Single Cell Analysis of Human Heart Failure with Preserved Ejection Fraction

Abstract

Background:

Heart failure with preserved ejection fraction (HFpEF) is a poorly understood, multi-system disease with high morbidity and mortality. To improve understanding of its pathobiology, we analyzed single-nucleus RNA sequencing (snRNA-seq) in human HFpEF myocardium versus controls.

Methods:

Septal myocardial biopsies from 19 HFpEF and 24 non-failing controls were analyzed using the 10X Genomics Chromium platform, with nuclei isolated from combined samples (6 patients/pool). Genotype-based demultiplexing was performed with souporcell, and gene expression quantified with CellRanger and CellBender. After quality control, nuclei were annotated by cell types and differential expression (DE) was performed between HFpEF vs controls using limma-voom. Functional analysis was performed using Gene Set Enrichment Analysis. Data were compared to prior snRNA-seq in dilated cardiomyopathy (DCM) versus controls.

Results:

We successfully demultiplexed pooled myocardial biopsies, assigning >70% of nuclei to individuals. After quality control, we recovered 48,886 nuclei and identified 14 cell types. Many differentially expressed (DE) genes across cell types were detected in HFpEF versus controls (Fibroblasts, 5905; cardiomyocytes, 5159; endothelial cells, 2143; pericytes, 1812; macrophages, 1405). Enriched pathways common to multiple cell types included immune activation, transcription/translation, metabolism, and protein quality control. They were particularly shared between cardiomyocytes and fibroblasts. VSMCs had a more synthetic, proliferative phenotype. Immune cell analyses suggested enhanced T cell activation and reduced macrophage clearance programs. Comparative analysis between HFpEF and DCM identified transcriptional differences primarily in cardiomyocytes. Two of three cardiomyocyte DE genes unique to HFpEF were validated to have concordant protein expression changes in HFpEF (MAP2K6 and PLPP3).

Conclusions:

Our findings reveal a distinct, cell-type-specific transcriptomic landscape in the human HFpEF myocardium. While HFpEF and DCM share significant molecular pathways across most cell types, the profound divergence within cardiomyocytes suggests a unique pathological driver for HFpEF. These signatures may provide a high-resolution roadmap for identifying precision therapeutic targets in HFpEF.

BACKGROUND

Heart failure affects nearly 7 million adults in the United States alone imposing over $30 billion in health-care costs per year. ¹ Worldwide, the incidence approaches 65 million. Approximately half of affected patients have heart failure with preserved ejection fraction (HFpEF), a syndrome where resting systolic function appears normal, yet patients exhibit similar symptoms and outcomes as those with reduced systolic function. ² While long viewed as a disorder of diastolic function, HFpEF has become dominated by severe obesity and cardiovascular-kidney-metabolic syndrome, impacting multiple organs and cell types. Current HFpEF animal models often incorporate a metabolic and hemodynamic stress (e.g., two-hit), yet still fail to fully recapitulate the complex, multi-system pathophysiology and severity of human HFpEF disease. Critically, there are limited molecular data from human myocardium to serve as a benchmark for animal models or identify high-value therapeutic targets, underscoring the need for human tissue analyses.

Our laboratory has previously reported human HFpEF myocardial transcriptomics by bulk RNA-seq³, metabolomics⁴, proteomics⁵, integrated analysis of glycolysis⁶, ultrastructural analysis⁷, and myocyte sarcomere function⁸. Collectively, they revealed signatures for depressed fat and glucose metabolism, reduced protein translation and quality control, lipid accumulation, and immune activation. An unresolved question is how these abnormalities segregate among different myocardial cell types based on single nucleus RNA sequencing (snRNA-seq), analysis that has been applied to other cardiac diseases^9–13. snRNA-seq is invaluable for identifying key signatures pertinent to cardiomyocytes as well as less abundant and/or transcriptionally less active cell-types such as immune and vascular cells. This is important as systemic markers of inflammation/fibrosis may not reflect cellular changes in the myocardium. Prior snRNA-seq studies of human dilated cardiomyopathy (DCM) and hypertrophic cardiomyopathy (HCM) report a canonical activated fibroblast population⁹, and altered immune cell subclusters.^9,10 If this pertains to human HFpEF is currently unknown.

