Multiomic analyses uncover immunological signatures in acute and chronic coronary syndromes

Abstract

Acute and chronic coronary syndromes (ACS and CCS) are leading causes of mortality. Inflammation is considered a key pathogenic driver of these diseases, but the underlying immune states and their clinical implications remain poorly understood. Multiomic factor analysis (MOFA) allows unsupervised data exploration across multiple data types, identifying major axes of variation and associating these with underlying molecular processes. We hypothesized that applying MOFA to multiomic data obtained from blood might uncover hidden sources of variance and provide pathophysiological insights linked to clinical needs. Here we compile a longitudinal multiomic dataset of the systemic immune landscape in both ACS and CCS (n = 62 patients in total, n = 15 women and n = 47 men) and validate this in an external cohort (n = 55 patients in total, n = 11 women and n = 44 men). MOFA reveals multicellular immune signatures characterized by distinct monocyte, natural killer and T cell substates and immune-communication pathways that explain a large proportion of inter-patient variance. We also identify specific factors that reflect disease state or associate with treatment outcome in ACS as measured using left ventricular ejection fraction. Hence, this study provides proof-of-concept evidence for the ability of MOFA to uncover multicellular immune programs in cardiovascular disease, opening new directions for mechanistic, biomarker and therapeutic studies.

Fig. 1. Study overview and patient characteristics.

a, Study design. In the Munich cohort, blood was analyzed from patients with ACS (total of n = 28; total of four timepoints, TP1M–TP4M), patients with CCS (n = 16) and patients with no CCS (n = 18; single timepoint, TP0M). A joint multiomic dataset was created from the Munich cohort by including clinical blood tests (cl), scRNA-seq (SC), flow cytometry, cytokine assay (cy), and plasma proteomics (p) and neutrophil (pmn) prime-seq. This was followed by data integration, MOFA model estimation (Y, input data matrices from each data modality; W, resulting weight matrix; Z, resulting matrix of factor values for each sample) and subsequent downstream analysis such as factor analysis, pathway enrichment, cell–cell communication and prediction. Findings from the Munich data cohort were evaluated in the Groningen data (V2 chemistry) as an independent validation cohort in which blood was analyzed longitudinally from patients with ACS (total of three timepoints TP1G–TP3G, total of n = 24 patients) as well as from a control group (TP0G, n = 31). Created with BioRender.com. b, X-ray images of a coronary catheterization of a patient with ACS: occlusion of the left circumflex artery, indicated by red arrow (left image); intervention, stent implantation (middle image); reperfusion (right image). c, Clinical blood tests. Individual timepoints for sterile ACS (TP1–TP4) compared with those for CCS (TP0M). Mean ± s.e.m. values are shown. d, UMAP plot of scRNA-seq data from PBMCs showing cell-type clusters used for subsequent analyses (n = 148,275). e, Analysis of CLR-transformed cell type abundance based on the scRNA-seq dataset. Data are shown using box–whisker plots (box, median and 25th to 75th percentile; whiskers, minimum to maximum). f, MOFA. Variance decomposition showing the percentage of explained variance per view and factor of the MOFA model with 20 factors. For each view, the heatmap shows the percentage of the variance that is explained by the respective factor. The color coding on the left indicates the data type of each view: green, plasma proteomics; blue, neutrophil prime-seq; orange, cytokine measurements; dark orange, clinical values; purple, scRNA-seq data. The greyscale grading in the heatmap depicts the percentage of variance. The bar plot (right) shows the total percentage of explained variance by all 20 factors. In c and e, parametric-distributed data were analyzed using ordinary one-way ANOVA with correction for multiple comparisons by Dunnett’s test; nonparametric-distributed data were analyzed using the Kruskal–Wallis test with correction for multiple comparisons by Dunn’s test. *P ≤ 0.05; **P ≤ 0.01; ***P ≤ 0.001. In cases in which only the ordinary one-way ANOVA or Kruskal–Wallis test, but not the multiple comparison, was significant, graphs are marked with a vertical bar on top. Exact P and n values are summarized in Supplementary Tables 13 and 14, respectively.

