under investigation (data not shown). All of these clones are being maintained in culture for further analysis. My approach is based on new siRNA-expressing plasmids termed pEBVsiRNA. The intrinsic features of these plasmids, such as their nuclear retention, their high stability and their tethering to endogenous chromosomes, greatly improve short- and long-term gene silencing in human cells (for review see ). Because the transactivating function of the oriP/EBNA1 complex is essential for activation of promoters carried by the plasmid, the reverse position of the shRNA cartridge related to EBV components could enhance the long-term transcription of shRNA sequences (). This transactivating process involves EBNA1-induced changes in the chromatin structure, including DNA looping and nucleosome destabilization (). In these conditions, 2 weeks after transfection (and hygromycin B selection) of pEBVsiRNA plasmids, ∼100% of transfected cells were silenced. This was the case for numerous genes tested, such as or . As a consequence, these cell populations are easily propagated for >1 month and maintained an impressive gene silencing. Selected clones elicited this marked gene silencing for an undetermined period in culture, even after several freeze–thaw cycles. Interestingly, all clones described here are still being maintained in culture, after several months or >1 year for my older clones (XPA, XPC or DNA-PKcs). I initially developed these new plasmids to compare short- and long-term gene silencing in DNA repair. I thought that transient transfection assays could either amplify or mask the real effects of specific gene silencing, due to an overproduction of siRNA in cultured cells, which could be due to (i) high concentrations of siRNA duplexes, (ii) very high numbers of copies per cell of an integrative plasmid or (iii) elevated virus titers. I targeted genes of the main DNA repair processes (NHEJ, NER, HR and BER) and of crucial DNA damage signaling pathways (ATM, ATR and MRN). Stable HeLa clones were established for all targeted genes. We could then assess the short- and long-term effects of specific DNA repair deficiency on different biological endpoints, such as chromosomal and telomeric abnormalities. Experiments are under way to compare genomic stability and telomere maintenance as a function of the loss of one specific DNA repair pathway (e.g. NHEJ versus HR), after short- and long-term gene silencing. The high efficiency of pEBVsiRNA vectors in different cell lines affords us the opportunity to rapidly obtain new knock-down clones that will be stable over time, in particular for certain genes where to date no mutant lines are available. Most of these clones display very low or undetectable levels of the protein of interest, as shown by western blot or immunocytochemical staining. However, the major drawback of the RNAi approach is underestimation of the biological role of a protein, because in theory a residual amount of the targeted protein remains. This raises two questions: (i) is the residual protein sufficient to ensure its biological functions, and (ii) can we detect a loss-of-function in knock-down clones? At present, my results reveal that knock-down clones exhibited the expected phenotype with an associated loss-of-function. For instance, XPA and XPC cells displayed an impaired UDS with an enhanced UVC sensitivity (), XRCC4, LigIV and DNA-PKcs cells failed to accurately join DNA ends during NHEJ and were highly sensitive to DSB-generating agents (). The cellular response of knock-down clones to diverse genotoxic injuries seems to be in agreement with the expected phenotype ( and ). Another point to be considered is the existence of redundancy in some functions in mammalian cells. For instance, Rad54B protein could compensate for the silencing of Rad54 (). My data also raise some intriguing and provocative questions. For instance, (i) do XPC and hHR23B proteins participate in the cellular response to ionizing radiation? (ii) does Lig III protein participate in the low-speed B-NHEJ? (iii) is the NHEJ pathway required for cellular recovery after etoposide-induced DNA damage? and (iv) are MRN proteins required during HR or NHEJ or both? Many other interesting questions emerge from these data, and the aforementioned stable knock-down clones afford us the opportunity to answer them. Because my approach is based on long-term culture, I cannot exclude the possibility that suppression of gene expression over a long period may provoke compensatory cellular responses. In culture, an ‘’ may mask the true biological consequences of specific gene silencing. However, this is also true during the multistage process leading to tumorigenesis, where a normal cell encounters serial genetic changes, including initiation, clonal expansion, pre-malignant lesions and malignant progression, before acquiring a tumoral phenotype. This adaptive period is also present in other strategies to switch off gene expression, as in knock-out mouse models where clonal cell expansion is required. It is therefore necessary to characterize the expected phenotype in knock-down cells during the culture, and also to compare short- and long-term experiments. In my hands, long-term silencing of genes of the NER or NHEJ pathways, mediated by pEBVsiRNA vectors, always gave the expected phenotypic modifications. To conclude, the combination of RNAi technology with pEBV-derived vectors offers an exceptional opportunity to rapidly create a set of stable knock-down clones covering various fields of genetics. We could mimic human diseases in a comparable genetic background. This approach may have several applications in drug screening, and in the early stages of testing new therapies. Now I have to develop my approach in other human cells, in particular normal cells such as stem cells. A recent study reinforces my strategy by demonstrating that the pEBV vector could effectively enhance RNA interference efficiency in human stem cells (hES) (). Abrogation of DNA repair genes in hES could offer new opportunities, because genetic modifications of stem cells have great therapeutic potential. Another fundamental aspect of this work is the validation of a specific siRNA sequence for several weeks or months in culture. This could ensure unfailing gene silencing before the beginning of time-consuming experiments, such as assays on hES.

Chromosomes in human cells are capped at both ends with non-coding tandem (TTAGGG) DNA tracts called telomere. Telomere DNA plays important role in maintaining chromosome integrity and stability. They not only prevent the loss of coding information by buffering against the incomplete DNA replication, but also protect the natural chromosome ends from being recognized as double-stranded breaks (). The DNA replication machinery produces two telomeres on each chromosome with different end structures, i.e. a blunt-ended leading strand telomere and a lagging strand telomere carrying a single-stranded 3′ G-rich overhang. The G-rich telomere overhang serves as substrate for telomerase that adds telomere repeats to telomere DNA and maintains telomere length homeostasis in germ lines and 85–90% of the tumor cells. The telomere overhang can also form distinct higher order structures (). It can fold into a four-stranded structure called G-quadruplex () and inhibit telomerase activity (). It can also swing back and invade the double-stranded telomere region to form circle-shaped t-loop structure (). Except the single-stranded overhang, the majority of the telomere DNA is double-stranded. It is not clear at present whether quadruplex can form in these regions. The formation of quadruplex in these regions would have to compete with the formation of the classical Watson–Crick duplex. Several studies have shown that the competition is affected by pH, temperature, cation species and strand concentration (). All these studies were carried out using an inter-molecular system in which the G- and C-rich strands were free separate molecules and duplex was found to be the exclusive or dominant structure. As an inter-molecular reaction, the formation of duplex between two free complementary strands strongly depends on their concentration () and, as a result, quadruplex formed at very low strand concentration can be fully converted to duplex with increase in strand concentration (). In cells, the two DNA strands in chromosomes are well constrained in close vicinity. Therefore, the results obtained using separate G- and C-rich strands may not truly reflect the competition . In this work, we studied quadruplex/duplex competition in an intra-molecular system in which the G-rich strand was linked via five thymines to the C-rich strand to better mimic the situation. Our data shows that the oligonucleotides only formed duplex in dilute solution containing physiological concentration of K (150 mM). However, they preferentially formed quadruplex under molecular crowding condition, a reality of intracellular environment that has been shown recently to promote and stabilize quadruplex formation (,). The formation of quadruplex was confirmed by gel electrophoresis, fluorescent dye staining, circular dichroism (CD) spectroscopy and fluorescent polarization. To simulate the telomere end in chromosomes, we constructed double-stranded DNA (dsDNA) carrying four consecutive copies of TTAGGG/CCCTAA at one or both ends. Preferential quadruplex formation in the dsDNA was also observed under molecular crowding condition by atomic force microscopy (AFM) and gel electrophoresis. These results suggest that quadruplex might possibly form or be induced at the blunt-ended telomere termini as an alternative form of telomere end structures. The 5′-(CCCTAA)T(TTAGGG)-3′ ( = 4, 3, 2, 1, 0) and relevant sequences () were purchased from Sangon Biotech (Shanghai, China). Fluorescent 3′ FAM-labeled (TTAGGG)TTA was purchased from TaKaRa Biotech (Dalian, China). In all experiments, oligonucleotides or DNAs were made in 10 mM Tris–HCl (pH 7.4) buffer containing 1 mM EDTA, 150 mM K and 40% PEG 200 or no PEG, heated at 95°C for 5 min and cooled down to room temperature before use. In some experiments, oligonucleotides were labeled at the 5′ end with [γ-P]ATP using T4 polynucleotide kinase (Fermentas, Lithuania). The dsDNAs (Ctrl-dsDNA and Telo-dsDNA) were obtained by polymerase chain reaction (PCR) from the pGEM plasmid. The telomere sequence (TTAGGG/CCCTAA) at one or both end of the Telo-dsDNAs was introduced using (CCCTAA)-containing primers. For the experiments in , 5′-end-P-labeled oligonucleotides (∼3 nM, 2000 c.p.m) were loaded onto 15% polyacrylamide gel containing 40% (w/v) PEG 200 or no PEG and electrophoresed at 4°C, 8 V/cm, in 1× TBE buffer containing 150 mM KCl. Gels were autoradiographed on a Typhoon phosphor imager (Amersham Biosciences, Sweden). For the experiments in , electrophoresis were carried out as above and the oligonucleotides were visualized with ethidium bromide (EB) or 3,6-bis(1-methyl-4-vinylpyridinium)carbazole diiodide (BMVC). For the experiments in , 100 ng (∼1 pmol) dsDNA or 5′-P-labeled ssDNA was mixed with 100 pM mjaSSB, a 76 kDa single-stranded DNA-binding protein from archaeon (), loaded onto 8% polyacrylamide gel containing 40% (w/v) PEG 200, and electrophoresed at 4°C, 10 V/cm, in 1× TBE buffer containing 150 mM KCl. Gels for dsDNA were stained with EB. Samples prepared in PEG were run in PEG-containing gels. The DNAs were visualized by autoradiography or EB staining. The CD spectra of different oligonucleotides (5 μM) were collected from 320 to 220 nm on a CD6 spectropolarimeter (HORIBA Jobin Yvon, France) at 25°C with 1 mm pathlength cylinder quartz cuvette. CD-melting profiles were recorded at 265 or 295 nm while temperature was increased at the rate of 1°C/min. Buffer blank correction was made for all measurements. Fluorescence polarization measurements were carried out on a Spex Fluorolog-3 spectrofluorometer (HORIBA Jobin Yvon, France) at room temperature (20°C). (TTAGGG)TTA-FAM (50 nM) was incubated with each of the indicated oligonucleotide (100 nM) for 5 h at room temperature. Measurements were carried out using excitation and emission wavelength at 480 and 520 nm, respectively. For each sample, 10 parallel measurements with integration time of 2 s were averaged. Buffer blank was subtracted. Polarization values were calculated according to the equation = ( − )/( + ), where the first subscript indicates the orientation of the excitation polarization and the second of the emission polarization (V for vertical and H for horizontal). is a calibrating factor related to the instrument settings calculated from = /. The UV melting profiles were obtained on a DU-640 UV-VIS spectrophotometer (Beckman, Fullerton, CA, USA) equipped with a digital circulating water bath. The absorbance of oligonucleotides at 1 μM was monitored at 260 nm while temperature was simultaneously measured using a thermal probe immersed in the sample cell. The average heating rate was about 1°C/min. Atomic force microscopy imaging was conducted with a Picoscan atomic force microscope (Molecular Imaging, Tempe, AZ, USA) (). PCR product purified by gel electrophoresis was dissolved in 10 mM Tris (pH 7.4) containing 2 mM Mg and 40% PEG 200 or no PEG. About 5 μl of DNA solution was dropped onto freshly cleaved ruby muscovite mica substrate (Digital Instruments, Santa Barbara, CA, USA). After 5 min, the mica surface was rinsed with ultrapure water and gently blew dry with nitrogen. Freshly prepared samples were mounted on AFM stage and imaged under MAC mode in air using Type II MAC Clever (spring constant = 0.95 N/m, Molecular Imaging, Tempe, AZ, USA). Typical scan rates were 1–3 Hz. The images were rastered at 512 × 512 pixels, unfiltered and flattened when needed. To better mimic the situation where the G- and C-rich strands of telomere DNA are constrained in close proximity, we synthesized 5′-(CCCTAA)T(TTAGGG)-3′ (named G4C4) and similar oligonucleotides. The structures of these oligonucleotides were studied in the absence and presence of 40% (w/v) PEG 200, an agent that has been used to mimic the molecularly crowded intracellular environment (). In G4C4, the G-rich motif has the potential to either fold into G-quadruplex or form hairpin duplex with the C-rich motif. Two sequences, G4C4R and G4T4, designed to form duplex and G-quadruplex, respectively, were used as references (). The structures of these oligonucleotides were first examined by electrophoresis using P-labeled oligonucleotides (). Gel electrophoresis showed that (TTAGGG) conjugated with non-complementary sequence of various length formed G-quadruplex in the presence of K (A). As judged from the mobility, the G4C4 formed duplex rather than G-quadruplex in the absence of PEG because it co-migrated with G4C4R that forms duplex, but migrated faster than the G4T4 that forms quadruplex (B, lanes 1–3). In the presence of PEG, the G4C4 seems to have formed quadruplex as reflected by the appearance of a new band with mobility similar to that of the G4T4 (C, lanes 1–3). From the intensity of the two bands, it is estimated that more than 70% of the G4C4 formed quadruplex. The formation of quadruplex became more dominant when the competition of duplex formation was reduced by decreasing the length of the complementary C-rich strand (C, lanes 6, 9). In the absence of PEG, no quadruplex formation was observed in these oligonucleotides (B), suggesting that quadruplex formation shown in C was induced by PEG. The structure of G4C4 in PEG solution was further verified by fluorescent dye staining with the reference oligonucleotides (). EB, a dye for dsDNA, identified the duplex fraction in the G4C4, which had a same migration and staining efficiency as the duplex G4C4R (A). BMVC is a carbazole derivative that preferentially stains the quadruplex formed by the human G-rich strand (). When the samples were pre-stained with BMVC and resolved on gel, an alternative band was observed for the G4C4, which was as efficiently stained as the G4 and G4T4 that are known to form quadruplex. On the contrary, the duplex formed by G4C4R was not stained (B). Post-staining of this gel with EB revealed the duplex bands of G4C4 and G4C4R below the quadruplex band of G4C4 (C). This staining assay confirmed the observation in the gel electrophoresis that the G4C4 formed quadruplex in the presence of PEG. The quadruplex formed by G4C4 in PEG solution was further analyzed by CD spectroscopy (). In a separate work, we provided evidences by CD, gel electrophoresis, fluorescence spectroscopy of 2-aminopurine substituted oligonucleotides that the core sequence of the G-rich strand of human telomere DNA (GTA)G adopts parallel-stranded conformation in 150 mM K and 40% (w/v) PEG solution (manuscript submitted), which features a negative peak near 245 nm and a positive peak near 265 nm in CD spectrum as the many reported parallel-stranded quadruplexes (). Here we decomposed G4C4 into three components, G4, C4 and T-linker, and recorded their CD spectrum separately. The G4 sequence alone produced a typical spectrum characteristic of parallel quadruplex with a negative peak near 245 and a positive peak near 265 nm. The G4C4 displayed a spectrum that is almost identical to the spectrum overlay of its composing components, G4, C4 and T5-linker (A). Moreover, the spectrum of G4C4 resembled that of the G4T4 that forms quadruplex, but differed from that of the G4C4R that forms hairpin duplex (B). In the majority of publications, it has been reported that anti-parallel quadruplex is characterized by a negative peak near 260 nm and a positive peak near 295 nm, while parallel quadruplex displays a negative peak near 240 nm followed by a positive peak near 265 nm (,,). The CD analysis () suggests that the G4C4 adopted parallel quadruplex structure in the G-rich region as the G4. However, this conclusion regarding the folding topology may not be definitive because exceptions have been reported where anti-parallel quadruplex formed by some sequences also showed positive peak near 260 nm (), typical of parallel quadruplex. The PEG-induced formation of quadruplex in G4C4 should liberate the C-rich strand from base paring with the G-rich strand. This expected outcome was examined by a fluorescent probe (TTAGGG)TTA-FAM which is complementary to the C-rich strand (). Upon binding to target strand, a probe is expected to show an increase in fluorescence polarization due to the decreased rotational freedom and this technique has been used in hybridization-based assays (). In our work, the probe was added to solutions containing G4C4R, G4C4 and C4, respectively. In the absence of PEG, the probe only hybridized with the C4 as reflected by the increase in its fluorescence polarization but did not bind to the G4C4R and G4C4 since no change in its fluorescence polarization was detected. In the presence of PEG, the probe hybridized with both G4C4 and C4 resulting in an increase in polarization in both cases, indicating that the G4C4 formed quadruplex leaving the C-rich strand available to the probe. To examine how molecular crowding could affect the quadruplex and duplex formation separately, we studied the thermal stability of the two structures formed by G4T4 and G4C4R, respectively, in the absence and presence of PEG by thermal melting. Telomere quadruplex in K solution without PEG is characterized by a positive peak at 295 nm in its CD spectrum (), therefore the denaturation of quadruplex of G4T4 was monitored by CD at 295 and 265 nm, respectively, in the absence or presence of PEG. The denaturation of duplex is characterized by hyperchromicity at 260 nm () and was monitored by UV absorbance. The results in A shows that the quadruplex formed by G4T4 was melted at 61.9°C in the absence of PEG, but was much more stable in the presence of PEG. On the other hand, PEG destabilized the duplex formed by G4C4R, resulting in a decrease of 14.5°C in melting temperature (B). The effects of PEG on the stability of both structures is in agreement with the data reported in a recent work (). To better simulate the telomere DNA in chromosomes, we constructed by PCR a 1.2 kb blunt-ended double-stranded DNA (Telo-dsDNA) carrying four consecutive copies of TTAGGG/CCCTAA at one end. Normal blunt-ended double-stranded DNA (Ctrl-dsDNA) containing no telomeric repeats was used as reference. The end structures of the dsDNAs were examined by AFM. Without PEG treatment, the two dsDNAs showed a typical shape of dsDNA edge at both ends (A and B). In the presence of PEG, the formation of quadruplex could be recognized at one end of the Telo-dsDNAs as spherical dot (C) with a height of 2.03 ± 0.48 nm which is similar to the reported values for quadruplex under AFM (). The height of 0.65 ± 0.13 nm of the duplex region also matches those reported values for dsDNA (,). Such structure was observed in the majority of the Telo-dsDNAs, but not seen in the PEG-treated Ctrl-dsDNA (D). The end structures of dsDNAs were also examined by gel electrophoresis. To increase sensitivity, a shorter dsDNA of 200 bp carrying four telomere repeats at both ends was constructed. The DNAs were incubated in PEG in the absence or presence of a 76 kDa single-stranded DNA-binding protein (SSB) before electrophoresis in PEG-containing gel. As judged from the mobilities, the SSB bound single-stranded DNA (ssDNA) (A), but not the Ctrl-dsDNA (B, lanes 3 and 4). In contrast, the Telo-dsDNA showed three distinct bands that can be explained by the formation of three different structures that had no quadruplex at both ends, had one quadruplex at one end or had quadruplex at both ends, respectively (Figures 7, lane 5). The two slower bands seemed to carry quadruplex. The formation of quadruplex at the end of the DNAs released the C-rich strand into single-stranded form, therefore SSB could bind to this strand and shifted the DNA to the much slower smears (Figure 7, lane 6). The intracellular environment is crowded with high concentration of macromolecules whose total concentration can reach 400 g/l (). Studies have shown that molecular crowding induced transition from anti-parallel to parallel G-quadruplex (), dissociation of duplex () in GTG telomeric DNA, and transition from intra-molecular G-quadruplex to long multi-stranded G-wire in (TG)TG but not in human G(TAG) telomeric DNA (). Recently, we reported that molecular crowding can induce quadruplex formation under salt-deficient conditions and greatly enhance its competition with duplex formation in 150 mM K solution using separate G- and C-rich strands (). In this work, we extended the study using DNAs in which the C- and G-rich strands of human telomere DNA were kept in close proximity to better resemble the situation. In either the linked G- and C-rich strand or the dsDNAs, quadruplex was observed in the presence of PEG 200 (, , ). Moreover, the dominant quantity of quadruplex over that of duplex indicating that quadruplex was more competitive than duplex. Molecular crowding played a bifacial role in the competition by stabilizing quadruplex and, in the meantime, destabilizing duplex. While the hairpin duplex is more stable than quadruplex ( of 80.2 versus 61.9°C) in the absence of PEG, it became much less stable than the later ( of 65.7 versus ∼>85°C) in the presence of PEG (). In our study, quadruplex was only observed when the DNAs were heat denatured to open the double stranded structure that was already present before the molecular crowding condition was applied. It is not clear whether such a structure can form . Quadruplex-forming sequences are present in many locations other than telomere in genomic DNA, for example, the promoter of BCL-2, retinoblastoma gene, hypoxia inducible factor 1α, oncogene [for review, see ()]. So far, there are several evidences supporting the presence of telomere quadruplex in cells () which may occur at the single-stranded G-rich overhang. The question whether quadruplex can form in the double-stranded region is not clear although the work on the promoter of the gene is consistent with quadruplex formation (). The observed quadruplex formation in double-stranded telomere DNA in this work suggests it may potentially occur or be induced by exogenous molecules at least at telomere termini. In cells, the double-stranded telomere DNA is associated with several proteins such as TRF1 and TRF2 (,), which hold the DNA in place. However, the structure of dsDNA is dynamic. It has to be opened in many biological DNA-processing events, such as replication, transcription and promoter recognition. Spontaneous and transient openings of DNA duplex known as DNA breathing occur under physiological conditions which creates bubbles with size of up to few tens of base pairs (). These events might provide opportunities for quadruplex to form at the blunt-ended telomere, which may present a possible alternative form of structures at telomere ends. Specific proteins may also participate to facilitate the opening of duplex or the formation of quadruplex. Small molecules that stabilize quadruplex or/and destabilize duplex may potentially induce such quadruplex formation, thus offer a possibility of manipulating the structure of double-stranded telomere DNA. It is believed that the blunt-ended telomere produced by leading strand synthesis is processed afterwards to generate a single-stranded G-rich overhang (), which is involved in the formation of t-loop that provides protection to the telomere end (). Our observation of quadruplex formation at the blunt-ended telomere leaves several questions open that may deserve further exploration: whether the quadruplex could form under conditions; how would it affect the processing of the blunt telomere ends and what effect it would produce if the end-processing mechanism fails to produce single-stranded overhang under abnormal conditions.

The area of molecular diagnostics is at the forefront of modern bioanalytical research and a sector of growing importance is the interface with materials science on the nanoscale and the conjugation of metallic nanoparticles with biomolecules (). The ability to detect DNA sequences in an highly selective manner for the rapid detection of genetic mutations and disease states is of ever growing interest and central to a number of new approaches to disease management. Specific DNA sequences can be detected by hybridization to complementary oligonucleotide probes conjugated with nanoparticle substrates in a number of different ways (). Gold nanoparticles have been most widely used in this manner with oligonucleotide adsorption typically achieved through thiol modification, following Brust's observation that alkylthiols stabilize gold colloidal solutions (). Bioanalytical probes based on those observations have been shown to be capable of discriminating between fully complementary and single base mismatched sequences (). This approach relies on the hybridization-induced reversible aggregation of the nanoparticles resulting in a distinctive red-shifting of the plasmon of the nanoparticles. The nanoparticles have unusually high extinction coefficients in the visible region of the spectrum which makes them easy to visualize colourimetrically, by eye or instrumentation, when a change in the plasmon relating to hybridization takes place. A particular disadvantage of the thiol adsorption strategy to immobilize the oligonucleotide probe on the gold is its lability under certain conditions such as prolonged or cycled elevated temperatures, high NaCl concentrations and treatment with biological buffer additives e.g. dithiothreitol (DTT) or mercaptoethanol (). Upon thiol desorption, irreversible aggregation occurs and the probe system is rendered inactive. To minimize these undesirable aggregation events, structurally complex multiple thiol linker systems have been investigated but were only ever reported as being used with gold nanoparticles (,). Thioctic acid is an inexpensive, commercially available and structurally simple disulphide species. Herein we report its preliminary employment as a linker molecule for oligonucleotide nanoparticle functionalization of greater stability than standard thiol analogues. Thioctic acid has received great attention in recent years in the area of gold and silver nanoparticle functionalization for a variety of applications (), ranging from the immobilization of tetrathiofulvalenes as cation sensors on gold electrodes () and the investigation of metal ion chelation (), to the probing of nanoparticle surface adsorption by nitroxide-modified thioctic ester spin labelled probes (). Indeed, it has also been used to immobilize molecules for bioanalytical applications; carbohydrates for non-specific protein interactions (), antibodies () and transition metal complexes for protein capture (). From an oligonucleotide perspective thioctic acid oligonucleotide templates have been generated by attachment to gold nanoparticles via mid-sequence amino-modified sites () and via multiple-step post-synthetic modification (), however, there was no data presented on the subsequent improved properties of the oligonucleotide–nanoparticle conjugates (). Until now, the stability of thioctic acid functionalized oligonucleotide–nanoparticle conjugates has not been reported and their ultimate application to oligonucleotide sequence analysis has not been realized. italic sub #text -(3-Dimethylaminopropyl)--ethylcarbodiimide hydrochloride (EDC.HCl) (1.840 g, 9.6 mmol, 1.2 eq) was dissolved in DCM (anhydrous, 20 ml), which, 1.7 ml diisopropylethylamine (DIPEA) (anhydrous, 9.6 mmol, 1.2 eq) was added and stirred for 10 min. To the solution -hydroxysuccinimide (1.290 g, 11.2 mmol, 1.4 eq) was added and the reaction mixture stirred in an ice bath. Thioctic acid (1.643 g, 8.0 mmol, 1.0 eq) was dissolved in anhydrous DCM (10 ml) and added to the reaction over 5 min and stirred overnight. The reaction mixture was washed with HCl (5% (v/v), 50 ml × 2) and water (distilled, 50 ml). The organic layer was dried over NaSO. Solvents were concentrated to yield 2.236 g of ester , as a yellow, solid. The residue was dry-loaded on NaSO and the product purified by flash column chromatography giving the product in 83% yield. δ(400 MHz; DMSO) 1.4–1.7 (6H, , CH × 3), 1.84–1.93 (1H, , CH), 2.38–2.46 (1H, m, CH), 2.68 (2H, t, 7.2, CH), 2.81 (4H, s, succinimidyl CH × 2), 3.12–3.21 (2H, m, CH), 3.56–3.65 (1H, m, CH). δ(400 MHz; DMSO) 24.415, 25.830, 28.026, 30.425, 34.217, 38.485, 56.011, 169.287, 170.612. 321.0937 ([M+NH] CHNOS requires, 321.0937). The NHS ester, , was added to the 3′-end of oligonucleotides (5′- FAM CAT TGA AGC TTC (Pr1), 5′- FAM CAT TGA AGC TTC TTT TTT TTT T (Pr2), 5′-FAM CAT TGA AGC TTC AAA AAA AAA A (Pr3) and 5′- ATC CTG AAT GCG AAA AAA AAA A (Pr4)) using 3′-amino-modifier C7 CPG solid supports (1 μmol). The Fmoc protecting group was removed from the 3′-amino group by treatment with 20% (v/v) piperidine/MeCN for 30 min. The NHS ester, (20 mg in 1 ml of MeCN) was applied to the column overnight then washed with MeCN (3 × 1 ml) before being used as the solid support for oligonucleotide synthesis. The oligonucleotide sequence was prepared on a MerMade 6 Nucleic Acid Synthesiser with the FAM introduced as a commercially available terminal phosphoramidite. The oligonucleotides were cleaved from the CPG supports by incubation in 1 ml of conc. NHOH for 3 days at room temperature. The ammonium hydroxide was removed (room temperature) and re-dissolved in HO (1 ml, distilled) and purified by reversed-phase HPLC on a Dionex UVD170U detector fitted with a P680 pump through a Phenomenex Clarity column. Buffer A: TEAA (0.1 M, pH 7), Buffer B: CHCN; T = 0, 95% TEAA: 5% MeCN, with a gradient of 1% min B over 15 min and held at 20% B for 5 min at a flow rate of 1 ml/min. After HPLC purification the fractions were collated and concentrated (room temperature) and redissolved in HO (1 ml, distilled), they were then freeze-dried and redissolved in HO (1 ml, distilled) and stored at 4°C. HPLC analysis of 3′-thioctic acid, 5′-FAM-modified oligonucleotides indicated modification had occurred. Confirmation of this was achieved by reference to MALDI mass spectrometry of a known thioctic acid modified sequence, 3977.6 ([MALDI] CHNOS-ATGCTCAACTCT requires, 3977.4). Sequences were prepared on a MerMade 6 Nucleic Acid Synthesiser with commercially available 3′-thiol-modified universal support (propyl) and FAM introduced at the 5′-end by a commercially available phosphoramidite. The oligonucleotides were cleaved from the CPG supports by incubation in 1 ml of conc. NHOH for 3 days at room temperature. The ammonium hydroxide was removed (room temperature) and re-dissolved in HO (1 ml, distilled) and purified by reversed-phase HPLC on a Dionex UVD170U detector fitted with a P680 pump through a Phenomenex Clarity column. Buffer A: TEAA (0.1 M, pH 7), Buffer B: CHCN; = 0, 95% TEAA: 5% MeCN, with a gradient of 1% min B over 15 min and held at 20% B for 5 min at a flow rate of 1 ml/min. After HPLC purification, the fractions were collated and concentrated (room temperature) and redissolved in HO (1 ml, distilled), they were then freeze-dried and then redissolved in HO (1 ml, distilled) and stored at 4°C. Thirteen nanometre gold nanoparticles were prepared citrate reduction of HAuCl. Thirty seven nanometre silver nanoparticles were prepared citrate reduction of AgNO. A number of options were considered for the attachment of thioctic acid to oligonucleotides. Ultimately, the synthesis of an active ester of the thioctic acid was determined as being the most versatile approach to functionalizing the termini of oligonucleotides. Primary amine groups are readily added to the 3′- or 5′-termini of oligonucleotides and will react with the active ester of thioctic acid either on a solid support or in solution to provide the disulphide functionalized oligonucleotide. In addition the thioctic acid was shown to be stable to the conditions used for oligonucleotide synthesis and deprotection indicating its compatibility with routine oligonucleotide synthesis. The -hydroxysuccinimidyl ester, , of thioctic acid was prepared in good yield by the esterification of the carboxylic acid with hydroxysuccinimide using EDC (). This active ester was then used in two different approaches. Modification at the 3′-terminus was achieved by using an amino-modified solid support. Initially, the Fmoc protecting group was removed using standard piperidine deprotection to give the free amine on the CPG. The active ester was added to the CPG in acetonitrile and left overnight (). Following washing with acetonitrile, the thioctic acid solid support was used with standard phosphoramidite chemistry. For this reason the surface coverage of each of the conjugate systems was assessed (). This was carried out by a method devised by Mirkin . () Triplicates of the FAM-modified oligonucleotide conjugates were treated with DTT (10 mM) and left to completely aggregate. After which they were centrifuged to ensure that all nanoparticulate sediment was separated from the supernatant. Aliquots of the supernatant were taken, again in triplicate, and fluorescence measured. These fluorescence values when correlated with a standard calibration curve gave a concentration for FAM-labelled oligonucleotide in solution. This FAM-oligonucleotide concentration divided by the starting molar concentration of nanoparticle gives a value of oligonucleotides per nanoparticle and dividing by the surface area of the sphere gives normalized surface coverage values which are easily converted to pmol·cm. Interestingly, it was found that D1 actually has less oligonucleotide surface coverage than both TT1 and T1 7.4 ± 0.3 pmol·cm cf. 17.5 ± 0.5 pmol·cm and 12.6 ± 0.6 pmol·cm, respectively. This shows that with no spacer bases the surface coverage by thioctic acid modified oligonucleotide–Au nanoparticle conjugates is less than that of the standard thiol conjugates. In turn, the surface coverage of standard thiol conjugates is found to be less than those that are not treated with DTT prior to conjugation (when the oligonucleotides are added directly to the nanoparticles without treating with DTT and purifying with size exclusion chromatography). This indicates that the conjugation process can be improved when using alkyl-thiolated oligonucleotides as there is no trade-off in conjugate stability or surface coverage as a result of omitting reduction by DTT. With the polyT spacer, D2, there is greater surface coverage of the Au nanoparticle by the oligonucleotide, 59.9 ± 6.7 pmol·cm compared with 12.0 ± 0.3 pmol·cm and 21.1 ± 1.2 pmol·cm for the TT2 and T2 samples. Therefore, it is difficult to argue that the enhanced stability was due to an increased surface coverage since in the previous case, without spacers, there is decreased surface coverage and the enhanced conjugate stability is still observed. Indeed, moving to the polyA spacer, D3, the surface coverage is in line with the thiol systems at 12.6 ± 0.8 pmol·cm cf. 21.1 ± 1.2 pmol·cm and 12.2 ± 0.7 pmol·cm for TT3 and T3. Again, as this disulphide conjugate has the same surface coverage as the standard monothiol the enhanced stability cannot be attributed to surface coverage effects. To demonstrate the versatility of the thioctic acid modified oligonucleotides, silver nanoparticles were used in a similar study. Silver nanoparticles are considerably less stable than gold and as a consequence have been subject to less success in DNA sensing primarily due to the lack of robust surface chemistry. A limited number of studies have been reported but use homo-oligonucleotides or direct hybridization approaches (). shows the degradation of thioctic acid- and thiol-terminated oligonucleotide sequences immobilized on 37 nm citrate-reduced silver nanoparticles. again shows the progressive red-shift of absorbance as a result of aggregation. The thiol system taken at 1 min intervals is clearly less stable than the disulphide which was monitored at 10 min intervals. It should be noted that with oligonucleotide–Ag conjugates rather than a new peak appearing at red-shifted wavelength due to a change in plasmon on the nanoparticle surface, the plasmon broadens and there is a loss of absorbance at 407 nm. For that reason it is less informative to plot an ‘emergence’ profile for the silver conjugates and their stability was assessed by reference to the 407 nm peak. For ease of comparison the ‘half-lives’ of the conjugate systems were calculated, taken as the time required for half the total absorbance change to occur at 407 nm, the results are shown in . With both the thiol and disulphide–Ag conjugate systems aggregation commences immediately upon treatment with DTT (10 mM). There is a marked difference in the rate of aggregation, however, with all of the thiol systems having a half-life of less than a minute, compared with the disulphide examples which have 15–30 min half-lives. Whilst it is difficult to compare gold and silver conjugate systems, we can say that the overall ‘conjugate’ stability displayed by the silver–disulphide systems is more stable (with respect to DTT-induced aggregation) than the ‘standard’ thiol–gold system (see ). Due consideration must be given to the surface coverage of the conjugates as a high surface coverage could explain enhanced stability. As with the gold conjugates the surface coverage data is variable depending upon whether there are spacer bases and what those spacer bases are. For example, the surface coverage of oligonucleotide on Ag nanoparticles for D1–Ag conjugates was found to be 21.1 ± 1.3 pmol·cm, compared with TT1–Ag which have a greater surface coverage, 144.7 ± 14 pmol·cm. Even T1–Ag has a greater surface coverage than the disulphide species at 31.2 ± 1.5 pmol·cm, albeit less than the standard thiol sample. Once again, the disparity in surface coverage does not impact the conjugate stability since we have seen that the disulphide systems are by far more stable than both of the thiol conjugates and yet has lower surface coverage. This surface-coverage-independent stability is also observed with the polyT and polyA sequence conjugates (refer to ). Comparing the results for stability of gold and silver ‘conjugates’ it can be seen that as anticipated, the disulphide does not stabilize the silver nanoparticles to the same extent as gold, due to weaker thiol–silver interactions. Surface coverage effects can again be dismissed as the stabilizing factor when comparing samples D2–Au (with a surface coverage of 59.9 ± 6.7 pmol·cm) and T2–Ag (with a surface coverage of 64.4 ± 3.0 pmol·cm) and yet there are vastly differing stabilities. Similarly, D1–Ag (surface coverage of 21.1 ± 1.3 pmol·cm) is considerably less stable than D3–Au (surface coverage of 12.6 ± 0.8 pmol·cm) despite their surface coverages being quite similar. It is worthy to note, however, that the disulphide on silver remains more stable than the ‘standard’ monothiol linker systems on gold and this is in spite of similar surface coverages in some cases e.g. D1–Ag (surface coverage of 21.1 ± 1.3 pmol·cm) and TT3–Au (surface coverage 21.1 ± 1.2 pmol·cm). This is highly significant as it now allows oligonucleotide–silver nanoparticle conjugates to be exploited in a similar manner to gold nanoparticles. As shown in , hybridization between the disulphide-immobilized oligonucleotide and a fully complementary sequence induces a characteristically sharp melting profile (monitored at 413 nm). This is a particularly exciting result since the hybridization was not carried out in the normal ‘sandwich’ fashion. Instead silver nanoparticles conjugated with a thioctic acid modified sequence, 5′- ATC CTG AAT GCG AAA AAA AAA A MOD 3′ (D4) were hybridized with the ‘unconjugated’ complement. This has been observed previously with gold nanoparticles, but not silver (). This shows the efficacy of the thioctic acid terminated oligonucleotide–Ag conjugates and their potential for employment in DNA detection. Furthermore, investigations are underway to establish oligonucleotide–Ag nanoparticle conjugates for DNA detection purposes in a manner similar to that of the established gold analogues. This is particularly attractive as silver nanoparticles have a higher molar extinction coefficient than gold.