This study used snRNA-seq to define cell-type-specific signatures in human HFpEF myocardium. Endomyocardial biopsies from prospectively well-phenotyped HFpEF patients were compared to myocardium obtained from non-failing (NF) donor controls (NF-control) using differential gene expression with functional annotation and sub-cluster analysis. Comparisons were also made to a prior snRNA-seq analysis of human DCM ⁹ to test if cell-type-specific transcriptional profiles are shared by DCM and HFpEF.

METHODS

Data Availability.

The final processed dataset is available in the Single Cell Portal at the Broad Institute under project ID SCP3342 (https://singlecell.broadinstitute.org/single_cell/study/SCP3342).

Study Population and Study Design.

This was a cross-sectional study using myocardial tissue samples from two biobanks (HFpEF tissue from Johns Hopkins University and non-failing donor control tissue from the Perelman School of Medicine at the University of Pennsylvania). Patients presenting to the JHU HFpEF clinic were referred for clinically indicated right heart catheterization with research endomyocardial biopsy. HFpEF diagnosis was based on established consensus criteria², including right heart catheterization demonstrated elevation of pulmonary capillary wedge pressure at rest or with supine exercise, as described previously³. Detailed inclusion and exclusion criteria are in the Supplement. Ventricular septal endomyocardial biopsies and location-matched non-failing (NF) control tissue from organ donor subjects were obtained as described¹⁴, frozen in liquid nitrogen, and stored in a −160° freezer. The study was approved by the Johns Hopkins University Institutional Review Board and the Perelman School of Medicine Institutional Review Board. All participants gave written informed consent and the study procedures were in accordance with institutional guidelines. Demographics and clinical characteristics between HFpEF and controls were compared by Mann-Whitney tests (continuous) or Fisher’s exact tests (categorical) using SciPy v1.13.1. P value <0.05 was considered significant.

Nuclei Isolation and Library preparation for single-nucleus RNA-sequencing.

Nuclei were isolated using previously published protocols^9,15 with some modifications. Due to small biopsy size (2–3 mg), six biopsies were pooled prior to nuclei isolation. Control samples were cut to match the size of the HFpEF biopsies so the proportion of control and HFpEF nuclei would be similar in each pool. The 6 samples (3 controls, 3 HFpEF) were embedded together in OCT and sectioned prior to nuclei isolation as described in the Supplement. Nuclei were loaded into a 10x Genomics Chip G targeting 3000–10,000 nuclei per pool and the 10x Genomics Chromium 3’ v 3.1 kit was carried out with slight modifications (Supplemental Methods). This approach relies on encapsulating single nuclei into microfluidic droplets containing uniquely barcoded beads, called Gel Beads-in-emulsions (GEMs), and subsequently referred to as “droplets.” Library quality control (QC) and quantification were carried out using qPCR and Fragment analyzer, followed by sequencing on NovaSeq 6000, targeting ~1 billion reads per library.

Single-nucleus RNA-sequencing data processing, demultiplexing, and QC.