Fig. 2. Multivariate integration and factor analysis reveal comprehensive immune signatures that explain variance among patients in ACS.

a, Overview of IAI (Factor 2). For each view, the heatmap shows the percentage of the variance that is explained by the factor. The bar plots show the total amount of features (left) and the relative amount of features (right; in respect to the number of view-specific features) among the top 1% of the highest-ranking features that influence the factor. The color coding on the left indicates the data type of each view: green, plasma proteomics; blue, neutrophil prime-seq; orange, cytokine measurements; dark orange, clinical values; purple, scRNA-seq data. The greyscale grading in the heatmap depicts the percentage of variance. b, IAI (Factor 2). Comparison of the factor values for each timepoint for sterile ACS, non-CCS and CCS. Mean ± s.e.m. values are shown. c, Replication of IAI in the Groningen cohort. Comparison of the factor values for each timepoint for ACS with controls. Mean ± s.e.m. values are shown. d, IAI (Factor 2). Normalized expression values of the top 0.5% of features for cluster 0 CD4⁺ T cell for sterile ACS and CCS. A longitudinal comparison of the normalized expression values (heatmap) and the weight of the features (bar plot) are shown. Plus and minus signs indicate the direction of the feature factor weight. e, Longitudinal comparison of normalized gene expression values of selected features for sterile ACS and CCS. For the following comparisons, only the post hoc test was significant: HINT1 T_reg cells (cluster 11). Plus and minus signs indicate the direction of the feature factor weight. f, Phenotyping by flow cytometry of the effects of plasma obtained from ACS and CCS on monocytes isolated from healthy donors. Individual timepoints for sterile ACS are compared with those for CCS. Mean ± s.e.m. values are shown for mean fluorescence intensities (MFIs). No post hoc analysis was performed when the Kruskal–Wallis test was not significant. For b, c, e and f, parametric-distributed data were analyzed using ordinary one-way ANOVA with correction for multiple comparisons by Dunnett’s test; nonparametric-distributed data were analyzed using the Kruskal–Wallis test with correction for multiple comparisons by Dunn’s test. In the case where only the ordinary one-way ANOVA or Kruskal–Wallis test, but not the multiple comparison, was significant, graphs are marked with a vertical bar on top. *P ≤ 0.05; **P ≤ 0.01; ***P ≤ 0.001. Exact P and n values are summarized in Supplementary Tables 13 and 14, respectively.

Fig. 3. IAI is characterized by distinct interleukin signatures in monocytes and T cells.

a, Positively enriched REACTOME immune system pathways in IAI (Factor 2) across all data dimensions for which at least 50% of genes have been included within the feature set. False discovery rate-adjusted P < 0.05. Coverage indicates the percentage of genes of the pathway that have been included in the analysis. Greyscale depicting P values. b, Factor weights of features (top 25%) in IAI (Factor 2) belonging to enriched interleukin pathways averaged across all views in the upper part of the heatmap (‘Pathway’) and shown per view in the lower part of the heatmap (‘View’). Heatmap depicts factor values of pathway genes. c, Normalized expression values of genes and cytokines belonging to the interleukin-6 signaling pathway for sterile ACS (clusters 0, 2 and 5) and CCS. A longitudinal comparison for CD4⁺ T cell clusters (clusters 0, 2 and 5) and cytokine features is shown. In cases in which only the ordinary one-way ANOVA or Kruskal–Wallis test, but not the multiple comparison, was significant, graphs are marked with a vertical bar on top. Plus and minus signs indicate the direction of the feature factor weight. d, Normalized expression values of genes belonging to the interleukin-6 signaling pathway for sterile ACS and CCS. A longitudinal comparison for CD14^high monocyte genes (clusters 4 and 7) is shown (plus and minus signs indicate the direction of the feature factor weight). In c and d, parametric-distributed data were analyzed using ordinary one-way ANOVA with correction for multiple comparisons by Dunnett’s test; nonparametric-distributed data were analyzed using the Kruskal–Wallis test with correction for multiple comparisons by Dunn’s test. e, Efferocytosis (n = 7), survival (n = 8), ROS production (n = 12) and chemotaxis (n = 7) of monocytes activated ex vivo with and without IL6ST inhibition. Comparison between control and treatment with the IL6ST-inhibitor group (paired dataset). Parametric data were analyzed using paired t-test (two sided); nonparametric data were tested using the Wilcoxon test (two sided). *P ≤ 0.05; **P ≤ 0.01; ***P ≤ 0.001. Data points of paired data are connected by a line. Exact P and n values are summarized in Supplementary Tables 13 and 14, respectively.