Genetic variation drives evolution. Nature shapes life by selecting genotypes with increased ‘fitness’ for the encompassing environment. In a rather general sense, a gain in understanding of the association between genetic variation and its phenotypic effects is therefore a step toward grasping how nature acts and how life evolves. More practically, the study of this association may lead to understanding what causes various disorders, such as diabetes or cancer, to appear. A very large portion of genetic variation is represented by single nucleotide polymorphisms (SNPs). It has been estimated that as many as 93% of all human genes contain at least one SNP and that 98% of all genes are in the vicinity (±5 kb) of a SNP (). SNPs occur in both regions that code for proteins (coding SNPs) and in regions that do not [non-coding; note that many non-protein-coding regions are likely to be transcribed as active RNAs (,)]. While non-coding SNPs are trivially more prevalent, an estimated 24 000–60 000 coding SNPs are found in the human genome (,). Protein-coding SNPs can be further divided into synonymous and non-synonymous (nsSNPs): synonymous SNPs, due to degeneracy of genetic code, do not change the amino acid sequence of resulting protein while non-synonymous SNPs do. Although non-synonymous SNPs generally have the most obvious functional/biochemical effects, they do not necessarily associate with functional or structural consequences. Non-synonymous SNPs are known to cause numerous diseases. For example, a point mutation in the hemoglobin beta gene (substitution of glutamic acid by valine) is one cause for sickle cell anemia (). Other diseases, such as diabetes, have been correlated with a number of SNPs, but their main genetic factors have yet to be selected from pools of available candidates. Disease association studies produce lists of SNPs implicated in a particular disease, but typically these ‘implications’ span tens of genes and non-coding regions. Experimentation designed to narrow down these long stretches to a set of the most likely candidates is generally reliable but not guaranteed to succeed. Additionally, it requires significant amounts of time and may be costly. For instance, one study of nsSNPs in leukemia-associated Rho guanine nucleotide exchange factor (LARG) (), thought to be involved in type 2 diabetes, required the completion of various health assessments and genotyping of selected SNPs in DNA of over 1600 individuals. In addition, functional assays of the single selected mutant thought to be functionally non-neutral (Y1306C) merited the transfection of mutants into specifically maintained cell lines. Breaking through from a large pool of candidate SNPs into the light of identifying the mutations causing a disease is a laborious quest. One challenge on this path is the prioritization of suspected nsSNPs according to their likely effects on function. Meeting this challenge requires the detection of SNPs that may have little effect in isolation but do damage in concert with others. While experimental methods are more reliable, they are also more cumbersome than computational studies. In fact, they are likely to benefit in efficiency and speed from application of some pre-filtering with predictions of nsSNP effects. Such predictions could be applied in general to studies of mouse genetics; they would benefit the elucidation of human mono-SNP and complex phenotype disorders, as well as evolutionary genetics. Computational methods may never be accurate enough to replace wet-lab experiments; however, they may help in selecting and prioritizing a small number of likely and tractable candidates from pools of available data. Recent studies () have shown that computational evaluation of certain protein character changes associated with nsSNPs is capable of giving good estimates of their functional effects. A range of approaches to classification was considered by these studies including use of machine-learning [SVM (,), decision trees (), neural networks (), random forests (), Bayesian models (), statistical approaches (,,)], and rule-based systems (). Similarly diverse were the types of input information used for the prediction: Some methods are applicable to all sequences and variations (e.g. SIFT (), which uses only mathematical computations for making inferences from alignments), while others require specific types of information, such as the coordinates of three-dimensional (3D) protein structures (e.g. SNPs3d (), which utilizes SVMs to recognize structural patterns). Some methods combine available information to improve the classification [e.g. PolyPhen (), which uses a rule-based cutoff system on available data including 3D-structures, SWISS-PROT annotations, and alignments]. All methods perform relatively well when applied to the data sets on which they were developed. Despite all of those solutions, the problem of predicting functional effects of nsSNPs is not solved. Aside from the desired improvement in accuracy of prediction (which, if significant enough, would make these methods useful in medical applications such as genetic counseling), the field would benefit from a comparison of performance across methods. As is, we cannot compare the performance of different methods from existing publications, because of the variety of testing sets and evaluation measures. One of the goals of this work was to compile a larger and more diverse data set that could fairly be used for evaluation of methods that require similar types of inputs. Despite all limitations of existing methods, quite a few experimentalists () describe using them for facilitating their research practices. For others, the reluctance to rely on results produced through computation is likely the relative lack of control over the ‘black box’ predictions. Making the basis of predictions clear, as might be possible with decision tree-based algorithms, would allow the researchers to select only those cases, which they are willing to believe. However, the accuracy of classification of the existing tree-utilizing methods [e.g. the Krishnan . () implementation] appears to be lower than that of other available tools. Yet, even these improved methods do not evaluate all nsSNPs equally well. For experimentalists this translates into a real possibility of getting the wrong prediction for the one mutant that they might really be interested in, without even a hint at a possible misclassification. As an alternative to providing user control at the cost of accuracy and applicability, we propose utilizing a reliability scale of predictions. While this approach may not explain the reasoning behind assignment of a mutant to either functional class, it will simplify the choices made on the basis of predictions. Here we described a novel method, SNAP (creening for on-cceptable olymorphisms), that could potentially classify all nsSNPs in all proteins into non-neutral (effect on function) and neutral (no effect) using sequence-based computationally acquired information alone. For each instance SNAP provides a reliability index, i.e. a well-calibrated measure reflecting the level of confidence of a particular prediction. SNAP is a neural network-derived tool that accurately predicted functional effects of nsSNPs in our newly compiled data set by incorporating evolutionary information (residue conservation within sequence families), predicted aspects of protein structure (secondary structure, solvent accessibility), and other relevant information. All information needed as input was obtained from sequence alone. SNAP refined and extended previous machine-learning tools in many ways, e.g. by the extensive data set used for the assessment, by the particular approach to data handling, and by its ubiquitous applicability (to sequences from all organisms, proteins with and without known structures, and entirely novel SNPs in scarcely characterized and un-annotated families). Additionally, SNAP outperformed the competitors throughout the spectrum of different accuracy/coverage thresholds and correctly estimated its own success through the reliability index. The importance of the later is that users will be able to focus on the subset of predictions that are more likely to be correct; they will also know if one of the mutants implicated in a malfunction was predicted to be deleterious with low confidence. #text xref #text Many studies noted that the location (e.g. buried/exposed) of a residue within the 3D structure (,,,) is relevant for the effect of a particular substitution on function. We used this observation by dividing the available mutants into three sets based on predicted solvent accessibility (buried = <9% exposed surface area, intermediate = >9 and <36%, exposed = >36%). Different evolutionary pressures exist for residues of different accessibility; this in turn requires the use of slightly different input features for the prediction. Although the thresholds chosen for this split were relatively arbitrary, they provided a good estimate of actual classes of accessibility. The numbers of mutants belonging to each set were ∼35 000, ∼25 500 and ∼21 000, respectively. Notably, the fractions of neutral to non-neutral substitutions were markedly different by class (0.75, 1.06 and 1.6, respectively). While our solution suffered from mistakes in predicting accessibility, it had the important advantage of generating a data set that was many orders of magnitude larger than any other set that has ever been analyzed with respect to the accessibility of mutants. For all testing purposes we split each of the three accessibility-grouped data sets, as well as the full data set, into ten subsets such that no protein in one set had HSSP-values >0 to any protein in another set (note that for alignments of >250 residues this implied that no pair of proteins had over 21% pairwise sequence identity). No other limitations were imposed on contents of each set. For each group of ten, we then used eight data sets for training (optimizing the free parameters), one for cross-training (determining the point at which training was stopped), and one for testing. Finally, we rotated through all sets such that each protein was used for testing exactly once. We used standard feed-forward neural networks with momentum term described elsewhere in detail (). All free parameters of the networks were chosen without ever considering the performance of the test sets. Instead, free parameters were optimized on the training (optimizing connections) and cross-training (optimizing architectures/stop training) sets that had no significant overlap to the test sets. We also applied support vector machines [SVMs ()], however, this worked slightly worse in our hands. Note that we trained SVMs using the same features as those selected for the best-performing neural network and attempted to optimize some of the free parameters on the cross-training data set. While the resulting SVM-based method was very accurate it performed somewhat worse than a comparable neural network-based method. sub disp-formula xref #text italic xref disp-formula #text disp-formula #text Finally, we trained a single network for all classes of solvent accessibility (buried, intermediate, exposed); we included all features that helped in any of the classes. The final SNAP network architecture included the following input features: explicit PSI-BLAST frequency profile, relative solvent accessibility predictions (PROFacc), secondary structure predictions (PROFsec), sequence-only predictions of 1D structure (PROFsec/PROFacc), Pfam information, PSIC scores (Methods section), predicted residue flexibility (PROFbval), and transition frequencies (likelihood of observing the mutation particular mutation imposed by the SNP). With a window length of five consecutive residues, this yielded neural networks with 195 input and 50 hidden units. Note that none of those parameters were optimized on the test sets for which we report performance; instead they all were optimized on the cross-training set (Methods section). Evaluated on the PMD/EC data (data sets produced by merging PMD and enzyme data; Methods), SNAP reached a higher level of overall two-state accuracy [78%, Equation ()] than SIFT (74%) and PolyPhen (75%; ). Given an estimated standard error below two percentage points, this suggested that SNAP outperformed SIFT and PolyPhen. The inclusion of SWISS-PROT annotations and SIFT predictions into the input vector of SNAP (SNAP in ) improved performance even further for both un-annotated proteins (79%; note: only SIFT predictions) and annotated ones (81%). Other methods appeared to reach similar levels of accuracy suggesting that the majority of mutants in the data set were classified similarly by all methods. This, however, turned out to be a rushed inference, e.g. SNAP correctly predicted when the one or both of the others (SIFT/PolyPhen) were wrong ∼1.7 times more often than vice versa (SNAP right 10 124 times when SIFT or PolyPhen were wrong; SNAP wrong 6117 times when SIFT or PolyPhen were right.) In other words, for a large subset of the PMD data set (16 241 mutants) for which at least one method was wrong (i.e. mutants with hard-to-establish functional effects), SNAP achieved an overall accuracy of 62.3%, while both PolyPhen (7966 right and 8275 wrong) and SIFT (7566 right and 8675 wrong) attained levels <50% accuracy. All our optimizations were performed on PMD/EC data sets (Methods). We carefully avoided over-optimistic estimates by a full rotation through three-way 10-fold cross-validation: training set for optimization of network connections, cross-training set for optimization of all other free parameters (hidden units, type of input, etc.), and the test set for assessing performance. Despite having applied this time-intensive caution, we wanted to test yet another independent data set. Overall, the PMD/EC-based estimates of performance for SNAP were confirmed by the other data sets (), namely for the LacI repressor, bacteriophage T4 lysozyme, HIV-1 protease, and human Melanocortin-4 receptor data (Methods). Although these data sets comprehensively sampled the space of possible mutations and carefully evaluated their effects, they covered a minute fraction of the entire sequence space (four proteins; only one from a mammal) and may therefore be less representative than the PMD/EC data. While these additional data were too limited to suggest firm conclusions, they helped to confirm trends. All methods performed slightly less accurately in terms of the average over all these data than over our PMD/EC data. The only overlap between these data sets and the PMD/EC data was in about one quarter of the LacI mutants (they all were contained in the data used for the development/assessment of SIFT and PolyPhen). SNAP still outperformed the competitors on LacI repressor, Lysozyme, and Melanocortin-4 data. The performance was radically different for the viral sequence: PolyPhen produced no predictions for any of its mutants and SNAP performed clearly worse than SIFT. This disparity might originate from the different features used by each method: SIFT bases its predictions only on alignments. In contrast, PolyPhen and SNAP also consider other characteristics (e.g. estimates of secondary structure, functional regions) that may have been misleading for this particular case. The SNAP predictions were not binary (neutral or non-neutral); instead, they were computed as a difference between the two output units (one for , the other for ; difference ranged from −100 to 100). ‘Dialing’ through different decision thresholds (which difference yields a prediction of ‘neutral’?) will enable users to decide which of the two flip sides of the same accuracy-coverage coin is more important to them. On the one end of the spectrum, predictions with very high differences are very accurate but will cover very few of the SNPs. On the other end, predictions with very low differences will capture all SNPs but this will be paid for with a significantly reduced accuracy. Thus, dialing through the thresholds generated ROC-like curves for accuracy versus coverage (). While the predictions of SIFT are also scaled, the SIFT performance has been optimized for the default cutoff (arrow in ). PolyPhen predictions do not have numerical values; instead they are sorted into four categories (benign, possibly damaging, probably damaging and unknown). Thus, given our assumption of classifying unknowns into the benign group, only two points exist on PolyPhen's graph: one that sorts all damaging values into the non-neutral class and another that assigns ‘possibly damaging’ SNPs to the benign category. Since ROC-like curves were only available for our methods and only partially for SIFT we didn't compare methods by their ‘area under the curve’ values. However, SNAP clearly outperformed both PolyPhen and SIFT throughout the ROC-like curves (). SNAP was only outperformed by SNAP, i.e. the version that also considered SIFT predictions and—when available—SWISS-PROT annotations. The SNAP's two output units provide for an additional measure of confidence of prediction. Intuitively, it is clear that a smaller difference between the values is indicative of lower confidence in the prediction. This is also the reason why accuracy and coverage of predictions are always at a tradeoff. Higher accuracy predictions are received by sampling more at more reliable cutoffs, thereby reducing the total number of trusted samples. To define this reasoning more precisely we introduced the reliability index measure [RI; range 0–9, Equation ()]. Higher reliability indices correlated strongly with higher accuracy of prediction. However, the majority of predictions is made in the middle of the index range (e.g. RI = 5, ). The SNAP performed better at correctly predicting non-neutral SNPs in the core of proteins (buried residues) than those at the surface (exposed residues, Figure SOM_2, Supplementary Data). This may be due to better-defined constraints responsible for functional consequences of altering buried residues (e.g. structural constraints). Additionally, it may be due to the fact that there were significantly more non-neutral training samples in the data set localized to the buried regions. Different shape of the accuracy/coverage curves for the neutral and non-neutral samples is the result of different ratios of neutral to non-neutral samples in various accessibility classes (Methods section). Although many studies have confirmed the importance of structure for functional integrity, none have analyzed mutants in different of solvent accessibility (buried, intermediate, exposed). The observation that the optimal input features differed between these three classes (Figure SOM_1, Supplementary Data) supported the intuition of any structural biologist, namely that different biophysical features govern the type of amino acid substitutions that disrupt function of surface and of core residues. For instance, the entire profile/PSSM was useful to determine the effect of substitutions on the surface, while it sufficed to consider the frequencies of the original and the mutant residues at the SNP position to determine the effects in the core. This differential behavior could partially be explained by that internal residues are more constrained by evolution; therefore, it would suffice to know whether or not the mutated residue ever appears in an alignment of related proteins. Surfaces, on the other hand, tend to be less conserved, i.e. many alternatives may be valid; hence, all of these (full PSSM) must be considered to determine the effect of a substitution. Another example of the difference is the importance of predicted 1D structure (secondary structure and relative solvent accessibility) features for the prediction of SNP effects on the surface (Figure SOM_1, Supplementary Data). Similar findings have been reported by other studies focusing on the prediction of very different aspects of protein function (,,). These 1D features may be particularly important to determining whether a given position is part of a functional site. Active site involvement of a buried residue, on the other hand, can hypothetically be approximated through assessing its flexibility. Indeed, this would also explain why the predicted flexibility appeared relatively more relevant for buried residues. Fewer and less-descriptive features required by the intermediate residue network may indicate the need for structural or functional characteristics that were not tested and/or for finer gradation of the data in terms of accessibility. The types of predicted features of protein structure and function that we used as input contributed to the significant improvements of SNAP. However, the most important single feature, other than the biophysical nature of the mutant and wild type amino acids, was the conservation in a family of related proteins (as measured by the PSIC conservation weight). This finding confirmed results from statistical methods applied by Chasman and Adams (). If more detailed aspects of structure and function were available for all proteins, these could be input to networks to improve performance. For instance, Yue and Moult (), have demonstrated that an SVM (Support Vector Machine, i.e. another machine-learning algorithm) can distinguish mutants involved in monogenic disease by considering structural features of the protein affected by the mutation (e.g. breakage of disulfide bond or over-packing). PolyPhen also uses available structural features of either the query protein or its homologue (>50% pairwise sequence identity to experimental structure required); its structural features include solvent accessibility, secondary structure, phi-psi dihedral angles, ligand binding, inter-chain contacts, functional residue contacts (as annotated in SWISS-PROT), and normalized B-factors. For proteins for which this data is available, PolyPhen slightly out-performed both SNAP and SNAP (B, ‘Structure’). However, these cases constituted a very small fraction of all the examples for which we had experimental information about SNP effects (A). The performance of SNAP improved significantly through using available annotations (). SWISS-PROT indications of active site, mutagen, transmembrane, binding, and otherwise important regions (Methods) allowed for better identification of possibly non-neutral mutations. For instance, using SNAP to predict the effects of SNPs in Melanocortin-4 corrected 2 of the 13 incorrect SNAP predictions of non-neutral effects (data not shown). In the absence of annotations for structure and function, the most valuable information is extracted from alignments and family/domain data (). A similar finding has been reported previously; namely in the absence of known 3D structure, evolutionary information was most relevant to predict SNP effects (). The distinction of mutants according to the severity of functional effects illustrated the performance of SNAP from a different angle. The difference between the two output nodes of SNAP ranges from −100 to 100; differences ≥1 are considered non-neutral by default. Higher differences correspond to more reliable predictions (). SNAP was trained only on experimental data of binary nature, i.e. the supervised output was either labeled as non-neutral or as neutral. Did the network learn implicitly to distinguish between more and less severe effects? We did not have enough data to analyze this question rigorously. We did however have limited data sets (LacI repressor data set containing 4041 mutants) and a limited grading of severity (neutral/slightly damaging/damaging/severe) to explore this question. SNAP clearly performed better on more severe effects (, red bars dominate on the right hand side) and clearly ‘more neutral’ (, green bars dominate left hand side) changes than those with intermediate effects. This suggested that the difference between the output values did not only reflect the reliability of predictions, but also the severity of the change. Put differently, more severe effects corresponded to stronger SNAP predictions. The reason why SNAP implicitly learned about the severity of effects was likely of statistical nature: the most severe and most neutral mutations were most consistent in the data set. There is a number of nsSNPs in the PMD data set that are known to introduce ‘gain of function’ for a particular protein. Unfortunately, many mutations entail a gain of one function, but a loss, or retention at same levels, of another. Additionally, modifications that lead to loss of function in one protein may very well correlate with a gain of function in another (e.g. increased structural flexibility may suppress or promote function). Given this reality, we chose not to separate out directions of effects of mutation: for the purposes of SNAP a gain of function is treated as a non-neutral sample. However, to illustrate that SNAP can recognize these mutants consider an example of a few nsSNPs of the metalloendopeptidase thermolysin (EC.3.4.24.). A study () of this enzyme, described in PMD, showed that set of 18 nsSNPs of thermolysin increase its activity (‘gain of function’). SNAP correctly identified all of these mutations as non-neutral, with reliability index range 0–5. Although SNAP is already capable of recognizing gain of function changes, it would likely benefit from seeing more of this sort of mutants. However, extensive data of this type is not currently available. Two features of SNAP make it particularly useful to researchers who want to scan large experimental data sets. The first is SNAP's particular strength is the correct predictions for the least obvious cases (those for which existing methods disagree). For such mutants SNAP was 13–17% points more accurate than other methods. Furthermore, these mutants are generally also the hardest to pinpoint experimentally since their subtle effects are more likely to contribute to a phenotype rather than fully account for it. In a typical experimental scenario in which experimental observations are likely to have already been subjected to various analysis tools, this improvement is likely to be extremely relevant and significant. For example, for the Melanocortin-4 receptor mutants (Methods and ; C. Vaisse personal communication), SNAP recognized significantly more damaging mutations than other methods. For instance, the Arg18Cys mutation, known to decrease protein basal activity (), was found to be non-neutral only by SNAP. Similarly, PolyPhen and SIFT also failed to recognize the deleterious Asn97Asp mutant, known to strongly affect ligand binding (). Overall, SNAP was wrong in only two of the nineteen instances in which at least one of the four methods tested was right; in comparison SIFT was wrong 10/19, PolyPhen 13/19, and SNPs3D 8/19. Had SNAP been available earlier, it would have been a significantly better choice for selecting candidates than any other method. The analysis of the Melanocortin-4 receptor appeared representative in light of a similar analysis for our large data set. The second important novel feature of SNAP is the introduction of a reliability index that correlates with prediction accuracy () and allows filtering out low accuracy predictions. In addition to providing an estimate about the accuracy, the reliability index also reflects the strength of a functional effect (). This feature is entirely unique for SNAP. We established that training separate networks on separate classes of solvent accessibility was beneficial, in principle. Unfortunately, the split yielded too small data sets. We expect to need 2–5 times more accurate experimental samples to improve performance. Furthermore, even the available data were not ideal as illustrated by the differences of functional assignments for the 55 mutants from the LacI repressor between the data set used for SNAP and those used for SIFT: 51 of these were considered as in training SNAP, while they were classified as by the authors of SIFT. We found that these assignments came from different studies and not from annotation mistakes. Different experimental methods and/or interpretations of results can introduce noise. Moreover, these data tend to be particularly inconsistent across species and across protein families, since a qualitative description of what constitutes an important change often differs across these experimental territories. d e v e l o p e d S N A P , a n e u r a l - n e t w o r k b a s e d t o o l t o b e u s e d f o r t h e e v a l u a t i o n o f f u n c t i o n a l e f f e c t s s i n g l e a m i n o a c i d s u b s t i t u t i o n s i n p r o t e i n s . S N A P u t i l i z e s v a r i o u s b i o p h y s i c a l c h a r a c t e r i s t i c s o f t h e s u b s t i t u t i o n , a s w e l l a s e v o l u t i o n a r y i n f o r m a t i o n , s o m e p r e d i c t e d — o r w h e n m a d e a v a i l a b l e o b s e r v e d — s t r u c t u r a l f e a t u r e s , a n d p o s s i b l y a n n o t a t i o n s , t o p r e d i c t w h e t h e r o r n o t a m u t a t i o n i s l i k e l y t o a l t e r p r o t e i n f u n c t i o n ( i n e i t h e r d i r e c t i o n : g a i n o r l o s s ) . A l t h o u g h s u c h p r e d i c t i o n s a r e a l r e a d y a v a i l a b l e f r o m o t h e r m e t h o d s , S N A P a d d e d i m p o r t a n t n o v e l t y . A m o n g s t t h e n o v e l a s p e c t s w a s t h e i m p r o v e d p e r f o r m a n c e t h r o u g h o u t t h e e n t i r e s p e c t r u m o f a c c u r a c y / c o v e r a g e t h r e s h o l d s a n d t h e p r o v i s i o n o f a r e l i a b i l i t y i n d e x t h a t e n a b l e s u s e r s t o e i t h e r z o o m i n t o v e r y f e w v e r y a c c u r a t e p r e d i c t i o n s , o r t o k n o w i n g l y b r o a d c a s t l e s s r e l i a b l e o n e s . T h e i m p r o v e d p e r f o r m a n c e t r a n s l a t e d t o m a n y u n i q u e a n d a c c u r a t e p r e d i c t i o n s i n o u r d a t a s e t . W e b e l i e v e t h a t b e t t e r f u t u r e e x p e r i m e n t a l d a t a w i l l d i r e c t l y t r a n s l a t e t o b e t t e r p e r f o r m a n c e o f a n y p r e d i c t i o n m e t h o d . I n t h e m e a n t i m e , e x p e r i m e n t a l i s t s m a y a l r e a d y s p e e d u p t h e i r r e s e a r c h b y u s i n g o u r n o v e l m e t h o d . p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

xref italic #text The C’10 and A’7H cell lines were cultured in DMEM medium (GIBCO BRL, Eragny, France) as previously described (). The C’10pI2 and A’7HpI4 clones, the C’10μ2 and A’7Hμ2 clones and the C’10μDB and A’7HμD5 clones were obtained after transfection with the empty pIRESpuro2 vector (Clontech, Heidelberg, Germany) or the pIRESpuro2 vector containing the cDNA coding for the human wild-type DNA polymerase μ (polμ) or the inactive form of DNA polymerase μ (polμD), respectively. Individual clones were obtained after transfection with jetPEI (Qbiogen, Illkirch, France) and puromycin selection (5 μg/ml). Polμ expressing plasmids pIRES-polμ and pIRES-polμD were constructed by PCR amplification from a P17-His vector plasmid carrying the cDNA sequence of the polμ gene (). The polμ catalytically inactive mutants were constructed from P17-His plasmid, using the Quick Change mutagenesis kit (Stratagene, La Jolla, CA, USA) according to the manufacturer's instructions. Residues D330 and D332 in polμ were replaced by Ala using two primers also containing a new BsmI site used for mutation selection. The resulting constructs were sequenced and error-free vectors were used to transfect cells. The WT and catalytically mutated forms of polμ on p17 plasmid were sequenced and introduced into BL21(DE3) bacteria before purification. The different polμ proteins were purified by Nickel chelation chromatography according to Novagen's suggestions with the following modifications. Cells were lysed by freezing for 16 h at −70°C and slowly thawing on ice in the presence of 20 mg/ml lysozyme and 1% Triton X100. Proteins were concentrated and stored in buffer containing 50 mM Tris pH 7.5, 100 mM NaCl, 0.1 mM EDTA, 1 mM DTT and 20% Glycerol. Protein purity was estimated as >90% by visual inspection of Coomassie Blue-stained 12% SDS–polyacrylamide gels. Standard primer extension reactions were performed at 37°C for 1 h as previously described (). Total cellular protein extracts (50 μg) were separated by 10% SDS-PAGE (PolyAcrylamide Gel Electrophoresis) and transferred to PVDF membrane. Polμ was detected using polyclonal antibody (1/500) (AbCam, Cambridge, UK) followed by incubation with horseradish peroxidase (HRP)-conjugated anti-rabbit IgG, and revealed using the ECL system (Amersham, Freiburg, Germany). Equal loading was determined using monoclonal anti-actin antibody (1/5000) (AC10, Sigma, Lyon, France). Cells (5 × 10) were plated and allowed to attach overnight. Expression of the meganuclease I-SceI in cells was achieved by transient transfection of the expression plasmid pCMV –I-SceI () using Jet-PEI (Q-BIOgen, Illkrich, France). Seventy-two hours after transfection, cells were dissociated with PBS/EDTA (50 mM), washed in PBS and fixed in PBS/2% PAF for 20 min at room temperature. Cells were then stained with anti-H2Kd (1/30 mouse isotype, SF1-1.1, Pharmigen), or anti-CD4 (1/30 rat isotype, H129.19, Pharmigen), or anti-CD8 (1/30 rat isotype, 53-6.7, Pharmigen) for 30 min at room temperature in PBS/0.5% BSA. Cells were then incubated with anti-mouse-FITC (1/530 mouse isotype, F-2761, Molecular probe) for 30 min at room temperature. The frequency of H2K events allowed us to estimate the efficiency of I-SceI transfection and activity. Scoring of the NHEJ events affecting CD4 and CD8 was performed by flow cytometry using FACScan (Becton Dickinson, San Diego, CA, USA). Cells were dissociated with PBS/EDTA (50 mM) and washed with PBS/0.5% BSA/2 mM EDTA. Cells (1 × 10) were stained with anti-CD4 for 15 min at room temperature, then incubated with beads coated with goat anti-rat IgG (Miltenyi Biotec, Bergisch Gladbach, Germany) in PBS/0.5% BSA/2 mM EDTA for 15 min at room temperature. After washing, the positively stained cells were separated onto miniMACS columns and enriched by ∼30%. Genomic DNA was prepared from a population of CD4 enriched cells (Puregene, Gentra Systems, Hameenlinna, Finland), and junctions of deletion events (CD4+) were amplified by PCR with the following primers (CMV1: 5′-TGGCCCGCCTGGCATTATGCCCAG-3′ and CD4int: 5′-GCTGCCCCAGAATCTTCCTCT-3′). PCR products of 800 bp were obtained after 35 cycles of amplification with Taq polymerase (Biolabs, Beverly, MA, USA) (). The PCR products were cloned into Topo-TA system (Invitrogen, Carlsbad, CA, USA), which allows isolation of individual clones, and sequenced on one strand. Cytotoxicities were determined by clonogenic assay (). For IR, cells (400 or 800) were plated in 25 cm flasks and allowed to attach overnight. Cells were subsequently irradiated in growth medium with a Co-source irradiator (IBL 437C type H, Oris industries SA, Gif sur Yvette, France). For camptothecin (CPT), mitomycin C (MMC) and methyl methanesulfonate (MMS) (Sigma, France) cytotoxicities, cells were plated in 6-well plates and allowed to attach overnight. Cells were treated in growth medium with increasing concentrations of drug for 24 h (for CPT) or 1 h (for MMS and MMC). After 8 days post-treatment incubation, the plates were washed with PBS and colonies were fixed with methanol, stained with crystal violet solution and colonies of more than 50 cells were counted. Survival was expressed as the plating efficiency of treated cells relative to the untreated control cells. The results are the mean ± SD of three independent experiments. Where absent, the error bars are smaller than the symbols. Cells were irradiated at 2 Gy, and metaphase spreads were prepared as described (). Chromosomal aberrations were analysed using a Nikon microscope (TE300) (objective × 100). Results are the sum of four separate preparations of metaphases resulting from two independent irradiation experiments. Student's paired analysis was used to examine differences between two sets of results for the measurement of frequencies of NHEJ events. The genetic instability frequency was assessed for significance by χ testing. Trend -values of less than or equal to 0.05 were considered significant. An inactive form of polμ (polμD) was generated by changing the amino acids Asp330 and Asp332 to Ala. Primer extension assay confirmed the absence of DNA polymerase activity of this protein (B) (). We generated stable cell lines over-expressing the different forms of polμ in Chinese Hamster Ovary (CHO) cell lines. We used two CHO cell lines (C’10 and A’7H) that contain an integrated chromosomal substrate designed to allow specific generation of DSB by the rare cutting enzyme I-SceI (). The different isoforms of polμ were expressed approximately four times more as compared to the corresponding parental cell lines (A). Since the antibody we used was not previously characterized, we performed western blotting with purified polμ protein and extracts from cells over-expressing polλ, the homolog of polμ, as positive and negative controls (see Supplementary Data). We first examined sensitivity to IR because increased sensitivity to IR is a hallmark of defective DSB repair by non-homologous end joining (NHEJ). Cells expressing polμD showed increased sensitivity to IR compared to control cells (C and D). Surprisingly, cells over-expressing the WT form were as sensitive to IR as the polμD expressing cells (C and D). We also tested cytotoxicity to additional DNA damaging agents such as MMS, MMC and CPT, and we did not observe any differences in cell toxicity between control cells and cells expressing the different forms of polμ (). Since NHEJ deficiency leads to chromosomal aberrations, we examined genetic instability by performing karyotype analyses on metaphases from cells exposed to IR. Scoring of chromosomal aberrations included chromosome breaks, triradial, dicentric and end-to-end fusion of chromosomes (). Chromosomal analysis showed that the expression of polμD significantly increased ( < 0.05) the genetic instability observed in IR-treated cells as compared to control cells (). Of note is that over-expression of polμ WT form did not induce a statistically significant increase of genetic instability as compared to control cells ( > 0.05). We use two different cell lines to analyse the involvement of polμ on different types of DSB repaired by NHEJ. Both cell lines possess two integrated I-SceI sites. In the absence of I-SceI expression, only the H2-Kd gene is expressed. After transient transfection with a plasmid coding for I-SceI, cleavage at the I-SceI sites leads to excision of the internal fragment H2-Kd/CD8. Subsequent joining of the DNA ends then leads to two different measurable events. Either the pCMV promoter is ligated to the CD4 fragment, leading to deletion of the excised fragment and expression of the CD4 gene (left portion of A), or the pCMV promoter is ligated to the excised fragment after inversion, leading to expression of the CD8 gene (right portion of A). In the C’10 cell line the two I-SceI sites are in direct orientation, creating complementary ends for deletion events that generate CD4 gene expression, but incompletely complementary ends for inversion events (CD8) (B). No significant differences were observed between cells expressing the different isoforms of polμ regarding the frequency of both CD4 ( = 0.2 and = 0.35 for polμ and polμD expressing cells compared with parental cells, respectively) and CD8 events ( = 0.3 and = 0.17 for polμ and polμD expressing cells compared with parental cells, respectively) (D). In the second cell line (A’7H) the two I-SceI sites are in inverted orientation, creating incompletely complementary ends for deletion events (CD4), but complementary ends for inversion events (CD8) (C). As observed for the C’10 cell line, no significant differences were observed with the A’7H cell line regarding the frequency of both CD4 ( = 0.45 and = 0.1 for polμ and polμD cells compared with parental cells, respectively) and CD8 events ( = 0.16 and = 0.34 for polμ and polμD cells compared with parental cells, respectively) (E). Although no significant differences in the generation of CD4 and CD8 events were observed between cell lines expressing the different isoforms of polμ, we decided to examine the quality of the repaired end junctions. We sequenced the I-SceI break site from CD4 positive cells generated from complementary (C’10 cell line) or partially complementary (A’7H cell line) DNA ends. The analysis of the junction sequences obtained from C’10 cell lines enriched for CD4 gene expression, focused on the repair of complementary ends for which no polymerization event is normally required (A). The sequences obtained showed no significant differences in junction fidelity. In all cell lines, we observed the same proportion (62%) of error-free repair (B). In three cell lines, a lower proportion (38%) of sequences showed erroneous repair. In most cases, deletions were observed. However, the largest deletions (more than 200 bp) were detected only in the control cell line (B). We observed the addition of one nucleotide at the break site only in one sequence from the polμD expressing cells. In summary, we consider that the repair junctions are similar in the cell lines over-expressing either the WT polμ or the polμD isoform, revealing that polμ has no impact on the repair of fully complementary ends. We then sequenced the repair junctions from CD4 enriched A’7H cells for which the junction involved partially complementary ends (A). Sequence analysis revealed an increased percentage of deletions (10–29%) in cells over-expressing the WT or the polμD as compared to the control cell line (C). These results showed that in contrast to the C’10 cell line, both forms of polμ induced a decrease in the use of the four protruding nucleotides. Precise examination of the processing of the four protruding nucleotides can be done and three classes of events delineated, depending on the partial pairing of the four protruding nucleotides, leading to gaps (class 1), mismatches (class 2) and flap structures (class 3) (B). We observed that the over-expression of the WT polμ form increased the percentage of class 1 events (from 74 to 90%), whereas such events were reduced in the polμD cells (from 74 to 50%) as compared to the control cell line (D). Concerning the class 2 events, we observed the opposite result: cells expressing the WT polμ displayed a decrease (from 21 to 10%) of such events, which were conversely increased (from 21 to 40%) in polμD cells. Analysis and interpretation of the class 3 events were not possible due to the low number of sequences obtained. This analysis showed symetrical opposite effects of the expression of the two different isoforms of polμ, the WT form of polμ facilitating generation of class 1 events that are the events requiring gap filling by DNA synthesis. On the other hand, the expression of polμD preferentially drove processing of the protruding nucleotides to the class 2 events that do not require DNA synthesis, but rather mismatch pairing. The use of an inactive form of polμ allowed us to investigate the involvement of polμ in the processing of DNA DSB. As previously observed for polλ (), expression of an inactive form of polμ sensitized cells to death by IR. We surmize that the absence of phenotype to IR treatment of polλ, polμ and polλ /polμ cells (), could be due to redundancy of other DNA polymerases, or to the adaptation and compensatory effect of another DNA repair pathway that handles DNA damage. Cell death by IR was similar in cells expressing the WT form of polμ and cells expressing the polμD (C and D). This could result from an inhibitory effect of polμ, preventing the repair of DNA damage by other DNA polymerases specialized in DSB repair, such as polλ, or by an alternative DNA repair pathway. Indeed, if the presence of a BRCT domain in polμ is responsible for the interaction with the NHEJ factors (,), a dominant negative effect could be caused by the over-expression of the BRCT domain of wild-type polμ that could titrate some of the NHEJ factors. The absence of an effect on cytotoxicity to other DNA damaging agents (MMS, MMC and CPT), favours the hypothesis of polμ being a specialized DNA polymerase for processing IR-induced damage. Interestingly, Mahajan . observed an induction of polμ expression after exposure to IR, but not after exposure to UV or MMC (). Although the cells over-expressing either form of polμ showed the same profile of IR sensitivity, only cells expressing the inactive form of polμ presented a significant higher chromosomal instability in response to IR (). Such a discrepancy between the two cell lines suggests that over-expression of the WT polμ form did not produce the same clastogenic events as polμD expression. However, the two types of events produced were equally deleterious in the two cell lines. We subsequently analysed the frequency and quality of some NHEJ events following specific induction of DSB by the I-SceI rare cutting enzyme. The scoring of inversion and deletion events after induction of DSB did not show significant differences in cells expressing the WT or the inactive form of polμ. No differences were observed in the frequency of NHEJ events, whether the DNA ends generated in the two different cell lines were complementary or partially complementary. The results obtained with the polμD form are in accordance with the previous results obtained when comparing polλ, polβ and polμ in the same system (). Altogether, these results show that the joining efficiency of the different DNA ends is not affected, providing evidence that the over-expression of the different forms of polμ is not acting as a dominant negative of the overall NHEJ process due to titration of NHEJ factors. In contrast, we previously observed that the expression of a polλ inactive form () decreased the efficiency of partially complementary end processing. This inhibitory effect could result from a higher affinity of polλ for this type of DNA ends as compared with polμ. In this case the inactive polλ could be stuck on the DNA ends, preventing its processing by another DNA polymerase such as polμ. On the contrary, we propose that the polμD has a lower affinity for this type of DNA ends as compared with polλ and could then be displaced by the endogenous polλ. The sequencing of the repair junctions confirms that in the presence of complementary ends, the expression of both WT and inactive forms of polμ has no effect on the processing of the DNA ends (). This suggests that no DNA polymerase is recruited at the break site. In contrast, when the DNA ends are partially complementary, the over-expression of both forms of polμ increases the percentage of deletions (from 10 to 29%) (C). Therefore, both forms of polμ decrease the efficiency of processing of the four protruding nucleotides, leading to an increase in deletion frequency that could explain the higher sensitivity to IR of both cell lines. A close examination of the events resulting from processing of the four protruding nucleotides at the break site, revealed significant changes in the proportion of the three classes of events accounting for such processes. Events involving gap filling (class 1) were increased in cells over-expressing the polμ WT form, whereas they were decreased in cells expressing polμD. One hypothesis is that excess of WT polμ protects the four protruding nucleotides and improves gap filling, because in this case the DNA polymerization step is no longer the limitating step. On the other hand, presence of the polμD inhibits only partially the process of gap filling, because the endogenous polλ could still perform some gap filling due to a stronger affinity for the DNA ends as compared with polμ. As demonstrated by the structural data obtained by Moon . (), the 8-kDa domain of polμ binds the 5′-phosphate of the downstream strand in gapped DNA. This interaction is certainly lower than that of polλ due to fewer interactions (). Thus, cells could compensate the effects of polμD expression by increasing the proportion of class 2 events, which do not require gap filling and DNA synthesis. In the case of over-expression of the WT polμ form the proportion of class 2 events is decreased, perhaps because there is more gap filling due to the presence of more available DNA polymerase. Comparison with previous results obtained from over-expression of the WT polλ, show a different processing of the four protruding nucleotides of the DNA break. Over-expression of polλ decreases the proportion of class 1 events, has no effect on class 2 events, but increases class 3 events () (). This could result from the ability of polλ to produce deletions at high rate when filling short gaps and to generate -2 base frameshifts (). Altogether, our results show that polμ participates in the cellular response to IR, but with an impact different from that of polλ as regards the processing of DNA breaks. In our system, the lesser impact of the inactive polμ form as compared to the polλ inactive form, could result from a lower affinity of polμ to the template (). While one of the defining catalytic properties of polμ is the repair of substrates lacking a template strand, our system does not allow the examination of these kinds of events. The study of substrate specificity for each DNA polymerase on DSB should elucidate whether both polλ and polμ cooperate to repair DSB, or if they are mutually exclusive. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Several medically important enveloped viruses that infect the respiratory tract, such as influenza (), respiratory syncytial virus (RSV) (), and parainfluenza () virus, form both spherical and filamentous virions at the surface of infected cells. This has been observed both in cell culture models with high passage, laboratory strains, and from virus isolated directly from nasopharyngeal secretions or observed in pathology samples (). It has been shown that in influenza, filamentous virions have a higher specific infectivity (), and it has been theorized that the filamentous virus morphology may be more effective for both infecting cells and evading the host immune responses, particularly in the respiratory tract (). One important question about the filamentous virion is: What are the mechanisms by which these virions bud from the plasma membrane of an infected cell? To answer this question, we chose as our model system the A2 strain of human respiratory syncytial virus (hRSV), grown in non-polarized Vero cells. Since many aspects of virion assembly and replication have been studied with RSV, this system is ideal for the study of filamentous virion egress. To image the live-cell dynamics of the genomic viral ribonucleoprotein (vRNP) of hRSV, a molecular beacon (MB)-fluorescent probe was designed to target specifically the vRNA. This allowed for direct observations of both the morphology and mechanics of the processes leading to viral egress within the cellular context. This approach has significant advantages over the DIC (differential interference contrast) method used in the previous dynamics study () of RSV, which lacked molecular specificity and the contrast needed to observe vertically oriented virions. Molecular beacons are dual-labeled, nucleic acid probes with a reporter fluorophore at one end and a quencher at the other. They are designed to form a stem-loop hairpin structure so that fluorescence emission occurs only when the probe hybridizes to a complementary target, resulting in a high signal-to-background ratio (SBR) (). Although MBs have been used in limited live-cell mRNA studies (), their potential for the analysis of viral RNA in living cells has only recently been demonstrated (,). Other methods for imaging RNA in live cells, including the use of fluorescently labeled full-length RNAs or RNPs () and GFP-fused RNA-binding proteins (GFP-MS2) (), do not allow for the evaluation of unmodified viral particles. For the imaging experiments, a single chimeric MB, with a DNA backbone stem and 2′-O-Methyl RNA backbone hybridization domain, was designed to target a gene-end-intergenic-gene-start sequence, 3′-UUU UUA CCC CGU UUA U-5′, that has three exact repeats (A) (,). Successfully used in antisense experiments (,), this site was considered the most accessible and therefore a prime site for probe hybridization. Targeting the repeated sequence provided signal amplification, resulting in a significantly increased SBR. In addition, since the target RNP is concentrated in cytoplasmic granules, often called inclusion bodies, and in viral filaments, which contain multiple copies of the vRNP, the signal was further enhanced. A 2′-O-Methyl RNA/DNA chimera design was used for the MB primarily because of its generally higher SBR () when binding to RNA. This enhanced our ability to observe viral RNA being packaged into filamentous virion (SBR ranging from 11 to 30 using chimera versus less than 5 using DNA), and is likely due to 2′-O-Methyl RNAs higher affinity for RNA than DNA (). The enhanced nuclease resistance is also a positive feature of using 2′-O-Methyl RNA for the hybridization domain (), but we did not ever observe evidence of active probe degradation with DNA or chimera probes. The lack of probe degradation is likely due to the use of streptolysin O-based delivery, which avoids the endocytic pathway (). Vero cells (ATCC CCL-81) were grown in DMEM (Sigma Aldrich, St. Louis, MO) with 10% FBS (ATCC) with 100 U/ml of penicillin and 100 mg/ml of streptomycin. Virus used was the A2 strain of RSV (ATCC VR-1544) at a titer of 1 × 10 TCID/ml. The titer was evaluated by serial dilution and immunostaining, 4-days post-infection (PI). All data shown was at day 2 PI and with a multiplicity of infection (MOI) of 0.5. At 48 h numerous synctia had formed, but no cell death had occurred. All cells were infected at 80–90% confluence, by removing the media, followed by washing with 1× PBS (without Ca and Mg), and then adding virus to the cells for 30 min at 37°C. After the 30-min incubation, regular media were added. The MB used in this study was a DNA/2′-O-Methyl RNA chimera, 5′-Quasar570- AAAAAUGGGGCAAAUA –BHQ2 -3′ (Biosearch Technologies, Novato, CA) where the underlined sequences represent the MB stem, lower case are DNA, and the upper case are 2′-O-Methyl RNA. The beacon's SBR ratio in solution (beacon to target ratio was 1:2 in 1× PBS buffer) was measured to be ∼50 using a Tecan Safire fluorescent platereader. Molecular beacons were delivered into infected and uninfected live, Vero cells on day 2 PI using a reversible permeabilization method with streptolysin O (SLO) (Sigma). Cells grown in normal medium were first washed with serum-free medium and then incubated with a mixture of 0.2 U/ml of SLO and 1 μM of MB in an appropriate amount of serum-free medium for 10 min at 37°C. The SLO/MB/serum-free medium was then removed and replaced with fresh, normal medium. For live-cell imaging, the cells were imaged via epifluorescence microscopy, 20 min after incubation in normal growth medium. Using SLO-based delivery, MBs were delivered into Vero cells with ∼100% efficiency. Total RNA was isolated from infected and non-infected Vero cells, day 2 PI using the Qiagen RNeasy Mini Plus kit, as per the manufacturers instructions. Reverse transcription was performed using the Thermoscript reverse transcriptase with typical conditions. Primers called BC6: 5′-TAA TTT TCA GGC TCC ATC TG-3′, and OD1: 5′-TGT TTG ACA ATG ATG AAG TA–3′ were used to amplify the gene end-intergenic-gene start region between the viral genes NS1 and NS2 (). The cycle conditions were as follows: stage 1: 95°C 5 min 1×; stage 2: 95°C 30 s; 58°C for 45 s; 72°C 90 s stage 3: 72°C 10 min 1×, using the Invitrogen Thermoscript RT-PCR system with Platinum Taq DNA Polymerase. In order to further understand the role of actin in the dynamics of viral genomic RNPs, MBs were delivered into Vero cells 2-day PI. Thirty minutes after MB delivery the cell media were removed and replaced with their normal growth media with 3 μg/ml of cytochalasin-D bodipy FL (Invitrogen). Time-lapse epifluorescence imaging of the cells was performed at 37°C and 5% CO2, immediately after the media were changed for ∼15 min. Out of focus light was removed using ImageJ ‘background subtraction’ in order to improve the contrast of the filaments shown. No other alterations were performed. Fixed cell images (in 4% paraformaldehyde in 1× PBS) of vRNPs or vRNP/myosin Va co-stained samples were imaged using a Zeiss LSM 510 Meta confocal microscope with a 100X Plan-Neofluor objective (NA = 1.3). The pinhole for the confocal microscopy was set such that each image is 0.5 μm thick, with 0.5 μm between images, and up to 31 images (∼15 μm in the z direction) were taken. Bitplane was utilized to reconstruct the z-stack images. This enabled the actual length of the vRNPs to be measured, not just the 2D projection length. All of the live-cell imaging performed was with a Zeiss Axiovert 200 epifluorescence microscope equipped with an internal shutter, and a Zeiss Axiocam MRm cooled CCD camera. Control of the internal shutter and image acquisition was performed using Zeiss Axiovision 4.4. Chroma filter sets 41007a and 49002 were used to image the Quasar570 and Alexa 488 fluorescence, respectively. All images were taken with a 100X EC-Plan-Neofluor objective (NA = 1.3). Fixed cell imaging was performed in 4-well, Nunc Labtek II chambered coverslips (#1.5), while all live-cell imaging was performed with Bioptechs DeltaT black dishes with a #1.5 coverslip for a bottom. The live cells were kept at 37°C using the Bioptechs DeltaT4 system and objective heater, and in a 5% CO in air environments via a port in the heated lid of the DeltaT4 system. Molecular beacons were delivered into a confluent monolayer of Vero cells 2-day PI with the A2 strain of hRSV, as well as non-infected Vero cells, using reversible cell membrane permeabilization with streptolysin O (), in order to determine both the specificity of the MB probe () and characterize the morphologies of the vRNPs. As seen in B, infected cells exhibited strong fluorescence with signal concentrated in spherical inclusion bodies (average SBR of 80) and individual viral filaments (average SBR of 11), while the non-infected cells were quite dark, demonstrating excellent probe specificity. The infection was confirmed using PCR; infected cells exhibited a very high level of hRSV amplicon, unlike the non-infected cells (C). To demonstrate probe accessibility for vRNPs that were in the process of being packaged, we fixed the hRSV-infected cells after delivering MBs, and then co-stained for the RSV F protein with indirect immunofluorescence. The F or fusion protein is a membrane glycoprotein that mediates viral fusion and entry, and also promotes fusion of the infected cell membrane with adjacent cell membranes, leading to the formation of syncytia. We observed significant image overlap of the filamentous MB signal with that from the anti-F-protein antibody on the surface of the cells, indicating that MBs can bind to vRNPs localized to the inside of the plasma membrane in the midst of the packaging process (D). In addition, the F protein and MB signal colocalization implies that the RNPs are located throughout the filament. In order to better describe the morphologies observed using MB imaging in the infected cells, 30 min after MB delivery, the cells were fixed, imaged with confocal microscopy and later reconstructed in three dimensions using Imaris Bitplane. In A, four images showing only the fluorescent information at distances greater than 12, 8, 4 and 0 μm from the glass surface are shown. The 0 μm view represents all of the signal from the bound MBs within the infected cells displayed in two dimensions; the cross section is ∼15 μm thick from the coverglass to the tops of the viral filaments (as represented by the fluorescent signal). Both cytoplasmic inclusions and viral filaments are easily viewed and measured, with the upper 3 μm (perpendicular to the image plane) dominated by filaments. In B, one and cross section are shown. In these cross-sectional views, the filaments are easily seen protruding from the curved, apical side of the cells, while the inclusions are distributed throughout the cytoplasm, but often closer to the apical membrane than the coverslip. The filament lengths from the 3D reconstructions were measured in Bitplane, and the distribution is plotted in C. The distribution is log-normal, ranging from ∼2.6 to 8.1 μm, with an average filament length of 4.6 μm and a SD of 1.54 μm. This is very consistent with prior measurements of filament length measured using electron microscopy; Armstrong . () quoted lengths up to 2.5 μm, Norrby . () greater than 2 μm, Bachi and Howe () up to 10 μm, Berthiaume . () up to 5 μm, and measurements by Roberts . () quoted a range from 4 to 8 μm. In addition, we found that some filamentous vRNPs (B, white circles) are connected at one end, with average angles of ∼71.4 ± 3.5°, which is very similar to morphologies formed by the actin network near the plasma membrane (,). Time-lapse imaging experiments were performed on Vero cells, 2-day PI, in order to observe the dynamics leading to virion egress. Images of infected cells were taken using 100–200 ms exposures, with 5 s between exposures for up to 15 min, after which photobleaching became significant. Twelve image sequences were taken per experiment and 10 experiments were performed, yielding 120 time-lapse movies. After analyzing the movie sequences, three primary modes of motion were identified: (i) filament projection and rotation, (ii) migration, and (iii) non-directed motion. As shown in A and Supplementary Movie 1, filamentous vRNPs on the surface of the cells rotated dramatically on one end after a period of little motion. A quantitative analysis indicated that the vRNPs in A exhibited little rotation for ∼250 s, then rapidly rotated to a near vertical position relative to the cell membrane within 50 s (B). The average rate of rotation after the vRNPs appeared to separate from the membrane was ∼1.67°/s. Once the vRNPs became vertical, they either continued to rotate rapidly (C and Supplementary Movie 2), with a rate ≤6°/s, or they migrated over short distances, exhibiting the second mode of motion observed (migration). It should be noted that the filamentous groups of vRNPs observed while rotating are flexible in nature. An example of a flexible filament can be observed in D, panel two (5 s, where the filament appears ‘bent’), and panel three (10 s, where the filament appears curved) and in Supplementary Movie 2. The flexibility of RSV nucleocapsids has previously been noted (), but were not observed in real-time during the egress process. A typical example of the migration mode can be seen in A and Supplementary Movie 3, where a vertically oriented vRNP migrated ∼2.8 μm in 50 s; the path of the vRNP is plotted in B. Prior to and after this migration, the particle is static. The path taken by the particle is very similar in morphology to that of lipid rafts (see inset in B) as imaged using an Alexa 488-labeled cholera toxin probe. We hypothesize that the raft acts as a guide for the particle path, which is consistent with the observation that vRNP packaging occurs in the lipid rafts (). The speeds observed of ∼25 migrating filaments ranged from ∼29.5 to 102 nm/s with an average speed of 56 nm/s, while the path lengths traveled ranged from ∼1 to 4 μm, with an average path length of 2.36 ± 0.9 μm. The speeds measured are commensurate with myosin transport (); myosin has been implicated in RSV viral protein trafficking () and given the connection between lipid rafts and the actin network (), myosin-driven transport likely plays a role in RSV virion egress. In addition, many vRNPs exhibited non-directed motion over periods greater than 15 min, as illustrated in C and Supplementary Movie 4 (A control video, where MBs were delivered into non-infected Vero cells, can be seen in Supplementary Movie 5). In order to compare the motion seen in A with that observed in 4C, and gain insight into the nature of the motion, the MSD or 〈 〉 of each trajectory was calculated (D). It was determined that for the motion in A, the MSD (D), could be fit by a parabola ([〈 〉 = 4 + ()], where = 0.0033 μm/s, and = 29.05 nm/s, with an = 0.984, representing directed motion; while the motion in C could be fit ( = 0.7) by a power law function 〈 〉 = 4, where 4 = 0.013 and α = 0.78 ( = 0.7), which is between 0.5 < α < 0.9 representing anomalous diffusion or obstructed diffusion (,,). The calculations shown in D are for the particles shown, but are representative of the populations exhibiting these modes of motion. In E, the fraction of filaments exhibiting each mode of motion is presented. Approximately 36% exhibit projection and rotation, 12% migration and 52% non-directed motion. In the previous section, evidence of the directed motion of the RSV filamentous vRNPs was presented. This evidence, in conjunction with previous studies of RSV filaments (,), points to motor driven transport on the cytoskeleton as the likely driver of the observed migratory motion, and possibly the rotational motion. Previous studies of RSV have demonstrated that actin filaments (,) and their growth, mediated by profilin () and rhoA activation (), are vital to virion egress. In many of those studies, infected cells were exposed to cytoskeletal depolymerizing agents. The amount of free virion or viral RNA in the cell culture media, or changes in virion morphology, as measured via immunofluorescence imaging, were used to gauge the effect of these agents on virion egress. In previous studies, actin depolymerization via cytochalasin D had the most adverse effects on virion egress, therefore making actin/myosin the most likely cytoskeletal/motor combination involved in the motion observed in our investigation. In addition, since assembly occurs at the lipid rafts and the rafts are linked to the actin network (), depolymerzing them should effect the observed motion. Live-cell MB hybridization, cell fixation and permeabilization, followed by immunostaining of myosin Va, and confocal microscopy imaging was performed in order to confirm the presence and possible role of myosin. Myosin V has previously been implicated in trafficking messenger RNP (), granules on the actin network near the plasma membrane (), and RSV proteins to the apical membrane in polarized cells (). The results from our simultaneous labeling experiments can be seen in , where the blue represents myosin Va, the red represents vRNPs (hybridized MBs) and the purple represents their colocalization. In A, like A, four images showing only the fluorescence information at distances greater than 9, 6, 3 and 0 μm from the glass surface are shown. The 0 μm view represents all of the fluorescent signal from within the infected cells, displayed in two dimensions; the cross section is ∼12 μm thick from the coverglass to the tops of the viral filaments. In B, and cross-sectional views are shown, with crossed lines representing where they were taken in the plane. From the data in A and B, it can be seen that the filaments are largely distributed along the apical surface, and that they colocalize well with myosin Va (purple color). Throughout the volume of the cell monolayer, both myosin Va (blue), represented as small dots, and vRNPs (MB, red), predominately in cytoplasmic inclusions, do not colocalize and appear independent. It is clear from this figure that the strongest colocalization occurs predominately in the filaments and near the surface of the monolayer, implicating myosin Va's role in the assembly and egress process. Vero cells 2-day PI and hybridized with RSV targeted MBs, were exposed to cytochalasin D, an actin depolymerising agent, in order to provide further evidence for the role of actin on the vRNP morphology and dynamics. We observed that many filamentous vRNPs exposed to cytochalasin D, tended to aggregate and form circular, inclusion-like structures of vRNPs within ∼1 min (A and Supplementary Movie 6). Two minutes later, these circular aggregates tended to move from the surface of the cell into the cell cytoplasm. After 15 min, almost all of the vRNPs had separated from the cell surface and were concentrated in granules. In B, a typical syncytia is shown; high concentrations of filamentous structures surround the syncytia, while a lower density of filaments and particles lie along the apical membrane. However, as shown in C, a syncytia 15 min after cytochalasin D exposure composed of predominately circular granules. This experiment helps confirm filamentous actin's role in governing vRNP morphology, and explain why virion egress is decreased when actin filaments are depolymerized. In order to further explain the observed dynamics of filamentous vRNPs in live cells, we propose a biophysical model. Presented in A–D are sketches of the possible actin network/nucleocapsid interactions for all three types of motion discussed above. A and B represents how vRNP rotation might be achieved. Myosin motors, attached to the vRNP and moving along the actin network, could cause the virion to flip up on one end (A), given that actin filaments are quite flexible. It is possible that this motion frees much of the virion from the plasma membrane, allowing for more rigorous rotation (C) or migration (A). The rigorous rotation may cause the separation of daughter actin strands within the virion from mother strands at the Arp2/3 interface (,). The dissociation constant for Arp2/3 with the mother strand is ∼50 times higher than with the daughter strand, and thus the weakest part of the structure (). This separation would then allow the virion to leave the cell. The initiation of migration (A) may be the result of myosin motors as a part of the virion, and their intermittent function once the vRNPs are physically separated from the actin network. This is supported by our observation of myosin Va colocalization with vRNPs (see A and B). Non-directed motion or anomalous diffusion (C) may occur for several reasons; for example, if vRNPs lack the ATP necessary to cause directed motion or if the vRNPs are separated from the actin network via the ADF/cofilin mechanism () too early and not by mechanical motion, they may be left trapped within the plasma membrane. It is also possible that these virions were released from an infected cell earlier and reabsorbed, or they are involved in genome replication, and not competent for egress. Future studies of the protein/vRNP interactions at the membrane are required to reveal the exact reasons of the non-directed motions, and identify conclusively the proteins involved in this process. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Thiopurine drugs, 6-thioguanine (6-TG), 6-mercaptopurine and its prodrug azathioprine, are common therapeutic agents for the treatment of acute leukemia, inflammatory bowl disease, autoimmune hepatitis and other pathological conditions (). The primary activation pathway of the thiopurine drugs is through the formation of thioguanine nucleotide (TGN) and its subsequent incorporation into DNA (,). Although extensive studies on 6-TG metabolism have been carried out (,), the biochemical mechanisms for its cytotoxicity remain largely unclear, partially because of the lack of a strong mutagenic effect of 6-TG (,). On the other hand, further transformation of 6-TG in DNA might be involved in the cytotoxicity of the thiopurine drugs. In this regard, 6-TG in DNA can be methylated by -adenosyl-L-methionine (-AdoMet) to give -methylthio-2-aminopurine (2-AP-6-SCH, ) (). In addition, 6-TG in DNA can be oxidized to 2-aminopurine-6-sulfonic acid (2-AP-6-SOH, ) upon UVA irradiation (). Both metabolites of 6-TG may miscode during DNA replication (,,). In this respect, it was found that 2-AP-6-SCH could block slightly the replication by 3′-5′-Exonuclease-free Klenow fragment (KF) (), and, while the primer extension experiments were carried out in the presence of dCTP or dTTP alone, the insertion of dTMP opposite the lesion by KF was found to be nearly as efficient as the incorporation of dCMP (). In addition, 2-AP-6-SOH can be bypassed by a Y-family DNA polymerase, human pol η (); on the other hand, a very recent study revealed that this lesion blocked DNA synthesis by KF (). The NMR structural data and thermodynamic studies suggested that the replacement of a guanine with a 6-TG perturbs slightly the normal helical form of DNA (), where the sulfur atom in 6-TG exists in keto form and assumes weakened Watson–Crick hydrogen bonding with the opposing cytosine. However, the other two important metabolites of 6-TG, namely, 2-AP-6-SCH and 2-AP-6-SOH, have not been investigated thoroughly for their effects on duplex stability. We set out to examine, in detail, how the two modified thionucleobases affect DNA replication and duplex stability by using structurally defined substrates. In this context, the steady-state kinetics assay has been commonly used for determining the cytotoxic and mutagenic properties of DNA lesions (). In these experiments, the efficiency and fidelity of nucleotide incorporation are determined by measuring the rates for the incorporation of one type of nucleotide at a time and fitting the rate data with the Michaelis–Menten equation (). Recently, Guengerich . (,) introduced an elegant LC-MS/MS method to investigate the multiple bypass mechanisms of polymerases toward DNA lesions . This method provides an efficient way for determining the identities and distributions of various replication products resulting from the polymerase reaction in the presence of all four dNTPs, which, relative to the conditions used for steady-state kinetic measurements, mimics better the replication conditions . The previous LC-MS/MS quantification of replication products was based on the relative ion abundances of the composing oligodeoxynucleotides (ODNs) observed in ESI-MS under an assumption that different ODNs, regardless of their lengths and nucleobase compositions, have the same ionization efficiency (,). This assumption, however, may not be valid owing to the fact that the hydrophobicity and free energy of solvation for different ODNs can vary, which can affect their signal intensities in ESI mass spectrum (). Thus, accurate LC-MS/MS quantification of the replication products requires the consideration of the different ionization efficiencies for different ODNs. In the present study, we prepared ODNs containing a 6-TG, 2-AP-6-SCH or 2-AP-6-SOH at a defined site and carried out the replication studies by using an improved LC-MS/MS method, which takes into account the ionization efficiency differences of ODNs. We also measured the thermodynamic parameters for the formation of duplexes bearing a 6-TG and its oxidized/methylated derivatives. The results from this study provide insights toward understanding the biological implications of 6-TG and its major metabolites. The phosphoramidite building block of 6-thio-2′-deoxyguanosine was obtained from Glen Research (Sterling, VA, USA). Unmodified ODNs used in this study were purchased from Integrated DNA Technologies (Coralville, IA, USA). [γ-P]ATP was obtained from Amersham Biosciences Co. (Piscataway, NJ, USA). All other chemicals unless otherwise noted were obtained from Sigma-Aldrich (St Louis, MO, USA). 3′-5′-Exonuclease-free KF and uracil-DNA glycosylase (UDG) were from New England Biolabs (Ipswich, WA, USA). The C-terminal catalytic core of yeast pol η, which was an N-terminally His-tagged fusion protein (), was kindly provided by Prof. John-Stephen A. Taylor at Washington University in St Louis. We first synthesized 6-TG-bearing ODNs by using phosphoramidite chemistry. After solid-phase synthesis, the controlled pore-glass (CPG) support was treated with 1.0 M DBU (1,8-Diazabicyclo[5.4.0]undec-7-ene) in anhydrous acetonitrile at room temperature for 5 h to remove the cyanoethyl protecting group for the thionucleoside, followed by treatment with 50 mM NaSH in concentrated NHOH solution at room temperature for 24 h to complete the deprotection. The synthesized d(ATGGCGCGCTAT) (‘G’ represents 6-TG) was purified by HPLC and its identity was confirmed by ESI-MS and tandem MS (MS/MS). The 2-AP-6-SCH-containing substrates were prepared following previously published procedures (). Briefly, d(ATGGCGCGCTAT) (20 nmol) was treated with 10% CHI/CHCN (v/v) in 0.05 M phosphate buffer (pH 8.5) at room temperature overnight, and the methylation reaction was quenched by addition of HCl until the solution pH reached 7.0. The reaction mixture was separated by HPLC on a 4.6 × 250 mm Apollo C18 reverse-phase column (5 μm in particle size and 300 Å in pore size, Alltech Associates Inc., Deerfield, IL, USA). The composition of buffer A was 50 mM TEAA (pH 6.5), and buffer B contained 50 mM TEAA and 30% acetonitrile (v/v). The gradient program for the mobile phase was: 0 min, 0% B; 5 min, 20% B; 45 min, 50% B; 50 min, 100% B; 55 min, 0% B. The flow rate was 0.8 ml/min, and a UV detector was set at 260 nm to monitor the effluents. The identity of the methylated ODN d(ATGGCGCGCTAT) (‘G’ designates 2-AP-6-SCH) was confirmed by ESI-MS and MS/MS. Selective oxidation of 6-TG in ODNs was performed according to previously published procedures for the oxidation of 2-AP-6-SCH-bearing ODNs (). In this respect, d(ATGGCGCGCTAT) (50 nmol) was incubated with 50-μl of 3.56 mM magnesium monoperoxyphthalate (MMPP) at room temperature for 1 h. The oxidation mixture was separated by HPLC with the conditions described above. The identity and purity of 2-AP-6-SOH -containing ODN were confirmed by ESI-MS and the sequence of the ODN was verified by MS/MS. The presence of 2-AP-6-SOH in the resulting ODN was also supported by its characteristic fluorescence spectrum. To further confirm the structure of the oxidation product of 6-TG, the 2-AP-6-SOH-containing dodecamer (1 nmol) was digested by four enzymes, i.e. nuclease P1, calf spleen phosphodiesterase, snake venom phosphodiesterase and alkaline phosphatase, to give mononucleosides (,). The nucleoside mixture was separated by using a 0.5 × 150 mm Zorbax SB-C18 column (particle size, 5 μm, Agilent Technologies, Palo Alto, CA, USA). The HPLC gradient was 0–15% acetonitrile in 20 mM ammonium acetate in 60 min, and the flow rate was 6.0 μl/min, which was delivered by using an Agilent 1100 capillary HPLC pump (Agilent Technologies). The effluent from the LC column was coupled directly to an LTQ Linear ion-trap mass spectrometer (Thermo Electro Inc., San Jose, CA, USA). The spray voltage was 4.0 kV, and the capillary temperature was maintained at 225°C. The mass spectrometer was set up to monitor the fragmentation of the [M–H] ion ( 330) of 2-AP-6-SOH 2′-deoxyribonucleoside, i.e. d(2-AP-6-SOH). Standard d(2-AP-6-SOH) was also injected in a separate LC-MS/MS experiment with the identical experimental setup as that for the digestion sample. For the modified ODNs used in replication studies, a 20-mer ODN, d(ATGGCGCGCTATGATCCTAG), was first synthesized. This ODN was then selectively methylated or oxidized by using the same experimental protocols as described above for the dodecameric substrates. For primer extension assays, the 20-mer lesion-containing or unmodified ODN (20 nM) was annealed with a 5′-P-labeled 15-mer primer, d(GCTAGGATCATAGCG) (10 nM) and the resulting ODN solution was incubated with 100 μM of each of the four dNTPs and a DNA polymerase (KF or yeast pol η) in a buffer containing 10 mM Tris–HCl (pH 7.5), 5 mM MgCl and 7.5 mM DTT. The reactions were continued at 37°C for 60 min and then quenched by adding a 2-volume excess of formamide gel-loading buffer [80% formamide, 10 mM EDTA (pH 8.0), 1 mg/ml xylene cyanol and 1 mg/ml bromophenol blue]. The replication products were resolved on 20% (1:19) cross-linked denaturing polyacrylamide gels containing 8 M urea. Gel images were obtained by using a Typhoon 9410 Variable Mode Imager (Amersham Biosciences Co.). We replaced the primer used in gel electrophoresis with an unlabeled, uracil-containing ODN, d(GCTAGGATCAUAGCG), which facilitated the production, after UDG treatment, of short ODNs containing the extended portion of the primer (). These short ODNs were readily amenable to sequencing analysis by MS/MS (). The lesion-containing template and primer (100 pmol each) were annealed and incubated at 37°C in the presence of 1 mM dNTPs and a buffer containing 10 mM Tris–HCl (pH 7.5), 5 mM MgCl and 7.5 mM DTT. KF (2 U) or yeast pol η (1.6 μg) was added and the reaction was continued overnight. The replication reaction was stopped by heating to 65°C for 10 min, and the resultant mixture was then incubated with UDG (4 U) in a buffer containing 20 mM Tris–HCl (pH 8.0), 1 mM DTT and 1 mM EDTA at 37°C for 5 h. The UDG cleavage reaction was quenched by adding piperidine until its final concentration reached 0.25 M. The resulting mixture was then incubated at 60°C for 1 h, the proteins in the mixtures were removed by chloroform extraction, and the aqueous layer was dried by using a Savant Speed-Vac (Thermo Savant Inc., Holbrook, NY, USA). The dried residue was redissolved in 100-μl HO for the following LC-MS/MS analysis (A 25-μl aliquot was injected in each run). The conditions for LC-MS/MS analysis were similar as described above. The gradient for the HPLC elution was as follows: 0–5 min, 0–20% methanol in 400 mM 1,1,1,3,3,3-hexafluoro-2-propanol (HFIP, pH was adjusted to 7.0 by addition of triethylamine); 5–40 min, 20–50% methanol in 400 mM HFIP. The capillary temperature was maintained at 300°C to minimize the formation of the HFIP adducts of ODNs. MS/MS data were acquired over an range of 500–1500. To correct for the effect of varied ionization efficiencies of different ODNs on quantification, we introduced a relative-ratio method. To this end, we first obtained the standard phosphorylated ODNs by treating 20 nmol unphosphorylated ODNs with 20 U of T4 polynucleotide kinase (New England Biolabs) in a buffer containing 50 mM Tris–HCl (pH 7.5), 10 mM MgCl, 1 mM ATP, 10 mM DTT and 25 μg/ml BSA at 37°C for 1 h. Immediately after the reaction, the reaction mixture was extracted with chloroform to remove the enzymes. The aqueous layer was dried, redissolved in water, and subjected to HPLC separation, where the gradient program was 0–40% methanol in 50 mM phosphate buffer (pH 6.8) in 60 min. The phosphate buffer, instead of TEAA buffer, was employed to avoid the loss of terminal phosphate group from the 5′-phosphorylated ODNs. A mixture composed of 5 pmol each of the standard phosphorylated ODN substrates, which were identified in each set of replication mixture (shown in and ), was dispersed in the same buffer as that used in the extension assays and injected for LC-MS and MS/MS analyses with the same experimental setup as that used for the analysis of the replication mixture. The integrated area of peak, which was found in the total-ion chromatogram (TIC) plotted for the production of the most abundant deprotonated molecular ion for each standard ODN, or in the selected-ion chromatogram (SIC) plotted for the formation of three abundant fragment ions of the ODN, was normalized against that of one specific ODN. The corresponding normalized ratios for the replication samples were also determined and combined with the ratios obtained from the analyses of standards to calculate the percentage of each product in the replication mixture. Here we use the KF-mediated reaction of the 2-AP-6-SOH-bearing substrate as an example to illustrate how the method works. In this reaction mixture, we found eight replication products, and the identities of these substrates are listed in . To determine the percentage of the un-extended primer (4mer) in the reaction mixture, we first determined the ratio of the peak area for the 4mer over that for the 11C found in the SICs for the analysis of the mixture of standard ODNs, i.e. (4mer) (the SICs are shown in Figure S8): We then calculated the corresponding ratio for the analysis of the replication mixture: The ° and represented the peak areas found in SICs for the analysis of standards and replication samples, respectively. The ratio determined for the replication mixture was then normalized against that determined for the standards, which gave normalized ratio for the 4-mer, i.e. (4mer): The normalized ratio was then calculated for each identified replication product, and the percentage of 4mer in the replication mixture was calculated from the ratio of the normalized ratio for the 4mer over the sum of the normalized ratios for all the replication products by using the following equation: Strand 1 5′- ATGGCCGCTAT -3′ Strand 2 3′- TACCGGCGATA -5′, where ‘’ represents an unmodified guanine, 6-TG, or its methylated/oxidized derivative. The UV absorbance-versus-temperature profiles were recorded on a Varian Cary 500 spectrophotometer (Varian Inc., Palo Alto, CA, USA), and the ODNs were dispersed in a 1.2-ml solution containing 10 mM phosphate (pH 7.0), 100 mM NaCl and 0.1 mM EDTA at a varying total ODN concentration (Ct) of 1.0, 1.8, 3.2, 5.6, or 10 μM. The absorbance was recorded in the reverse and forward directions for a temperature range of 80–10°C at a rate of 1°C/min, and the melting temperature (m) value was obtained by the derivative method (). The error limits for ▵°, ▵° and ▵° derived from fitted parameters were calculated by using previously the described equations (). We employed traditional phosphoramidite chemistry and synthesized two 6-TG-containing ODNs, d(ATGGCGCGCTAT) and d(ATGGCGCGCTATGATCCTAG). The identities of these two substrates were confirmed by ESI-MS and MS/MS analyses (Figures S1 and S4). As reported previously (), 6-TG in ODNs can be selectively methylated to 2-AP-6-SCH by treatment with methyl iodide (CHI) in a phosphate buffer (pH 8.5, ). In addition, 2-AP-6-SCH in ODNs can be oxidized selectively to 2-AP-6-SOCH upon treatment with MMPP (magnesium monoperoxyphthalate) (). We employed similar procedures and isolated the desired 2-AP-6-SCH- and 2-AP-6-SOH-containing ODNs from the reaction mixtures by HPLC ( and Figure S2). The yields for the selective formation of these two products, as estimated from peak areas in the HPLC traces, were ∼45 and 75%, respectively (Figure S2). The molecular masses of the 2-AP-6-SCH- and 2-AP-6-SOH-containing dodecameric ODNs were measured by ESI-MS to be 3690.9 and 3724.8 Da, respectively, which are in accordance with the corresponding calculated average masses of 3691.3 and 3725.0 Da, respectively. Moreover, the sites of the lesions were confirmed by the product-ion spectra (MS/MS) of the [M–3H] ions of these ODNs (). The 2-AP-6-SOH-containing substrate was further examined by fluorescence spectroscopy and LC-MS/MS analysis of the enzyme-generated nucleosides of the ODN (see Materials and Methods). The oxidized ODN gave identical fluorescence spectrum (Figure S3) as the authentic d(2-AP-6-SOH), which had an excitation maximum at 324 nm and an emission maximum at 410 nm (). LC-MS/MS of the nucleoside mixture showed a fraction eluting at the same time (∼8 min) and exhibiting the same MS/MS fragmentation pattern as the standard d(2-AP-6-SOH) (data not shown). We next examined how the presence of 6-TG or its oxidized/methylated derivative affects DNA replication by carrying out primer extension assays with a replicative polymerase, KF and a Y-family DNA polymerase, yeast pol η. As depicted in Figure S5, 6-TG and 2-AP-6-SCH inhibited slightly the replication by KF, whereas the presence of a 2-AP-6-SOH inhibited KF-mediated DNA replication to a greater extent. The latter result was consistent with what was reported by Zhang and coworkers (). Yeast pol η, on the other hand, could bypass 6-TG and its modified derivatives to give full-length products. We further assessed the bypass and miscoding properties of 6-TG and its metabolites by LC-MS/MS following the previously described method (,) with some modifications. In this context, we adopted the HFIP buffer system, which was first reported by Hancock . () for the LC-ESI-MS/MS analysis of ODNs. It turned out that this buffer system resulted in high efficiency in both HPLC separation and electrospray ionization of ODNs with minimal cation adduction. To illustrate this, we use the analysis of the yeast pol η-catalyzed primer extension products of the 2-AP-6-SCH-containing substrate as an example. As shown in the TIC (a), the lesion-bearing strand elutes at 29.5 min. After UDG cleavage and hot alkaline treatment, the 5′ portion of the primer is produced in two forms, namely, d(GCTAGGATCAp) and d(GCTAGGATCAXp) (‘X ’ represents an abasic site), eluting at 26.5 min and 26.0 min, respectively. The latter form results from the incomplete cleavage of the abasic site by hot piperidine treatment, which might be attributed to the relatively low temperature (i.e. 60°C) employed for the treatment. The 3′ portion of the primer strand, which carried the extension products and included 10C, 10T, 10A, 11C, 11T, 11C_T, 11T_T, 11A_T, elutes at 26.9 min (ESI-MS averaged from this retention time is shown in b, and the sequences for the identified products are listed in ). Other than these full-length products, we also found the un-extended primer (4mer) and a frame-shift product (8C), which elute at 20.8 min and 26.5 min, respectively. The identities of the above ODNs were determined from ESI-MS and MS/MS measurements. For instance, the MS/MS of several replication products, including a frameshift product, are shown in Figure S7, and the MS/MS for the [M–4H] ions of the 5′ segments of the primer are shown in Figure S6. The same LC-MS/MS method also allows us to identify the replication products from the other five replication reactions (the identified products are listed in and ). To quantify accurately the relative amounts of these extension products, we first examined the relative ionization efficiencies of ODNs in negative-ion ESI-MS. To this end, we injected a mixture, which contained 5-pmol of each standard 5′-phosphorylated ODN in the same buffer as that used for extension assays, for LC-MS and MS/MS analyses. As depicted in , the seventeen 5′-phosphorylated ODNs identified from the pol η-mediated replication mixture of the 2-AP-6-SOH-bearing substrate indeed exhibit substantially different efficiencies in forming the most abundant molecular ions, revealing the importance of considering the varied ionization efficiencies of different ODNs in the LC-MS/MS quantification of replication products. In this respect, the relative ionization efficiencies were assessed by normalizing the total-ion current observed for the most abundant deprotonated molecular ion found for each ODN against that found for the [M–3H] ions of 10C, for the yeast pol η-mediated reaction, and 11C, for the KF-catalyzed reaction (See Figure S8 for an example). In the viewpoint that MS/MS provides improved signal-to-noise ratio for measuring the relative amounts of different ODNs present in the replication mixture, we also examined the relative efficiencies for the formation of three abundant fragment ions, from the injection of an equimolar mixture (5 pmol each) of the 17 authentic ODNs as mentioned earlier. It turned out that the ratios obtained for most ODNs are similar to those found based on molecular ions, and the efficiencies for the formation of three abundant product ions were again markedly different for these ODNs (). In this respect, -test showed that, at 99.9% confidence level, the mean ratio obtained for 10A or 10G was significantly different from that determined for 4mer, 5C, 5G, 6C, 6A, 6G, 8A, 10A, 11C or 11T_T. It is worth noting that two isomeric ODN products were found in the pol η-catalyzed replication mixture of all three thionucleoside-containing substrates ( and Figure S10), i.e. 11T or 11A_T, which exhibit very similar retention time. Therefore, the quantification of these two products has to rely on MS/MS. In this context, it is worth noting that the quantification of these two ODNs also necessitates the use of fragment ions with distinct values for the two ODNs. In this case, we employed the [a-G], [a-C] and [a-C] ions, which have values of 870.6, 1035.1 and 1179.7, respectively, for 11T and 875.1, 1039.7 and 1184.1, respectively, for 11A_T [Nomenclature for fragment ions follows that described by McLuckey . ()]. By using the relative ratio method described in ‘Materials and Methods’ section, we quantified the percentages of individual ODNs in the reaction mixtures and summarized the results in and . Our results showed that KF could bypass 6-TG efficiently, and the most abundant product was found to be the products with a dCMP being inserted opposite the 6-TG and with the blunt-end addition of a dA, d(AT) or d(ATA) ( and Figure S9d). Consistent with the results from the primer extension monitored by PAGE analysis, the 2-AP-6-SCH-containing substrate can also be bypassed; however, the 2-AP-6-SCH induced much more nucleotide misincorporation than 6-TG. Among all the full-length products, dTMP (∼53%) is inserted opposite the 2-AP-6-SCH much more preferentially than dCMP (∼20%), followed by dAMP (∼5%) and dGMP (∼5%, and Figure S9e). This result is significantly different from the similar efficiency of dCMP and dTMP incorporation in primer extension experiments where only dTTP or dCTP was present in the replication mixture (). The 2-AP-6-SOH-bearing substrate blocked KF-mediated primer extension more readily as represented by the formation of much smaller amount of the full-length products (a total of ∼18%, and Figure S8). In contrast to what we found for the 2-AP-6-SCH-containing substrate, KF preferentially incorporates a dCMP opposite 2-AP-6-SOH. Considering the full-length products formed from this reaction, the incorporation of dCMP and dAMP constitute 17% (11C, ) and 0.8% (11A, ), respectively, whereas the insertion of dTMP was barely detectable. For the primer extension with