Single-Cell Splicing Isoform Atlas of the Adult Human Heart and Heart Failure

Abstract

Background: Alternative splicing plays crucial roles in normal heart development and cardiac disease by influencing protein-coding sequences, functional domains, and molecular networks. However, a detailed characterization of the human heart isoform landscape remains incomplete.

Methods: Leveraging long-read single-nucleus RNA sequencing and computational analysis, we dissected full-length isoform heterogeneities, expression patterns, and usage shifts across cell types, cell states, and cardiac conditions of the adult left ventricle. We applied in silico approaches to assess the functional relevance of identified isoforms; validated isoform compositions of representative cardiac genes using reverse transcription quantitative polymerase chain reaction and targeted amplicon sequencing; and developed a web server for interactive navigation of our results.

Results: The data revealed that isoform heterogeneity is widespread in the cardiac cellular system, serving as a posttranscriptional buffer system that calibrates the molecule reservoirs in human hearts. In healthy left ventricles, ≈30% of cell type-specific genes were polyform, using multiple isoforms tailored to cell type-specific programs. Among ubiquitously expressed genes, >300 showed differential isoform usage with cell type specificity. Compared with heart failure, 379 genes in cardiomyocytes demonstrated marked isoform usage shifts, most of which are predicted to change protein coding outcomes through direct changes in protein coding sequences and switches between intron retention and non-protein-coding biotypes. In contrast, cell state-specific programs tend to operate on monoform genes associated with changes among cell states. In addition, our data revealed heart failure-associated differential isoform usage events in stromal and immune cell types in the cardiac microenvironment.

Conclusions: We present a comprehensive atlas of splicing isoforms in the normal adult heart and heart failure through long-read single-nucleus RNA sequencing and comprehensive computational analyses. The results suggest crucial roles of isoforms in buffering core cellular programs and contributing to disease-associated cell states. The full-length details of these cell-specific isoforms serve as an important reference for downstream translational and mechanistic studies and are available on our online data portal at https://github.com/gaolabtools/heart-isoform-atlas.

Keywords: heart; heart ventricles; sequence analysis, RNA.

Figure 1.. Complex cellular ecosystem in the left ventricle revealed by long-read single-nuclear transcriptomics.

A, Workflow of tissue collection and sequencing of 6 non-diseased (ND) and 6 end-stage heart failure (HF) donors. B, UMAP of 30,683 nuclei of 10 major cell types, labeled by marker genes. C, Comparison of cell type compositions between ND and HF samples (n=6 per group). D, Heatmap of top differentially expressed genes in cell types in both ND and HF samples. E, Top gene set enrichment results comparing HF to ND samples. F, Volcano plots of differential gene expression analysis in 3 cardiac cell types. Stars highlight splicing related genes. Two-sided Wilcoxon tests were used in (C) with **P ≤ 0.01. LV, left ventricle; CM, cardiomyocyte; FB, fibroblast; EC, endothelial cell; PC, pericyte; SMC, smooth muscle cell; EDC, endocardial cell; Mye, myeloid; Lym, lymphoid; LEC, lymphatic endothelial cell; Neu, neuronal.

Figure 2.. Isoform heterogeneity pervades diverse cellular programs in the normal left ventricle.

A, Number of isoforms by structural category. FSM, full splice match; ISM, incomplete splice match; NIC, novel in catalog; NNIC, novel not in catalog; Other includes antisense, genic intron, genic genomic, and intergenic. B, Frequency of genes based on the detected number of associated isoforms. C, UMAP of 30,683 nuclei showing 8 Louvain clusters of single-cell isoform expression data. D, Number of upregulated isoforms (DEIs) in each isoform-based cluster. E, Alignment between gene expression-derived cell types and isoform derived clusters. The Adjusted Rand Index (ARI) is shown. F, Calculation of isoform diversity and threshold of monoform and polyform genes. prop (p), proportion. G, Distribution of isoform diversities of cell type-specific DEGs from non-diseased samples. The number of monoform (m) and polyform (p) cell type-specific DEGs displayed in an inset bar plot. H, Number of genes involved in top enriched ontologies in each cell type, colored by isoform diversity categories. Significance corresponds to term enrichment. *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001, **** P ≤ 0.0001. GO:BP, gene ontology: biological process. I. Heatmap of isoform diversities of genes in housekeeping ontologies. Small circles indicate polyform genes. J, Heatmap showing pair-wise interactions between cell types. K, Bar plot of the numbers of interacting genes and isoform diversity categories in each cell type in normal hearts.

Figure 3.. Cell type-specific isoform usage in ubiquitously expressed genes.

A, Heatmap of pairwise differences in the proportions of isoforms exhibiting cell type-specific isoform usages. B, Quantification of genes with DIU events by cell type, with colored bars indicating DEG classification. C, Relative portions of two usage shift patterns of dominant isoforms (dMDI and sMDI) in DIU events, grouped by DEG classification. dMDI, distinct most dominant isoform; sMDI, stable most dominant isoform. D, Comparison of isoform diversity scores of genes with DIU events characterized by dMDI and sMDI from all samples. Wilcoxon test. **** P < 0.0001. E and F, (left) Gene expression and (right) isoform proportions of TNNI3 and ACTG1 in each cell type. Isoforms having less than 5% proportion are collapsed as ‘Other’. Isoform names are annotated with transcript biotype. PC, protein coding; RI, retained intron; NMD, nonsense-mediated decay. G and H, Sashimi plots overlayed by donor depicting consensus read counts of genomic regions covering the TNNI3 and ACTG1 isoforms from (E and F). Transcript structures are shown at the bottom.

Figure 4.. Cell type-specific isoform usage shifts in heart failure compared to normal hearts.

A, Volcano plot illustrating DIUs between ND- and HF-derived cardiomyocytes. Each dot represents a tested gene, labeled with the isoform with the greatest difference in isoform proportions. B, Sunburst plot showing the number of DIUs that are DEG, non-DEG, sMDI, and dMDI in cardiomyocytes. C, Proportion of transcript biotype switches identified between ND and HF of DIUs associated with a dMDI. D, Relative proportions of isoform diversity shift patterns in cardiomyocyte DIU events. E, Comparison of alternative splicing events in isoforms involved in DIU compared to isoforms with no DIU in cardiomyocytes. F and G, (top, left) Single-cell gene expression, (top, right) isoform proportions, and (bottom) selected transcript structures of FRY and RSRP1. Isoforms having less than 5% proportion are collapsed as ‘Other’. H, Sunburst plots showing the number of DIUs that are DEG, non-DEG, sMDI, and dMDI in endothelial cells, fibroblasts, mural and myeloid cells. SE, skipping exon; MX, mutually exclusive exon; A5, alternative 5’ splice site; A3, alternative 3’ splice site; RI, retained intron; AF, alternative first exon; AL, alternative last exon.

Figure 5.. Isoform expression patterns contribute to heart failure-associated cell states.

A, UMAP of 5 cardiac cell types and states characterized by gene expression data. B, Relative cell proportions of cell states/subpopulations comparing ND and HF samples (n=6 per group). Bars represent mean + SEM. C, Upregulated genes in heart failure-associated cell states/subpopulations, colored by isoform diversity classification. Number of monoform and polyform classifications of significant differentially expressed genes are shown. Gray data points represent genes with no available isoform data. Odds ratio (OR) and P-value of monoform to polyform enrichments are shown. D, Heatmap of top differentially expressed isoforms in each cell state/subpopulation. Statistical analysis was performed using two-sided Wilcoxon tests in (B) and Fisher’s exact test in (C). *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001, **** P ≤ 0.0001, not significant (ns).