Fig. 4. T-cell- and plasma-mediated changes in monocytes in ACS.

a,b, Spearman correlations (|cor| ≥ 0.4) between ligand and target genes across all samples (n = 128). Target genes are selected as belonging to the top 1% of features with positive (a) or negative (b) feature weight in IAI (Factor 2). Ligands were selected based on a minimum regulatory potential score of 0.0012 for the shown targets according to the NicheNet model (corresponding to the 97% quantile of the regulatory potential score). Interactions mentioned in the main text are highlighted with a darker color. c, Spearman correlation scores (Cor) of selected ligand–target pairs from a and b. d, A longitudinal comparison of normalized expression values of selected genes for sterile ACS and CCS. Parametric-distributed data were analyzed using ordinary one-way ANOVA with correction for multiple comparisons by Dunnett’s test; nonparametric-distributed data were analyzed using the Kruskal–Wallis test with correction for multiple comparisons by Dunn’s test. Plus and minus signs indicate the direction of the feature factor weight. e, Factor weights of the top 15 ligands with the highest factor weight in IAI (Factor 2). f, Monocyte phenotyping by flow cytometry after incubation with sterile ACS (TP4M) plasma with anti-IL-6 antibody or isotype control. All MFIs were normalized to the respective CCS plasma incubation average of the marker. Paired plasma incubation data with isotype (green dot) and IL-6 (red dot) inhibition are shown (TP4M n = 7). Parametric data were analyzed using the multiple paired t-test (two sided). *P ≤ 0.05; **P ≤ 0.01; ***P ≤ 0.001. Data points of paired data are connected by a line. Exact P and n values are summarized in Supplementary Tables 13 and 14, respectively.

Fig. 5. Factor analysis identifies distinct immune signatures in ACS subtypes.

