Skip to main content
. 2023 Nov 22;12:e93161. doi: 10.7554/eLife.93161

Figure 3. Multicellular programs associated with myocardial remodeling.

(A) Each factor forming the latent space reconstructed by multicellular factor analysis when applied to single-cell data can be interpreted as a higher-level representation of coordinated molecular processes across cell types, here referred to as multicellular programs. The specific cell-type signatures from these programs can be recovered from the feature weights across cell types, where the expression changes of one cell type relate to the other. Moreover, since each multicellular program associates with the variability of samples (e.g. differentially active across conditions), cell-type signatures can also be interpreted in the same context. (B) Log10(number of genes) for each cell-type Factor 1 signature associated with the sample conditions of the human myocardial infarction data. Signatures were divided into ischemic-like or myogenic-like signatures based on the weight of each gene. (C) Jaccard index across myogenic-like (upper triangle) and ischemic-like (lower triangle) cell-type factor signatures associated with the sample conditions of the human myocardial infarction data. (D) Functional enrichment of MSigDB’s hallmarks in cell-type signatures. Enrichment is quantified as normalized weighted means, using the gene weights of each cell-type signature. Top 25 pathways based on mean absolute enrichment score are shown.

Figure 3.

Figure 3—figure supplement 1. Coordinated transcriptional programs across cell types in myocardial remodeling inferred with multicellular factor analysis and differential expression analysis.

Figure 3—figure supplement 1.

(A) Gene weights of Factor 1 estimated from the human myocardial infarction single-cell dataset across cell types, representing the multicellular program associated with myocardial remodeling.
(B) Log10 number of genes belonging to distinct cell-type-specific factor signatures. (C) Similarity of gene weights of Factor 1 to traditional differential expression analysis. Left panel shows the correlation between log fold changes of differential expression testing of three contrasts with the gene weights of Factor 1. Right panel shows the proportion of differentially expressed genes with a false discovery rate ≤ 0.05 for each cell type and contrast from the genes used in the multicellular factor analysis model.
Figure 3—figure supplement 2. Cell-state-dependent and -independent transcriptional deregulations upon myocardial infarction.

Figure 3—figure supplement 2.

(A) Cell-state markers captured in cell-type Factor 1 signatures of endothelial (Endos) and myeloid cells, cardiomyocytes (CMs), and fibroblasts (Fibs). In each panel is shown the adj. p-value of the enrichment of cell state markers of the indicated state in cell-type-specific factor signatures (left) and the group’s mean proportion of cell states across patient groups calculated from single-cell data (right). The most overrepresented state in myogenic (left column) and ischemic (right column) samples is shown per cell type. Sample size per patient group is shown in the figure. (B) Distribution of log-normalized gene expression across samples of the human myocardial infarction atlas of CCAR1, MARCH1, GUCY2C, and HIF1A from pseudobulk profiles of Endo and myeloid cells, CMs, and Fibs, respectively. The left column is the pseudobulk expression of the gene in a cell type for each sample, while the right column comes from the pseudobulk expression of the gene in a cell state. Sample size per patient group is shown in the figure. (C) Proportion of variance (eta-squared) of gene expression of genes belonging to the Factor 1 signatures of CM, Endo, Fib, and myeloid cells explained by the patient’s conditions or cell states. Colors represent significance of the association of pseudobulk expression variability to patient’s conditions or cell states (ANOVA adj. p-value <0.05). (D) Distribution of the log2 ratios across cell types between the proportion of variance explained by the patient condition and cell states. Each point in each distribution is the ratio of a gene of the cell-type Factor 1 signature (see C). CM, n=1234, Endo, n=279, Fib, n=584, myeloid, n=230. Data information: In A, composition bar plots, error bars show one standard deviation from the mean. In B data is presented as box plots where the middle line corresponds to the median, the lower and upper hinges correspond to the first and third quartiles, and the whiskers extend no further than 1.5× interquartile range (IQR).
Figure 3—figure supplement 3. Spatial mapping of multicellular programs associated with myocardial remodeling.

Figure 3—figure supplement 3.

(A) Quantification of the relative area of the expression of cell-type-specific Factor 1 signatures separated by their gene weights in spatial transcriptomics data of human myocardial infarction for cardiomyocytes (CMs), endothelial cells (Endos), fibroblasts (Fibs), and myeloid cells. Adjusted p-values of Wilcoxon tests between conditions are shown for differences where the value was lower than 0.05 (n=28). Myogenic, n=14, fibrotic, n=5, ischemic, n=9. (B) Spatial mapping of the activation of CM, Fib, Endo, and myeloid cell-type Factor 1 signatures in representative examples of myogenic, ischemic, and fibrotic samples. The color of each spot represents the combination of the expression of myogenic and ischemic-like programs mapped to the red-green-blue color space (red = ischemic like, blue = myogenic like). Patient identifiers as in Kuppe et al., 2022a, are indicated in the slide header.