snRNA-seq libraries were processed using CellRanger v4.0.0¹⁶ with a pre-mRNA version of the GRCh38–2020-A transcriptome reference provided by 10X Genomics (Supplemental Methods). The resulting count matrix was processed with CellBender remove-background v0.2.0¹⁷ to identify non-empty droplets and remove ambient RNA, which is often enriched for cytoplasmic RNA, and chimeric library fragments. Three libraries (Pools 1, 6, and 6b) were considered experimental failures and excluded from further analysis due to some combination of low valid barcodes, low rates of mapping confidently to the genome/transcriptome, low fraction of reads in CellRanger determined cells, and/or a low number of total genes detected (Supplemental Table 1, Supplemental Figure 1). We assigned nuclei to their patient of origin by demultiplexing pooled libraries using souporcell v2.0¹⁸ and aligning genotype signatures with reference genotypes derived from bulk RNA-seq³ (Supplemental Methods). Next, we jointly aggregated 87,430 souporcell singlets to perform extensive nuclei QC as previously reported^9,15 and detailed in the Supplement. All single-cell analysis was performed using scanpy v1.9.1 unless otherwise stated. ¹⁹ We corrected for a batch effect for the patient of origin for each nucleus using Harmony v0.1.7 with default settings.²⁰ To identify low quality nuclei, the following per-nucleus QC metrics were considered: 1) total number of unique molecular identifiers (UMIs), 2) total number of unique genes detected, 3) percent of UMI mapping to mitochondrial genes, 4) proportion of reads mapping exclusively to exons, 5) transcriptional entropy as estimated using the ndd library in Python (https://github.com/simomarsili/ndd/tree/master), and 6) doublet score as estimated in Scrublet v0.2.3²¹. We tested for differences in cellular composition between HFpEF and controls using scCODA v0.1.9 (Supplemental Methods).

Differential expression between HFpEF and control.

Differential gene expression between HFpEF and control samples was performed across all cell types (pseudo-bulk) and by cell type, similarly to Chaffin et al., 2022⁹ and Simonson et al., 2023¹⁵ and as detailed in the Supplement. The limma-voom pseudo-bulk approach was used for cell-specific differential expression testing. Gene counts from all nuclei of a given cell type were summed within each patient to account for the correlation of nuclei from the same patient and differences in the number of nuclei per patient. Therefore, each individual had one value for each gene + cell-type combination. This approach controls for differences in the number of nuclei per sample and reduces false positives from pseudo-replication bias when treating each nucleus from an individual as a separate value. ^22–24 Differential expression models were adjusted for sex and pool to control for batch effects. Multiple testing correction was performed within each cell type using Benjamini-Hochberg at false discovery rate (FDR)=0.05. Genes with a directionally concordant and significant (adjusted p-value < 0.05) association with both CellBender and CellRanger expression matrices, along with a background contamination heuristic ≤ 0.40 (see Supplemental Methods), were considered significant. Estimates of logFC were compared to prior bulk RNA-seq in controls and HFpEF from the same biobanks, with overlapping samples in 19 HFpEF and 20 controls. ³ Covariate adjustment was performed as a sensitivity analysis. Functional annotation was performed with Gene Set Enrichment Analysis (GSEA) of the Reactome database using clusterProfiler 3.18.1, ReactomePA 1.34.0, and a Jaccard similarity index to identify redundant pathways as described in the Supplemental Methods. Comparisons of the bulk RNA-seq and single nuclear RNA-seq and gene counts by race were performed in R version 4.0.3.

Sub-clustering of major cell types.

Sub-clustering of 6 major cell types (cardiomyocytes, fibroblasts, endothelial cells, mural cells, macrophages, and lymphocytes) was performed using single-cell Variational Inference (scVI) 0.17.3²⁵ (Supplemental Methods). Marker genes for each sub-cluster were calculated as for the global map. Representative markers were selected as genes expressed in at least 15% of nuclei from the sub-cluster, with a logFC > 0.5 and FDR-adjusted p-value < 0.05 compared to all other sub-clusters of the cell type. To compare sub-clusters to previously reported populations, ^9–11,26,27 we scored the transcriptional signatures from these previously published studies in our nuclei (Supplemental Methods).