a, IAI (Factor 2). Longitudinal comparison of factor values in ACS subtypes (sterile ACS, ACS with hospital-acquired infection, ACS with delayed recanalization after vessel occlusion). b, Comparison of CK, CK-MB and troponin at early timepoints (TP1 and TP2) of ACS subtypes. c, Comparison of CRP levels between patients with sterile ACS and patients with ACS with hospital-acquired infection. d, Comparison of good and poor outcomes during hospitalization using EF levels. e,f, Longitudinal comparison of IAR factor (Factor 4) values (e) and CK levels (f) between patients with good and poor outcomes. g, ROC AUC plot for prediction of good versus poor outcomes for IAR (Factor 4), GRACE score, normalized CK levels, normalized CRP levels and normalized troponin levels at TP1M. h, ROC AUC plot for validation of prediction of good versus poor outcomes in the Groningen cohort. A lasso model was trained on the top features of IAR at TP1M and applied to the Groningen cohort (TP1G good outcome and poor outcome). i, Overview of IAR (Factor 4): for each view, the heatmap shows the percentage of the variance that is explained by the factor. The bar plots show the total amount of features (left) and the relative amount of features (right, in respect to the number of view-specific features) among the top 1% of the highest-ranking features that influence the factor. The color coding on the left indicates the data type of each view: green, plasma proteomics; blue, neutrophil prime-seq; orange, cytokine measurements; dark orange, clinical values; purple, scRNA-seq data. The greyscale grading in the heatmap depicts the percentage of variance. j, Factor values of the top 1% of features in IAR (Factor 4), showing only features belonging to clinical values (indicated by an asterisk), cluster 3 NK cell, cytokines and plasma proteomics. k, Normalized expression values of selected top features from NK cells in the Munich and Groningen cohorts, comparing patients with good and poor outcomes at TP1. The minus sign indicates the direction of the feature factor weight. l, ROC AUC plot showing the prediction results of a logistic regression model using selected NK features (CD53, GZMB, TXNIP) trained on the Munich dataset and applied to the Groningen dataset. For a–c and f, the data were analyzed using mixed-effects analysis with correction for multiple comparisons by Tukey’s test (a and b) or Šidák’s test (c and f). For d, e and k, the parametric datasets were analyzed using an unpaired t-test (two sided). For a–f, mean ± s.e.m. values are shown. *P ≤ 0.05; **P ≤ 0.01. Exact P and n values are summarized in Supplementary Tables 13 and 14, respectively.

Fig. 6. Patients with CCS are characterized by high IC values.

a, IC (Factor 1). Comparison of factor values for patients with CCS, coronary sclerosis (non-CCS) and healthy coronaries (non-CCS). Mean ± s.e.m. values are shown. b, Overview of IC (Factor 1). For each view, the heatmap shows the percentage of the variance that is explained by the factor. The bar plots show the total amount of features (left) and relative amount of features (right; in respect to the number of view-specific features) among the top 1% of highest-ranking features for the factor. The color coding on the left indicates the data type of each view: green, plasma proteomics; blue, neutrophil prime-seq; orange, cytokine measurements; dark orange, clinical values; purple, scRNA-seq data. The greyscale grading in the heatmap depicts the percentage of variance. c, Factor values of the top 1% of features showing only features belonging to CD4⁺ T cells (clusters 0 and 2) in IC (Factor 1). d, Normalized expression values of selected top features of IC (Factor 1) for samples classified as CCS and non-CCS (plus or minus signs indicate the direction of the feature factor weight). e, Factor weights of the top 10 ligands with the highest factor weight in IC (Factor 1; plus or minus signs indicate the direction of the feature factor weight). f, Normalized expression values of selected top ligands of IC (Factor 1) for samples classified as CCS and non-CCS (plus or minus signs indicate the direction of the feature factor weight). For a, Parametric-distributed data were analyzed using ordinary one-way ANOVA with correction for multiple comparisons by Tukey’s test. For d and f, parametric data were analyzed using an unpaired t-test (two sided) and nonparametric data were tested using the Mann–Whitney test (two sided). *P ≤ 0.05; **P ≤ 0.01; ***P ≤ 0.001. Exact P and n values are summarized in Supplementary Tables 13 and 14, respectively.

Extended Data Fig. 1. Changes in clinical values and cell-type abundance across time.

