examples/Spatial_multi-omic_map_of_human_myocardial_infarction.txt

Coronary heart disease driving acute myocardial infarction is the larg- est contributor to cardiovascular mortality, which in turn is the leading cause of death worldwide1. Substantial progress has been made in the acute therapy of myocardial infarction, focusing primarily on percu- taneous coronary intervention resulting in decreased acute mortality. However, the morbidity and mortality caused by left ventricular cardiac remodelling after myocardial infarction remain unacceptably high2.
Cardiac remodelling after myocardial infarction involves immune cell recruitment and demarcation of the infarcted area followed by resorp- tion of necrotic tissue, phagocytosis, myofibroblast activation, scar formation and neovascularization3. Understanding the exact cellular and molecular mechanisms of cardiac remodelling processes from the acute ischaemic event to the chronic cardiac scar formation in their spatial context will be key to developing novel therapeutics.

Here we used a combination of single-cell gene expression, chro- matin accessibility and spatially resolved transcriptomics to study the events of cardiac tissue reorganization and to characterize the cell-type-specific changes in gene regulation, providing an integrated spatial multi-omic map of cardiac remodelling after myocardial infarc- tion. Our multi-omic data-driven approach, including spatial context, enables us to understand how a given cell state changes based on the cells’ neighbourhood and how this relates to transcriptional and regu- latory variations. By deconvoluting spatial transcriptomics spots into cell-type abundances, we characterized cell niches occurring in differ- ent stages following acute myocardial infarction. We identified differ- ent cell states of cardiomyocytes, endothelial cells, myeloid cells and fibroblasts that are associated with disease progression on the basis of the integrated single-cell multi-omics data. Moreover, we inferred the gene-regulatory networks differentiating these cell states and pro- jected this information onto specific tissue locations, thus mapping putative regulators controlling gene expression on specific myocardial tissue zones and disease stages. This enabled us to gain novel insights into the gene-regulatory programmes driving injury of cardiomyocytes, activated phagocytic macrophages and their relation to myofibroblast differentiation in cardiac tissue remodelling. Our results provide a comprehensive spatially resolved characterization of gene regulation of the human heart in homeostasis and after myocardial infarction. We have released our spatial multi-omics data through publicly avail- able platforms to enable users to interactively explore the dataset. We anticipate that this data will be a reference map for future studies and ultimately for the development of novel therapeutics.

We applied an integrative single-cell genomics strategy with single- nucleus RNA sequencing (snRNA-seq) and single-nucleus assay for transposase-accessible chromatin sequencing (snATAC-seq) together with spatial transcriptomics from the same tissue mapping human cardiac cells in homeostasis and after myocardial infarction at unprecedented spatial and molecular resolution (Fig. 1a–c and Supplementary Table 1). We profiled a total of 31 samples from 23 individuals, including four non-transplanted donor hearts as con- trols, and samples from tissues with necrotic areas (ischaemic zone and border zone) and the unaffected left ventricular myocardium (remote zone) of patients with acute myocardial infarction (Fig. 1a). These acute myocardial infarction specimens were collected from heart tissues obtained at different time points after the onset of clini- cal symptoms (chest pain), before the patients received an artificial heart or a left-ventricular assist device because of cardiogenic shock and as a bridge to transplantation (Supplementary Fig. 1a–c). We also analysed nine human heart specimens at later stages after myocardial infarction (fibrotic zone; Fig. 1b) that exhibited ischaemic heart disease and were available from heart transplantation recipients at the time of orthotopic heart transplantation.
For each cardiac sample, we obtained 10-μm cryo-sections and isolated nuclei from the remaining tissue directly adjacent to the cryo-section with subsequent fluorescence-activated nuclei sorting (FANS) for snRNA-seq and snATAC-seq (Fig. 1c). After filtering out low-quality nuclei, we obtained a total of 191,795 nuclei from all samples for snRNA-seq, with an average of 2,020 genes per nucleus, together with chromatin accessibility data from 46,086 nuclei overall with an average of 28,066 fragments per nucleus (Supplementary Fig. 2a,b and Supplementary Tables 2–5). After controlling for data quality, the spatial transcriptomics datasets contained a total of 91,517 spots (average of 3,389 spots per specimen and 2,001 genes per spot) (Sup- plementary Figs. 2c,e–g and 3a,b). Quantification based on histology revealed an average of four nuclei per spatial transcriptomic spot from all slides (Supplementary Fig. 2c and Supplementary Table 6). Sam- ples from the ischaemic zone had the lowest abundance of nuclei and
an enriched expression of genes associated with cell death and the regulated necrosis pathway, suggesting increased necrotic cell death (Supplementary Fig. 2d). This integrated dataset represents, to our knowledge, the largest and most comprehensive multi-modal profil- ing of human myocardial infarction tissue including spatial informa- tion and samples at distinct disease progression stages. We devised an integrative data analysis approach spanning all three modalities of our single-cell experiments to study cardiac cell-specific information and cell-specific interactions in their spatial and disease progression context (Extended Data Fig. 1a).
We established a map of major human heart cell types using the snRNA-seq and snATAC-seq datasets independently. First, we clustered cells on the basis of the integrated snRNA-seq data from all samples after batch correction (Extended Data Fig. 1b). Clusters were annotated with curated marker genes from the literature4–6 and ten major cardiac cell types were identified (Fig. 1d,e). We also found an additional cluster with enriched expression of the cell-cycle marker gene MKI67, which showed a high score of cell-cycle G2/M and S phases and was mainly recovered in ischaemic zone samples (Extended Data Fig. 1c,d). To validate the annotations, we compared the data with a recent study on healthy human hearts4 and an independent novel dataset of ischaemic heart samples (n = 3, generated during this study) and observed a high agreement and correlation in terms of molecular profiles and cellular composition (Extended Data Fig. 1e–g). Of note, the cycling cells were also captured in the independent ischaemic dataset (Extended Data Fig. 1f ).
We next integrated and clustered the snATAC-seq data from all sam- ples (Extended Data Fig. 2a). These clusters were annotated on the basis of gene chromatin accessibility with the same markers as for snRNA-seq. This approach identified eight major cell types, matching all cell types from snRNA-seq data with the exception of two rare cell types (that is, mast cells and adipocytes) (Fig. 1f,g). Label transfer from snRNA-seq to snATAC-seq indicated that the annotations between these two modalities were consistent (Extended Data Fig. 2b,c). This was fur- ther supported by a high correlation of cellular composition between snRNA-seq and snATAC-seq and the presence of the same eight cell types in the majority of samples (Extended Data Fig. 2d,e). To explore regula- tory information provided by the snATAC-seq, we performed transcrip- tion factor footprinting analysis using cell-type-specific pseudo-bulk ATAC-seq profiles. This revealed footprinting-based binding activity of known transcription factors such as MEF2C (ref. 7) in cardiomyocytes, CEBPD)8 in myeloid cells, FOS–JUNB9 in fibroblasts and SRF10 in vascular smooth muscle cells (vSMCs), which correlated with the expression of their predicted target genes in snRNA-seq data (Extended Data Fig. 2f ). Together, our integrative analysis of transcriptomic and chromatin accessibility data defined a robust catalogue of cell types in the adult human heart across multiple modalities and samples.