yeast pol η, substrates containing 6-TG and its metabolites can all be bypassed with different levels of efficiencies, which is consistent with the results obtained from the above gel electrophoresis experiment. For all three lesions, the most abundant products carry the correct nucleotide (dCMP) opposite the lesion (, the SICs for monitoring the formation of replication products and a summary of the identities and percentages of those products are shown in Figure S9). Aside from the correct nucleotide incorporation, the insertions of dAMP and dTMP were also observed (). In this respect, if we only consider the full-length products, dCMP, dTMP and dAMP were inserted opposite 6-TG at frequencies of 74, 5.2 and 4.7%, respectively (), whereas these three nucleotides were incorporated opposite 2-AP-6-SCH at frequencies of 52, 17 and 8%, respectively (). The respective frequencies for the insertion of these three nucleotides opposite 2-AP-6-SOH are 29, 1.2 and 5.8%. These results, therefore, demonstrated that 2-AP-6-SOH also blocked pol η-mediated polymerization more effectively than 6-TG or 2-AP-6-SCH (). In keeping with our findings for the KF-mediated reaction, pol η also misincorporated dTMP more frequently than dAMP opposite 2-AP-6-SCH, whereas the wrong nucleotide dAMP was inserted opposite the 2-AP-6-SOH with greater efficiency than the incorrect nucleotide dTMP. It is worth mentioning that a single LC-MS/MS experiment facilitates us to gain insights into both the nucleotide incorporation opposite the lesion and the primer extension beyond the lesion site. In this context, our data demonstrated the presence of a significant amount of frame-shift products in the reaction mixture, particularly in the mixtures resulting from the replication of 2-AP-6-SOH-harboring substrate induced by yeast pol η (6C, 6A, 6G, 8C and 8A, a total of ∼40%, Figure S9c) and KF (6C and 8C, a total of ∼20%, Figure S8). To gain insights into the effects of 6-TG and its metabolites on duplex stability, we further determined the thermodynamic parameters for duplex formation by measuring the melting temperatures for the thionucleoside-containing duplex ODNs (see ‘Materials and Methods’ section and Figure S10). It turned out that the replacement of a guanine with a 6-TG resulted in the destabilization of the duplex by 2.4 kcal/mol in free energy at 37°C (). In addition, 2-AP-6-SCH and 2-AP-6-SOH caused even more destabilization to duplex DNA, with being 4.2 and 4.3 kcal/mol, respectively (). The cytotoxicity of the thiopurine drugs involved mostly the formation of 6-TG nucleotide upon metabolic activation and its subsequent incorporation into DNA (). In DNA, 6-TG can be methylated by -AdoMet to form 2-AP-6-SCH () and converted to 2-AP-6-SOH upon UVA irradiation (). Both 2-AP-6-SCH and 2-AP-6-SOH may affect DNA replication and/or repair, thereby exerting their cytotoxic and mutagenic effects (,). In this article, we obtained pure ODNs containing a structurally defined 6-TG, 2-AP-6-SCH, or 2-AP-6-SOH at a specific site, and confirmed the identities of these ODNs by ESI-MS, MS/MS and, for the 2-AP-6-SOH-bearing substrate, by fluorescence spectroscopy. In addition, the 2-AP-6-SOH-containing ODN was digested by enzymes to mononucleosides, and LC-MS/MS analysis confirmed the presence of 2-AP-6-SOH 2′-deoxyribonucleoside in the nucleoside mixture. We then carried out the replication studies of 6-TG and its metabolites with KF and yeast pol η. Consistent with previous studies (,), we found that 6-TG and 2-AP-6-SCH slightly blocked the replication by KF, whereas 2-AP-6-SOH blocked substantially the primer extension by KF. All three thionucleosides, however, could be bypassed by yeast pol η with varying efficiencies. The miscoding and polymerase stalling properties of DNA lesions are frequently assessed by the steady-state kinetics assay (). Recently, LC-MS/MS, because of its high efficiency and its capability in offering sequence information for many ODNs in a single experiment, has been developed as a new tool to investigate the mutagenic and cytotoxic effects of DNA lesions (). Different from the steady-state assay, the LC-MS/MS method can allow for the analysis of the replication products from the reactions with the mutual presence of all four dNTPs, instead of adding one type of dNTP at a time. It may better represent the real polymerization reaction conditions , thereby providing more accurate measurements. However, the LC-MS/MS analysis of extended ODNs in the replication mixture was somewhat limited because common HPLC mobile phases are not compatible with both the HPLC separation and the electrospray ionization of ODNs. Here we found that the HFIP buffer reported by Hancock . () allowed for the effective separation and efficient ionization of ODNs in the replication reaction mixture. In this newly developed LC-MS/MS method, the quantitative analysis of individual ODNs in the reaction mixture is complicated by the different ionization efficiencies of the composing ODNs. In this article, we quantitatively assessed the ionization and detection efficiencies of different ODNs through the analysis of authentic compounds. Our results revealed that different ODNs could exhibit significant differences in the magnitude of signals produced in MS or MS/MS, which calls for the need of assessing the relative ionization efficiencies of different ODNs while LC-MS/MS is used for this type of analysis. With the consideration of different ionization efficiencies of different replication products, our LC-MS/MS results revealed that dTMP is inserted opposite 2-AP-6-SCH with much greater efficiency (2.5-fold) than dCMP by KF. The resulting 2-AP-6-SCH/T base pair may trigger the DNA mismatch repair pathway (). If not repaired efficiently, the lesion may lead to a high incidence of G→A transversion mutation. Our LC-MS/MS data also demonstrated that 2-AP-6-SOH can block significantly the extension by KF as represented by the formation of a large amount of the unextended primer (4mer, 32%, Figure S8) and pentameric products (5C, 5A and 5T, larger than 30% in total, Figure S8), suggesting that this lesion may introduce significant structure distortion to duplex DNA. However, the translesion synthesis polymerase, yeast pol η, can bypass all three thionucleosides, including 2-AP-6-SOH, and give full-length products. Although dCMP is the most favorite nucleotide being inserted opposite 2-AP-6-SCH and 2-AP-6-SOH, we observed significant frequencies of misincorporation of dTMP and dAMP opposite 2-AP-6-SCH and 2-AP-6-SOH, respectively. Other than nucleotide misincorporation opposite the lesion sites, we also found a substantial amount of -2 frameshift products in the replication mixtures for 2-AP-6-SCH- and 2-AP-6-SOH-containing substrates. Remarkably, this type of products account for ∼40% of the products in the pol η-mediated replication of the 2-AP-6-SOH-containing substrate. We reason that three possible mechanisms may contribute to the formation of these products (). In this respect, yeast pol η incorporates both dCMP and dAMP opposite the lesion, and the resulting 2-AP-6-SOH/A and the 2-AP-6-SOH/C base pairs may distort the local double helix structure, which may cause the two bases on the 5′ side of the lesion to flip out. The polymerase can then continue to add one or three correct nucleotides in the presence of template (, left). Other two mechanisms involve the looping-out of the lesion together with flanking 3′ or 5′ nucleotide followed by the incorporation of the correct nucleotides by the polymerase (, middle and right). The latter two mechanisms may account for the high occurrences of 6C and 8C in the replication mixtures of 2-AP-6-SOH-containing substrate mediated by KF or pol η. In this context, it is worth noting that different flanking sequences may give rise to different types of frameshift products. The thermodynamic studies revealed that the presence of a 6-TG caused an increase in ° at 37°C by 2.4 kcal/mol relative to that of the parent duplex. The ° was close to that induced by one mismatched base pair (). The presence of 2-AP-6-SCH and 2-AP-6-SOH resulted in even greater destabilization to duplex DNA, i.e. by 4.2–4.3 kcal/mol in Gibbs free energy at 37°C. The increased destabilization to double-stranded DNA induced by 2-AP-6-SCH and 2-AP-6-SOH than by 6-TG may facilitate the more efficient recognition of the two lesions by DNA repair enzymes, though other factors, e.g. substrate specificities of DNA repair enzymes and base pairing, may also contribute to the recognition of these lesions during repair (,). Since the approval of the thiopurine drugs by FDA in the 1960s, azathioprine, 6-mercaptopurine, and 6-TG have been widely used as therapeutic agents in the treatment of a variety of human diseases (,). However, there is a high occurrence of certain cancers in long-term survivors of these patients (). For example, 20 years after transplant, about 60–90% of the patients who have taken azathioprine as an immunosuppressant develop squamous cell carcinoma (). Such great prevalence of skin cancer in transplant patients is not found in the general population. The methylation and oxidation of 6-TG, due to the increased potential in miscoding and in inducing frameshift mutations, may contribute to the development of cancers in those patients. In this respect, it was shown that aberrant processing of 2-AP-6-SCH/T mismatch was more toxic in mismatch repair-proficient than in deficient cells (). Future studies on the quantification of these lesions formed and mutagenesis study of these lesions using shuttle vector technology should offer more insights into the roles of these lesions in the development of cancers in those patients who have been treated with the thiopurine drugs. Such studies are currently being pursued in our laboratory. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Gene regulation is governed by a number of biomolecules competing for DNA-binding sites, recognizing each other, assembling on the double helix, binding ligands on their ‘backs’, forming sophisticated DNA structures, etc. This picture is further complicated because DNA is tightly packed , and because proteins or drugs may link DNA segments separated by large distances along the sequence. Knowing the information about all molecular players and the rules of their interaction, Nature ‘calculates’ the transcription level for each gene. Is this biological LEGO game solvable on a computer? Let us take this as a working hypothesis. It is now believed that most of the binding events involved in gene regulation are reversible and governed by the thermodynamic equilibrium (). Nowadays, high-throughput microarray technology allows us to determine thousands of thermodynamical parameters from a single experiment (). In addition, the bioinformatics sequence analysis methods provide a way to predict the protein–DNA-binding affinities (). There is a growing understanding now that a statistical–mechanical methodology is required to predict gene regulation based on this large amount of data (). Some methods consider just several predefined binding sites to find a solution for comparatively simple gene regulatory systems (). For example, the 's lac-operator containing several binding sites for LacI and C-reactive proteins (CRPs) that may multimerize and assist DNA loop formation, can even be described analytically (). The combinatorial regulation at a single eukaryotic enhancer is also a solvable task (,). More complex systems may be accessed with the help of different network approaches (). However, once we identify all the important states (which is not a trivial task), a huge number of states, parameters and computation time make many interesting systems practically incalculable without special tricks. Several methods of solving one-dimensional lattice models have been developed in the past, including the generating functions (,), the transfer matrix method (,), the combinatorial approaches (,) and other modifications (). In the case of non-site-specific binding, the problems of site overlapping, competitions and contact interactions may be solved analytically by any of these methods. The McGhee–von Hippel (MvH) approach is probably most widely used for the description of typical DNA–protein and DNA–drug experiments (). However, site-specificity requires calculations according to the real polymer sequence, which rules out any analytical solutions (,,). Taking into account the long-range interactions between the proteins bound to the DNA, poses additional difficulties that cannot be easily resolved by the combinatorial approaches (,). For example, the recent GOMER algorithm allows treating long-range interactions between a protein and a DNA promoter, but not the long-range interactions between two DNA-bound proteins (). The generating functions method is a more general tool that has been extensively tested for many kinds of one-dimensional problems (,). At first, this method seemed inapplicable for the case of long-range interactions (). Later studies have showed that the generating functions method still allows treating long-range cooperativity, but it fails if more then one type of large protein exists in the system (). On the other hand, the transfer matrix method allows treating site-specificity (), long-range interactions () and multiple binding (). Yet there are other basic binding features such as the multilayer assembly, DNA looping, nucleosome sliding, etc. for which none of these methods have been tested. It seems from the literature analysis that only the transfer matrix method is left as a potential approach to solve the whole complexity of DNA–protein–drug lattice models in a unified systematical way. Up to now, there were no attempts to apply this method to the biophysical characterization of complex gene regulatory systems. On the other hand, a complementary field of mathematical analysis of DNA sequences now actively uses matrix methods. This provides an additional argument for choosing the transfer matrix formalism as a general systematic tool. At this point, we have to make several methodological comments. All models for DNA-ligand binding mentioned above belong to the class of the so-called Ising models. In his doctoral thesis in 1924, Ernest Ising was studying ferromagnetism and introduced the model of a linear chain of magnetic moments, which are only able to take two positions, ‘up’ and ‘down’, and which are coupled by interactions between the nearest neighbors. Later the Ising model became popular in many fields of physics. Naturally, when a number of physicists moved to biophysics inspired by Schrödinger's definition of DNA as a one-dimensional aperiodic crystal, they brought the Ising model to the new field, in particular to the study of DNA melting and DNA-ligand binding (). At that time, the field of bioinformatics did not yet exist. When bioinformatics emerged later, it was mostly driven by the biologically inspired mathematicians who came with their own concepts such as the Markov chains. In the 1920s, a pioneer of cybernetics, Norbert Wiener, performed a first rigorous study of a continuous Markov process. This work and the later Kolmogorov's probability theory popularized the ideas of the 19th-century mathematician Andrei Markov, who studied the sequences of random variables in which the future variable is determined by the present variable but is independent of the way in which the present state arose from its predecessors. The Markov chains are now widely used in bioinformatics, in particular in the DNA sequence analysis (, ). Both the Markov chains and the Ising lattices may be formulated with the help of the transfer matrices containing the probabilities of transition of a system between different states. Consequently, the Ising model may be converted into the Markov model, and vice versa. The general transfer matrix formalism is evidently the ancestor of both the Markov chains and the Ising lattices. However, the differences between the transfer matrices employed in the biophysical and bioinformatical studies of DNA–protein interaction are not just historical. Bioinformatics is mostly interested in DNA sequence analysis, motif finding, etc. Therefore the different states in the Markov chains are either (A, T, G, C) or the occurrences of dinucleotides, trinucleotides, etc. or some other ‘words’ composed from the nucleotide dictionary (, ). On the other hand, in the biophysical models, the DNA sequence is fixed, and the different states are ‘free’ or ‘bound’ depending on the presence or absence of a protein at a given DNA site (). In the matrix models considered below, the states are also divided into ‘free’ or ‘bound’, but these states are further subdivided into a number of microstates allowing us to treat complex models of DNA–protein–drug interaction. Hopefully our work will help to join the efforts of biophysics and bioinformatics in the description of gene regulation. #text xref fig #text The binding events at O operator of bacteriophage λ control the famous genetic switch from the lysogenic state (when λ peacefully lives inside the infected ) to the lytic state (when λ duplicates itself in a large number of copies leading to the lysis of the host cell). Two regulatory proteins, the CI and Cro repressors, act at O operator. In the lysogenic state, the gene coding the Cro protein is ∼95% suppressed, while the gene coding the CI protein is on. The CI protein aims to maintain its own expression and to switch off all other genes. CI domination determines that the phage is in the lysogenic state. When the host cell is damaged or irradiated, its SOS system activates RecA protein that stimulates self-cleavage of CI. This leads to induction of the lytic state of phage λ. In the lack of CI, Cro dominates at O, maintaining its own expression, switching on early lytic genes, and suppressing the gene (). The CI and Cro proteins homodimerize in solution due to C-terminal domains and bind DNA as dimers using a helix-turn-helix motif in the N-terminal domains. The dimers may adsorb on DNA non-specifically () or bind sequence-specific sites (). The specific binding constants are orders of magnitude larger then the non-specific ones. The CI and Cro dimers cover 17 bp upon binding to DNA. The structure of O operator is shown in . The O operator consists of three 17-bp specific binding sites for Cro and CI proteins, enumerated O1, O2 and O3. Each site may bind Cro and CI with different affinities. The O site overlaps with the promoters P and P. The σ subunit of the RNA polymerase (RNAP) binds to the –10 and –35 recognition regions at the promoter, making ∼35 bp inaccessible for binding by other proteins. RNAP binding to P starts transcription of CI protein (direction to the left from O in A). RNAP binding to P starts transcription of Cro and other yearly lytic proteins (to the right from O). P overlaps with O1 and O2. Thus, binding of repressor proteins to any of O1 and O2 sites precludes RNAP binding at P. P overlaps with O3 and borders upon O2. RNAP bound at P contacts with CI dimer bound at O2. This contact activates transcription from P. The Cro–Cro and CI–CI contacts are also energetically favorable (). The binding scheme shown in A allows 40 distinguishable arrangements of three proteins, Cro, CI and RNAP among three binding sites O1, O2 and O3 (). Several years ago it seemed natural that this picture completely describes the regulatory events at O operator (,). However, then it was shown that it should be further complicated to take into account the interaction of O with another operator O situated ∼2.4 kb from O. O may be linked to O thought DNA looping due to bridging by CI proteins (B). The O operator consists of three binding sites O1, O2, O3 similar to O, and overlaps with promoter P symmetrical to P. Most of the binding energies for this system are known from the experiments. The energies of non-specific binding of Cro and CI are –4.2 and –4.1 kcal/mol correspondingly (). Cro binds O1, O2 and O3 sites with the energies –12.0, –10.8 and –13.4 kcal/mol (). CI binds O1, O2 and O3 sites with the energies –12.5, –10.5 and –9.5 kcal/mol correspondingly (). sub #text S u p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Seed plants contain the largest mitochondrial genomes investigated so far. Although their sizes reach up to about half of the genome, the mitochondrial genomes more or less encode only the same small set of ∼60 genes. This gene collection is found with minor variations in the mitochondrial genome sequences of , sugar beet, rape seed, tobacco, wheat, rice and maize (). The transcription units, mono- as well as polycistronic, are spaced across the entire genome lengths and are separated by large spacer sequences without any obvious function. Transcription of mitochondrial genes frequently starts from multiple promoters of various types, generating precursor RNAs that have to pass through various processing steps such as RNA editing, splicing of group II introns, 3 end trimming as well as formation of secondary 5 termini (). Many or all of these steps are required to generate a mature translatable mRNA or functional transfer as well as ribosomal RNAs and can potentially contribute to the regulation of mitochondrial gene expression. But up to date it is still unknown whether or to which extent these processes control or even regulate the realization of the mitochondrial genetic information. Transcription initiation is one of the most important levels to regulate gene expression in bacteria, archaea as well as nuclei from eukaryotes. This process has also been intensively examined in mitochondria of various plant species. In for instance at least two conserved promoter motifs have been found and in addition a number of non-conserved transcription initiation sites are present (). The situation might be even more complex since not all mitochondrial transcription units have been analyzed in this respect. Up to date it is still unclear to what extent if at all plant mitochondrial gene expression is regulated or controlled at the transcriptional level. Run on transcription studies showed that mitochondrial genes are transcribed at different rates, most likely determined by differing promoter strength (,). But so far there is no clear evidence that expression of individual genes is actively regulated during transcription initiation events. Several reports provided convincing evidence that also post-transcriptional processes influence plant mitochondrial mRNA steady state levels (,). This was demonstrated by comparing transcriptional rates with the steady state RNA levels. The observed discrepancies for several genes were interpreted to originate from post-transcriptional processes influencing RNA stability. Still little is known how the stability of a transcript is determined. Of course there must be -elements as well as -factors, which probably determine RNA stability in a concerted action. Stem–loop (SL) structures are good candidates for -acting processing signals and such structures at or near the 3 ends of several plant mitochondrial RNAs have been found to influence the mRNA stability and (), most likely preventing exonucleolytic degradation (). However, the nature or features of structures determining 3 ends and/or RNA stability are still unknown. Another important ()-factor is the polyadenylation state on an RNA. Short oligo(A) tails at mature 3 ends have been found to destabilize plant mitochondrial RNA both and and are thus expected only at a minor fraction of the steady state pool (,,,). The 5 ends of mRNAs can be generated directly by transcription initiation or by subsequent 5 processing events. Support for the existence of the latter has been obtained by mapping of such ends of various genes in mitochondria of different plant species. It is still unclear however, how these ends are generated. Up to now no evidence has been reported for a 5 to 3 exonucleotlytic activity. Consequently the generation of secondary 5 termini has been attributed to (an) endoribonuclease(s). So far two different endonucleolytic RNase activities have been described in plant mitochondria, both being involved in the maturation of tRNAs. An RNase P-like activity has been found to cut precisely at the mature tRNA 5 end, while RNase Z cleaves directly at or one nucleotide downstream of the discriminator nucleotide at the 3 end (). The prerequisite for the cleavage of precursor molecules by these activities is the formation of the tRNA secondary structure (,). tRNA-like elements (t-elements) with non-canonical cloverleaf structures are also substrates for these enzymes at least . Such t-elements have been found dispersed in wheat mtDNA and if transcribed as part of an RNA molecule these structures could potentially be involved in cleavage of such long primary transcripts to generate secondary ends (,). But so far only little evidence has been found for the function of such t-elements . To gain more information about 5 and 3 ends of mRNAs in plant mitochondria, we analyzed mRNA extremities of all protein-coding genes annotated in the mitochondrial genome of . Single 3 ends are found for almost all transcription units, while multiple 5 ends are found in several cases. Only a few of the 5 ends are found at conserved promoter sequences, while most 5 termini are most likely generated post-transcriptionally. Analysis of the sequences surrounding the 5 and 3 termini revealed that several of the ends coincide with termini of t-elements or stem–loop structures suggesting that RNase Z and RNase P are involved in mRNA 5 and 3 end processing. An ecotype Columbia cell suspension culture was cultivated on a shaker (120 r.p.m.) in the dark at 23°C. The ecotype of this culture was recently confirmed by analyzing corresponding informative genetic markers (). Mitochondria were isolated from cultures six days after the transfer to fresh medium according to a method described previously (). RNA from these organelles (mtRNA) was extracted following previously established protocols (). Alternatively, 100 mg frozen mitochondria (fresh weight) were disrupted in a mortar. The fractured organelles were suspended in the lysis buffer of an RNeasy Plant Mini kit. RNA was isolated following the instructions of the manufacturer (Qiagen). CR–RT–PCR analyses were either performed as described before (A) () and/or carried out using a modified protocol (B). Briefly, large-scale self-ligation was performed with up to 50 µg RNA in a total volume of 100 µl. After ligation, samples were desalted using Microcon YM-10 or -30 micro concentrators (Millipore) and stored as aliquots of 15 µl at −20°C. First strand cDNA synthesis was done with 5 µg of total RNA and 2 µg of mitochondrial RNA, respectively, and 200 U M-MLV RNase H Minus (point mutant) reverse transcriptase under conditions recommended by the manufacturer (Promega). The RNA template was then degraded by adding 1/5 volume of 1 M NaOH and an incubation of 10 min at room temperature. The sample was subsequently neutralized with an equal amount of 1 M HCl and the cDNA was purified with the GFX™ PCR DNA and Gel Band Purification Kit (GE Healthcare). This purification step also removed the primer used for cDNA synthesis. This oligonucleotide could cause the amplification of multiple PCR products when present in the first amplification reaction containing another primer with the same orientation. One-fifth of the cDNA sample was used as template in a single RT–PCR. In some cases amplification was done in a two-step PCR using primers with melting temperatures of about 73°C. This allows annealing and synthesis to be performed at 68°C. The extremities of the and t-elements, respectively, were determined by using mitochondrial 5S and 18S rRNAs as linker molecules, since 5 and 3 end mapping of small RNA species with conventional CR–RT–PCR is rather inefficient. The rRNAs were linked to the target RNAs by bulk ligation of 5 µg of mitochondrial RNA as described above. The use of 18S instead of 5S rRNA as anchor molecule permits an amplification of larger products, which increases PCR efficiency. Further details are given in and Supplementary Figures 29, 30 and 31. For northern blot experiments 3–10 µg of total or 1–5 µg of mitochondrial RNA were size fractionated on 1% (w/v) agarose gels using glyoxal as denaturing agent. Gel electrophoresis, blotting and hybridization were done as described previously (). Nucleic acids were blotted onto Duralon UV (Stratagene) or Hybond-XL membranes (GE Healthcare) and hybridized with radioactively labeled probes as outlined in the manufacturers’ guidelines. Primer extension analyses were performed according to standard protocols (). Sequencing of PCR products was commercially obtained (MWG Biotech and 4baselab). Individual cDNA clones were sequenced using Thermo Sequenase Primer Sequencing kits as recommended by the manufacturer (GE Healthcare) and an ALF sequencing system (GE Healthcare). sequence analyses were done at the NCBI server using various blast tools (). Secondary structure predictions were done the with program ‘stemloops’ from the Wisconsin GCG software package version 10.2, with a program searching for palindromes () and a web interface RNA-fold program (). Searches for conserved sequence motifs were done with the MEME search tools (). The mitochondrial genome encodes 32 protein-coding genes. To gain more information about the major termini of the transcripts of these genes in this model plant species we analyzed the RNAs covering these loci by CR–RT–PCR. This experimental approach allows the simultaneous determination of 5′ and 3′ termini, the detection of non-encoded nucleotides and delivers unambiguous information by sequencing of the amplified cDNA products. In addition this method allows fine mapping of the ends on the nucleotide level. Generally two different approaches were used, both starting from gene-specific cDNA synthesized on self-ligated RNA (). In the first approach, which was mainly used in the initial phase of the project, products were amplified from mitochondrial RNA isolated from ecotype Columbia (Col) cell suspension culture. A first PCR was performed with the primer used for cDNA synthesis, which anneals in the 5′ terminal region of the reading frame and a forward primer complementary to sequences located in the 3′ terminal part of the gene. Products of this reaction were inspected on agarose gels and the prominent fragments were selected for further amplification in (a) nested PCR(s). The resulting fragment(s) was (were) cloned and about 20 cDNA clones were sequenced for each PCR product (A). In the alternative approach, gene-specific cDNA synthesis was performed on both mitochondrial and total RNA from the Col cell suspension culture. After cDNA synthesis a single PCR was performed with oligonucleotides annealing to 5′ and 3′ terminal sequences of the reading frame (B). In this PCR, the primer annealing to the 5′ part of the gene is different from the oligonucleotide used for cDNA synthesis. In some cases primers with melting temperatures above 73°C were used to increase the specificity in these reactions. Prominent products which appeared identical in the PCR patterns of both RNAs were selected for sequencing from both directions using the PCR primers. This delivers unambiguous sequences up to the ligation site. Downstream of this site several sequences are superposed with multiple peaks at each nucleotide position, which in many cases results in non-readable sequences. Usually the product obtained from mtRNA was used in the sequence reaction. As an example for the mapping procedure the analysis of the 1 gene is described in detail here. This gene is flanked in 5′ by 10 kb and in 3′ by 17 kb non-coding sequences, this isolated location suggesting a monocistronic 1 mRNA transcribed from a cognate promoter (A). A transcription initiation site has indeed been identified 355 nucleotides upstream of the ATG (). The transcripts were analyzed by both CR–RT–PCR approaches (A and B). In the first approach cDNA synthesis was initiated from oligonucleotide Atcox1-1 complementary to sequences from position +156 to +138 relative to the TG (+1). The same primer and oligonucleotide Atcox1-3 (+1473 to +1493) were used in the first amplification reaction yielding a strong product of 500 and two weak products with sizes of about 400 and 650 nucleotides, respectively (A and B, left panel). These were subsequently used as DNA templates in three separate amplifications (second PCR), each performed with primer pair Atcox1-2 (+118 to +98)/Atcox1-4 (+1501 to +1520). While no product was obtained in the reaction with the smallest DNA template, strong fragments of 500 and about 300 nucleotides were generated in the reaction with the 500-bp template. A relatively weak cDNA product of 500 bp is obtained from the 650-bp template. The 500-bp cDNA from the second PCR was then cloned, which yielded cDNA clones with sizes between 270 and 500 bp (data not shown). Sequence analysis of 32 clones of various sizes identified 3′ ends in 23 clones 47, in six clones 46 and in one clone 45 nucleotides downstream of the stop codon, respectively (Supplementary Table 1). A single cDNA clone has a 3′ end in the reading frame 40 nucleotides upstream of the stop codon while another clone has a 3′ end 64 nucleotides downstream of the reading frame. This analysis suggests a single dominant 3′ end for all mRNAs 47 bp downstream of the reading frame with some slightly scattering 3′ ends around this position (349 783 in complete sequence nc_001284.2). Non-encoded nucleotides, i.e. four adenosines followed by a fragment of the mitochondrial 18S rRNA were found in a single cDNA clone. In contrast to the homogenous 3′ terminus, the 5′ ends found in individual cDNA clones vary considerably. The 5′ termini in the largest cDNA inserts are found in five clones 241, in two clones 240 and in three clones 239 nucleotides upstream of the ATG start codon. All other 20 clones have 5′ ends between −216 to −20 relative to the ATG with two minor clusters between −187 and −179 and −48 and −43. This suggests that a major 5′ terminus of the steady state mRNA is located 241 nucleotides upstream of the ATG (at position 351 654). The other clones are either minor ends or represent degradation products. In the second approach, cDNA synthesis was again initiated from primer Atcox1-1 but was followed by a single amplification reaction with primer pair Atcox1-2/Atcox1-4. A 500-bp cDNA was obtained as dominant product from both RNAs (C). This PCR product was directly sequenced with oligonucleotide Atcox1-2 directed to the 5′ end, yielding a sequence across the ligation sites of the steady state mRNA pool (D). The chromatogram shows a clean sequence up to the adenosine at position 351 654 followed by an overlay of minor signals at all positions but with a dominant unique sequence starting with an adenosine at position 349 783. This transition marks the ligation site and determines the 5′ and 3′ transcript termini. The majority of the transcripts thus starts with a thymidine 241 nucleotides upstream of the ATG and ends with another thymidine 47 nucleotides downstream of the reading frame. That the sequence can be clearly followed across the ligation site indicates that the vast majority of the mRNA molecules has identical 5′ and 3′ ends. The appearance of minor signals beyond the ligation site shows the presence of a minor fraction of 1 RNA molecules with slightly differing 5′ ends. Both approaches consistently identify the dominant mRNA 5′ end located 241 nucleotides upstream of the ATG and a 3′ terminus corresponding to a 47 nucleotide 3′ UTR. To check these CR–RT–PCR mapping results by independent methods, primer extension reactions were carried out with primers Atcox1-2 and Atcox1-5 (−129 to −148) using mtRNA as template. Atcox1-2 is elongated to a major product of about 360 nucleotides corresponding to the above detected 5′ end at −241. Additional weak products indicate minor ends up- and downstream of the main terminus (A). Higher resolution by separation of Atcox1-5 extension products alongside sequencing reaction products from the same primer shows major products at positions −241 to −239 with the same quantitative distribution as seen in the CR–RT–PCR (B). For further analysis by yet another independent method a northern blot experiment was performed. This assay detects a single transcript of about 1900 nucleotides in each RNA preparation consistent with the size of 1872 nucleotides calculated on the basis of the mapping data (C). In summary, these results clearly show that in the major transcripts range from 241 bp upstream of the ATG to 47 bp downstream of the stop codon consistent with an mRNA of 1872 nucleotides. Thus, the major 5′ terminus is located 114 bp downstream of the previously identified promoter 355 bp upstream of the ATG (). In addition, these data document again the reliability of the CR–RT–PCR analysis, particularly of the approach with direct sequencing of PCR products. Using the experimental approaches shown in we analyzed the transcripts of the mitochondrially encoded proteins in . Details of the individual mapping procedures and their results are shown in the gene-specific supplementary figures (Supplementary Figures 1–27). These indicate the primers used for cDNA synthesis as well as the amplification reactions and document the results of the PCRs including the positions of the 5′ and 3′ ends. The two CR–RT–PCR strategies detect those ends, which are most abundant and which are located within distances of a few hundred nucleotides from the primers used. This assumption was independently confirmed for many transcripts, either by northern or primer extension analyses for instance for and . All 3′ termini detected are given in . For all mRNAs single slightly scattering 3′ ends are found, the only exception being the transcript for which two ends were detected. One of these ends (around +115) shows an unusually wide scattering over several nucleotides (). The identified 3′ ends are mostly located downstream of the reading frame with 3′ UTRs up to 498 nucleotides found for . The major 3′ termini of the and genes are found within these reading frames (: −46 and : −17 before the stop codon) confirming recently published results (). Additional 3′ ends downstream of these reading frame-internal major 3′ ends can be detected only at particular processing products. Two other 3′ termini were identified within the two pseudo genes ψ and ψ, the latter corresponding to the 3′ end of the mRNA (). A MEME analysis of sequences flanking the 3 ends (−/+ 20 bp) did not reveal any conserved sequence motif. However, it is noteworthy that transcripts ending with a guanidine are under-represented. Nine transcripts each end with cytidines and uridines, respectively, while the mRNAs of six genes end with adenosines, which is roughly the random distribution. Sequences flanking the 3′ termini were also screened for the presence of inverted repeats with the potential to form stem–loop structures (SL). The most striking structure is a double stem–loop found exactly upstream of the 3′ end of the mRNA (). This has been observed in a previous study, which suggested this structure to have a function in the exonucleolytic maturation of this end (). Parts of this double-inverted repeat are duplicated in the 3′ flanking region of exon e with the downstream-located SL being identical. The mature transcript termini being at identical positions as the mRNA 3′ ends indicate a common function of this duplicated SL in formation of these 3′ ends. A single stem–loop with different primary sequence is found in an analogous position directly upstream of the 3′ terminus, which suggests a similar function of this SL in the generation of this end (). Inverted repeats in the 3′ UTRs of the and genes form single SLs and are positioned several nucleotides upstream of these ends (data not shown). Two SLs separated only by 10 nucleotides are present in the 3′ UTR. Interestingly the 3′ terminus of the latter transcript maps one nucleotide downstream of the 3′ end of a short stable RNA (Ath-377) identified in a general RNomics analysis in (). Similarly the 3′ end of the - mRNA coincides with the terminus of Ath-290. The 5′ termini identified in our analysis are listed in . These ends could originate directly from transcription initiation (referred to as primary ends) or could be formed by post-transcriptional processing (known as secondary ends). The mechanism of RNA ligation allows only the connection of 5′ monophosphate ends and excludes 5′ termini with two or three phosphates which are expected at primary ends derived from transcription initiation (,). However, the triphosphate ends are rather unstable and thus primary ends can also carry 5′ monophosphate groups, which allows their direct ligation and detection without treatment of the RNA. This was likewise observed previously (). Several major 5′ ends detected in our experiments have been identified previously as primary ends in mitochondria (). In agreement with the results of this study we identified identical 5′ termini for -1 (−200), (−157), (−228) and (−205) mRNAs. For we found an additional minor end at position −224 instead of −226 as detected before. The major 5′ end of the mRNA at position −140 could as well be of primary origin, since it maps within a perfectly conserved promoter motif (). In addition, a minor end of (−1898) was found at the predicted site within the CNM1 promoter motif. This terminus was only observed in a CR–RT–PCR experiment designed to identify mRNA ends in this specific region (Supplementary and E). Considering the variety of different promoter sequences found in mitochondria () a clear assignment of some of the ends identified in this study is difficult. Nevertheless, it is reasonable to assume that most of the 5′ ends detected are the result of post-transcriptional processes since the surrounding sequences do not show any similarity to previously characterized promoter motifs (,,). Of the 30 major ends potentially derived from post-transcriptional processing only four transcripts start with a cytidine, seven with guanidine, nine with uridine and 10 with adenosine. As for the 3′ ends no conserved general sequence motifs emerge at or within the sequence flanking the secondary ends, while the MEME analysis identified the promoter motif when the primary ends were analyzed (data not shown). However, we notice a striking similarity between the primary structures surrounding the 5′ terminal nucleotides of the 26S rRNA, the and the 9 mRNAs (). Here 24 nucleotides are perfectly conserved between 26S rRNA and , of which 22 are identical also in the sequence. This suggests that these primary sequences might be important for the generation of these ends and/or that they are part of a conserved secondary structure element involved in signaling this processing. The 5′ UTRs extend up to 645 nucleotides () as described earlier. The mRNA of has no 5′ UTR at all, the major 5′ terminus being identified at position +2 within the ATG. Since an approach to map further upstream-located 5′ ends failed, it remains unclear whether this gene is functional or not. The mitochondrial genome encodes a number of genes that are potentially transcribed within dicistronic transcripts. This includes --- and -. All these genes were examined by CR–RT–PCR specifically assaying for monocistronic mRNA or dicistronic transcripts. Clear PCR products and ends were found for potential monocistronic mRNAs of (3′ end at +106), while no distinct ends were detected for monocistronic transcripts of the other nine genes found within tandem arrangements. The CR–RT–PCR assays of the downstream genes of the - and - arrangements detected 5′ ends exclusively upstream of the 5′ located genes suggesting a co-transcription of these genes. Likewise, the analysis of the upstream gene of - detected a 3′ end exclusively downstream of the 3′ located gene. The potentially dicistronic genes were also investigated by CR–RT–PCR assays in which one primer annealed to the 5′ terminal region of the upstream reading frame and the other to the 3′ terminal part of the downstream-located gene. In all instances, these primer combinations yielded distinct and clean PCR product. This together with the lack of clear ends of potential monocistronic RNAs suggests dicistronic transcription for these genes. In the case of additional monocistronic mRNAs seem to be present in the steady state RNA pool, while no such transcript is detectable for . As indicated in the previous section most 5′ ends detected in our transcript end analysis are most likely generated post-transcriptionally. Since in plant mitochondria no 5′ exonucleolytic activity has so far been detected these ends are most likely derived from endonucleolytic cleavage (). To obtain experimental evidence for an endonucleolytic 5′ processing reaction we used an experimental approach in which the mitochondrial 5S rRNA was used as an ‘anchor’ molecule. This rRNA was ligated to the 3′ end of the potential 5′ cleavage product (for details see Materials and Methods section and A). For the 5′ cleavage product a 195-bp cDNA fragment is expected in an amplification reaction with primers Atcox1-lm.H (−329 to −291) and At5S-mega.R (+107 to +82) on a cDNA template synthesized from oligonucleotide At5S-5 (+118 to +100). When products of this PCR were separated on an agarose gel, the predominant cDNA fragment indeed showed the expected size (B). This product was thus cloned and sequenced. In a total number of 20 clones, 19 cDNA inserts contained 3′ ends corresponding to nucleotide position −242. This is the position exactly upstream of the major mature 5′ end of the mRNA at −241 (D and , and ). An additional end was found at −239 for which the corresponding 5′ end of the 1 mRNA at −238 has not been detected, indicating that the ends at these positions might vary slightly more than seen in the above described analysis of the 5′ mRNA extremities ( and ). The results from this approach strongly suggest that the major 5′ end of the transcript around position −241 is generated by endonucleolytic cleavage. Interestingly 13 cDNA clones of the 5′ leader RNAs contained non-encoded cytidines and adenosines at the 3′ termini. These nucleotides might be added by a terminal tRNA nucleotidyltransferase, which adds the 5′ -CCA-3′ triplet to tRNAs and indeed a perfectly conserved pseudo-uridine arm is found upstream of this 3′ end. In fact, 74 nucleotides of the 5′ leader molecule can be folded into a tRNA-like structure (C), a so-called t-element (). This clearly indicates that the 5′ end of the mRNA at position −241 is generated by a cleavage reaction most likely catalyzed by RNase Z. This result raises the question whether also other termini of mitochondrial mRNAs might be generated by RNase Z or by RNase P. But also mRNA ends not associated with a t-element are generated by an endonucleolytic cleavage. Analogous experiments performed to detect potential processed 5′ leader molecules revealed that also the major 5′ end at position −83 is derived from an endonucleolytic cut (Supplementary Figure 28). To check whether other mitochondrial 5′ and also 3′ mRNA ends could be derived from cleavage by RNase Z or RNase P we screened the complete sequence of the mitochondrial DNA for the presence of the highly conserved 5′-GGTTCRANYCC-3′ motif of the pseudo-uridine arm of tRNAs. Apart from the canonical tRNAs this motif is present upstream of the 5′ ends of and as well as downstream of the gene-internal 3′ ends of and , indicating that these conserved sequences could be involved in the generation of 5′ ends as well as 3′ termini. Folding of these sequences revealed that the conserved motifs are parts of t-elements with ends coinciding with the experimentally mapped transcript termini (). To gain more experimental evidence for the importance of these t-elements in RNA processing, we searched for 3′ ends mapping downstream of the reading frame-internal major 6 and 3′ termini. Such ends would be expected if the t-elements are cleaved off from respective precursor RNAs as postulated. For the identification of the 5′ and 3′ ends of the t-element we used the same approach as described above for the detection of the 3′ end of the 5′ leader. To identify the extremities of the t-element, 18S rRNA was used as anchor molecule instead of 5S rRNA in an otherwise identical approach (Supplementary Figures 29 and 30). Direct sequencing of the PCR products revealed 5′ and 3′ ends of the t-element at position −36 (240 021) and +37 (239 951), consistent with the t-element model presented in . The t-elements contain single non-encoded cytidines at their 3′ termini. An experimental approach to detect the +37 end at the mRNA failed. Obviously t-element processing is very efficient, so that this 3′ end cannot be considered as a genuine mRNA terminus. In the case of the t-element, a 5′ terminus is found two nucleotides downstream (−15) of the reading frame-internal mRNA 3′ end (−17) strongly suggesting a cleavage by RNase P at the site predicted by the model (). Likewise, the 3′ terminus of this t-element was mapped at position +40 also consistent with the predicted t-element secondary structure and a cleavage by RNase Z (). Like the t-element derived cDNAs, also cDNA clones originating from the t-element contained non-encoded cytidines and adenosines (Supplementary Figure 29K). In summary, these results suggest that both tRNA processing enzymes generate 5′ and 3′ ends of plant mitochondrial mRNAs. This does not only occur at canonical tRNAs such as tRNA and tRNA located upstream of and and tRNA encoded downstream of the genes, but also at t-elements. RNase Z and RNase P are thus responsible for endonucleolytic processing of at least a subset of plant mitochondrial mRNAs. In our study, we have analyzed the 5′ and 3′ mRNA ends of all annotated mitochondrial protein-coding genes in . This comprehensive study reveals several general features of these organellar transcripts in plants. First, almost all transcripts have a single 3′ end that in the vast majority of mRNAs scatters over only few nucleotides. The only exception is the mRNA, which has two 3′ ends 105 nucleotides apart. Second, for many genes we find mRNAs with different 5′ ends. Third, with the exception of the primary 5′ ends within promoter sequences, there is no conserved sequence motif evident at the 5′ termini or at the 3′ ends. Thus other determinants govern the post-transcriptional generation of these ends. Fourth, the majority of the 5′ ends are most likely derived from processing. There are several arguments that support this conclusion. Most of the ends identified do not terminate in any known promoter motif (). In addition, upstream of many of the 5′ ends detected in our study promoter sequences have been found for instance in the case of () and potential conserved promoters can be predicted upstream of (−640 in ecotype C24, here 5′ ends have been found at this position), (−1320 and 1230), (−1400) and (−890). Moreover, our analysis shows that the 5′ ends of the and the mRNAs are generated by endonucleolytic cleavage. Analogous cleavages may generate other secondary 5′ and 3′ ends (for a detailed discussion see subsequently). Our analysis also revealed non-encoded nucleotides most likely attached to the 3′ ends. Mostly only single adenosines are found and less frequently cytidines, only in rare instances thymidines (corresponding to uridines in the RNA) or guanosines. The extensions can be grouped into two categories as observed previously (,,,). They are either oligohomopolymeric adenosine stretches (up to 24 adenosines) or short extensions of adenosines and cytosines. However, the vast majority of the mRNAs do not have non-encoded nucleotides consistent with previous assumptions (). Our analyses also confirm that RNA editing in the 5′ and 3′ UTRs is a rare event. Apart from the editing sites found previously (), we detected new editing sites only in the 5′ UTR of . However, a blast search revealed that these editing sites are located within a 186-bp duplication of a part of the reading frame, where exactly the same cytidines are edited. The three C to U conversions are found at positions −286 (position 241 044 = 77069 in ), −308 (241 066 = 77091) and −316 (241 074 = 77099). We also identified a new editing site in the reading frame at position 302 265, which is only edited at a rate of about 35%. In contrast, some postulated editing sites were never seen edited in our analysis. This includes cytidines at position 219 244 in the peudo gene and at position 260 938 in . Taken together, this comprehensive investigation of mitochondrial mRNA ends in provides new general insights into the nature and the generation of these ends and thus represents a broad basis for further detailed studies of individual mRNA ends. As mentioned earlier most or at least a large portion of the 5′ termini could be derived from post-transcriptional processing. We provide clear experimental evidence for an endonucleolytic generation of four 5′ mRNA ends (). While this kind of processing was expected for and , where tRNAs are located upstream of the genes, we now find that also a t-element upstream of acts as -element most likely directing cleavage by RNase Z. A similar scenario can be expected for the 5′ ends of and , where also t-elements can be found 5′ of the mature mRNA ends. In line with the role of such elements in 5′ end formation, it can be assumed that even minimal secondary structures such as simple stem–loops can be recognized as processing signals by tRNA processing enzymes. Three such structures might direct RNase Z to generate two 5′ termini (−459 and −406) and an -2 5′ end (−268), while a simple SL downstream of the (−375) 5′ terminus suggests RNase P to generate this end (). But there must be an additional mode of processing since no t-element or any other obvious secondary element is found at the 5′ cleavage site of mRNA and many other potential cleavage sites. Maybe a totally different set of proteins including a so far unknown endonuclease generates these ends. But it might be also possible that specific binding proteins allow an action of the tRNA processing enzymes on substrates that do not resemble a tRNA or parts of the tRNA secondary structure. While exonucleolytic processes had been suggested for formation of at least some 3′ ends, we now found evidence for direct endonucleolytic generation of 3′ ends or at least for the involvement of endonucleolytic cleavage in 3′ end formation. This has been similarly found during the investigation of Ogura cytoplasmic male sterility (CMS) in Brassica cybrids. Here a t-element was identified downstream of the , which is responsible for the CMS phenotype. Most likely cleavage by RNase P generates the 3′ end of mRNAs. In addition also 3′ cleavage of the t-element is observed giving rise to stable t-element RNA (). Analogous to 5′ end generation, direct endonucleolytic 3′ end generation is expected when a canonical tRNA is located 3′ to an mRNA, which results in a cleavage by RNase P as for instance for both mRNAs (). At least an essential role of endonucleolytic cleavage in 3′ end formation can be assumed for and mRNAs. For the former we found the 3′ end of the mRNA and the 5′ end of the respective t-element to be separated by a single nucleotide, indicating that this adenosine might be removed in a trimming process involving RNR1. Likewise, we identified the 5′ terminus of the t-element 10 nucleotides downstream the respective 3′ mRNA ends suggesting these nucleotides also to be removed by RNR1. It would thus be interesting to see whether such a concerted action of endonucleolytic cleavage and exonucleolytic trimming is frequently observed in the formation of mitochondrial mRNA ends. Interestingly, both ends derived from this kind of processing are located within the reading frame. Also in analogy with 5′ end processing, it is also possible that stem–loops might be -elements recognized by tRNA processing enzymes resulting in an endonucleotlytic generation of a 3′ end. Such a scenario might be expected for the 3′ end formation of the mRNA directed by a SL (), but studies are required to discriminate between endo- and/or exonucleolytic generation of this end. The involvement of tRNA maturation in dismantling large precursor RNAs occurs in mitochondria of animals (). Here protein-coding genes are usually interspersed by at least a single tRNA. After transcription of a complete strand of the mitochondrial genome endonucleolytic removal of the tRNAs results in the generation of individual mRNAs. This schema of processing seems at least partially conserved in plant mitochondria although the protein coding genes are often separated by large spacer sequences. The detection of many secondary 5′ ends raises the question why some mRNAs retain their original primary 5′ ends while most others are processed. A size reduction of the 5′ UTRs would be a simple explanation, but even mRNAs with short original 5′ UTR (for instance ) are processed, while some of the processed ends () define larger 5′ UTRs. Of course the nature of the 5′ UTRs could influence translation and it is possible that some mRNAs are only accessible for translation after 5′ processing. Here two scenarios are possible. Either a potential ribosome entry sequence is blocked in the non-processed RNA or binding of specific translation factors is not possible prior to processing. The first scenario seems to be rather unlikely since no Shine–Dalgarno-like sequence has been identified in plant mitochondria. Although conserved sequence elements have been found in some mRNAs () the restriction of this motif to certain mRNAs excludes a general function. Moreover, there is so far no experimental evidence for the importance of these conserved sequences. More likely seems the second scenario. mRNA-specific translation factors have indeed been found in mitochondria from () and in chloroplasts from () and higher plants (). In the latter case. the CRP1 protein was found to specifically interact with 5′ UTRs of and psaC mRNAs and this protein activates the translation of these RNAs. CRP1 belongs to the PPR protein family of which many members are transported to mitochondria, where they could exhibit similar functions. In there seems to be only little variability at the 3′ ends of mitochondrial mRNAs and apart from a slight scattering over a few nucleotides the termini are determined very clearly. This provokes two questions: how are 3′ ends generated and defined and what mechanisms exist that stabilize mRNAs and prevent them from degradation? Both questions might be closely related since exonucleolytic 3′ end maturation can be considered as a process leading from a precursor RNA towards as stable mature mRNA. Recently, two 3′ exoribonucleases have been identified as important -factors necessary for the generation of both and mRNAs () with the PNPase also being responsible for the degradation of dispensable RNAs (). An important role of these exoribonucleases in 3′ end processing is highly likely and protection of an RNA from exonucleolytic degradation is probably crucial for stability of a plant mitochondrial mRNA. In chloroplasts, SLs located in the 3′ UTRs, are important for both 3′ end trimming and stability of the mRNA (). Likewise, such structures have been suggested to protect plant mitochondrial mRNAs in some instances (,) but the comprehensive survey now clearly reveals that the presence of such structures at the 3′ end of mitochondrial mRNA is an exception rather than the rule. The vast majority of the mRNA 3′ UTRs does not contain any obvious single or double SLs suggesting that other -elements determine the stability of an mRNA. Thus, in plant mitochondria there seems to be an mRNA stabilizing and also 3′ processing system that is different from the one that has been suggested for chloroplasts. This discrepancy, obvious at least in terms of -elements, is somehow surprising since both organelles contain very similar and identical, respectively, 3′ exoribonucleases. Thus one might speculate that different auxiliary factors might communicate between the different -elements and the exoribonucleases. But what determines 3′ end formation and mRNA stability in plant mitochondria? Our data clearly show that there is neither a conserved primary structure nor a conserved secondary structure element. Thus, individual mRNA-specific determinants might exist, either at the primary sequence or secondary structure level and it might well be that also individual -factors might be involved. This might be also reflected by the ribonucleases involved in 3′ end formation. While some ends might be generated exclusively by exoribonucleases, we now found clear evidence for the participation of endonucleolytic activities in this process. This is apparent for the two mRNAs, whose 3′ ends are directly created by the cleavage of RNase P at the 5′ end of tRNA (UGA). But also the formation of the 3′ ends of the truncated and mRNAs requires endonucleolytic cuts. However, these 3′ termini are formed by a concerted action of both endo- and exoribonucleases. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

In plants, cytosolic-free calcium level (Ca) plays pivotal roles as an intracellular second messenger in response to a variety of stimuli, including light, phytohormones, oxidative stress, drought, cold and pathogens (,). One of the earliest events in response to pathogen attack is an increase in the [Ca], with complex changes in its amplitude, frequency and duration (,). A Ca influx has been shown to be essential for the activation of defense-related genes, phytoalexin biosynthesis and hypersensitive cell death (). Ca signals are sensed by intracellular Ca-binding proteins and transduced by downstream effector proteins that regulate cellular processes (,). Calmodulin (CaM), one of the best characterized Ca-binding proteins, contains four helix–loop–helix Ca-binding motifs referred to as EF hands. Ca-bound CaM transduces Ca signals by modulating the activity of numerous diverse CaM-binding proteins such as metabolic enzymes, transcription factors, ion channels, protein kinases/phosphatases and structural proteins, which generates physiological responses to various stimuli (,). Mammalian cells have only a few CaM genes encoding one or a few isoforms. In contrast, plants possess a large repertoire of CaM and CaM-like genes that encode several CaM isoforms (,). In all plants examined, CaM genes, even those encoding the same isoform, are differentially expressed in response to various external stimuli such as touch, heat shock, cold, light, auxin and pathogens (). We previously cloned five CaM isoforms (GmCaM1–5) from soybean (). The conventional isoforms GmCaM1, −2 and −3 share more than 96% identity with mammalian CaM, while the divergent isoforms GmCaM4 and −5 share only 78% identity to GmCaM1; these are the most divergent isoforms reported thus far in the plant and animal kingdoms (,). The cellular level of GmCaM4 rapidly and dramatically rises in response to specific stimuli such as pathogen exposure and salinity (). Furthermore, ectopic expression of GmCaM4 or GmCaM5 in tobacco and confers increased pathogen resistance through the formation of spontaneous hypersensitive response-associated lesions (even in the absence of pathogens), which is not mediated by elevated levels of salicylic acid but by increased levels of systemic acquired resistance gene expression (). In addition, these transgenic plants exhibit salt tolerance and can accumulate high levels of proline (). These results demonstrate that Ca signaling mediated by specific CaM isoforms contributes to pathogen and salt stress resistance in plants. Homeodomain proteins in the homeobox gene family play important roles as transcription factors in plant, animal and fungal development (). Mutant analyses of monocot and dicot plants showed that leaf development involves the down-regulation of meristem-specific homeobox genes such as the knotted-like homeobox (knox) genes (,). Ectopic expression of these genes leads to altered cell fates in the leaf, suggesting a pivotal function in early leaf development (). However, it is not well established that homeodomain proteins are involved in plant stress signaling. Pathogenesis-related homeodomain proteins from parsley and pathogenesis-related homeodomain proteins from (PRHP and PRHA), members of the PHD finger subfamily, were isolated on the basis of their interaction with a 125-nt region within the promoter, which is rapidly stimulated by fungal or bacterial elicitors (). We are interested in understanding how plants recognize biological and environmental stresses such as pathogens and salinity, and how signals are transduced to activate transcription of the gene in response to such stresses. For this purpose, it is crucial to identify -acting elements and -acting factors that regulate expression. We previously identified two regions (−1286 to −1065 and −858 to −728) in the promoter that are bound by proteins induced by pathogen or NaCl exposure (). The GT-1 -acting element in the −858 to −728 bp region of the promoter is partially involved in expression by interacting with a GT-1 like transcription factor in response to pathogen and salt stresses (). However, -acting element(s) in the −1286 to −1065 region of the promoter and their DNA binding proteins remained to be identified and characterized. Here, we report the presence of two repeats of a conserved homeodomain binding site, ATTA, within the −1286 to −1065 region of the promoter. We used the yeast one-hybrid system to isolate two transcriptional regulators, GmZF-HD1 and GmZF-HD2, which specifically bind to a 30-nt A/T-rich -acting element within the −1207 to −1128 region. Supershift and transient expression assays confirmed that GmZF-HD1 is a component of DNA-nuclear proteins complexes formed and that it functions as an transcriptional regulator of pathogen-responsive expression of the gene. Soybean suspension cells ( L. cv. Williams 82; W82) were cultured in MS medium supplemented with 0.75-mg l benzyl adenine and maintained at 25°C in the dark with shaking at 130 r.p.m. pv carrying () was grown and manipulated as previously described (). The strain XL-1 Blue (Stratagene La Jolla, CA, USA) was used for cloning. Glutathione S-Transferase (GST) fusion proteins were expressed in BL21 (pLys S) DE3. The yeast strain YM4271 () was used as a host for transformation with a reporter vector in the yeast one-hybrid screen (Clontech). Yeast one-hybrid screening was performed using the MATCHMAKER one-hybrid system (Clontech). To make a target reporter construct, four tandem repeats of the A2 region of the promoter (positions −1207 to −1128 bp) containing putative -acting elements were inserted into the I and I sites of and the I site of . The two expression plasmids were simultaneously transformed into YM4271. Cells from a 300-ml culture were transformed with soybean cDNA libraries and plated on synthetic minimal medium containing 20 mM 3-aminotriazole (3-AT) but lacking His and Leu. The soybean cDNA library was constructed using the yeast expression vector (Stratagene) and RNA isolated from apical and elongating regions (0.3–1.3 cm from the cotyledon) of 4-day-old etiolated soybean hypocotyl tissue. After incubation at 30°C for five days, positive colonies were cultured in YPD and selected on medium containing 20 mM, 45 mM or 60 mM 3-AT but lacking His and Leu. Colonies were then transferred to filter paper and tested for β-galactosidase activity. Plasmids were extracted from positive colonies, amplified in , and purified for sequencing. Nuclear protein extracts were prepared as described () from soybean suspension culture cells (W82). Aliquots were taken for determining protein concentrations with the Bradford protein assay kit (Bio-Rad), and aliquots of the extracts were frozen in tubes at −70°C. EMSA was performed as described (,) using [P]-labeled DNA probes. Soybean nuclear extracts or GST fusion proteins were pre-incubated in a binding buffer (20 mM HEPES, pH 7.9, 0.5 mM DTT, 0.1 mM EDTA, 2 μg poly dI/dC, 50 mM KCl, 15% glycerol) for 10 min at room temperature and then incubated with 40 Kcpm of end-labeled probes for 20 min. The resulting protein–DNA complexes were electrophoresed in non-denaturing 8% polyacrylamide gels in 0.5 X TBE buffer. The gels were dried and exposed to X-ray films. The DNase I footprinting assay was performed as described (,) using a 118 bp DNA fragment containing −1207 to −1128 bp of the promoter and Easy vector (Promega) sequences between the SphI and SpeI sites. The fragment was labeled with α-P isotope at the SpeI site. A DNA-nuclear protein binding mixture (100 μl) containing 10 mM HEPES pH 7.9, 60 mM KCl, 1 mM EDTA pH 8.0, 7% glycerol and 1 mM DTT was pre-incubated for 20 min at room temperature, one unit of DNase I cofactor solution (50 μl) was added, and incubation was continued for 3 min at room temperature. Stop solution (100 μl) with 100 μg of proteinase K was immediately added and incubation was continued for 30 min at 55°C. Total reaction mixtures were extracted with Tris-saturated phenol, and products were precipitated by adding 25 μl 3 M sodium acetate, pH 5.2, and 500 μl 100% ethanol. The final pellet was re-suspended in sequencing loading buffer (0.1 M NaOH:formamide [1:2], 0.1% bromophenol blue, 0.1% xylene cyanol), and the mixture was heated at 100°C for 2 min, chilled on ice, and loaded onto 8% polyacrylamide-urea DNA sequencing gels. Full-length and cDNA clones were cloned into the BamHI and XhoI sites of the expression vector (Amersham Pharmacia Biotech), respectively. The resulting constructs were confirmed by sequencing and introduced into BL21 (pLys S) DE3, and GST fusion proteins were expressed and purified using glutathione–agarose beads according to the manufacturer's instruction (Amersham Pharmacia Biotech). Various tissues harvested from 4-day etiolated soybean ( L. cv. Williams) seedlings and W82 cells collected on filter papers by vacuum filtration were used for isolation of total RNA. RNA gel blot analysis was carried out as previously described (). Total RNA (20 μg each sample) prepared from apical hypocotyls, elongating hypocotyls, mature hypocotyls, plumule, seed, root and pathogen-treated soybean suspension cells () was separated on 1.5% formaldehyde/agarose gels, transferred onto a nylon membrane (Hybond N+, Amersham), and hybridized with the cDNA probe. To verify equal loading, rRNA was visualized by staining with ethidium bromide. Hybridization and washing were performed under high stringency conditions. A GmZF-HD1 antibody was raised against the GST::GmZF-HD1 protein in rabbits (Animal Culture Facility, Gyeongsang National University). A protein A-Sepharose column was used to separate the IgG fraction of the rabbit serum. After adjusting the pH of the crude serum to 8.0 by adding 1/10 volume of 1.0 mM Tris–HCl (pH 8.0), the antibody solution was washed extensively with 10 column vol of 100 mM Tris–HCl (pH 8.0) and in turn 10 mM Tris–HCl (pH 8.0). The antibody was eluted with 0.1 M glycine–HCl (pH 2.5) and immediately neutralized with 0.1 vol of 1 M Tris–HCl (pH 8.0). The supershift assay was performed by pre-incubating affinity-purified anti-GST::GmZF-HD1 antibody and nuclear extracts together, and the samples were then processed as for EMSA. Control experiments were performed by incubating the same amount of IgG fraction of preimmune antiserum in the binding reaction mixtures. A tetramer of the A2 region in the promoter was fused to a minimal −46 promoter-GUS reporter gene (A2 region-m35S-GUS) (). To construct the effector plasmid, full-length cDNA was inserted into a plant expression vector () containing the promoter and terminator (). Isolation of mesophyll protoplasts and polyethylene glycol-mediated DNA transfection were performed as previously described (). In each transformation, an protoplast suspension (5 × 10 per ml) was transfected with 10 μg of reporter construct alone or together with 15 μg of the effector construct or a vector DNA control (). Transfected protoplasts were incubated in W5 solution for 16 h in the dark. Glucuronidase (GUS) expression was detected fluorometrically with the substrate 4-methyl umbelliferyl glucuronide, as described (). LUC assays were performed using the Promega luciferase assay system according to the manufacturer's instructions. To normalize the transfection efficiency, the promoter- control vector was cotransfected in each experiment as described (). Based on DNA binding assays (electrophoretic mobility shift assay, or EMSA) and transient expression assays in protoplasts with deletion constructs, we previously showed that two regions of the promoter are involved in pathogen-mediated gene regulation (). One is a -acting element within the −858 to −728 region of the promoter. However, another region upstream of the -acting element, from −1286 to −1065, remained to be further characterized. In this study, we undertook to identify core -acting elements upstream of the -acting element that specifically interact with nuclear extracts from pathogen-treated cells by dividing this 220-nt region into three overlapping fragments (A1, A2 and A3), which were used as probes for EMSA (). The EMSA data showed that a stable binding complex formed only between the A2 probe (−1207 to −1128) and nuclear extracts from pathogen-treated cells (A). To first prove that the complex originated from the interaction of the A2 probe with proteins but not with contaminants in the extracts, we also performed EMSA using the A2 region as a probe with extracts heated for 5 min at 65C, or treated with RNase A or proteinase K. We found that the binding complex was heat-labile and sensitive to proteinase K but not RNase A, confirming that it was a DNA–protein complex (data not shown). We then designed seven overlapping double-stranded oligonucleotides (A2–1 to −7) to further map the binding site and performed competition assays using the A2 region as a probe (B). Remarkably, a 100-fold molar excess of the unlabeled A2–2 oligonucleotide completely inhibited the binding of nuclear proteins to the A2 region, but an excess of any of the other six oligonucleotides did not have this effect (C). We then employed a general and powerful tool for identifying transcription factor-binding sites in promoter sequences, the DNase I footprinting assay. We used the A2 region as a probe with various concentrations of nuclear extracts from pathogen-treated cells. Similar to what we found with the competition assay (C), a region covering ∼20-nt (AGTAATTAAATAATAAATTA) and almost identical to the A2–2 sequence was protected from DNase I digestion by nuclear extracts (). These results suggest that a -acting element within the A2–2 region of the promoter interacts with pathogen-activated nuclear protein(s). The 20-nt sequence identified by the DNase I foot-printing assay, AGTAAATAATAA, contains two core consensus sequences (underlined) that have been identified as motifs bound by homeodomain proteins. As shown in the competition assay (C), neither the A2–5 nor the A2–6 fragment, which contain a partial or full core sequence, respectively, competitively inhibited the binding of nuclear proteins to the A2 probe, suggesting that the full length of the A2–2 fragment is required for binding nuclear proteins. To determine the overall sequence sufficient for binding nuclear proteins, we tested fragments of varying length that include the A2–2 sequence at the 5′ end. The result of this experiment indicated that an additional 10 bp of sequence flanking the A2–2 fragment was necessary for the maximal binding of pathogen-induced nuclear proteins (data not shown). Therefore, we used this 30-bp sequence for further experiments. We used the yeast one-hybrid screening system () to isolate cDNAs encoding DNA binding proteins that interact with the 80-nt A2 region in a sequence-specific manner. As a dual reporter system, we first constructed a yeast strain (YM4271) carrying integrated copies of and with four tandem repeats of the 80-nt A2 fragment of the promoter (A). In the resulting strain, the gene is transcribed at a basal level, permitting growth on minimal medium lacking histidine but not in the presence of 20 mM 3-aminotriazole (3-AT), a competitive inhibitor of the gene product. YM4271 was then transformed with soybean cDNA libraries by the lithium acetate/polyethylene glycerol method, and a ∼1.5 × 10 transformants were screened. Forty-two positive clones were selected on minimal medium lacking histidine and containing 20 mM 3-AT. Plasmids were recovered and electroporated into an host for propagation. The cDNAs from the isolated plasmids were subjected to restriction and sequencing analysis, which allowed the 42 cDNA clones to be classified into 16 distinct groups (). To select true positive cDNAs that encode transcriptional regulators among the 16 representative cDNA clones, we examined their activity by a filter-replica method using 5-bromo-4-chloro-3-indolyl-β--galactopyranoside (X-gal) as a substrate. inc inger-omeoomain protein1 and -2, respectively) formed blue colonies on filter papers containing X-gal (B). With the selected 16 cDNA clones, we also performed a cell growth assay on selection medium without histidine but with a higher concentration of 3-AT (60 mM). The data again showed that only clones 5 () and 1 () survived on medium containing 60 mM 3-AT. These results confirmed that the GmZF-HD1 and −2 proteins specifically and strongly bound to the A2 region of the promoter and activated transcription of the dual reporter genes in yeast, suggesting that they were good candidates as transcriptional regulators of pathogen-induced gene expression. To examine the structures of the and − cDNA clones, we sequenced the two cDNAs, ∼770-nt and 660-nt, respectively, which revealed that both were full-length. The cDNA encodes an open reading frame of 182 amino acids, which specifies a putative protein with a predicted molecular mass of about 20 kDa. The cDNA encodes an open reading frame of 176 amino acids, which specifies a putative protein with a predicted molecular mass of about 19.4 kDa (A). We submitted the nucleotide sequences of the two cDNAs to the GenBank/EMBL/DDBJ database (accession numbers AY695729 and AY695730 for the cDNA and cDNA, respectively). We then searched databases for sequences homologous to those of the GmZF-HD1 and −2 proteins. We found that each GmZF-HD protein has two highly conserved putative zinc finger domains in the N-terminus and a DNA binding homeodomain in the C-terminus, which is characteristic of homologs in the plant ZF-HD family, including AtHB24 and −26 of the C3 plant and FtHB1 of the C4 plant (A). The ZF-HD class of homeodomain proteins may be involved in the photosynthesis-related mesophyll-specific gene expression of phosphoenolpyruvate carboxylase in C4 species of the genus (). More recently, it was proposed that ZF-HD transcriptional regulators play overlapping regulatory roles in floral development (). However, there is no prior evidence that ZF-HD homeodomain proteins are involved in pathogen signaling and plant defense mechanisms. Interestingly, PRHA in the PHD finger sub-family has been reported to be involved in the transcriptional regulation of the gene (). A phylogenetic tree constructed by comparing deduced amino acid sequences indicates that GmZF-HD1 and −2 are more closely related to FtHB1 than to PRHA (B). Therefore, these observations imply that the GmZF-HD1 and −2 proteins represent a new class of ZF-HD proteins that is involved in pathogen signaling in soybean and that the characteristic zinc finger domains play a role in protein regulation. The copy number of the gene in the soybean genome was estimated by Southern blot analysis. Simple patterns of hybridization were observed under high-stringency conditions, suggesting that is present in single- or low-copy number in the soybean genome (data not shown). The tissue-specific expression pattern of were previously investigated in soybean plants by northern analysis () and by expression of a promoter-β-glucuronidase (GUS) reporter in transgenic plants (). These studies showed that is primarily expressed in the apical hook and elongating hypocotyls in both soybean and transgenic . To determine the profile of expression, we performed northern analysis with tissue-specific RNAs. The transcript was highly abundant in apical hooks and roots, but much less abundant in elongating hypocotyls. It was almost undetectable in mature hypocotyls, plumules, and seeds (A). These results are very similar to the expression patterns of in soybean seedlings, suggesting that the GmZF-HD1 protein may be involved in regulating the expression of . It was previously reported that expression of the gene is dramatically induced by pathogen treatment (). Therefore, we investigated the expression of in response to pathogen treatment. Although was highly expressed in soybean suspension cells under normal conditions, it was not significantly induced by pathogen treatment (B). This result suggests that the activity of in the induction of gene expression in response to pathogen may be controlled at post-transcriptional instead of transcriptional levels. To evaluate whether GmZF-HD1 and -2 bind to the 30-nt sequence containing the two repeats of a conserved ATTA homeodomain binding site in the promoter, we produced recombinant full-length GmZF-HD1 and −2 proteins fused to GST in . The two full-length recombinant proteins were similarly capable of binding to the 30-nt sequence in the gel shift assay (B). Then, to test which domain (ZF or HD) is involved in DNA binding, we performed EMSA with C-terminal HD deletion derivatives (ΔHD) and N-terminal ZF deletion derivatives (ΔZF) of each protein (A). As shown in B, EMSA performed using the 30-nt sequence as a probe with the purified deletion mutants showed that the N-terminal ZF deletion mutants (i.e. those containing only the homeodomain) strongly bound to the 30-nt sequence, but that the C-terminal HD deletion mutants did not (i.e. those containing only the zinc finger domains) (B). We then performed a competition assay and found that DNA–protein complex formation was significantly reduced by an up to 200-fold molar excess of unlabeled 30-nt competitor (data not shown). Since the binding affinity to the 30-nt sequence of proteins containing only the homeodomain (GmZF-HD1ΔZF and −2ΔZF) was much stronger than that of the full-length proteins (GmZF-HD1 and −2), and since GmZF-HD1ΔZF and GmZF-HD2ΔZF bound with similar affinity to this sequence, GmZF-HD1ΔZF was chosen for further EMSA experiments to characterize its interaction with this region of the promoter. To more precisely map the binding site of the GmZF-HD1 protein, we designed six mutated versions of the 30-nt sequence (M1 to M6), each with 5-nt mutations as outlined in A. The ability of the recombinant GmZF-HD1(ΔZF) fusion protein to bind wild-type or mutated 30-nt sequences was examined using EMSA and competition assays. As shown in B, the GmZF-HD1(ΔZF) protein strongly bound to the M3, M5 and M6 sequences and weakly bound to M4, but not to M1 or M2. To confirm this result, we performed a competition assay using each mutated DNA sequence as a competitor. A 200-fold molar excess of M3, M5 or M6 almost completely inhibited binding to the 30-nt sequence, similar to the inhibition conferred by the wild-type sequence. On the other hand, M1, M2 and M4 as competitors did not affect the formation of DNA–protein complexes, which were similar to those observed in the DNA binding assay (C). These results strongly suggest that GmZF-HD1 binds in a sequence-specific manner to the promoter and that the 20-nt sequence AGTAAANNNNNA containing two ATTA repeats may be a putative core -acting element in this promoter. To confirm that GmZF-HD1 is an active component of the protein-DNA complex observed in EMSA experiments with the 30-nt sequence of the promoter and nuclear extracts from pathogen-treated soybean suspension cells, we raised anti-GmZF-HD1 polyclonal antibodies against the affinity-purified GST::GmZF-HD1 fusion protein. We used these antibodies for supershift assays together with nuclear extracts from pathogen-treated cells (A). The specificity of the antiserum was confirmed by western analysis with recombinant GST::GmZF-HD1 proteins and total and nuclear soybean proteins. The antiserum showed no cross-reactivity to any other proteins derived from total bacterial cell extracts or to any other soybean proteins (data not shown). Incubation of 0.5–1 μg of antiserum, nuclear extracts and the 30-nt probe in the gel shift reaction clearly resulted in a supershift of a DNA–protein complex (A, lanes 6–8), which was not observed in the presence of preimmune serum (A, lane 1) or in the absence of antiserum (A, lane 3). These observations strongly confirm that the GmZF-HD1 protein is a component of the nuclear protein–DNA complex formed with the 30-nt sequence of the promoter and that this complex forms when cells are stimulated by pathogens, but not under basal conditions. After demonstrating that GmZF-HD1 is a component of this pathogen-induced complex, we then examined whether the protein activates downstream transcription in a manner dependent on the 30-nt sequence. To do this, we used leaf mesophyll protoplasts harboring a GUS reporter plasmid, which was constructed by fusion of the GUS gene with four tandem repeats of the A2 region of the promoter cloned upstream of the cauliflower mosaic virus ( minimal promoter in (). We also constructed an effector plasmid by inserting the full-length cDNA sequence after the 35S promoter (B). As shown in C, cells transfected with the reporter construct alone exhibited an increase in GUS activity of about 2.9-fold, compared to control cells transfected with the vector, indicating that endogenous proteins may recognize the promoter sequence in the reporter construct. Co-expression with the effector protein GmZF-HD1 increased GUS activity about 5.1-fold. These results support the conclusion that the GmZF-HD1 protein actively functions as a transcriptional regulator and that it is involved in pathogen-induced gene expression. It is important to understand how plants recognize and transduce stresses such as pathogen infection and high environmental salinity to activate transcription of key cellular-response mediators such as calcium binding proteins, or CaMs. We previously showed that gene expression induced by environmental stresses such as pathogen and salinity is partially mediated by the binding of a GT-1-like transcription factor to the -acting element (GAAAAA) in the −858 to −728 region of the promoter (). In addition, we reported that the −1286 to −1065 region upstream of the GT-1 -acting element is also involved in pathogen-stimulated gene regulation (). In this report, we identified repeats of a conserved homeodomain-binding site, ATTA, in the promoter that are relevant to pathogen responses. To isolate transcription factors that bind the -acting element, we screened soybean cDNA libraries using a yeast one-hybrid system and cloned two cDNAs, and −, encoding homeodomain DNA binding proteins that specifically interact with the A2 region (80 bp) of the promoter (). A previous yeast one-hybrid screening performed in plants had a success rate of 87.8% (36 true/41 total clones) (). In this study, we obtained a success rate of 66.7% (28 true/42 total clones), indicating that our screen was effective for isolating new proteins that bind to the promoter. The homeodomain is one of the most important DNA-binding motifs in diverse organisms such as humans, and plants, and it has been widely studied as a model system for protein–DNA interactions (). The first crystal structure of the homeodomain–DNA complex was revealed in 1990 by Kissinger . (), and subsequent structures of homeodomain–DNA complexes have been found to be very similar (). Our data also show that the homeodomain but not the zinc finger domain is involved in sequence-specific DNA binding (B). Most homeodomains bind to ATTA as a core consensus sequence (). To precisely map the GmZF-HD1 and −2 binding site in the promoter, we performed EMSA using bacterially expressed recombinant GmZF-HD proteins (). Recombinant GmZF-HD1 and −2 proteins specifically bound to the 30-nt sequence of the promoter (B), which is consistent with the results obtained with nuclear extracts from pathogen-treated cells (A, C and ). A supershift assay using an anti-GmZF-HD1 antibody confirmed that GmZF-HD1 is indeed a component of nuclear extracts that form DNA–protein complexes with the 30-nt sequence of the promoter (A). More detailed EMSA and competition assays with a series of mutant oligonucleotides revealed that two repeats of the conserved ATTA homeodomain binding site in the promoter are required for full binding activity with the GmZF-HD1 and −2 proteins (A, B and C). The functional role of GmZF-HD1 and −2 in gene regulation was also confirmed by a transient expression assay performed with protoplasts (C). The zinc finger homeodomain protein ATHB33 was recently reported to bind to a 20-bp sequence (AGTGTCTTGTAAAA), which contains an ATTA core sequence (underlined) that is similar to that of other homeodomain proteins in plants (). In contrast, we showed here that both of the ATTA core sequences in the 30-bp sequence are required for full binding of the GmZF-HD1 protein (). These data suggest that GmZF-HD1 proteins may recognize distinct promoter motifs that other plant homeodomain proteins do not. These distinctions may play pivotal roles in the regulation of different target promoters. It is not yet clear how the GmZF-HD1 and −2 proteins are regulated after pathogen stimulation. To better understand the mechanism of this response, the localization of GmZF-HD1 and −2 before and after pathogen exposure must be determined. Members of the ZF-HD family have two highly conserved N-terminal zinc finger domains and a C-terminal homeodomain (). Interestingly, Windhövel . () demonstrated that the two conserved N-terminal motifs constitute a novel dimerization domain that is involved in homo- or heterodimer formation. Mutagenesis of the second zinc finger domain showed that highly conserved cysteine residues are essential for protein–protein interaction (). In the case of GmZF-HD1 and −2, their binding to ATTA repeats in the promoter may not require dimerization (either homo- or heterodimerization) since a mutant deleted for the N-terminal ZF (i.e. containing only the homeodomain) still strongly bound to the 30-nt sequence containing this site. This result is perhaps unexpected because dimerization through the zinc finger could stabilize binding. Here we speculate that two monomers with N-terminal zinc finger domain truncation may have easier access to the closely spaced binding sites because of reduced steric hindrance. We propose a model in which GmZF-HD1 translocation to the nucleus is important for regulating its transcription of the promoter. This proposal is based on three observations: (i) there is no significant increase in transcript level upon pathogen treatment (B); (ii) the GmZF-HD proteins are present in nuclear extracts from pathogen-treated cells but not in extracts from untreated cells (A); and (iii) post-translational modifications do not seem to play a major role in the binding of GmZF-HD to the promoter because recombinant GmZF-HD1 and −2 expressed in bacteria, in which proteins are not post-translationally modified, exhibit DNA binding activity ( and ). Therefore, a possible scenario is that in response to pathogen attack, an unknown inhibitory protein present in the cytosol no longer prevents the GmZF-HD1 and −2 transcription factors from entering the nucleus where they can interact with the promoter. GmZF-HD1 and −2 are small proteins (20 and 19.4 kDa, respectively) so in principle they could easily diffuse into the nucleus even without a nuclear localization signal sequence. Nuclear proteins extracted from unstimulated cells did not bind to the conserved homeodomain-binding site in the promoter, suggesting that GmZF-HD1 and −2 are not present in the nucleus under basal conditions. The best example of this model would be the activation of the NF-κB transcription factor in animal cells exposed to bacterial pathogens. NF-κB is released from the inhibition of IκB in the cytosol and is then translocated into the nucleus, where it regulates the expression of immune response genes (). This suggestion is consistent with our data, but alternative models can be proposed. For example, post-translational modifications may be important for optimal GmZF-HD activity but not for DNA binding. There may also be an increase in protein expression upon pathogen infection although transcriptional up-regulation does not occur (as judged by northern analysis). At present, it is not known whether discrete -acting elements respond to different developmental and environmental cues or whether single or overlapping DNA motifs bind multiple regulatory factors, for example, as postulated for the parsley phenylalanine ammonia lyase1 () gene () and the bean gene (). We observed that the expression patterns of the and genes were very similar soybean seedling tissues (A), consistent with a role for the GmZF-HD1 protein in binding and controlling the activity of the promoter. Their overlapping patterns of gene expression are also important for the transcriptional activation of in response to the transduction of external pathogen-specific signals into the nucleus. To summarize, we have shown that a new class of plant homeodomain proteins, GmZF-HD1 and GmZF-HD2, specifically bind to a 30-nt A/T-rich DNA sequence that contains two repeats of the conserved homeodomain binding site ATTA in the promoter following exposure to plant pathogens. Thus our results strongly support regulation of the gene by the GmZF-HD transcription factors as a significant component of the plant defense-signaling pathway.