(a) Clinical blood tests. Individual timepoints for sterile ACS (TP1-4M) compared to CCS (TP0M). CRP; Neutrophils. Parametric distributed data were analyzed using the Ordinary One-Way ANOVA with correction for multiple comparisons by Dunnett’s test; non-parametric distributed data were analyzed using the Kruskal-Wallis test with correction for multiple comparisons by Dunn’s test. *p≤0.05, **p≤0.01, ***p≤0.001. Mean values with +/- SEM are shown. (b) Analysis of centered log ratio (CLR) transformed cell type abundance of phenotypically defined immune cells to CD45+ leukocytes in PBMCs based on flow cytometry. Individual timepoints of sterile ACS compared to CCS. Parametric distributed data were analyzed using the ordinary one-way ANOVA with correction for multiple comparisons by Dunnett’s test; non-parametric distributed data were analyzed using the Kruskal-Wallis test with correction for multiple comparisons by Dunn’s test. *p≤0.05, **p≤0.01, ***p≤0.001. Mean values with +/- SEM are shown. In case only the ordinary one-way ANOVA or Kruskal–Wallis test was significant, graphs are marked with a vertical bar on top. (c) Analysis of centered log ratio (CLR) transformed cell type abundance based on scRNA-seq dataset. Individual timepoints of sterile ACS compared to patients with CCS. Parametric distributed data were analyzed using the Ordinary One-Way ANOVA with correction for multiple comparisons by Dunnett’s test; non-parametric distributed data were analyzed using the Kruskal-Wallis test with correction for multiple comparisons by Dunn’s test. For the following comparisons, only the post hoc was significant: CD4+ T cells – cluster 5. *p≤0.05). Data are shown as a Box-Whiskers plot (box: median, 25th to 75th percentile; whiskers: minimum to maximum). (a)-(c) Exact p-values (Supplementary Table 13) and n-numbers (Supplementary Table 14) were summarized in the supplementary tables.

Extended Data Fig. 2. Gene expression changes across time and disease states in IAI.

(a) Integrative ACS Ischemia (Factor 2). Normalized expression values of top 1% features for sterile ACS and CCS patients in longitudinal comparison visualized within heatmap and weight of the features visualized as barplot (+/- indicates the direction of the feature factor weight). N-numbers (Supplementary Table 14) were summarized in the supplementary tables.

Extended Data Fig. 3. In vitro ACS plasma induced changes in monocyte phenotype and function.

(a) Monocytes isolated from healthy donors incubated with patient plasma, and phenotyped by flow cytometry. Individual timepoints of sterile ACS compared to CCS. Parametric distributed data were analyzed using the ordinary one-way ANOVA with correction for multiple comparisons by Dunnett’s test; non-parametric distributed data were analyzed using the Kruskal–Wallis test with correction for multiple comparisons by Dunn’s test. *p≤0.05. In case only the ordinary one-way ANOVA or Kruskal–Wallis test was significant, graphs are marked with a vertical bar on top. Mean values with +/- SEM are shown. (b) Monocyte efferocytosis, ROS production and survival after plasma incubation. Individual timepoints of sterile ACS compared to CCS. Non-parametric distributed data were analyzed using the Kruskal-Wallis test with correction for multiple comparisons by Dunn’s test. *p≤0.05, **p≤0.01, ***p≤0.001. Mean values with +/- SEM are shown. (a) – (b) Exact p-values (Supplementary Table 13) and n-numbers (Supplementary Table 14) were summarized in the supplementary tables.

Extended Data Fig. 4. Analysis of medication mediated effects on IAI.