Using these data, we first identified overrepresented biological pro- cesses for each major histomorphological region (control, remote zone, border zone, ischaemic zone and fibrotic zone) using spatially variable genes (Supplementary Table 7). We identified cardiac muscle contrac- tion in remote zones and controls, with adaptive immune system in the border and ischaemic zones and with matrisome processes in the fibrotic zones (Extended Data Fig. 2g). Overall, this analysis confirmed that the spatial data clearly reflect typical zones of biological processes following acute human myocardial infarction.
Since each spatial transcriptomics spot captured a group of cells, we increased its resolution by estimating the cell-type compositions of each spot. To this end, we deconvoluted each spot on the basis of the annotated snRNA-seq data from the same sample (Fig. 1h, Supplemen- tary Figs. 2e–g and 3a,b, Supplementary Tables 8 and 9 and Methods). The estimated cell-type compositions from spatial transcriptomics of each patient generally agreed with their respective observed composi-
activity, and in ischaemic regions, increased myeloid cell abundance occurred in areas of higher NFκB signalling activity (Fig. 1h,i).
tions in the snRNA-seq and snATAC-seq data (Extended Data Fig. 2h). We
then estimated signalling pathway activities with PROGENy (Methods)
Mapping the information obtained from the snATAC-seq data to for each spot from the spatial gene expression data. The comparison of
space resulted in spatially resolved footprinting-based transcription factor binding activity, as exemplified by the previously described tran- spatially localized pathway activities with the estimated cellular abun-
dance per spot enabled us to link the information on spatial cell com-
scription factors associated with cardiomyocytes (for example, MEF2C; position to cellular function for each slide. For example, in areas with
ref. 7), myeloid cells (for example, CEBPD8 and ATF111), fibroblasts (for example, FOS–JUNB9) and vSMCs (for example, SRF10) (Fig. 1j). To test the association of genetic variants with cell types, we performed enrich-
ment analysis based on cell-type-specific pseudo-bulk ATAC-seq pro-
files and cardiomyopathy-related single nucleotide polymorphisms
(SNPs) obtained from genome-wide association studies12 (GWAS). We
focussed on SNPs relevant to left ventricular function, since we hypoth-
esized that these might provide the most biologically relevant infor-
mation for the cellular composition of myocardial tissue. This analysis
revealed that SNPs associated with stroke volume and left ventricular
end-diastolic volume were enriched in endothelial cells (Extended Data
Fig. 2i), consistent with the role of the endothelial cells in cardiac relaxa-
tion and dilation . SNPs associated with left ventricular end-systolic 13
volume and left ventricular ejection fraction were enriched in cardio- myocytes, supporting the relationship between contraction and these left ventricular measures. We also visualized the spatial distribution of GWAS signals by mapping SNPs associated with left ventricular ejec- tion fraction to each spot from spatial transcriptomics (Extended Data Fig. 2j). In summary, our integrated spatial atlas enabled us to map cell-type abundance, signalling pathway activities, transcription factor binding activity and GWAS signals across the complete spectrum of cardiac tissue zonations, providing an in-depth view at tissue remodel- ling processes following myocardial infarction in humans.