It is now widely accepted that RNA molecules participate actively in virtually all cellular metabolic processes. Unlike DNA, in which double stranded complementary Watson–Crick (WC) base pairs are obligatory to maintain genetic fidelity, RNA molecules are rich in structural elements comprising non-canonical (i.e. mismatched) base pairs, base triples, junctions, turns, bulges and loops [recently reviewed by Leontis and Westhof ()]. These structural elements are essential for RNA biological function. Therefore, knowledge of the physicochemical properties of these structural elements will help us understand the molecular basis of RNA folding, stability and function. One such structural element is the G·U wobble base pair (from here on referred to as a G·U pair), which is the most common non-WC base pair present in RNA. First hypothesized by Crick in 1966 to account for codon degeneracy (), the G·U pair has been found in nearly all forms of RNA including transfer (t)RNAs (,), small nuclear (sn)RNAs () and ribosomal (r)RNAs (,). Several ribozymes also contain G·U pairs in their structures, such as the Group I and Group II introns () and the Hepatitis Delta Virus (HDV) ribozyme (). Furthermore, most of the G·U pairs are highly conserved. Substitution of a G·U pair with other base pairs often has detrimental effects on RNA function. For example, the single G3·U70 pair at the acceptor arm of the tRNA is the identity element for aminoacyl-tRNA synthetase recognition. Substitution of this G·U pair with WC base pairs completely abolishes aminoacylation with alanine both and (,). The G1·U37 wobble pair at the P1 helix of HDV ribozyme is critical for its cleavage reaction. Only an A·C wobble pair, but not WC base pairs, could partially replace the G·U in terms of reactivity (,). A G·U pair located in stem I of the mRNA coding for ribosomal protein S15 is an example of the U·G/C·G motif, which serves as the recognition determinant for the binding and autoregulation of S15 (). Studies have identified several properties that contribute to the diverse biological function of G·U pairs. First, the geometry of the G·U pair provides unique chemical groups in the major and minor grooves. Among them, the non-hydrogen-bonded amino group of guanine in the minor groove was thought to be important for the 5′ splice site selection of the Group I intron (,). Second, the continuous presence of three electronegative groups (guanine N7, guanine O6 and uracil O4) creates a broad region of negative electrostatic potential in the major groove of the G·U pair. This region of electronegative potential is proposed as the recognition site for the binding of metal ions and other positively charged ligands (). In fact, a recently compiled database of metal ion binding sites in RNA structures has shown that the major groove of the G·U pair is the most common metal ion binding motif (). Third, it has been noted that the presence of the G·U pairs in the A-form RNA helices introduces certain sequence-dependent changes to the structural parameters like the helical twist (). It was proposed that the distortion of the backbone by the G·U pair positions the functional groups of the HDV ribozyme for efficient catalysis (). Detailed reviews on the structural, chemical and biological properties of RNA structure with G·U pairs can be found elsewhere (,). G·U pairs occur in nearly every class of RNA as single base pair or in tandem form (i.e. adjacent G·U pairs). Most of the G·U pairs found in tRNA are in single form, whereas tandem G·U pairs are commonly observed in rRNA. Analysis of the tandem G·U pairs in rRNAs showed that the sequence 5′-UG-3′/3′-GU-5′ (Motif I) is the most prevalent, followed by 5′-UU-3′/3′-GG-5′ (Motif III) and 5′-GU-3′/3′-UG-5′ (Motif II). Thermodynamic stabilities of RNA structures containing these three motifs follow the same order (). NMR and X-ray studies have explored the structural features of different motifs for tandem G·U pairs (,). In Motif I, the six-member ring of the guanine base of the G·U pair stacks directly on top of the six-member ring of the guanine of the opposite strand. In contrast, there is considerable intra-strand stacking between the tandem G·U pairs in Motif II, with the purine ring overlying the pyrimidine ring. Motif III exhibits an intermediate situation between Motif I and Motif II, with a mix of inter- and intra-strand stacking. Data from structural studies suggest that electrostatic interaction, base stacking, hydrogen bonding and the neighboring WC base pairs all contribute to the thermodynamic stabilities of different tandem G·U motifs (,). As a result of the strong electrostatic field associated with the high-charge density of the RNA sugar-phosphate backbone, RNA structure and function are highly influenced by electrostatic interactions. The molecular surfaces of RNA molecules display unique electrostatic patterns that are important for recognition and binding of cationic species [for example, see references (,)]. Therefore, careful quantitative characterization of electrostatic features of RNA structural elements, such as the G·U pair, is vital for a better understanding of RNA ligand recognition events. Previous studies of the electrostatic features of the G·U pair were focused on the general properties of the major groove of a limited number of structures. In this research, we seek to provide a more in depth, quantitative analysis of the electrostatic properties of individual G·U pair motifs in order to achieve a detailed understanding of effects contributing to the total electrostatic landscape. The ultimate goal is to predict accurately the likelihood of entrapment of cationic ligands. To accomplish this, we employ the non-linear Poisson–Boltzmann approach, which delivers precise surface and site electrostatic potential values. Our calculations indicated that, in some cases, the major groove of G·U pairs demonstrated enhanced electronegativity over standard GC and AU base pairs. For single G·U pairs, it is sequence dependent, but for tandem G·U pairs, it largely depends on motif conformation. Furthermore, we propose that both the sugar-phosphate backbone and the G·U base atoms contribute significantly to the electronegativity at the major groove of the G·U pairs. The structures used in the calculations were selected from the Protein Data Bank [PDB ()] with the help of a database of non-canonical base pairs found in known RNA structures (). The PDB codes of both NMR and X-ray structures used in this paper are listed in . The structures contain either single G·U pair or tandem G·U pairs in different motifs. For NMR structures with multiple models, calculations were done on two models with the lowest potential energies as assessed using AMBER8 (). In all the cases, as the two models showed very similar trends, only the results from the model with the lowest potential energy were presented. All physical coordinates of helical structures used in this study were incorporated into our calculations exactly as experimentally determined (except 1IKD, as noted in ). For each helix containing G·U pair(s), we generated a set of new helices in which both WC GC and AU base pairs replaced the G·U pair(s). These model ideal A-RNA helices () were generated using the AMBER8 () nucgen utility and were for the purpose of comparison with their G·U counterparts. In one case included in our study, the original structure determination included both G·U (PDB: 434D) and GC (PDB: 435D) helices. The missing hydrogen atoms were added with the program REDUCE (). The electrostatic potentials at the surface (here taken as the solvent-excluded molecular surface defined by a probe radius of 1.4 Å) and atomic sites of isolated G·U, GC and AU base pairs were obtained with the fast multipole-accelerated () boundary element solution of the Poisson equation (). The interior and exterior dielectric constants were fixed at 2 and 80 (25°C), respectively. All default code parameters were employed as described in detail elsewhere (). For RNA helices, which are now embedded in a solvent medium that contains univalent salt ions corresponding to a salt concentration of 1 M (with an ion exclusion region of 2 Å radius), a novel linear/non-linear Poisson–Boltzmann algorithm was used (Boschitsch and Fenley, unpublished results). Very similar results were also obtained using the hybrid boundary element and finite difference Poisson–Boltzmann solver (with no ion exclusion region) which was employed previously (). The atomic partial charges and radii were taken from the AMBER94 force field (). Calculations repeated with the CHARMM27 force field parameters () showed the same trend. This similarity indicates that our conclusions were independent of the choice of radii and charge parameters. Formal RNA charges, with a −0.5 charge assigned to each non-bridging phosphate oxygen atom, were employed to examine sequence independent electrostatic features. When formal RNA charges were used, the atomic radii were assigned based on the AMBER94 parameter set. The molecular surfaces were color coded according to electrostatic potential derived from either the Poisson or the non-linear PBE and were rendered using the virtual reality modeling language (VRML) (). The 3D structures of the RNA base pairs and helices were displayed using the ViewerPro program (Accelrys, Inc., San Diego, CA, USA) and saved in the VRML file format. The 3D structure was then incorporated into the electrostatic potential maps for easy identification. In order to facilitate visual inspection, color mapping of the electrostatic potential was finely scaled as follows: green (most positive), followed by blue, white (neutral), red and yellow (most negative). The electrostatic potentials surrounding specific atom sites were calculated by taking the average electrostatic potential of sampled grid points outside the molecule around the particular atom. The grid points were sampled by generating two layers of spheres 1.2 and 2.4 Å away from the van der Waals surface of the atom. The electrostatic potentials of the grid points were obtained using the Poisson and PBE approach for the isolated base pairs and RNA helices, respectively, as described above. The electrostatic potential of the major and the minor groove was calculated by averaging the site electrostatic potential of the G·U pair(s) base atoms facing the major/minor groove surface. The atoms facing the major groove include N7(G), O6(G), N4(C), NH4(C) and O4(U). The atoms facing the minor groove include N2(G), N3(G), NH2(G), O2(C) and O2(U). The cross-strand inter-phosphate distance for a given base pair step was computed using the 3DNA program (). We began our study of the electrostatic features of the G·U pair by calculating the electrostatic potentials of an isolated G·U pair without the context of the RNA helix, and comparing the results with isolated WC GC and AU base pairs. The calculation was accomplished by use of the Poisson equation as described in Methods. shows the surface electrostatic potential maps for the isolated G·U and WC GC/AU pairs. As shown in the figure, the major groove of the AU base pair is mostly negative along the edge formed by the nitrogen and carbonyl oxygen atoms, with a positive electrostatic potential spot associated with the adenine amino group. Greater extremes in electrostatic polarity are observed in the major groove of the GC base pair. Deep electronegative potential is seen on the G edge, while the C edge is predominantly electropositive. The polarity observed in the WC base pair is due to the presence of electropositive amino group of adenine and cytosine in the major groove. In contrast, the major groove of the G·U pair is uniformly and strongly negative, since the co-planar N7(G), O6(G) and O4(U) atoms lining the major groove are all electronegative. On the minor groove edge, the non-hydrogen-bonded guanine NH group of the G·U pair created an electropositive region, which is absent in the AU pair and less pronounced in the GC pair. We calculated an average electrostatic potential at the major groove of the isolated G·U pair of −0.4 kT/e, as compared with average electrostatic potentials at the major grooves of GC and AU base pairs of −0.2 and −0.3 kT/e, respectively. The average electrostatic potential at the minor groove of the isolated G·U pair is neutral, whereas for both GC and AU pairs, the value is −0.3 kT/e. Therefore, the overall electrostatic potential of isolated G·U pair is somewhat more negative in the major groove and slightly more positive in the minor groove than for the WC GC/AU base pairs. In addition to wobble pairs, several X-ray and NMR structures have displayed bifurcated G·U interactions (,,,). In this arrangement, the O4 of U hydrogen-bonds with hydrogens attached to N1 or N2 of G. The bifurcated G·U positions H5 and H6 of U in the major groove, instead of the electronegative O4. Hence, the surface electrostatic potential of bifurcated G·U displays similar polarity as observed in the WC GC base pair (data not shown). We illustrate this point by calculating the electrostatic surface potential of two helices of the Group I intron that contain tandem G·U pairs [PDB: 1C0O () and 1AJF ()] using both the non-linear PBE and the linear PBE. A shows the results for the NLPB treatment for each helix, as well as the cobalt(III)-hexammine ([Co(NH)]) ion bound to the RNA in the NMR-derived structure (,). B shows the corresponding surface maps computed with the linear PBE. Comparison between A and B illustrates that the non-linear PBE, as opposed to its linearized version, predicts the metal ion binding sites with far greater precision. In order to examine the impact of base sequence on electrostatic properties of G·U pairs, we divided the RNA helices containing G·U pairs into four categories: (i) RNA helices with a single internal G·U pair; (ii) RNA helices with Motif I tandem G·U pairs (5′-UG-3′ vs. 5′-UG-3′, characterized by purine-purine cross-strand stacking); (iii) RNA helices with Motif II tandem G·U pairs (5′-GU-3′ vs. 5′-GU-3′, with purine-pyrmidine intra-strand stacking); and (iv) RNA helices with Motif III tandem G·U pairs (5′-GG-3′ vs. 5′-UU-3′, with a mix of intra- and inter-strand stacking). We chose three representative RNA helices for each of the four categories and, for the purpose of reference, generated corresponding canonical A-form RNA helices. These helices contained the identical sequence, except for substitution of G·U with both AU and GC (in different structures), each having parameters defined by fiber diffraction data (). lists the electrostatic potential values calculated for the G·U pairs embedded within RNA helices, compared with those of their canonical counterparts. The data in indicate that the major groove electrostatic potential of each standard A-form RNA helix is approximately −4.0 ± 0.3 kT/e. This value fluctuates only slightly with different base sequences and stacking patterns. In contrast, the major groove electrostatic potentials of the G·U pairs embedded in RNA helices vary from −4.0 to −6.3 kT/e. For RNA helices containing a single G·U pair, the major groove electrostatic potential is generally considerably more negative than that of the corresponding canonical RNA helices, depending on the specific base sequence. For RNA helices with tandem G·U pairs, the major groove electrostatic potentials are highly influenced by the stacking patterns. shows the surface electrostatic potential maps for one representative RNA helix from each motif compared with its canonical (GC) counterpart. We found that the electrostatic features of Motif I tandem G·U pairs are quite different from those of Motifs II and III (also see ). The electrostatic potential at the major groove of Motif I RNA helices is more positive than (one case) or slightly more negative than (two cases) that of the corresponding WC RNA helix, depending upon groove width (see ahead). However, all of the Motif II and Motif III RNA helices showed enhanced negativity at the major groove, compared with their WC counterparts (GC or AU). We also noted that helices with Motif I are less negative than that of Motif II and Motif III. The most dramatic example was observed between 1EKA and 1GUC. Despite similarities in helical length and the nearest neighbor base pairs, the major groove electrostatic potential of 1GUC is significantly more negative than that of 1EKA (−6.3 kT/e vs. −3.5 kT/e). Although the major groove of G·U pairs is uniformly negative in the sense that there is no positive amino group present, a close look at individual base atom sites revealed that electrostatic polarity still exists at the major groove. presented the electrostatic potential at N7(G), O6(G) and O4(U) of the G·U pairs from each RNA helix studied (for RNA helices with tandem G·U pairs, only one G·U pair was shown). For each G·U pair, the average electrostatic potential associated with O6(G) are the most electronegative. In most cases, O4(U) is the least electronegative atom. On average, the difference in potential between O6(G) and O4(U) is 1.7 kT/e. In 434D, the extreme case, the difference between O6(G) and O4(U) is as high as 2.6 kT/e. Next, we investigated the sources of the broad region of negative electrostatic potential at the major groove of the RNA helix containing the G·U pair(s). The two most likely sources are the partial charges of the G·U base atoms and the sugar-phosphate backbone charges. To investigate the contributions from the partial charges of the G·U base atoms to the major groove potential, we obtained two major groove electrostatic potential values. The first was calculated with all partial charges included and the other with the partial charges of the G·U base atoms set to zero. The contribution from the G·U base atoms was then defined as the difference between the two electrostatic potential values. As a control, we also calculated the contribution of the corresponding GC/AU base atoms to the major groove potential. illustrates the contributions of the partial charges of the G·U, GC and AU base atoms to the major groove potentials. As is shown, the partial charges of the G·U base atoms make a significant contribution to the total major groove electronegativity, while the base partial charges of the WC GC/AU base pairs have only a small effect. Overall, tandem G·U base atoms contribute around −2.0 kT/e to their major groove electronegativity, whereas the corresponding WC base pairs contribute only about −0.7 kT/e to the major groove electrostatic potential. We then examined the contributions of sugar-phosphate backbone to the electronegativity by calculating the ‘formal’ electrostatic potential. The calculation was performed by placing a -0.5 charge on each non-bridging phosphate oxygen atom. shows the major groove electrostatic potentials of the G·U pairs derived from these calculations, compared with their canonical counterparts. For the majority of the RNA helices containing G·U pairs, the formal major groove electrostatic potential of G·U pairs is similar to that of the standard A-form RNA helices. However, for some G·U pairs (for example, 1HLX and 1EKA), the formal major groove electrostatic potential is less negative than their canonical counterparts. In particular, we found that most RNA helices with Motif I tandem G·U pairs have less negative formal electrostatic potential than their canonical counterparts. On the other hand, the majority of RNA helices with tandem Motif II and III G·U pairs have similar formal electrostatic potential as their WC counterparts. The gradation of the potential is depicted very clearly in . Again, we noticed that helices with Motif I are less negative than that of Motif II and Motif III, when the electrostatic potential were computed with only the sugar phosphate backbone charges included (also see ). In order to investigate the origin of the difference in formal electrostatic potential between the different motifs, we evaluated the inter-strand phosphate-phosphate distance. As shown in , Motif I had a wider mean cross-strand interphosphate distance than either of the other motifs or than the canonical helices. A wider major groove will result in a lesser concentration of negative potential from the backbone and bases. In contrast, dimensions of the major groove in helices having Motifs II and III are relatively similar to their WC counterparts. In these motifs, the more negative potential is readily apparent. We focused our study of the electrostatic features of the G·U pair on the major groove, which is the primary binding site for catonic groups in RNA helices. Unlike the major groove, the minor groove potential of standard A-form RNA exhibits a higher degree of dependence on base sequences (), with the partial charges of the base atoms contributing heavily to the groove potential. In contrast, the G·U base atoms do not make significant contributions to the minor groove potential (data not shown). Although the minor groove electrostatic potential of an isolated G·U pair is more positive than that of isolated WC base pairs, when incorporated into RNA helices, the minor groove electrostatic potential of G·U pairs tends to be more negative than that of the WC base pairs in standard RNA helices (see ). Our results therefore disagree with those of Trikha (), who performed electrostatic potential analyses of an RNA helix with a Motif III tandem G·U pair and reached the conclusion that the minor groove presented a region of positive potential, which was absent in its canonical counterpart. We were not able to identify the source of discrepancy between our results. However, we noticed that the electrostatic patterns of the WC RNA helix used as the control in their study contradicts standardly accepted electrostatic profile of A-RNA (i.e. the major groove should be more negative than the minor groove). Divalent metal ions are critical for RNA structure and function (). Metal ions can facilitate folding of RNA into a variety of intricate tertiary structures by counter-balancing the repulsion between the negative charges of the sugar-phosphate backbone, and some metal ions participate directly in ribozyme catalysis. Although there is no definitive method for identification of metal ion binding sites in RNA under physiological conditions, location of site bound ions are often inferred from results of X-ray crystallography, NMR or phosphorothioate substitution experiments, among other methods. However, there is great value in being able to predict likely cationic binding sites based upon computational methods. Techniques such as valence screening, molecular and Brownian dynamics simulations, and microenvironment analysis have been used to identify possible metal ion binding sites (). Given the strength of the NLPB in accurate prediction of the electrostatic features of structured biopolyelectrolytes (), we have applied this technique to obtain detailed information about the electrostatic potential of G·U base pairs in different structural contexts.

In , as in many animals, the female state is defined by the presence of two X chromosomes, but males contain only a single X chromosome in addition to the gene-poor Y chromosome. A priori, the genome of male fruit flies appears unbalanced due to the halved dosage of X chromosomal genes. Re-establishment of proper balance requires a compensatory mechanism that raises the expression levels of the single male X chromosome to match the expression from the two X chromosomes in females (). Failure of such ‘dosage compensation’ is lethal to the affected males (for reviews see ()). Dosage compensation in flies is achieved through a male-specific ribonucleoprotein complex, the Dosage Compensation Complex (DCC, also known as the Male Specific Lethal or MSL complex), which is able to distinguish the X chromosome from the autosomes and to bind the X selectively. The DCC consists of five core proteins (MSL1, -2, -3, MOF and MLE) and two non-coding RNAs, and (for review see () and references therein). Gene activation involves modification of X chromosomal chromatin by the DCC-associated acetyltransferase MOF, which acetylates histone H4 at lysine 16 (), but contributions from more general factors, such as the H3 serine 10 kinase Jil1 (,), the supercoiling factor () and nuclear pore components () have been suggested. The targeting of these effects to the X chromosome relies primarily on MSL1 and MSL2, the two DCC proteins that are able to recognize a subset of sites on the X chromosome even in the absence of all other factors (). However, faithful occupancy of all sites on the X chromosome requires additional factors, such as the activities of the acetyltransferase MOF and the RNA helicase MLE, as well as the presence of the roX RNAs (,). How the DCC recognizes a single chromosome is a question of great interest. Ultimately, recognition must involve X-specific DNA sequences. Combining chromatin immunoprecipitation (ChIP) with probing of high-density oligonucleotide arrays ‘tiling’ the entire X chromosome, the interaction of the DCC with the X chromosome has recently been mapped (,). These studies revealed that about 25% of the X chromosomal DNA is bound by DCC in tissue culture cells or embryos and, notably, the majority of DCC binding is found within coding sequences, reaffirming an earlier suggestion that the DCC may act to facilitate transcription elongation rather than initiation (). However, despite this wealth of interaction data it has not been possible to distil a set of ‘consensus’ DNA sequences that define DCC binding (,). It therefore remains possible that the observed X chromosomal binding pattern of the DCC is governed by more than just DNA sequence. DNA sequence may just define a subset of ‘primary’ targeting sites from which the DCC is distributed to secondary sites in neighbouring chromatin. Because the DCC interacts preferentially with active genes it is possible that the process of transcription itself or a transcription-associated epigenetic modification of chromatin generates secondary sites (,). While the nature of those presumed secondary sites is entirely unclear, the existence of primary sites, defined by DNA sequences with autonomous DCC recruitment activity, has been inferred from P-element-mediated insertion of X-derived sequences into autosomes, where association of the DCC with these ectopic sites can be monitored on polytene chromosomes by immuno-histochemistry. The first DCC binding sites characterized in this way correspond to ∼200 bp sequences found in the coding regions of the and genes (,). A prominent feature in these sequences is an abundance of GA sequences, and mutation of GAGAG tracts significantly reduced the recruiting power of the element (). Unfortunately, bioinformatic efforts failed to detect related DCC targeting sequences on the X chromosome () and the third binding site characterized in some detail lacked GAGAG sequences altogether (). Primary DCC targeting elements thus differ in DNA sequence and may therefore belong to different classes. According to the currently accepted model, many targeting elements of varying affinity are spread across the X chromosome. Strong targeting elements are able to autonomously recruit the DCC to an ectopic integration site on an autosome even at reduced concentrations of MSL proteins. In contrast, weaker sites are only bound in the X chromosomal context, presumably because the density of targeting elements leads to an increased local DCC concentration (). This model was substantiated by the analysis of 11 X-chromosomal fragments of varying affinity for DCC isolated by ChIP (). An attempt to identify sequence motifs responsible for DCC recruitment from a subset of these ‘DCC Binding Fragments’ (DBFs) with high- and moderate-affinity highlighted a number of clustered motifs (). Accordingly, a DCC binding site may be composed of clusters of variable combinations of several degenerate sequence motifs. However, perhaps due to the relatively large size of the fragments analysed, it was not possible to predict further DCC binding based on the clustering of these motifs (). Thus, several previous studies indicate that the DNA sequences comprising high affinity DCC binding sites are diverse, possess varying affinities for the DCC, and can be dispersed over several kb (,,,,). Given the degeneracy of the DNA motifs seen in association with DCC interaction, one has to assume that the contributions of individual elements to overall DCC targeting may be small and hence difficult to document by established methodology. In order to monitor the effect of weak targeting determinants, we developed a sensitive transfection-based ‘one-hybrid’ assay that amplifies weak DNA interaction events into a strong transcriptional read-out. The assay allows rapid identification of sequence elements able to recruit the MSL2 protein. We applied the assay to localize the DCC binding determinants within several previously described high affinity DBFs (). We describe several sequence motifs that contribute to MSL binding and show that high affinity MSL binding sites can be generated by oligomerization of weaker elements. Fly genetic manipulation and crossing was performed as in ref. (). A more detailed description is provided in the Supplementary Data online. FISH and Immunostaining were performed exactly as described in (). For immunostaining, one of two affinity purified rabbit anti-MSL1 antibodies kindly provided by M. Kuroda () and E. Schulze () were used at a dilution of 1:200 and 1:400, respectively. DNase-I hypersensitive (DH) sites in adult flies were mapped as described (), except that nuclei from 2.5 g adult flies, sorted according to sex, were divided into seven portions and digested with a titration of up to 60 units DNase-I (Roche, Penzberg, Germany). Ten μg DNase-I-digested DNA per lane was digested with restriction enzymes as described in figure legends, run on 0.8% agarose gels, then blotted and hybridized as described (). Total RNA was recovered by grinding adult male or female flies under liquid nitrogen, then extracting the powder with Qiazol (Qiagen, Hilden, Germany) reagent according to manufacturer's instructions. Twenty μg of RNA from male and females was run on a 1.2% agarose gel containing formaldehyde, then transferred and hybridized according to standard laboratory protocols (). Plasmids made for use in this publication are detailed in Supplementary Table 2. SL2 cells were maintained in culture at 26°C and split every 3–4 days. For transfection, cells were diluted to 2.5 × 10 cells/ml at 1 ml per well of a 12-well tissue culture plate. The following day, the cells were transfected using the Effectene reagent (Qiagen) according to manufacturer's instructions, with the single exception that 20 μl Effectene reagent was used per μg of transfected DNA. Per well of cells, the following amounts of plasmid were transfected: pRL-TK, 15 ng; pGL3-TK and derivatives, 315 ng; pVP16 and other activator constructs, 160 ng. Each tested firefly reporter plasmid was transfected into duplicate wells, and on at least two occasions using DNA prepared from two different maxipreps. DNA was prepared using Qiagen or Promega maxiprep kits. Two days following transfection, cells were harvested by centrifugation and washed in 1 ml PBS. Cells were then lysed and luciferase activities determined using the Dual Luciferase Kit (Promega, Mannheim, Germany) according to manufacturer's instructions. Light emission was measured using a Lumat 9501 Luminometer (Berthold, Bad Wildbad, Germany). Western analysis was performed according to standard laboratory protocols () using antibodies directed against MSL1 (), MSL2 and HSV VP16 (Santa Cruz Biotech, SC7545, Santa Cruz, CA, USA). The previously isolated DBF5, DBF6, DBF7, DBF9 and DBF12 are considered to contain high affinity binding sites for the DCC (). However, given that these fragments are between 2.5 and 6.7 kb long, localizing the targeting elements requires further mapping. So far, DBFs have been characterized by integration of candidate fragments into autosomes through P-element-mediated gene transfer and monitoring of DCC recruiting power by MSL1 immunostaining on polytene chromosomes. In order to guide the construction of further clones for this analysis we first explored whether DNase-I hypersensitivity (DH) would highlight regions of interest. DNase-I hypersensitivity analysis indicates chromosomal loci where chromatin is disrupted due to the interaction of non-histone proteins (). The three known high-affinity DCC binding sites within the roX genes and at cytological position 18D10 (,,) all contain regions of male-specific DNase-I hypersensitivity, and in the case of the roX genes these isolated sites are alone able to recruit the DCC. Adult flies were sorted according to sex, nuclei prepared and treated mildly with DNase-I to digest only the most exposed sites in chromatin. DH sites within the DBFs were identified by indirect end-labelling (Supplementary Figure S1). DH sites were present in every DBF clone, but notably, only two contained male-specific sites. A strong and a weak general DH site (i.e. common to both sexes) were found in DBF12 within intronic sequences of the gene. A single, weak male-specific site was found in DBF6, spanning an intron and coding sequences of the gene. The remaining DH sites revealed in clones DBF-5, -7 and -9 (Supplementary Figure S1) are summarized as follows: Two general sites were found at the 5′ ends of two neighbouring genes (CG15892 and CG3815) in DBF5. In DBF7, a male-specific site was seen in the vicinity of two small introns of CG2025, and a general site was present between the CG2025 and CG1847 genes. Lastly, in DBF9, three DH sites present in both sexes were found in an intron of the gene. To examine whether DH sites found in both sexes could direct DCC targeting, the strong (DHS-S) and weak (DHS-W) general DH sites from DBF12 were chosen for rigorous testing of DCC recruitment potential. P-elements containing these sequences were generated and integrated into autosomes of transgenic flies. We determined the insertion sites by DNA FISH on polytene chromosomes (data not shown) and then monitored the recruitment of DCC to these ectopic sites by staining with MSL1 antibody. The 480 bp DHS-S was capable of recruiting MSL1 in both wild type and mutant flies (A and B), demonstrating that this sequence contains a high affinity DCC binding site. In contrast, a 700 bp fragment containing DHS-W failed to recruit MSL proteins in three transgenic lines under wild type DCC expression (data not shown). The single, weak male-specific site found in DBF6, spanning an intron and coding sequences of the gene, was also tested for DCC recruitment in transgenic flies. A 500 bp fragment containing the DH site and flanking sequences recruited MSL1 to an autosomal insertion site in wild type males (C), but was relatively weak compared to the DBF12 DHS-S. No MSL1 binding could be detected on the DBF6 DH insertion in flies carrying the mutation (data not shown) or at the reduced DCC levels obtained when MSL2 is provided only by the expression mutant SXB-1 or NOPU in females (data not shown) (,,). Only under conditions of MSL1 and MSL2 overexpression was robust binding to the DBF6 DH site seen (D). A hexamer of this sequence showed improved binding in wild type flies (E), but was still not capable of recruiting the DCC at lower DCC concentrations that characterize the SXB-1 and NOPU genetic backgrounds. Therefore in this instance, a male-specific DHS was not sufficient to define a high affinity binding site for the DCC, and the presence of a male specific site is a poor indicator of DCC recruiting ability. It had become clear that DHS mapping was of only limited diagnostic value for the identification of DCC interaction sites. Also, P-element-mediated transgenesis was considered too time consuming to attempt narrowing down the DBFs to minimal DNA elements. We therefore developed a novel strategy for the characterization of candidate DCC targeting sequences employing co-transfection of three plasmids in male SL2 cells. The assay is based on the activation of a reporter gene after transient transfection into SL2 cells (A). Candidate DCC binding sites are cloned in front of the firefly luciferase gene, which is driven by a minimal thymidine kinase (tk) promoter. To convert recruitment of a key DCC component (MSL2) into a robust signal, an MSL2-VP16 fusion protein is expressed from a second, co-transfected plasmid. This hybrid protein consists of the entire MSL2 protein, to which one of the strongest known transactivation domains, taken from the Herpes Simplex Virus VP16 protein C-terminus (,), is fused. The assay therefore does not measure dosage compensation, but recruitment of the MSL2 fusion protein to the promoter of the reporter plasmid. ‘One-hybrid’ strategies of this kind have been successfully employed in previous efforts to uncouple the DNA binding properties of a protein from its natural function (). If MSL2-VP16 is attracted to a target sequence, either directly or indirectly through incorporation into a binding complex, the VP16 activator will boost the expression of firefly luciferase. The resulting increase in luciferase activity can be expressed as activation over the light emission seen with the control activator, which consists of a mutant version deleted for most of MSL2, leaving the VP16 activation domain alone. A second, obligatory normalization control for non-specific effects, such as transfection efficiency, involves co-transfection of a Renilla luciferase expression vector. We first tested the functionality of the assay using the entire cDNA, and the 217 bp DHS, which are known to contain high affinity DCC binding sites (see above; B). As can be seen when comparing the empty firefly vector pGL3 to those containing the entire cDNA or the isolated DH site, the basal level of luciferase activity (in the presence of VP16 only) is heavily influenced by sequences inserted into the vector. The cDNA produced a drop in luciferase expression relative to empty vector, whilst the DHS sequence alone caused an increase. This phenomenon was seen with all sequences tested, but the sequences shown represent the most extreme cases of repression and activation observed. It most likely reflects the juxtaposition of both real and cryptic transcription factor binding sites to the promoter driving firefly luciferase expression, or of nucleosome exclusion sequences, and must be considered non-specific. The effect of recruiting MSL2-VP16 is therefore best expressed as the fold activation in luciferase activity in the presence of pMSL2-VP16 over that seen with pVP16 (B and ). The assay yielded a reproducible, ∼5-fold activation of transcription in the presence of cDNA, but the DHS led to a robust 24-fold activation. An additional MSL2 construct, fused to GFP instead of VP16 and previously shown to be capable of substituting the native MSL2 protein (), showed negligible activation over pVP16 in the assay, demonstrating the need for the artificial VP16 activator fusion to observe MSL2 binding (B). The stronger activation of the -DHS compared to the entire cDNA suggested that the distance an MSL2 recruiting element lies from the promoter may influence the level of luciferase activation. Within the roX1 cDNA, the DHS is the primary element responsible for DCC recruitment (), and in the cDNA lies ∼2.3 kb from the promoter. The effect of distance from the promoter was examined in greater detail with subclones derived from DBF12 (see below). Having determined that the one-hybrid assay could be used to isolate small sequences crucial to DCC binding, we applied it to isolate minimal MSL2 binding elements on three clones (DBF-6, -9 and -12) and performed less detailed analysis on DBF-5 and DBF-7. We first applied the one-hybrid assay to the ∼500 bp DBF12 DHS-S. A summary of the clones analysed to map an MSL2 recruitment site is shown in . From the first two clones splitting the DHS in two (DBF12-L2 and -L4), it was clear that the MSL2 recruiting sequences resided in the 5′ half of the DHS, within DBF12-L2. Restricting clone DBF12-L2 into ever-smaller fragments led to the identification of a 40 bp MSL2 binding element (DBF12-L15, see A), which retained the ∼4-fold activation potential of DBF12-L2. Trimerization of L15 led to an MSL2-dependent 24-fold enhanced luciferase activity (A), rendering this oligomer almost as potent as the native DHS (). Clearly, the interaction of MSL2-VP16 can be boosted by clustering of interaction modules. We used the one-hybrid assay to further define and mutate the minimal MSL2 binding sequence within DBF12. The sequence of DBF12-L15 contains runs of adenines and a GAGA sequence (B). A still shorter 16 bp element retaining these sequences (DBF12-L22) still promoted MSL2-dependent luciferase activation (B). Mutating blocks of 4–5 bp of this 16 bp ‘core’ sequence in the context of the original 40 bp L15 fragment (clones L18, L19 and L23) caused a complete loss of MSL2 recruitment, confirming the importance of this element for MSL2 binding. Conversely, mutation of a 5 bp sequence outwith this core did not abolish activity (clone L20). To explore whether the ‘GAGA motif’ in L22 was important we systematically mutated the last ‘GA’ dinucleotide. The adenine was shown to be dispensable (clones L30–L32). In contrast, a transversion of the adjacent guanine to a cytosine (L33) or a thymine (L35) resulted in a complete loss of MSL2 recruitment, whilst a transition to an adenine (L34) had little or no effect. These experiments therefore allow the requirements for MSL2 recruitment to be dissected at the single nucleotide level. Establishing general rules for MSL2 recruitment requires the isolation of a panel of binding elements. We therefore applied the one-hybrid approach to the DH sites of DBF6 and DBF9. The DBF6-L4 fragment, which contains the male-specific DH site, recruited MSL2-VP16 to a similar extent as DBF12 (A). Sequentially removing sequences from the 3′ end of the DH site first caused an increase in luciferase activity, most likely due to moving MSL2-binding sequences closer to the promoter, followed by complete loss when the intron in the centre of the DH site was removed (compare clones DBF6-L4, L5, L6 and L7). A clone containing 68 bp of only intronic sequences (L8) recruited MSL2, although the fact that clone L9 showed double this activity argued that additional 5′ coding sequences, incapable of attracting MSL2 on their own, contribute to recruitment (A). The presence of similar contributing elements in coding sequences 3′ to the intron is also suggested by the higher activity of clone L5 than L6. Removing 17 bp from the 5′ end of L8 did affect its activity somewhat (L10), but further removal of a CGAGAAA sequence (L11) almost abolished activity. Further deletion into a GA repeat resulted in complete loss of MSL2 recruitment (L12). Restriction from the 3′ end (L13 and L15) suggested that the central GA repeat may form a core element, flanked by other sequences that contribute to MSL2 recruitment. Fine mapping of DBF6 was therefore carried out on the DBF6-L15 fragment, which consists of a GA repeat flanked by two short sequences (a 5′ CGAGAAA and a 3′ TATA motif, B). Clones L18, -L19 and -L23 were constructed with dinucleotide substitutions within the 5′ CGAGAAA sequence. None of the dinucleotide mutants caused complete loss of activation compared to the parent clone, suggesting that, similar to the mutants of DBF12 discussed above, mutation of individual bases (or pairs in this case) can be tolerated within an essential motif. Likewise, mutation of the entire 3′ TAT motif (clone L17) revealed its importance, but single base substitutions (L20–L24) had little effect. Deletions and substitutions within the central dinucleotide repeat region highlighted the role of uninterrupted GA repeats (L26–L31). In order to explore whether GA repeats alone were able to recruit MSL2 we mutated three bases within L15 to construct a fragment consisting entirely of GA repeats (L16). This sequence did not support any MSL2 recruitment, confirming the importance of the CGAGAAA and/or TATA motifs flanking the GA repeat (B). However, a trimer of the 26 bp L16 GA repeat sequence demonstrated robust, 7.8-fold MSL2 recruitment, providing a second example for the earlier notion that weak elements, which by themselves are unable to recruit MSL2, may gain affinity by oligomerization and clustering (see also ). A similar approach was also applied to DBF9. Three DH sites, named A, B and C had been identified in intronic sequences at the 3′ end of this clone (Supplementary Figure S1C). Each of the DH sites was analysed, but only DHS-C exhibited modest MSL2 recruitment (A). Dividing DHS-C into clones L6 (or L9) and L10, revealed that most MSL2 recruitment partitioned with clone L10. Restriction of L10 from the 3′ end resulted in a modest increase in activity (L7), suggesting that MSL2 recruiting elements had been moved closer to the promoter. Further efforts therefore focused on restricting the L7 clone to a minimal activation element. Clones L11 and L12 defined a 5′ end for the MSL2 recruiting sequences. Trimming sequences from the 3′ end of L11 resulted in a sequential loss of MSL2 recruitment (L13–L15; see also B). These observations suggested that the 5′ end of L11 contains an essential ‘core’ MSL2 recruiting element, followed at the 3′ end by weaker elements, which are not essential but contribute in a cumulative fashion to MSL2 recruitment. Fine mapping of the DBF9 fragments therefore focused on mutating bases in the 5′ ‘core’ or 3′ accessory elements to confirm this hypothesis (B). Mutating two 5 bp blocks in the core of L13 abolished recruitment activity (clones L16 and L17), whilst disrupting two CACA elements in the 3′ end resulted only in reduced activity (clones L18 and L19). Therefore, these experiments confirmed that DBF9 contained an important element of ∼25 nucleotides, flanked by accessory elements that contribute to MSL2 binding. The analysis pointed to the CA dinucleotide as one such accessory motif. Both genes encode non-coding RNAs that span also the DHS sequences responsible for DCC recruitment, although in the case of very few, if any, transcripts read through the DHS (). At the 18D site, however, no RNA could be detected in the region of the male-specific DH site (). We therefore examined all the DBF sequences seen to recruit the DCC for transcripts in male and female adult flies by Northern blotting. In contrast to both roX DH sites, but similar to the 18D site, no male-specific transcripts could be detected (Supplementary Figure S3), although we cannot exclude that we have missed rare, large (>8 kb), developmentally regulated transcripts, or micro-RNAs. Comparison of the results presented here to high resolution ChIP-chip data for MSL1 () demonstrates that most MSL2-recruiting sequences are found within peaks of MSL1 binding (Supplementary Figure S4). These peaks are embedded in broader regions of MSL1 binding and it is currently unclear whether these interactions are entirely defined by DNA sequence or whether secondary targeting determinants, such as histone modification marks, contribute to the observed profile. In order to ascertain that the one-hybrid assay indeed selects high affinity DCC binding sites, we tested whether the DBF12-L15 fragment was able to recruit MSL1 to ectopic, autosomal locations in transgenic flies. An autosomal insertion of the 40 bp DBF12-L15 was not able to recruit the DCC in wild type flies (A). However, deletion of the L15 fragment from the DBF12 DHS-S clone caused loss of DCC recruitment, even in a wild type male background, confirming the essential nature of the 40 bp L15 sequence (B). Notably, a trimer of L15 recruited DCC to a similar or greater extent than the parent DHS-S construct in wild type males (C, compare to A), and also demonstrated robust recruitment at low DCC concentration (SXB-1 background; D). Thus, similar to the high affinity site at 18D (), multimerizing an essential element not sufficient to be classed as a high affinity site on its own created an artificial high affinity DCC binding site. Therefore, DBF12-L15 may well be the shortest DCC targeting element identified to date. How the DCC of recognizes the X chromosome for selective interaction is an unsolved question. Although there is ample evidence that DNA sequences are involved, defining consensus sequence elements that may serve as binding sites for DCC components has been difficult. The available evidence points to the existence of different sequence motifs, clustering in regions covering several kb, which form the highest affinity binding sites. Such a definition necessitates testing numerous candidate binding sites and extensive mutagenesis. So far, the established method to evaluate X chromosomal sequences for DCC recruitment is time consuming since it involves generating stable fly lines containing candidate sequences integrated into an autosome. The ‘one-hybrid’ strategy we introduced abbreviates this process dramatically. Fusing the VP16 transactivation domain to MSL2 leads to a robust activation of a reporter gene provided that MSL2 is targeted to the candidate DNA upstream of a minimal promoter. This strategy has several important features. First, the assay solely measures chromosome binding of MSL2 without constraints imposed by a requirement for normal function in dosage compensation. This allows mutating MSL2 regardless of potential consequences on functions other than recruitment. Second, for the assay to work it does not matter whether MSL2 binds the chromosome directly or indirectly via an adaptor molecule or even the entire DCC. Third, the assay appears to be more sensitive for the identification of minimal targeting determinants than the polytene chromosome recruitment assay. The DBF12-L15 fragment would have been missed in the chromosome recruitment assay because its affinity for MSL2 is too weak if present as a monomer. The element is nonetheless essential for DCC binding. However, the dramatic increase in MSL2-responsiveness upon trimerization led to uncovering its autonomous recruitment potential in flies. This enhanced sensitivity may be due to the fact that MSL2 is overexpressed in SL2 cells and hence present in artificially high concentrations that allow recognition of weak elements. Native high affinity sites may be composites of several weak elements (see below) that individually are unable to attract DCC autonomously, but which can be detected in the one-hybrid assay. Finally, the assay may be adapted to a high-throughput format, which should allow screening many DNA sequences in parallel. Notably, the DNA sequence elements identified as targeting determinants resemble those found earlier with the more established assay (see below (,,)). However, at this point we cannot exclude that the assay only detects a subset of DCC binding sites with special characteristics. For example, all the core sequences identified to recruit MSL2 in this analysis lie in non-coding regions of the genome, whereas the majority of DCC binding is seen in coding regions (,). According to a recent model (), the DCC may interact with chromosomes in two (or more) distinct binding modes: a primary mode, determined largely by DNA sequences, and a secondary mode employing transcription-associated epigenetic features. A similar model has recently been proposed for dosage compensation in (). Accordingly, distribution of the DCC over the X chromosome may involve primary recruitment to a subset of sites (possibly including those identified in this analysis) from which DCC is distributed to the majority of secondary sites (). The one-hybrid assay allowed the fast mapping of minimal MSL2 targeting elements within larger DBFs identified by conventional means. These are the smallest known binding sites for the DCC. Deleting sequences from the DBFs in our quest for minimal elements we noticed the existence of ‘accessory’ elements, which by themselves are not sufficient to recruit the DCC in the transfection assay, but in the vicinity of a ‘core’ element contribute to the overall affinity. One such accessory motif consists of short CA dinucleotide repeats. The core elements appear purine-rich on one strand, and although purine-pyrimidine transversion affected activity of one nucleotide position, it could not account for all observed changes in activity, nor was the length of purine tract required for activity consistent between the different clones. The results therefore suggested that despite a general tolerance of mutation, some nucleotide positions within the core may be more important than others. However, the only sequence motifs shared between the three shortest elements isolated are AGA, GAG and AAA, and attempts to build longer, more flexible motifs, did not produce convincing results. The importance of GA-rich sequences for DCC recruitment has already been established for the loci (), and they are also common in the DBF clones (). However, the removal of the GAG motif from the shortest clones isolated from DBF9 (clones DBF9-L13 and -L14) does not lead to complete loss of MSL2 recruitment, confirming that not even this is an essential motif. On the other hand, trimerization of an element that essentially only consists of GA repeats can recruit MSL2 in our assay. Together, these observations strengthen the earlier hypothesis that high affinity DBFs are composites of several distinct sequence motifs with variable DCC recruitment potential that synergise to generate a high affinity site. These motifs may be dispersed, but cluster to form high affinity DBFs. However, diversity in sequence appears not to be a fundamental requirement since high affinity sites can also be generated from homotypic elements by oligomerization, as shown here for two examples and as was also previously observed for a larger element (). The relative tolerance towards point mutations emphasizes that these elements are degenerate. The binding specificity of the DCC therefore seems surprisingly plastic, which may explain the failure of genome-wide binding analyses to define a single consensus. The observed degeneracy implies that each single targeting determinant has a relatively low affinity for DCC, and we have to assume that the sum of many weak interactions effectively generate high-affinity DBFs (). Ultimately, the future identification of a greater number of targeting elements should allow a better definition of the motifs recognized by the DCC. Our conclusions are in broad agreement with recent results from , where isolated or clustered motifs can render a high affinity DBF, but are not sufficient to explain all observed DCC binding (,). Emulating previous studies (), DH sites were found in all of the five high affinity DCC binding sites studied in this analysis. However, we found for the first time DH sites common to both sexes that overlapped with DCC recruiting elements, and only two of the five DBFs contained a male-specific DH site. The single example of a male-specific DH site tested in the transgenic fly for DCC recruitment was not capable of recruiting the DCC with high affinity, suggesting that male-specific DNase-I hypersensitivity is merely a crude indicator of chromatin accessibility and does not reflect the strength of the underlying sequence to attract the DCC. This is similar to the findings of the DBF at 18D, where a male-specific DH site was necessary for high affinity of a larger fragment, but was alone not sufficient (). Conceivably, regulatory elements attracting the DCC in males may have co-opted sequence elements and interacting factors that facilitate chromatin opening, and so perform this function also in females. Accessory elements may therefore not contribute to defining a DBF, but rather facilitate the interaction of the MSLs with DNA in chromatin. For example, runs of poly A/T tend not to be assembled into nucleosomes, which could aid interaction of the DCC (). Furthermore, GAGAG sequences are found in many regulatory elements, where the interacting proteins, such as GAF, recruit nucleosome remodelling factors that render nucleosomal DNA accessible (,). p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Protein synthesis is an essential activity that accounts for a large part of the ATP turnover in living cells. This process is performed by large macromolecular machines composed of proteins and rRNA molecules. These ribonucleoprotein complexes, called ribosomes, each comprise a small subunit and a large subunit that cooperate to decipher (translate) the information encoded in mRNA molecules. Recruitment of the small (40S) eukaryotic ribosomal subunit onto a cellular mRNA generally occurs via the capped 5′end. Since the AUG start codon of a gene can be located many hundreds (in some cases thousands) of nucleotides downstream of the 5′end, the 40S subunit then has to translocate (scan) along the mRNA to reach the initiation site (). This processive, sequence-independent scanning phase involves at least 11 eukaryotic initiation factors [eIFs ()], equivalent in to at least 21 proteins, that participate in a series of different complexes, as well as a number of (largely undefined) conformational states. The control of translation initiation is a key determinant of growth, is important to the stress response () and, when malfunctional, contributes to disease states (,). It is therefore important to understand rate control of this process at a quantitative level. Moreover, as understanding of the molecular mechanisms underlying eIF function progresses, we will be able to fit them into a quantitative framework that can serve as a robust model for the overall process (). Binding of the ternary complex (comprising Met-tRNA, eIF2 and GTP) to the 40S subunit is stabilized by eIF1A and eIF3 () (A). Yeast eIF3 binds to eIF2, thus promoting binding of mRNA to 40S; mammalian eIF3, on the other hand, seems to be capable of binding (in an RNA-dependent manner) directly to the 40S subunit (). eIF4F is anchored to the 5′end of the mRNA via the cap-binding protein eIF4E which, despite its small size, may play roles in a number of cellular processes (,). Genetic experiments in yeast have indicated that eIF1, eIF2 and eIF5 influence start codon selection (), while biochemical experiments have shown that eIF1 and eIF1A play roles in scanning and formation of the 48S complex, which comprises 40S, the eIFs and mRNA (,). A growing body of evidence indicates that, at least in budding yeast, eIF1, eIF2, eIF3 and eIF5 may bind to the 40S subunit as a preformed multifactor complex (MFC (); A). Thus the MFC components, together with eIF1A, play a key role in 40S-mRNA recruitment, scanning of the 5′ untranslated region, and start codon recognition (,). Recognition of the start codon in an mRNA leads to hydrolysis of eIF2-bound GTP, followed by 60S joining aided by eIF5B-GTP. Hydrolysis of the latter GTP precedes initiation of protein synthesis. The functioning of living cells depends on the coordinated action of many molecular processes, including translation. However, thinking on the rate control of processes such as translation has been influenced by the expectation that control is usually determined by a single rate-limiting step. For example, it has frequently been suggested or assumed that eIF4E acts as a rate-limiting factor in translation initiation (). This approach to posttranscriptional gene expression probably derives from early models of metabolic control, widely disseminated in biochemistry teaching texts (), in which it has been assumed that one enzyme-catalysed step would be responsible for rate control through each pathway. On the other hand, uncertainty about the role of eIF4E in rate control has been created by contrasting observations of the effects of artificially enhanced eIF4E synthesis in primary cell cultures as opposed to cell lines (,,). Up to now, translation has not been subjected to detailed quantitative control analysis, and therefore the mode of control in the translation initiation pathway could not be precisely elucidated. In this article, we address the issue of rate control in the translation initiation pathway using genetic titration combined with control analysis of eIF synthesis rates . We have chosen to perform this work in , the eukaryotic organism with the best-defined translation system as well as the eukaryotic organism of choice for attempts to achieve comprehensive characterization of metabolic and genetic pathways in the coming years. Studies of this relatively simple eukaryote are far more readily integrated into coherent models of rate control than are the largely qualitative results derived from the diversity of dissimilar higher eukaryotic systems under study. In our study, we find that control follows a distributed model in which all steps investigated contribute to the determination of overall rate, albeit to differing degrees. The approach taken here has broad relevance for research on gene expression. pCM225 () was employed as template for PCR amplifications of the promoter-substitution cassettes (B). Transformants containing the promoter-substitution cassette were selected on YPD-G418 plates and further tested via analytical PCR. For construction of PTC210, (encoding eIF4G2) was deleted using a PCR-based disruption method (). Complementation of the doxycycline-inducible phenotypes of strains PTC209, 210, 229, 230 was achieved via transformation using plasmids expressing the eIF-encoding genes. The eIF4E- and eIF4G1-encoding plasmids were YCp33Supex2- and YCp33Supex2- (). The eIF1A- and eIF5B-encoding plasmids, pRS316- and pRS316-, were constructed via PCR amplification of and from chromosomal DNA and cloning into pRS316 (), respectively. For western blotting, standardized amounts of recombinant eIF4E and eIF1A were applied to SDS-PAGE gels next to the extracts or, alternatively, extract amounts equivalent to different numbers of cells were used to calibrate the band intensities for eIF4G and eIF5B. Antibodies against eIF4E and eIF1A were generated by immunization of rabbits with the purified factors (Abcam, Cambridge, UK). Antibodies against eIF4G1 were generated as described previously (), while an anti-eIF5B serum was kindly provided by Tom Dever (NIH, Bethesda, MD, USA). The membranes were incubated with FITC-labelled anti-rabbit IgG (Sigma, St Louis, MO, USA), washed and then scanned using a Typhoon Imager (Molecular Dynamics, Piscataway, NJ, USA). Translation in cell-free extracts was studied as described previously (). Each quantitation experiment was repeated at least three times. We applied basic concepts derived from metabolic control analysis (MCA), which was originally developed to examine the relative control exerted by each step in a metabolic pathway (,), to the translation initiation pathway. Specifically, we applied the concept of the sensitivity coefficient (), which expresses the dependence of a variable of the system (in this case the flux through the pathway, i.e. the translation initiation rate) on the rate of a certain step in the pathway. The original term referred to the effects of changes in enzyme concentration on catalytic flux through a metabolic pathway. The component control coefficient, as defined here, refers to the effect of changes in the concentration of a component engaged in the assembly of a macromolecular complex that has catalytic activity. This coefficient is defined at the steady state as , where is the flux through the system (in this case the translation rate). Our overall aim was to analyse key steps of control at the three major phases of the initiation pathway: 40S recruitment, 40S scanning leading to AUG recognition, and ribosomal subunit joining (A). In order to do this, we placed transcription of the chromosomal genes encoding eIF4E and eIF4G ( and , respectively; 40S recruitment), eIF1A (; scanning and AUG recognition) and eIF5B (; subunit joining), under the control of the Tet-off operator system () (B). We chose both eIF4E and eIF4G because there has been a long-standing debate in the field as to whether eIF4E, as opposed to any other factor acting on the initiation pathway, acts as the ‘rate-determining’ factor in translation (). The doxycycline-dependent slow-growth phenotypes of the strains could be suppressed by genetic complementation upon transformation using expression plasmids bearing the corresponding wild-type genes (C). Additionally, PCR was used to confirm the structure of the chromosomal insertions (data not shown). We next tested whether the four strains constructed as described earlier would allow us to regulate synthesis continuously over a wide range of steady-state levels. Calibrated western blot analysis was used to quantitate the abundance of each eIF over a range of doxycycline concentrations (A, B, C, D and E). These ‘genetic titrations’ enabled us to investigate the relationship between intracellular eIF levels and the rates of cell growth and of protein synthesis over a range that does not become excessively restrictive to cell function (see below). In control experiments, we investigated whether the levels of eIFs whose synthesis was not subject to Tet-off regulation were affected by doxycycline (data not shown). The results revealed that, in each of the four strains, only the eIF encoded by the -regulated gene was subject to limitation. We next determined the doubling times of each strain in logarithmic growth as a function of intracellular eIF abundance. Growth limitation by doxycycline was explored down to a growth rate of <20% of wild type, thus providing a broad range of data points for analysis (F). Control experiments revealed no effects of doxycycline upon growth of a strain that did not have a gene subject to control by the Tet-off control system. We also followed the rate of protein synthesis as a function of eIF abundance in the steady state. Polysomal gradients revealed shifts in the distribution of ribosomes from the polysomal fractions into the monosomal fractions (see SupplementaryData), and indicated differences in the sensitivity of the initiation pathway to comparable changes in abundance of the four eIFs. In order to obtain a more exact picture (A, B, C and D), we determined the rates of S-methionine incorporation into cells. Plotting these rates versus relative abundance of each eIF, we were then able to identify the quantitative differences between them in terms of control (E). A remarkable feature of the plots in is that they all seem to approximate to a biphasic structure, with a break at 80% of maximum [eIF] or higher between two regions of distinct values. One possible explanation for this behaviour is that progressively reducing the amount of intracellular eIF at some point restricts the redundancy of the macromolecular assembly routes, thus limiting eIF binding to a route that has a higher value. Irrespective of the mechanistic basis for this effect, the most relevant region in terms of normal assembly of the translation initiation apparatus is the region with the smaller value that is observed at higher eIF concentrations—this tells us the contribution to control of each eIF at or near its wild-type level in the cell. Metabolic control analysis (,,,) was originally developed for analysing the behaviour of enzyme systems that interconvert metabolites. Here we have developed the concept of the component control coefficient () for a macromolecular assembly pathway (see Materials and Methods section). The strength of rate control (component control coefficients) declined in the order: eIF4G > eIF1A > eIF4E > eIF5B (E). The rationale for calculating the component control coefficients as we have done in this work is summarized in the Materials and Methods section. The largest value obtained at near wild-type levels was 0.50 for eIF4G1. This means that for every change of 2% in [eIF4G1], the rate of initiation changes by 1%. Rate control imposed by eIF5B, in contrast, is more than three times weaker. In control experiments, we examined whether the abundance of typical endogenous mRNAs in the constructed strains is affected by -regulated changes in eIF gene transcription rate. We quantitated the abundance of a typical long-lived mRNA () and of a typical short-lived mRNA () in the strains listed in as a function of doxycycline addition. In all cases, doxycycline-induced inhibition of -regulated eIF production had no effect on mRNA abundance (data not shown). We, therefore, conclude that the changes in protein synthesis rate associated with reductions in the intracellular availability of the selected eIFs are not likely to be attributable to reduced mRNA stability. This finding is fully consistent with our earlier observations that reduced eIF activities, at least over a certain range, do not cause general destabilization of mRNA in yeast (,). A truly rate-limiting factor would have a value of 1, and as we can see from , all of the eIF values determined here fall far short of 1 (in the near-wild-type concentration range). This unpredicted result therefore tells us that the eIFs share rate control of the overall pathway, and that no single eIF exercises complete rate control (). A further key issue relates to the summation rule (), which states that the sum of all control coefficients should equal 1. More recent results have indicated that this sum can in fact be >1 (approaching 2) for metabolic pathways if there is macromolecular crowding () or where the enzymes on a pathway exchange substrate groups (). Our data now show that rate control summation in the translation initiation pathway, and therefore probably in other macromolecular assembly pathways, follows a different rule to that applicable to most metabolic pathways. The objective of the straightforward control analysis performed in this paper is to relate flux through the overall system (the translation initiation pathway) to local activities of the eIFs. However, we need to bear in mind that as protein synthesis is attenuated, this affects cell growth and physiology that in turn could, at least theoretically, feed back on the rate control relationships of the respective factors. A useful indication of the global state of the cell as a function of changes in the rate of protein synthesis is provided by the plots shown in , which derive from the data presented in and in the Supplementary Data section. In all four cases, growth is tightly linked to protein synthesis rate, with some variation in the slope. This suggests that there is a common fundamental relationship between growth and protein synthesis, with only a small dependence on which eIF is mediating limitation, thus simplifying the analysis and interpretation of the rate control data. However, it should be noted that we have not characterized the influence of changes in the cell physiological state on the values estimated here, which would be necessary for hierarchical control analysis () of this system. Thus the control relationships we describe can only be assumed to apply to the log phase growth conditions specified. While the cellular system was the primary focus of this quantitative study, it is also evident that many studies of translation initiation have utilized cell-free extracts. We accordingly decided to make a comparative study of the rate control behaviour of two of the eIFs, namely eIF4E and eIF1A. Extracts were prepared from strains PTC209 and PTC229 () that had been growing in the presence of doxycycline. Purified eIFs were then added back to reconstitute progressively the respective extracts. The translation competence of the extracts was followed using a capped and polyadenylated luciferase-encoding reporter mRNA (). Analysis of translation rate as a function of total eIF present in each extract revealed that the apparent control effects exercised by eIF4E and eIF1A in the lower concentration ranges were much greater than in the comparable states. The relationship between translation rate and eIF1A abundance in the extract from strain PTC229 grown in the presence of doxycycline (B, and compare C) did not approximate to the type of biphasic form seen , showing instead a region of very strong dependence on factor concentration leading to a (non-responsive) plateau. On the other hand, the dependence of translation rate on eIF4E concentration in A includes a region of intermediate factor dependence. In control experiments, it was found that extracts prepared from the same strains grown in the absence of doxycycline showed relatively minimal responses to the addition of eIF4E or eIF1A (insets in A and B). This minimal response was slightly higher in the case of PTC209. This was most likely because the chromosomal construction in this strain directed a somewhat reduced maximal (non-suppressed) rate of transcription, and thus of eIF4E synthesis, compared to the level directed by the wild-type promoter (data not shown). Replotting the data in A and B together allows more direct comparison between the eIF4E and eIF1A experiments (C) and also between the data as a whole and the results of . Overall, an important feature of these data is that they emphasize the contribution of the crowded and compartmentalized environment of the living cell to rate control in translation. This study has provided the first set of component control coefficients for eukaryotic translation initiation, and thus the first quantitative framework for defining how the components of this pathway collectively contribute to its overall control. The procedure utilized here could be applied to all of the macromolecular assembly pathways based on intermolecular interactions and different conformational states, thus building up an increasingly accurate picture of control across the network of intracellular interactions that are involved in gene expression. Establishing this common quantitative framework for representation and modelling of the control and regulation of gene expression should be generally useful. Our study illustrates how quantitative control data provide important insight into the workings of a complex biological system. The data show that rate control is distributed over different steps (and different eIFs) in the initiation pathway. It was noted previously that distributed rate control could be an inevitable consequence of the action of evolutionary ‘forces’ on a pathway of this type (). An additional outcome is that there is no simple relationship between intracellular eIF concentration and a factor's contribution to control of the system. Thus, eIF4E is significantly more abundant in yeast cells than are eIF1A and eIF4G (), yet its value is far from proportionately smaller than those of these other two factors. In other words, it is unwise to base any judgments of likely control strength of a component of such a pathway on intracellular abundance alone. This consideration is also relevant to any models of disease caused by mutationally induced deficiencies in eIF function (), since the quantitative data presented here show that we need to abandon the assumption that any particular eIF commands full control over the rate of the overall pathway under any particular set of conditions. It is of course essential to remember in a study of this type that at least eIF1A and eIF4G may act at more than one site on the initiation pathway, and thus that each value for these proteins represents the sum of multi-site action. This does not detract from the usefulness of these values as quantitative indicators of factor-centric control as manifested by the system under the defined conditions, but does mean that any mechanistic interpretation must take the multi-site functions into account. In this context, for example, it is remarkable that despite being involved in several steps on the initiation pathway, eIF1A manifests a value that is only 16% greater than that of eIF4E, a factor that is thought to have a single site of action on the pathway. In relation to potential targets for regulation of translation initiation, it is notable that the comparable values of three of the eIFs studied in this work make them all potentially effective sites for targeting regulation. Thus, the observation of distributed control in the translation initiation pathway also tells us not to assume that regulation is likely only to be effective if targeted to one or two of the characterized eIFs. Finally, the current study sets the stage for comprehensive rate control analysis of this, and other, eukaryotic gene expression pathways. This will ultimately lead to the discovery of new general control principles that guide such systems. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

An obvious approach to studying a biological processes, such as the reaction of cells to a stimulus, is to measure the activity of the cell at a sequence of time points. However, when the measurements consist of high-throughput gene expression microarrays, it is not obvious how to extract biologically meaningful results. We describe a new computational method, called StepMiner, the primary goal of which is to assist biologists in understanding the temporal progression of genetic events and biological processes following a stimulus, based on gene expression microarray data. At the most basic level, StepMiner identifies genes which undergo one or more binary transitions over short time courses. It directly addresses the one of the more basic questions one can ask of time course data: ‘Which genes are up-regulated or down-regulated as a result of the stimulus?’ and ‘When does the gene transition to up- or down-regulated?’ StepMiner extracts three types of binary temporal patterns. The first type, shown in (a and b), describe ‘one-step’ transitions, where the expression level of a gene transitions from a high to a low value or from a low to a high value. The second type, shown in (c and d), describes two-step transitions. Genes in this category turn on then back off or vice versa. The third type consists of genes for which the one- or two-step patterns do not fit appreciably better than a constant mean value (the null hypothesis). This can result when the gene expression level is genuinely constant, or when the other patterns fit no better than the constant because the behavior of the gene is complex. The expression levels for up- and down-regulated are chosen that best fit the data. Fitting the patterns of one- and two-step transitions requires an algorithm that evaluates every possible placement of the transitions (or step) between time points, and chooses the one that gives the best fit. This process is called adaptive regression. The algorithm was evaluated on simulated time course microarray data with 15 non-uniform time points. Noise-free data was generated for both one-step and two-step categories; Gaussian noise was then added to the original data, and then StepMiner was used to recover the original behavior, with a -value threshold of 0.05. A total of 4000 genes with 15 time points were artificially created, with 2000 one-step genes and 2000 two-step genes. A describes the proportion of correctly identified gene expression patterns as a function of different step heights, where the position of the steps are fixed at certain time points. All single steps are fixed at the fifth position and all binary two steps are fixed at the fifth and nineth positions. As can be seen in the figure, when the step height is 5σ, StepMiner identifies genes correctly over 90% of the time. As the step height is reduced relative to the noise level, the proportion of correct identifications drops dramatically (as expected). The drop in accuracy is higher for two-step signals because of the greater degrees of freedom for those signals. B describes the proportion of correctly identified time courses using the same setting as A except that the steps are placed between random time points. As the figure shows, there is a small reduction in the accuracy compared to A. The behavior of StepMiner is similar in both A and B. Higher confidence matches occur if all constant segments in a curve have several time points. This result shows that most matches where the steps are not at the beginning or end of the time course are reasonably high confidence. Hence, it would be desirable to design experiments so that there are several points before the first interesting transition and after the last interesting transition. C shows the sensitivity of StepMiner to the number of time points and the -value threshold. As can be seen from the figure, accurate matching of two-step signals requires more time points than matching of one-step signals. The proportion of matches can be increased by increasing the -value threshold, but only at the cost of an increased FDR (which can be measured and adjusted as described in False Discovery Rate section). D describes the proportion of two-step genes correctly identified when the number of time points between steps is varied. This figure is based on 2000 genes with 15 time points. The first step is fixed at fourth position and the spacing between steps is varied from 1 to 9. The height of the step is varied from 1σ to 5σ to observe the desired effect. As can be seen from the figure, a spacing of at least three time points is required for over 95% accuracy, when the step height is > 3σ. Also, as the second step approaches the end of the time points, the proportion of correct identifications decreases. The steps are also required to be placed at least three time points from the end points to achieve 95% accuracy. It is important to demonstrate the value of the method on real microarray data for at least two reasons: the true signals may not be step functions, and the noise from the actual experiment may not be Gaussian. Hence, StepMiner was applied to a publicly available time course of microarrays monitoring gene expression levels in yeast during the diauxic shift in a glucose-limited culture . In this experiment, the yeast utilizes fermentative metabolism when glucose is abundant. As the glucose is depleted, the metabolism shifts abruptly to oxidative metabolism. RNA samples were collected approximately every 15 min and measured with microarrays. An analysis of the results was published in 2005 [Brauer ]. In that article, the data were analyzed using hierarchical clustering by gene [Gollub ]. Of the many clusters generated, the authors picked seven clusters that had fairly high correlations and that, by visual inspection of the dendrogram, appeared to consist of genes with temporal behavior related to the diauxic shift. In the original article, the sets of genes in the selected clusters were examined using GO-TermFinder to identify GO annotations of genes that are enriched. The article lists several GO annotations that had extremely small -values according to GO-TermFinder. Many of the annotations are obviously related to diauxic shift. Based on these annotations, three of the clusters of genes appeared to be highly relevant to diauxic shift, three were enriched with annotations of unknown relevance to diauxic shift and one cluster was not significantly enriched with any GO annotations. For comparison, it is possible to reanalyze the data using gene sets derived from StepMiner. Binary signals were extracted from the diauxic shift data, using a -value cutoff of 0.05, resulting in an FDR of 15%. Out of a total of 2284 genes in the diauxic shift data, 1088 were matched to single steps, 267 were matched to binary two steps and 929 did not match anything. The fitting step functions are shown for three genes in . A heat map of the genes expression profiles appears in . In the heat map, the top genes are those that change once, the rising genes first, and falling genes second. Lower, there is a group of genes that go up then down, and last, the genes that go down then up. Each of these groups is sorted by the time of first change. The ordered response of genes to stimuli is immediately evident when so depicted. The heat map also makes apparent two discontinuities, at 8.25 h and 9.25 h. These correspond to observed changes in the growth rate of the yeast around 9 h. The genes are then automatically collected into five generic gene sets: ‘up’, ‘down’, ‘up then down’, ‘down then up’. The generic gene sets are further divided into specific gene sets based on the position and the direction of the transition. This process resulted in 80 different generic and specialized gene sets, which were analyzed using GO-TermFinder with a -value cutoff of 0.001. A table of the 120 low -value GO annotations, in ascending order, is included in the Supplementary Data S3. Many of the GO annotations are directly related to metabolism. The GO annotations and FDR-corrected -values for the clusters reported in Brauer were recomputed with the latest yeast gene annotations from the Gene Ontology Consortium website. To compare with the results of Brauer , shows the GO annotations from that article that had low -values, and shows the corresponding -values from the StepMiner groups. The annotations that had the lowest -values in Brauer had even lower -values in the StepMiner groups. Further, the GO annotations are obtained fully automatically using StepMiner — it is not necessary to select interesting clusters manually. In most cases, the -values in the reanalysis are lower than Brauer 's, which suggests that grouping by time-of-change is at least as effective as hierarchical clustering at identifying relevant genes. Four GO annotations had significant results in Brauer 's analysis, but not in the StepMiner analysis: ‘siderophore transport’, ‘intracellular transport’, and ‘secretory pathways.’ Interestingly, these GO annotations were associated with clusters that, in the words of Brauer were ‘less interpretable in terms of diauxic shift’. StepMiner is also a model-based method, but the one- and two-step patterns are different from the models of other methods. The transition interval method from Hottes is perhaps the most similar, but their models have a transition interval segment between constant-level segments. The transition interval in their model is defined as the change from 25 to 75% of the maximum. The Boolean model proposed by Shmulevich binarizes genes without considering the time component. These methods do not provide -values, FDR or other statistically justifiable measures of confidence. Even when the gene expression level over time is only approximately binary, we find that the results produced by StepMiner are sensible. For example, consider the measurements for the genes in . In each case, the behavior of the gene may be complex or noisy, but StepMiner reports reasonable (and objective) results about when each gene becomes up-regulated. The -value for an individual gene captures the degree to which the binary model fits the temporal variation in gene expression. Large variations in the supposedly down-regulated and up-regulated intervals will lead to worse -values than approximately constant behavior. Signals that transition between two levels, but transition slowly, will have worse -values than signals that transition rapidly. For a slowly transitioning signal, the best placement of the transition is not obvious; StepMiner will tend to put it in the middle of the transition. In the extreme case of purely linear behavior, StepMiner will place a transition in the middle—but the -value will be poor and the gene is likely to end up in the ‘other’ category depending on the user-specified -value cutoff. There are two ways that a low -value match can occur: there could be several consecutive points that are consistently low or high, or there could be one or two measurements that deviate greatly from the others. In practice, a low -value from multiple points is more trustworthy than a low -value from large differences, because a single deviant measurement could be an outlier resulting from non-Gaussian measurement error. Very short time courses are problematic, because reliable low -value matches are unlikely to occur. There is simply too little evidence to support the matching of steps, even when steps exist. On the other hand, very long time courses are problematic because the data may actually have more than two steps, and neither the one-step nor two-step patterns will match well. There is currently an upper limit of two steps in StepMiner because the running time of adaptive regression algorithm increases exponentially with the number of steps. Replicated measurements at the same time point should not be averaged. Instead, they should be handled using the same matching algorithm as sequential measurements, except that the algorithm should not try to put a step between simultaneous measurements. With this processing, they can directly improve the -values of extracted signals. This conclusion is supported by simulation results shown in . Each of the four different step types was simulated, with time of each step from a uniform distribution over the entire interval. As discussed above, the measurements at each time point were taken, and Gaussian noise was added so that the step height is 5σ. When a step is found between time points and , the time of the step is estimated to be ( + )/2. The ‘time error’ of the step is | − ( + )/2|. The number of correctly classified steps is shown. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

The anthracycline doxorubicin () is among the most versatile chemotherapeutic agents currently used in the clinic (,). The proven clinical utility of doxorubicin, a DNA-directed drug, has been tempered by dose-limiting cardiotoxicity, and this prompted a search for analogues with comparable therapeutic efficacy yet lacking the characteristic cardiotoxicity (). The anthracenedione class of compounds were identified as good drug candidates designed to satisfy these criteria. The anthracenediones, most notably mitoxantrone (Novatrone™) and its 5,8-dehydroxy analogue ametantrone, are simplified anthracycline analogues, which retain the planar ring structure characteristic of anthracyclines that permits intercalation between base pairs of DNA (,) (). The biochemical mechanism by which mitoxantrone exerts its cytotoxic effects is likely to be multifaceted, however its role as a topoisomerase II poison and subsequent induction of cytotoxic double-strand DNA breaks has been well established (). Despite an improved clinical tolerability of mitoxantrone chemotherapy, it still exerts a range of toxic side-effects including myelosuppression and cardiotoxicity (). This cardiotoxicity may be attributed to the 5,8-dihydroxy substituents of mitoxantrone since mice treated with this drug exhibited a delayed mortality () yet those treated with ametantrone did not. A second-generation group of anthracenediones were prepared in an effort to develop compounds endowed with better therapeutic efficacy and reduced side-effects. These compounds retained the anthraquinone nucleus of mitoxantrone, however, the 5,8-substituents implicated in its cardiotoxicity were removed and nitrogen atoms introduced into the chromophore. Krapcho . () rationalized that these nitrogen atoms may provide basic sites or improved hydrogen bonding, therefore providing the analogues with a potentially greater affinity for DNA and topoisomerase II. A series of these novel anthracenediones were prepared and included compounds bearing either one (mono-aza) or two (di-aza) nitrogen atoms within the chromophore (). Interestingly, an and screen of these compounds for anti-tumour activity revealed that only mono-aza analogues comprising the nitrogen atom at position 2 of the chromophore demonstrated significant anti-cancer efficacy (). Within this select group of compounds, BBR 2778 (6,9-bis[(2-aminoethyl)amino]benzo[]isoquinoline-5,10-dione dimaleate) emerged as the most promising drug candidate. BBR 2778 (subsequently named Pixantrone™; ) demonstrated superior anti-leukemic activity in mice over a wide, well-tolerated range of doses when compared with mitoxantrone (). Further preclinical studies in mice showed that Pixantrone has a wide spectrum of anti-tumour activity, and a marked efficacy against haematological malignancies, particularly lymphomas and leukemias (). Histopathological evaluation of the heart tissue in these studies revealed that Pixantrone induced no detectable delayed cardiotoxicity (). These key findings prompted the entry of Pixantrone into clinical trials for further development. The drug is progressing through these trials with encouraging results as a single agent and in combination regimens, and is currently in Phase-III studies in the treatment of indolent and aggressive non-Hodgkin's lymphoma (). Like mitoxantrone, the mechanism of action of Pixantrone is not fully understood but likely to be multimodal. Pixantrone interacts with DNA with modest affinity via intercalation (). The drug can function as a topoisomerase II poison by stabilizing the normally transiently bound protein–DNA complex (,,), giving rise to double-strand DNA breaks. However, these breaks do not correlate directly with the potency of Pixantrone as a cytotoxic compound (,). This suggests that Pixantrone may be operating via an additional, currently unknown, mechanism of action. Although mitoxantrone functions as a topoisomerase II poison via its ability to intercalate within DNA, a novel form of mitoxantrone–DNA interaction has been identified. Mitoxantrone can be readily oxidatively metabolized to generate reactive species that bind covalently to DNA (). A common theme amongst these studies was that each of the oxidative systems utilized hydrogen peroxide which can generate formaldehyde by oxidation of substrates present in the system (). Subsequent studies using cell-free systems revealed that formaldehyde alone, rather than oxidative metabolism, was sufficient for activation of mitoxantrone and subsequent covalent binding of the drug to DNA (). Presently, it is believed that these DNA adducts are linked via a ‘secondary’ amino function of a single side-chain of mitoxantrone (). Although mitoxantrone and Pixantrone share close structural similarity, Pixantrone bears a ‘primary’ amino group in each of its side-chains and is therefore more susceptible to formaldehyde activation and consequently has a greater potential to form DNA adducts. The present study explored the potential of Pixantrone to bind covalently to DNA through pre-activation by formaldehyde. Pixantrone was provided by Cell Therapeutics Europe, Bresso, Italy. Mitoxantrone dihydrochloride and formamide were purchased from Sigma Chemical Co., St. Louis, MO, USA. Formaldehyde solution (40% v/v) was obtained from BDH. The plasmid pCC1 containing the UV5 promoter was constructed by Carleen Cullinane (Peter MacCallum Cancer Centre, Melbourne, VIC, Australia). A Maxi Plasmid Purification Kit was purchased from Qiagen, Valencia, CA, USA. Ultra-pure dNTPs, [αP] dATP (3000 Ci/mmol), [αP] dCTP (3000 Ci/mmol) and ProbeQuant G-50 micro-columns were purchased from GE Healthcare, Piscataway, NJ, USA. The restriction enzyme HindIII was purchased from Promega, Madison, WI, USA and calf thymus DNA was from Worthington Biochemical Corporation, Lakewood, NJ, USA. Klenow fragment from DNA polymerase I and BSA were both from New England Biolabs, Beverly, MA, USA. Tris-saturated phenol was obtained from Invitrogen, Carlsbad, CA, USA and glycogen was from Roche Molecular Biochemicals, Nutley, NJ, USA. The remaining chemicals and reagents were of analytical grade. Distilled water passed through a four stage Milli-Q purification system was used to prepare all solutions. Pixantrone and mitoxantrone stock solutions (stored at −20°C) were prepared by dissolving each in Milli-Q water to an approximate concentration of 2 mM. The precise concentrations of each drug were determined spectrophotometrically using = 19 200 M cm at 608 nm and = 296 at 641 nm for mitoxantrone and Pixantrone, respectively. Formaldehyde solutions were freshly prepared on the day of each experiment. HB101 cells containing the plasmid pCC1 were grown overnight in selective LB broth containing 50 μg/ml ampicillin. The plasmid was subsequently isolated using a Qiagen Maxi Plasmid Purification Kit. The plasmid was linearized by restriction digestion with the sticky-end generating enzyme HindIII. The 5′-overhang of the fragment was filled in using the Klenow fragment of DNA polymerase I in the presence of either [αP] dATP or [αP] dCTP. Unincorporated label was removed from the labelled fragment by passing the reaction mixture through a G-50 ProbeQuant chromatography column. The eluted fragment was subsequently subjected to phenol/chloroform extraction, ethanol precipitated and then resuspended in 1× TE (10 mM Tris, 1 mM EDTA, pH 8.0). The final DNA concentration was adjusted to 400 μM by the addition of sonicated calf thymus DNA. Covalent drug–DNA adducts were formed in a reaction mixture typically consisting of the following: end-labelled DNA (25 μM) was reacted with Pixantrone or mitoxantrone and formaldehyde in phosphate-buffered saline (pH 7.0) at 37°C. Intercalated drug (but not covalently bound drug) was removed from DNA by extraction with Tris-saturated phenol twice followed by a single chloroform extraction. DNA was then precipitated with ethanol and sodium acetate in the presence of glycogen as an inert carrier. Samples were subsequently resuspended in 10 μl 1× TE buffer and denatured in two volumes of loading dye (90% formamide, 10 mM EDTA, 0.1% bromophenol blue and 0.1% xylene cyanol) at 52°C for 5 min. DNA adduct stability studies were performed as described above, however an additional extraction procedure was employed following incubation of drug–DNA adducts for defined time periods. Samples were extracted once with phenol, once with chloroform and then ethanol precipitated to remove any non-covalently bound drug resulting from dissociation as a consequence of the thermal lability of the adducts. Samples were subjected to electrophoresis through 0.8% agarose gels overnight at 30–40 V in 1× TAE buffer. Gels were dried under vacuum in a Bio-Rad Model 583 gel drier and subsequently exposed to a phosphor screen overnight. Phosphorimaging analysis of each dried gel was performed using a Molecular Dynamics Model 400B PhosphorImager and the bands quantitated using ImageQuant software. xref #text xref #text