(a) Longitudinal comparison of the factor value of Integrative ACS Ischemia between patients with sterile ACS with and without pre-medication. ASA, ß-Blocker, ACE/AT1-inhibitor, Statin. The dataset was analyzed using the Mixed-effects analysis with correction for multiple comparisons by Šidák test. *p≤0.05. Mean values with +/- SEM are shown. (b) Integrative ACS Ischemia (Factor 2). Comparison of the factor values of each timepoint of sterile ACS (TP1-4M) and non-CCS (TP0M) with CCS (TP0M) patients. Only patients without pre-medication (upper part) of: ASA, ß-Blocker, ACE/AT1-inhibitor, Statin. Only patients with pre-medication (lower part) of: ASA, ß-Blocker, ACE/AT1-inhibitor, Statin. Parametric distributed data were analyzed using the Ordinary One-Way ANOVA with correction for multiple comparisons by Dunnett’s test; non-parametric distributed data were analyzed using the Kruskal-Wallis test with correction for multiple comparisons by Dunn’s test. *p≤0.05, **p≤0.01, ***p≤0.001. Mean values with +/- SEM are shown. (c) Comparison of the factor value (Integrative ACS Ischemia, Factor 2) in patients without ACS (pooled non-CCS and CCS) between with and without respective pre-medication: ASA, ß-Blocker, ACE/AT1-inhibitor, Statin. The parametric dataset was analyzed using an unpaired t-test (two-sided). *p≤0.05. Mean values with +/- SEM are shown. (d) Integrative ACS Ischemia (Factor 2). Comparison of the factor values of TP1-3M with TP4M of patients with sterile ACS. Only patients without pre-medication (upper part) of: ASA, ß-Blocker, ACE/AT1-inhibitor, Statin. Only patients with pre-medication (lower part) of: ASA, ß-Blocker, ACE/AT1-inhibitor, Statin. Parametric distributed data were analyzed using the Ordinary One-Way ANOVA with correction for multiple comparisons by Dunnett’s test; non-parametric distributed data were analyzed using the Kruskal-Wallis test with correction for multiple comparisons by Dunn’s test. *p≤0.05, **p≤0.01, ***p≤0.001. Mean values with +/- SEM are shown. (a) – (d) Exact p-values (Supplementary Table 13) and n-numbers (Supplementary Table 14) were summarized in the supplementary tables.

Extended Data Fig. 5. The effects of IL6ST inhibition on monocyte-driven cytokine secretion.

(a) Monocyte cytokine secretion upon inhibition of IL6 signaling analyzed by flow cytometry. Comparison between control and treatment with IL6ST-inhibitor group (paired dataset, n=8). Parametric data were analyzed using the paired t-test (two-sided), non-parametric data were tested using the Wilcoxon test (two-sided). *p≤0.05, **p≤0.01. Illustration of paired data by connecting the data points. Exact p-values were summarized in Supplementary Table 13.

Extended Data Fig. 6. Cytokine driven or further effects of ACS plasma on immune cells.

(a) Monocyte phenotyping after incubation with sterile ACS (TP1-3M) or CCS (TP0M) plasma upon IL6 inhibition by flow cytometry. All MFIs were normalized to the respective CCS plasma average of the marker. Comparison between isotype and IL6 inhibition (paired dataset). Parametric data were analyzed using the multiple paired t-test (two-sided). *p≤0.05, **p≤0.01, ***p≤0.001. Illustration of paired data by connecting the data points. (b) T cell phenotyping after incubation with sterile ACS (TP1-4M) or CCS (TP0M) plasma by flow cytometry. Individual timepoints of sterile ACS plasma (randomized n=8) compared to CCS, mean value of at least 5 PBMC donors incubated with each patient plasma. Parametric distributed data were analyzed using the Ordinary One-Way ANOVA with correction for multiple comparisons by Dunnett’s test; non-parametric distributed data were analyzed using the Kruskal-Wallis test with correction for multiple comparisons by Dunn’s test. In case only the ordinary one-way ANOVA or Kruskal–Wallis test was significant, graphs are marked with a vertical bar on top. For the following comparisons, only the post hoc was significant: T effector memory: CD4+ CCR7- CD45RO+. *p≤0.05, **p≤0.01. Mean values with +/- SEM are shown. (a) – (b) Exact p-values (Supplementary Table 13) and n-numbers (Supplementary Table 14) were summarized in the supplementary tables.

Extended Data Fig. 7. Lagged plasma–cell or cell–cell-mediated ligand–target interactions in IAI.

(a) Lagged spearman correlations (|cor| ≥0.4) between ligand and target genes across ACS samples (n=84). Ligand-target pairs have been lagged and mapped as follows before calculating the correlation: Ligand_TP1M ~ Target_TP2M; Ligand_TP2M ~ Target_TP3M; Ligand_TP3M ~ Target_TP4M [Target_TP1M and Ligand_TP4M values have not been mapped and included]. Shown target genes selected as top 2.5% of features with positive feature weight on Integrative ACS Ischemia. Shown ligands selected based on minimum regulatory potential score of 0.0012 for those targets according to the NicheNet Model (corresponding to 97% quantile of regulatory potential score). (b) Lagged spearman correlations (|cor| ≥0.4) between ligand and target genes across ACS samples (n=84). Ligand-target pairs have been lagged and mapped as follows before calculating the correlation: Ligand_TP1M ~ Target_TP2M; Ligand_TP2M ~ Target_TP3M; Ligand_TP3M ~ Target_TP4M [Target_TP1M and Ligand_TP4M values have not been mapped and included]. Shown ligand genes selected as top 2.5% of features with positive feature weight on Integrative ACS Ischemia. Shown targets selected based on a minimum regulatory potential score of 0.0012 of the selected ligands to the targets according to the NicheNet Model (corresponding to 97% quantile of regulatory potential score). (c) Exemplary visualization of lagged associations shown in circoplot in (b): Left: Violin plots showing normalized expression values of selected features for sterile ACS patients in longitudinal comparison. Timepoints were mapped as visualized by the violin plots: Ligand_TP1M ~ Target_TP2M; Ligand_TP2M ~ Target_TP3M; Ligand_TP3M ~ Target_TP4M. [Target_TP1M and Ligand_TP4M values have not been mapped and included as no corresponding ligand/target value is available, indicated by gray color]. Right: Scatter plots showing spearman correlation score of the visualized example based on mapping target molecule expression (CRP) to lagged ligand expression (IL6) as indicated by the violin plots. (d) Normalized expression values of cytokines are shown in comparison of individual timepoints of sterile ACS with CCS. Parametric distributed data were analyzed using the Ordinary One-Way ANOVA with correction for multiple comparisons by Dunnett’s test; non-parametric distributed data were analyzed using the Kruskal-Wallis test with correction for multiple comparisons by Dunn’s test. *p≤0.05, **p≤0.01, ***p≤0.001. In case only the ordinary one-way ANOVA or Kruskal–Wallis was significant, graphs are marked with a vertical bar on top. (+/- indicates the direction of the feature factor weight). Exact p-values (Supplementary Table 13) were summarized in the supplementary tables. (c) – (d) N-numbers (Supplementary Table 14) were summarized in the supplementary tables.

Extended Data Fig. 8. Gene expression changes across time and outcome in IAR.

(a) Integrative ACS Repair (Factor 4). Normalized expression values of top 1% features in longitudinal comparison visualized within heatmap and weight of the features visualized as barplot, all samples included (n=128). Divided by outcome; ‘NA’ in case no EF value has been available for the ACS samples (n=7) and for CCS and non-CCS samples. (+/- indicates the direction of the feature factor weight).

Extended Data Fig. 9. Gene expression changes across time and disease states in IC.

(a) Integrative CCS (Factor 1). Normalized expression values of top 1% features in longitudinal comparison visualized within heatmap and weight of the features visualized as barplot, all samples included (n=128). (+/- indicates the direction of the feature factor weight).

Extended Data Fig. 10. Ligand–target interactions in IC.

(a) Spearman correlations (|cor| ≥0.4) between ligand and target genes across all samples (n=128). Target genes selected as top 1% of features with negative feature weight on Integrative CCS (Factor 1). Ligands selected based on minimum regulatory potential score of 0.0012 on those targets according to the NicheNet Model (corresponding to 97% quantile of regulatory potential score). Interactions mentioned in the main text highlighted by darker color. (b) Spearman correlations (|cor| ≥0.4) between ligand and target genes across all samples (n=128). Target genes selected as top 1% of features with positive feature weight on Integrative CCS (Factor 1). Ligands selected based on minimum regulatory potential score of 0.0012 on those targets according to the NicheNet Model (corresponding to 97% quantile of regulatory potential score). Interactions mentioned in the main text highlighted by darker color.