xref italic #text Enzymatic probing was performed as described previously (). Specifically, after removal of 5′-triphosphate with CIAP (calf intestine alkaline phosphatase, Stratagene) RNAs were 5′-end-labeled with [γ-P]ATP using T4 PNK (New England Biolabs) and PAGE purified. Traces of labeled RNA were incubated with RNase S (Promega), V1, A or T1 (Ambion), either in the presence or absence of TPP in cleavage buffer (10 mM Tris pH 7.5, 100 mM KCl and 5 mM MgCl, 1 µg tRNA). Prior to the addition of nucleases, RNA was denatured by incubation at 95°C for 5 min and allowed to fold by slowly cooling to room temperature for 40 min in the presence of MgCl and TPP (Sigma) if indicated. Subsequently, the reactions were precipitated, dissolved in PAGE-loading buffer and analyzed on a 10% denaturing PAA-gel. RNA markers were prepared by partial digestion with alkaline solution and cleavage with RNase T1 under denaturing conditions. Visualization was done by PhosphorImaging (Fuji, FLA3000). Biotinylated thiM RNA (35 nM) was incubated with [P]-end labeled RNA hairpin N25.1.22 (1 nM), 35 nM soluble streptavidin, and varying concentrations of MgCl in binding buffer (10 mM HEPES pH 7.5, 100 mM KCl) for 30 min at RT. After incubation, the reactions were passed through 0.45 µm nitrocellulose membranes and washed four times with 200 µl binding buffer. Bound RNA was quantified by PhosphorImaging. All assays were performed in quadruplicates. Chemical probing was performed as described previously (). Modification with DMS or kethoxal: In a final volume of 10 µl, RNA (0.2 µg/µl) was incubated in 50 mM HEPES pH 7.8, 100 mM KCl, 10 mM MgCl with 0–4.5 mM TPP (Sigma) at 25°C for 30 min. Then, 1 µl of a 600 mM solution of DMS in ethanol or 200 mM kethoxal (ICN) in water, respectively was added, mixed and incubated 20 min at 25°C. After precipitation, 5′-end labeled primer was annealed and primer extension performed. Modification with CMCT: In a final volume of 10 µl, RNA (0.2 µg/µl) was incubated in 50 mM potassium borate pH 8.0, 100 mM KCl, 10 mM MgCl with 0–4.5 mM TPP (Sigma) at 25°C for 30 min. Then 1 µl of a 200 mM solution of CMCT in water was added, mixed and incubated 20 min at 25°C. After precipitation, 5′-end labeled primer was annealed and primer extension performed. In a 20 µl reaction 0.2 µg RNA in 50 mM Tris–HCl (pH 8.3), 75 mM KCl, 3 mM MgCl, 20 mM DTT, 0.5 mM dNTPs (each) was heated to 65°C for 5 min, then chilled on ice for 1 min. After addition of 1 µl (200U) of Superscript II Reverse Transcriptase (Invitrogen) the reaction was incubated at 42°C for 50 min, followed by inactivation at 70°C for 15 min. After precipitation the DNA fragments were separated on an 8% denaturing PAA gel at 2000 V and visualized by PhosphorImaging. The sequencing reactions were carried out using the PCR templates with the Sequenase Version 2.0 PCR Product Sequencing kit (USB). To monitor TPP-induced changes of the chemical modifications of 165 thiM we used probing. For this purpose, the respective DNA templates were transcribed in the respective probing buffer (50 mM HEPES pH 7.8, 100 mM KCl, 15 mM MgCl for DMS and kethoxal probing; 50 mM potassium borate pH 8.0, 100 mM KCl, 15 mM MgCl for CMCT probing) in the presence of 0–3 mM TPP for 3 h and used directly for chemical probing. The amount of RNA in the reactions was equal, as determined on 2.5% agarose-gels. ThiM riboswitches were prepared from dsDNA templates, generated by PCR using the appropriate primers, by transcription and purified by PAGE. 5′-biotinylated RNA was prepared by GMPS transcription, using 4-fold excess of GMPS over GTP, and subsequent treatment with iodo-acetyl biotin (Pierce). For the preparation of the RNA hairpin N25.1.22 the complementary oligos 22.1F and 22.1R were hybridized and used for transcription by T7 RNA polymerase. The Mutant M4 was generated by QuickChange (Stratagene) according to the manufacturer's instructions. We previously identified an RNA hairpin (N25.1.22) with a loop sequence which is complementary to nt 95–101 of the thiM riboswitch (). We showed that the RNA hairpin binds to the expression domain of the thiM riboswitch, thereby specifically competing with TPP but not with thiamine for binding. This allowed the conclusion that the binding region of the hairpin to the riboswitch undergoes fundamental structural rearrangements upon metabolite binding. A depicts the respective RNA hairpin N25.1.22 and the complementary region responsible for thiM binding highlighted in cyan. Binding of the hairpin to thiM strictly depends on the Mg-ion concentration (B), suggesting that the region targeted by the hairpin also forms a hairpin structure, thus allowing the formation of a kissing complex. Kissing complexes can be quite stable, even in cases where only two base pairs are involved, and frequently mediate molecular recognition events between RNA molecules (,). In addition, secondary structure changes required for proper thiM riboswitch function should fulfill the following criteria: (i) The region of the thiM riboswitch that binds to the RNA hairpin should be single stranded in the absence of TPP and become paired upon addition of TPP. (ii) In the TPP-free form of the thiM riboswitch the SD sequence should be accessible for ribosome binding, thus be essentially unpaired; and (iii) the nucleotides that reside downstream of the SD region should not be engaged in extensive base-pairing secondary structure in the TPP-free form of the thiM riboswitch, to avoid loss of translation efficiency. A secondary structure of the TPP-free thiM riboswitch that is consistent with these requirements is shown in A and B, respectively. In this fold, the formation of the hairpin nt 92–103, harboring the binding region (nucleotides shown in cyan) of the selected RNA hairpin N25.1.22, stipulates the pairing of residues 71–86 with nt 125–108. These pairings are consistent with both chemical (A) and enzymatic (B) probing data (see also Supplementary Figures S2 and S3). Chemical probing with dimethylsulfate (DMS) modifies the proposed ACCA-loop (A, nt 96–99) whereas the adjacent stem of the hairpin remains unmodified. The observed RNase S1 and RNase A cleavages at positions A96 to A99 (B) further indicate that the target region of the RNA hairpin indeed forms the proposed loop structure. The crystal structure of the aptamer domain of the thiM riboswitch revealed that the nt 71–86 form the paired regions P5, P4 and P1 in its TPP-bound variant, whereas the nt 108–125 are assumed to be implied in the formation of P8 and L8 of the expression domain leading to the sequestration of the SD-sequence (). In our model of the TPP-free variant of the thiM riboswitch, the anti-SD sequence pairs with nt 83–86 in the context of an extended helix, leaving the SD-sequence unpaired and thus accessible for binding to the ribosome. These findings are supported by the chemical probing data (A), which clearly show that nt 126–129 of the SD sequence are modified by kethoxal and DMS in the absence of TPP. Thus, in accordance with the function of the riboswitch the SD sequence is unpaired and allows initiation of translation by ribosome binding. Chemical and enzymatic probing data ( and ) indicate that the stems P2 and P3 of the aptamer domain are pre-organized in the absence of TPP. No modifications of the nucleotides comprising these stems were found by chemical probing and the formation of stem P2 is further supported by RNase V1 cuts that reveal that positions C15 to G17 and C49 to G51 are paired (Supplementary Figure S4B). However, chemical and enzymatic probing indicates that the positions U28 to U32 and U39 to U46 are unpaired in the absence of TPP (A, Supplementary Figures S2, S4B). The region ranging from nt 52–70 was highly modified by chemical probing (Supplementary Figure S2B, S2C), and enzymatic cleavages by RNases A and S were observed (B and Supplementary Figure S4B), indicating that the region from U52 to A70 is also largely unpaired. We next analyzed changes in the secondary structure of the thiM riboswitch that are induced by the addition of TPP. Changes that could be monitored by chemical and enzymatic probing are summarized in and , representative PAA-gels of these experiments are shown in Supplementary Figures S6 and S7. Chemical probing of the aptamer domain of the thiM riboswitch (Supplementary Figure S1) reveals changes within the aptamer domain that are in accordance with its recently reported X-ray structure (,). Most importantly, modifications of the nucleotides U39 to A43 that are essential for the recognition of the 4-amino-5-hydroxymethyl-2-methylpyrimidine (HMP) moiety of TPP are less abundant in the presence of TPP (Supplementary Figure S1). Additionally, reduced modifications of ntU28 to U32 and ntU39 to U46 in the presence of TPP were observed. This indicates that these residues become paired or involved in a more compact structure upon binding to TPP, and the data obtained by chemical probing are in agreement with the crystal structure of the complex of TPP with the aptamer domain (,). Chemical probing of the 165 thiM construct revealed similar changes for nucleotides G40–U46 as well as G60 and A61 in the presence of TPP (Supplementary Figure S6C). More importantly, in chemical probing of 165 thiM in the presence of TPP, we observed changes in the expression domain of the thiM riboswitch. For example, modifications or nucleotides that are part of the SD-sequence (D: G129 and to a lesser extent G126 and G127) were slightly decreased. This result suggests that the SD-sequence becomes involved in base pairing upon TPP binding to the aptamer domain. It is also in accordance with the proposed structural model and the generally assumed mechanism for translation inhibition by sequestration of the SD-sequence. Other TPP-induced changes can be observed within the proposed hairpin motif nt 92–103 (B). Modifications of nucleotides A96–A99 become less intensive in the presence of TPP (B) suggesting that this region is unpaired in the absence but becomes paired in the presence of TPP. This rearrangement results in the nearly consecutively base-paired stems P6 and P7. Notably, this conformational change explains why the RNA hairpin N25.1.22 is released from the thiM riboswitch by the addition of TPP. The modification intensities of nucleotides U87 and A85, A84 are decreased in the presence of TPP, indicating that these nucleotides become less accessible upon binding of the metabolite (Supplementary Figure 6C). The three nucleotides U87, A85 and A84 are involved in the formation of stem P1 which resides at the connection of the aptamer- to the expression domain of the riboswitch in its TPP-bound form (A). The adjacent nucleotide G86 however, is neither modified in the absence of TPP nor in its presence, consistent with G86 being involved in stable G–C base pairing (Supplementary Figure 6C). The enzymatic probing experiments also indicate TPP-induced changes within the aptamer domain of the thiM riboswitch, although less pronounced as compared to the chemical probing analyses ( and Supplementary Figure S7). The most prominent changes were observed within the helix J3-2, harboring the binding region of the HMP moiety of TPP, at nucleotides U46 and U39 (B). The X-ray structure of the aptamer domain revealed that nucleotide C24 of stem P2 is bulged out in the presence of TPP (). In accordance with this finding the cleavage at position C24 with RNase V1 becomes slightly protected in the presence of TPP (C). TPP-dependent protection is also observed for the single-strand-specific RNase A cuts at positions C38, U39 and U46 (B), consistent with the notion that these nucleotides are essentially unpaired in absence of TPP, but form a compact structure in its presence. Enzymatic cleavages of nucleotides in the expression domain were also influenced by addition of TPP: The formation of the loop L8 becomes perceptible by RNase S1 cuts at position U118 and U119 as well as RNase T1 cleavage at G117 that are well pronounced in the presence of TPP (D). The notion that G117 is only cleaved in the presence of TPP is consistent with G117 being embedded within a stable helical context in the TPP-free form and becoming exposed within a single-stranded loop (L8) in the TPP-bound form of thiM (A). TPP-induced cleavage by RNase T1 can also be detected for G123 and to a lesser extent for G122 (Supplementary Figures S7B and S7C). Positions 116–121 show strong RNase S cleavage in the presence of TPP but not in its absence (Supplementary Figure S7C). Cleavage by RNase A and S were observed at nt 96–99 which form the putative hairpin structure that is targeted by the RNA hairpin N25.1.22 (B). Notably, these cleavages are protected in the presence of TPP, leaving only C97 accessible for the nuclease (B). These findings are in accordance with the data that were obtained by chemical probing (B). Taken together, they explain why the binding of the RNA hairpin N25.1.22 to the thiM riboswitch is competed by the addition of TPP. In addition, on the basis of above results, we can suggest a secondary structure model of the TPP-free form of the thiM riboswitch and its structural transition to the TPP-bound form (). In this model, the stem regions P2 and P3 of the aptamer domain are preformed and allow the binding of the thiM riboswitch to the HMP moiety of TPP. The presence of Mg-ions complexed to pyrophosphate induces a conformational change, resulting in the formation of P4, P5 and P6–P8. Our model also explains why thiamine fails to induce the conformational changes necessary to gain genetic control, although it binds to the riboswitch, albeit with significant lower affinity. We have previously reported a riboswitch variant containing mutations in the region nt 95–101, which is complementary to our hairpin (). In mutant M4 the formation of stem P7 proposed for the TPP-bound riboswitch is disrupted (A). This mutation—the substitution of G100 and G101 by two Cs—caused loss of genetic control of the corresponding β-galactosidase fusion construct although the riboswitch still bound to TPP (K∼20 nM) indicating that besides stem P8, which harbors the SD sequence, also the formation of stem P7 may be important for proper function of the riboswitch (). To test whether the loss of the ability to exert genetic control in mutant M4 can be rationalized on the basis of secondary structure changes, we carried out chemical probing of this mutant. Results of chemical probing as well as the derived secondary structure for M4 are shown in and Supplementary Figure S8. Chemical probing data for M4 in the absence of TPP are in accordance with the secondary structure resembling the TPP-bound form of the wild-type 165 thiM riboswitch (B for nt 1–60, and lanes without TPP in C). However, base pairing cannot be entirely maintained in stem P7 due to the introduced mutations. The aptamer domain appears to be entirely preformed also in the absence of TPP except for stem P1 (nt 85–88) that appears to be unpaired (A and modifications in the region nt 85–88 in C). Furthermore, the data suggest that in M4 the SD sequence is mainly paired even in the absence of TPP (absence of modifications for nt 126–130 in C). Chemical probing in the presence of different concentrations of TPP reveals changes in the aptamer domain for nucleotides that participate either in TPP binding (e.g. U39, G40, G60 and to a lesser extent A41, A43 in C) or in formation of stem P1 (G86, U87 in C). Alterations of nucleotide accessibility comprising the SD sequence were not observed. In this study, we have elucidated the conformational changes that occur in the thiM riboswitch during the transition from the TPP-unbound to the metabolite-bound form. Based on previously reported results of our and other groups we derived the secondary structure of the thiM riboswitch of in the absence of TPP and its conformational conversion upon TPP addition (,). In the sense of communicating domains, portions of the aptamer domain pair with the parts of the expression domain in the absence of TPP. In previous studies, Winkler . have shown by inline-probing that the thiM riboswitch undergoes structure modulation upon binding of its ligand (). The authors found many single-stranded positions in the region of nt 39–80 — which is part of the aptamer domain — exhibiting a reduction in spontaneous cleavage in the presence of TPP. Serganov . conducted enzymatic probing of the thiM riboswitch using the nucleases V1 (helix specific) and T2 (single-strand specific) to further confirm the TPP-binding sites derived from their crystal structure of the aptamer domain of the thiM riboswitch (). TPP-induced changes in our chemical probings are consistent with TPP-binding sites known from the crystal structure of the aptamer domain observed by Serganov . Within the expression domain of the thiM riboswitch Winkler . observed low spontaneous cleavage by inline-probing for the proposed stems P2–P7, indicating that the corresponding nucleotides remain structured in both the presence and absence of TPP (). Consistent with our results, they found that within the expression domain of the thiM riboswitch the region comprising nt 126–130 becomes more structured upon TPP addition (). The conformational changes observed in our study are more dramatic than suggested by the in-line probing data, and include parts of the aptamer- and the expression-domain. In this manner, our data suggest that the expression domain initially pairs with a large part of the aptamer domain in the absence of TPP. The TPP-induced conformational transition results in a paired anti-SD sequence and renders the ribosome-binding site unpaired, thus allowing translation initiation. Upon TPP-binding the anti-SD sequence becomes unpaired and sequesters the SD sequence which leads to inhibition of the translation start. This inter-domain communication leads to global changes in the secondary structure of the thiM riboswitch, without significant changes in the ratio of paired versus unpaired regions, i.e. large portions of the riboswitch that are paired in the TPP-free riboswitch remain engaged in pairing in the TPP-bound form. The same is true for most of the unpaired regions. Enzymatic and chemical probing data in absence and presence of TPP support our secondary structure model. Our model reveals that the stem regions P2 and P3 of the aptamer domain are preformed in the TPP-free form of the thiM riboswitch from . These regions were shown to build up one of the helices responsible for binding to the HMP moiety of TPP. The crystal structure of the aptamer domain of the riboswitch reveals that this helix is well ordered in the presence of pyrithiamine which shares the HMP moiety with TPP but completely lacks the pyrophosphate group (). Sudarsan . showed that pyrithiamine binds to the thiM riboswitch with an affinity comparable to that of thiamine, but orders of magnitude weaker than that of TPP (). The pyrophosphate moiety of TPP binds to a second helix consisting of P4 and P5. This helix remains disordered in a crystal structure of the aptamer domain in complex with ligands that lack any phosphate (). The presence of one phosphate group leads to an ordered structure, and the overall structure of the aptamer domain becomes more compact compared with the complex that is formed by thiamine analog pyrithiamine (). We show that the stems P4 and P5 of the thiM riboswitch are not preformed in the absence of TPP. Instead, our data suggest that they base pair with parts of the expression domain to form a long helix with a terminal hairpin. Upon addition of TPP this structural element is disrupted and the P4, P5 helices are formed. This conformational change also explains why the short riboswitch-binding RNA hairpin N25.1.22 () is released upon the addition of TPP, but not by thiamine, suggesting that binding occurs largely via the HMP moiety, in accordance with structural data. However, N25.1.22 is not competed by the addition of thiamine. The formation of stems P4 and P5 is not possible since thiamine lacks the pyrophosphate moiety. Thus, thiamine is able to bind to the thiM riboswitch but does not induce the conformational changes that are necessary to repress gene expression. Taken together, the secondary structure model that we suggested above on the basis of the binding behavior of the RNA hairpin N25.1.22 was confirmed by the more direct evidence resulting from chemical and enzymatic probing data. We previously reported that a riboswitch variant M4 with mutations loosening stem P7 cannot exert genetic control any more (). Here we show by chemical probing that, in contrast to the wild-type 165 thiM riboswitch, the mutant M4 adopts a structure similar to the TPP-bound form of the wild-type riboswitch also in the absence of TPP. The expression domain forms a hairpin-like structure containing the stems P6, P8 and P7, which is partially unpaired due to the introduced mutations. Probing data indicate that the SD sequence is paired even in the absence of TPP and that no TPP-induced structural alterations occur within this region. This means that presence or absence of TPP does not affect the accessibility of the SD sequence and hence the initiation of translation. This is exactly what we observed using β-galactosidase fusion constructs (). Chemical probing thus provides an explanation for the loss of genetic control in M4. Interestingly, mutation of G100/101 to Cs seems to favor the TPP-bound form in respect to the TPP-free form of the riboswitch although these mutations disrupt two G–C base pairs in either case. It seems that the mutations and subsequent loss of base pairing destabilize the ultimate hairpin in the TPP-free form to such an extent that it cannot form any more and that this renders formation of the entire stem from nt 70–125 in the TPP-free form unfavorable. In contrast, the hairpin-like structure that is adopted by the TPP-bound form of the thiM riboswitch can still form even if there are two mismatches in stem P7. Stem P7 is flanked on either side by G/C rich stems (P6/P8) that contribute to the stability of this structure. Our data thus indicate a finely balanced equilibrium of the two forms of the riboswitch that is necessary for its ability to exert genetic control. Even sequence alterations that, at a first glance, seem to affect both forms to a similar extent can slightly favor one of the structures and cause loss of genetic control. In conclusion, our data provide insight into the conformational rearrangements and movements that occur in thiM riboswitches when the TPP-concentration in bacteria increases to a certain threshold level, resulting in gene repression by inhibition of translational initiation. The secondary structures and conformational changes introduced here lead to insights that will be important for the functional engineering of riboswitches and for their further exploration as potential drug targets (). Moreover, the knowledge about the structure of riboswitches in the absence of their ligands might allow the design of allosteric switches based on RNA that do not affect translation efficiency, but enable strong regulation by the addition of a ligand. Furthermore, the results indicate that short RNA motifs represent a useful and possibly generally applicable tool for detecting conformational changes in allosteric RNAs on a secondary structure level. Accordingly, phylogenetic analysis might provide insight into the conformational changes of expression domains from other ‘thi’-box containing riboswitches, although sequences of these domains are less well conserved than those of the respective aptamer domains. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Genome sequencing projects of higher eukaryotes have revealed a surprisingly low number of genes that fail to explain their organismic and developmental complexity (). Post-transcriptional processes that recode, diversify and fine tune the transcriptome are now regarded as the potential players leading to evolutionary variation. Alternative splicing and RNA editing are the key events leading to transcriptome diversification (,). Small non-coding RNAs, in turn, fine-tune RNA stability and translatability (). RNA editing by adenosine deaminases that act on RNA (ADARs) is widespread in metazoa (). ADARs deaminate adenosines to inosines (A-to-I) within double-stranded or structured RNAs. As inosines resemble guanosines, editing by ADARs can alter splice sites or change the coding potential of an RNA and, therefore, generate diversity in proteins that are encoded by a single gene (). Moreover, ADAR-mediated editing can affect non-coding sequences such as introns, UTRs or miRNAs thereby altering the secondary structure, stability or base-pairing potential of edited RNAs (). Mammals have two active editing enzymes, ADAR1 and ADAR2, which exhibit different substrate specificities (). A third protein, ADAR3, seemingly lacks enzymatic activity (). ADARs contain a conserved deaminase domain and two or three double-stranded RNA binding domains (dsRBDs). Both viral and cellular ADAR targets have been described. Most cellular substrates are found in the central nervous system, but also non-neuronal substrates are increasingly being discovered (). A well-studied substrate for ADAR editing is the pre-mRNA encoding glutamate receptor subunit B (GluR-B) (). This RNA is edited at two exonic sites termed the Q/R and R/G, respectively. Editing at these sites leads to codon exchanges and thus alters the properties of GluR-B-containing ion channels. Editing at the Q/R site, located in exon 11, changes a glutamine codon to an arginine codon and results in a lower Ca permeability of the channel (,). Furthermore, editing at the Q/R site is essential for proper tetramer assembly of AMPA receptors (). The R/G site is located in exon 13 upstream of an alternatively spliced region known as the flip/flop module (). Here, an arginine codon is converted to a glycine codon, which allows faster recovery of the receptor from desensitization (). Two additional editing sites are found in intron 11, called hotspot 1 (or +60 site) and hotspot 2 (or+262/263/264 site), respectively (). The Q/R site and hotspot 2 are solely edited by ADAR2, whereas hotspot 1 and the R/G site can be edited by ADAR1 and ADAR2 (,). Editing at the Q/R site is nearly complete. Mice lacking ADAR2 are prone to epileptic seizures and die about three weeks after birth; however, they can be completely rescued by introducing a ‘pre-edited’ GluR-B gene (). Editing levels at the R/G site reach 75% in adult mice, but are much lower in embryos and gradually increase throughout development (). In most coding targets, exonic editing sites depend on an intronic editing complementary sequence (ECS) (). The ECS, base-pairs with the editing site to form the double-stranded structure required for ADAR binding. Editing must, therefore, be a co-transcriptional event that occurs prior to intron removal (). The close proximity of editing and splice sites coupled with experimental data suggest that RNA editing and splicing are coordinated : for instance, mice lacking ADAR 2 are deficient in the removal of intron 11 in GluR-B pre-mRNA which neighbors the Q/R editing site (). Coordination of editing and splicing has also been observed at the R/G site of GluR-B pre-mRNA. Here ADAR2 inhibits splicing , but seemingly not (). The C-terminal domain of RNA Pol-II might play a role in coordinating editing at nascent transcripts as it is required for efficient autoediting of ADAR2 pre-mRNA, but not for splicing (). Moreover, in malignant gliomas, hypoediting correlates with alternative splicing in 5-HT serotonin receptors (). Finally, in , a strong correlation between alternative splice site choice and editing efficiency upstream of the alternative splice site was observed for two ADAR substrates (). In principle, RNA editing could affect splicing via three alternative mechanisms: first, editing might alter -acting signals that modulate splicing activity. Second, editing could destabilize a double-stranded structure, thereby allowing access for proteins to the splice site. Third, ADAR binding, but not editing itself, could help to recruit factors that regulate splicing. To distinguish amongst these possibilities, we investigated the correlation between splicing and editing at the R/G and Q/R sites in GluR-B pre-mRNA using a novel cell-based splice assay. Reporter constructs containing the editing sites and constructs that either mimic edited RNA, or in which editing and/or ADAR binding is prevented, were tested. Our results show that editing at the R/G site leads to a reduction in splicing efficiency in the adjacent intron and influences the downstream alternative splicing event. This phenomenon is caused by the inosine at the R/G site and does not require ADAR binding. Furthermore, editing of both the Q/R nucleotide and an intronic editing hotspot is required for efficient splicing of the intron adjacent to the Q/R site. The EGFP ORF was amplified from pEGFP-C2, (Clontech) by PCR using suitable primers and cloned in-frame, downstream of the RFP ORF in dsRed express. To separate the two tags, a flexible linker (an unstructured region of the β-galactosidase gene) was inserted between the two ORFs. Additionally, a nuclear localization signal (SV40 large T-antigen NLS) and a polylinker was inserted into the expression plasmid. This vector was the starting construct into which further DNA fragments for investigation were introduced. As a splice control, the adenovirus major late pre-mRNA (Ad1), was introduced (). An unspliceable Ad1 pre-mRNA was produced by mutating the 5′ and 3′ splice consensus sequences of the Ad1 intron using the method described in the QuickChange™ site-directed mutagenesis kit (Stratagene, La Jolla, CA,USA). GluR-B sequences were amplified by PCR from mouse genomic DNA. For sequences spanning exons 13 through 14, the following primers were used: 5′ TCGAGAATTCTTGCAGTGTTTGATAAAATGTGGA 3′ (forward, containing an EcoRI site) and 5′ CTTCGGTACCCACTCTCCTTTGTCGTACCACCA 3′ (reverse, containing a KpnI site). A ‘pre-edited’ version was generated by site-directed mutagenesis. An ‘uneditable’ version was produced by replacing the sequence of the ECS, thus preventing formation of a double-stranded RNA. To create an uneditable ‘binding but not editing’ R/G-containing construct, the cytosine in the ECS opposing the edited base was mutated to a guanosine (). A GluR-B DNA stretch that spans exons 13 through 16 was amplified using primers: 5′ TGTAGTCGACAATTGCAGTGTTTGATAAAA 3′ (forward) and 5′ ATAGGTACCTTAACACTCTCGATGCCATA 3′ (reverse). The Q/R editing constructs spanning exons 11 through 12, were amplified with primers: 5′ TCTTGTCGACGAGCCTTGGAATCTCTATCATG 3′ (forward) and 5′ AGGAGGATCCAACTCTTTAGTGGAGCCAGAGT 3′ (reverse). ‘Pre-edited’ variants of the Q/R site and intronic hotspot 2, respectively, were introduced by site-directed mutagenesis. Deletion mutants were created using internal restriction sites. All inserts were controlled by sequencing. HeLa and HEK293 cell lines were transfected using either Lipofectamine 2000 (Invitrogen, CA,USA) or Nanofectin (PAA, Pasching, Austria) according to the manufacturer's instruction. Cells were analyzed 24–48 h post transfection. RNA was isolated with Trifast reagent (PEQLAB, Erlangen, Germany) followed by two rounds of DNAseI digestion. Additionally, to remove residual transfected DNA, the sample was digested with DpnI. RevertAid M-MuLV Reverse Transcriptase (Fermentas, Lithuania) was used for reverse transcription using a GFP-specific primer: 5′ CCTCTACAAATGTGGTATGGCTG 3′. The 1/10th of the reactions was used for PCR reactions. Plasmids that were used for the transfections were used as PCR controls. The following primers were used for the amplifications: 5′ GGTGGAGTTCAAGTCCATCTACATGG 3′ (forward, in RFP); 5′ TCGACCAGGATGGGCACCAC 3′ (reverse, in GFP); 5′ ACCTCATATCCGTATACAAACCGTT 3′ (reverse, in intron 13 of mouse GluR-B); 5′ GCAGCAAGCTTGACAACAAAAA 3′ (reverse, in Ad1 intron); 5′ GCAGCTGCTGACATCTTTATAGTG 3′ (reverse, in intron 11 of mouse GluR-B). The following primers were used for amplification: 5′ TCTTGTCGACGAGCCTTGGAATCTCTATCATG 3′ (forward, exon 11); 5′ ATATGGATCCGTGGCGATGCCGTAGCCTTTGGAA 3′ (reverse, exon 12); 5′ TGTAGTCGACAATTGCAGTGTTTGATAAAA 3′ (forward, exon 13 with a SalI site) and 5′ ATAGGTACCTTAACACTCTCGATGCCATA 3′ (reverse, exon 16 with a KpnI site). PCR products were purified, cloned into the pCR2.1 vector (Invitrogen, CA,USA) and individual clones were sequenced. Cells grown on coverslips were fixed and stained as previously described (). Measurements of fluorescence intensities were carried out with the help of Quantity One software (Biorad, CA,USA). To determine whether variations amongst different samples are statistically significant. A student's -test was performed for all data sets using Microsoft Excel. To investigate the effect of RNA editing on splicing of the GluR-B pre-mRNA, a splice assay was established that allows the quantification of splicing efficiencies . cDNAs encoding red and green fluorescent proteins (RFP and GFP) were cloned in-frame into a tissue culture expression vector separated by a flexible region of the lac Z gene, an NLS and a polylinker. Genomic fragments containing splicing and editing sites and their flanking introns were then inserted into the polylinker separating the RFP and GFP open reading frames (ORFs) (Supplementary Figure S1, ). Upon transfection into tissue culture cells the RFP reporter was constitutively expressed, while expression of the GFP reporter depended on the removal of the intron. As a positive control, constructs were used that contained either only the flexible linker region between the RFP and GFP reporter, or a standard splicing substrate, the adenovirus major late pre-mRNA (Ad1). To further control for the effect of nonsense mediated decay (NMD) on RFP expression, a splicing-deficient Ad1 was also tested. Fragments of mouse GluR-B spanning the majority of exon 13, containing the R/G site, the adjacent intron 13 and a large piece of exon 14 were introduced into the reporter vector (mimicking ‘wild type’) and tested for splicing efficiency. A ‘pre-edited’ construct was generated by inserting a guanosine at the site of editing. To generate an uneditable construct, the sequence of the ECS was changed, preventing formation of a target editing site (Supplementary Figure S2). In HeLa cells, which show moderate levels of editing, the uneditable mutant exhibited the strongest GFP fluorescence when compared to RFP fluorescence (A). The wild-type construct displayed intermediate green signals while the ‘pre-edited’ R/G construct showed the weakest green signals. This result suggests that splicing of the mini-reporter construct is inhibited by editing. The wild-type construct, which can be edited by endogenous ADAR, showed a slight reduction in splicing while the ‘pre-edited’ construct was spliced least efficiently. For quantification, 100 cells were recorded in the green and red channels, respectively, and fluorescence ratios were calculated (C). Furthermore, splicing efficiencies were quantified by RT-PCR using sets of primers specific for intronic (unspliced) and exonic (spliced) regions (B and D). In both assays (fluorescence of 100 cells versus RT-PCR experiments), the ‘pre-edited’ construct showed reduced splicing efficiency, while an uneditable construct was most efficiently spliced. Student's -test indicated that the observed differences between wild-type and pre-edited or the uneditable construct were significant. Most interestingly both methods give similar results, indicating that either technique can be used to measure splicing efficiencies. To allow faster and more quantitative measurements of splicing efficiencies, we employed fluorescence activated cell sorting (FACS) to determine fluorescence intensities (). HEK293 cells, which have low endogenous editing activity, were used for subsequent cotransfection experiments (). Wild-type, ‘pre-edited’, and uneditable R/G constructs were cotransfected with plasmids expressing either ADAR1 or ADAR2, or an empty control plasmid. After 48 h red and green fluorescence of approximately 20 000 cells were analyzed by FACS. The gate was set to analyze cells that exhibited moderate red fluorescence (A: red fluorescence, -axis; green fluorescence, -axis). Within the gate, the ratio of green (spliced) to red (input) fluorescence was calculated for each event. The mean and SD of these ratios is given in 2B and is graphically displayed in 2C. These experiments confirm that a ‘pre-edited’ R/G fragment is spliced less efficiently than a wild-type or uneditable construct (B and C). Cotransfection of a wild-type construct with ADAR2 reduces splicing efficiency, suggesting that ADAR2 edits the R/G site. Bulk sequencing of RT-PCR products of transfected cells confirmed editing at the R/G site in the presence of ADAR2 (D). To test whether binding of ADAR2 to the R/G site had an effect on splicing, we tested a GluR-B mutant construct ‘BNE’ (binding not editing) that allows ADAR binding but prevents editing () (Supplementary Figure S2). This construct was spliced as efficiently as the wild-type construct and was not affected by ADAR2 cotransfection. This demonstrates that an inosine at the R/G site but not binding of ADAR2 is responsible for reduced splicing of intron 13 (B and C). GluR-B is alternatively spliced downstream of the R/G site. Two splice conformations exist . The flip conformation fuses exon 13 to exons 15 and 16 while in the flop conformation exon 13 is fused to exons 14 and 16 (). Conceptually, it is possible that editing in exon 13 preferentially suppresses splicing of the adjacent intron 13 but leaves splicing of intron 14 unaffected. If this was the case, editing might lead to preferential exclusion of exon 14 and inclusion of exon 15 (flip conformation). Therefore, to test whether R/G site editing influences alternative splicing, fragments spanning exons 13–16, resembling the edited, an uneditable, or wild-type state were introduced into the RFP-GFP vector and transfected into HEK293 cells. Using primers located in the RFP and GFP regions of the construct, two bands were detected by RT-PCR (A). Sequencing revealed that the upper band contained correctly spliced fragments in both, the flip and flop conformation at approximately the same frequency (exons 14 and 15 are of identical size). The faster migrating band represented an erroneously spliced product, fusing exon 13 directly to exon 16, leading to a premature stop codon. Interestingly, the erroneously spliced product was most prominent in wild-type and the uneditable R/G constructs, while the ‘pre-edited’ fragment was predominantly spliced correctly. Taken together with the finding that editing of the R/G site decreases the efficiency of splicing, our observation suggests that a decrease in the speed of splicing facilitates alternative splice site choice, at least in the context of the reporter construct. The R/G site is located two nucleotides upstream of the 5′ splice-site of intron 13. Editing could therefore impair the activity of this splice donor site and selectively slow down splicing of intron 13 thus promoting preferential inclusion of exon 15, the flip conformation. To test for a possible correlation between R/G site editing and alternative splicing , cDNAs from adult mouse brain were amplified using primers located in exons 13 and 16, respectively. Here, considerable variation was observed between brain cDNAs from different mice. While a positive correlation between editing and splicing in the flip conformation was observed in the cDNA of one mouse (mouse B in B), this correlation was rather weak in the cDNA obtained from another mouse (mouse A in B). This indicates that both splicing and editing vary amongst different individuals and neuronal tissues. Most importantly, these two post-transcriptional events lack an obvious functional correlation as previously suggested (). In contrast to the R/G site, lack of editing at the Q/R site selectively inhibits splicing of intron 11, and leads to nuclear accumulation of GluR-B pre-mRNA (). To investigate the influence of editing at the Q/R site on splicing of intron 11, Q/R-containing genomic fragments spanning exons 11 and 12 (‘wild type’) were cloned into the RFP-GFP vector. Despite the use of different cells lines (HEK293, HeLa, neuroblastoma) splicing was only detected in a few cells, using the cell-based fluorescence splice assay (data not shown). Introduction of a guanosine at the Q/R site or removal of the ECS did not increase splicing efficiency. However, removal of large parts of intron 11 resulted in an increase in splicing activity (data not shown). The region removed also included an intronic editing hotspot 2, a target of ADAR2. Therefore, a construct was made where both the Q/R site and hotspot 2 in intron 11 were ‘pre-edited’. Observation of red and green fluorescence under the microscope indicated efficient splicing of this reporter. Unfortunately, a golgi-like localization pattern of all Q/R constructs precluded their quantitative analysis by FACS. Thus, to quantify splicing efficiencies, RT-PCR experiments were performed using intronic and exonic primer combinations. To remain in the exponential phase of amplification, 25 PCR cycles of different cDNA dilutions were run (C). After electrophoresis (A), band intensities were measured, and splicing efficiencies calculated (B). Wild-type and constructs ‘pre-edited’ at the Q/R site showed only limited splicing. Cotransfection of the wild-type construct with ADAR2 clearly increased splicing efficiencies at a statistically significant level. Removal of the ECS also led to a moderate increase in splicing which was found to be statistically insignificant. A construct ‘pre-edited’ at the intronic hotspot 2 alone showed no increase in splicing, while concomitant ‘pre-editing’ at intronic hotspot 2 and at the Q/R site showed the highest splicing efficiency that was also statistically significant (A and B). The intronic editing site at hotspot 2 (AAA) is predicted to form a double-stranded structure by base pairing with a complementary region located at the intronic editing site at hotspot 1 (UUU) (Supplementary Figure S3). Editing of one of the two strands might help to open the double-stranded structure of this region, enabling access of factors that facilitate splicing. Therefore, a compensatory mutation was introduced around position 60, changing the three consecutive U residues opposing the editing hotspot to AAA, thus preventing base pairing with the adenosines to be edited. However, this mutant (compensatory mutation; ) displayed weak splicing even when combined with a “pre-edited” Q/R site (Q/R edited, compensatory mutation; ). This suggests that changes in the primary sequence rather than an alteration of the secondary structure are responsible for the observed increase in splicing efficiency upon editing of hotspot 2. Sequencing of the spliced product derived from the wild-type fragment cotransfected with ADAR2 showed that exon 11 is correctly joined to exon 12 and that the Q/R site is edited. Sequencing of the unspliced precursor showed that hotspot 2 is efficiently edited upon cotransfection of ADAR2 (D). In addition to the PCR product of expected size, another, lower molecular weight band was observed in most RT-PCR reactions. Sequencing of this fragment revealed an erroneously spliced product. A cryptic splice site in exon 11 was used instead of the proper 5′ splice site in intron 11. The same erroneous fragment was isolated from mouse brain cDNA, which indicates that this splicing error can also occur (data not shown). Whether this error is due to under editing of GluR-B RNA cannot be proven, since the editing sites are spliced out in these RNAs. Several findings suggest a tight co-transcriptional coupling of splicing and editing (, ). Here, we show that RNA editing within the pre-mRNA encoding the B subunit of the mouse glutamate receptor can have dual effects on splicing. While editing at the Q/R site and at an adjacent intronic editing hotspot is a prerequisite for splicing of intron 11, editing at the R/G site downregulates splicing, possibly facilitating alternative splicing. p p l e m e n t o r y D a t a a r e a v i a l a b l e a t N A R O n l i n e .