DCM and HFpEF combined analysis.

snRNA-seq data from the 48,866 nuclei generated in this study were combined with 343,887 nuclei from 11 DCM and 15 controls previously reported, ⁹ mitigating batch effects between the studies (Supplemental Methods). Differential expression analysis by cell type was performed by identifying genes in three categories, adjusted for sex: 1) differentially expressed between HFpEF (n_max=19) and their respective controls (n_max=24), 2) differentially expressed between DCM (n_max=11) and their respective controls (n_max=15), and 3) genes with a different effect in HFpEF versus control than DCM versus control. To achieve this, we used a similar pseudo-bulk testing approach as in the HFpEF versus control comparison. Three comparisons of interest were obtained: HFpEF – Control_HFpEF, 2) DCM – Control_DCM, and 3) (HFpEF – Control_HFpEF) – (DCM – Control_DCM). We adjusted for multiple testing correction using the Benjamini-Hochberg correction at FDR=0.05. Differentially expressed genes unique to HFpEF were identified by being both different to controls and either unaltered or changed in the opposite direction in DCM. LogFC estimates from the combined DCM vs control (LV) and HFpEF vs control (RVS) snRNA-seq analysis were compared to the prior bulk RNA-seq data³ that had directly compared DCM vs HFpEF.

Protein Expression Analysis.

Human heart tissue samples underwent protein isolation and quantification of protein expression using Western blotting as detailed in the Supplemental Methods. After testing for normality using a Shapiro-Wilk test, the three disease groups were compared using ordinary one-way analysis of variance and Dunnett’s multiple comparisons test in GraphPad Prism, v10. P value <0.05 was considered significant.

RESULTS

Study population

Thirty HFpEF and 29 NF-control biopsies were processed and n=19 and n=24 respectively remained in the final analysis. Eleven samples were excluded due to technical issues affecting two pooled samples (6 HFpEF, 5 NF-controls) or unsuccessful genotype-based demultiplexing (5 HFpEF). Clinical characteristics of the final study populations are provided in Supplemental Table 2. Compared to NF-controls, HFpEF were older (median age 65 vs 58.5 years, p=0.036), predominantly Black or African-American (84.2% vs 0%, p=3.7e-9), and had a higher prevalence of hypertension (100% vs 62.5%, p=0.0025) and diabetes (78.9% vs 12.5%, p=2.4e-5). HFpEF had a higher body mass index (BMI, 43.2 vs 27.8kg/m², p = 2.5e-7) and lower estimated glomerular filtration rate (eGFR 35 vs 87mL/min/1.73 m², p=0.009) vs NF-controls. Resting invasive hemodynamics (Supplemental Table 2) were similar to those reported in a similar cohort.¹⁴ Additional clinical characteristics for controls are provided in Supplemental Table 3. Controls had some comorbidities: 29% obese, 33% hyperlipidemia, 12.5% with diabetes, 62.5% with hypertension. Only 3 had atrial fibrillation, 2 coronary artery disease, and none had ventricular arrhythmias, pacemakers, defibrillators, coronary artery bypass or valvular surgery.

Single nucleus UMAP generation demonstrates feasibility of genotype-based demultiplexing

We began by confirming the feasibility of pooling biopsy samples and demultiplexing using genetic signatures from the snRNA-seq data (Supplemental Figure 2). Three DCM and three non-failing controls were used for the feasibility study (Supplemental Methods). There was consistency in cell types identified and differential expression in the pooled vs individually run samples. ⁹ Cell composition was similar other than an increase in cardiomyocytes with the pooled approach, possibly due to a higher rate of demultiplexing due to transcriptional complexity. After observing reliable data in our initial test, we next generated snRNA-seq for all HFpEF/NF-control pools. Two pools were excluded based on QC (Supplemental Figure 1) and a total of 87,430 droplets (73.3% of total non-empty droplets) were successfully demultiplexed/assigned to an individual (Supplemental Table 4, Supplemental Figure 3). There was a high concordance between known sex and inferred sex based on gene expression for patients in each pool (Supplemental Figure 4). Extensive QC was performed to remove multiplets (droplets with two or more nuclei), cytoplasmic contamination, and technical failures (Supplemental Methods, Supplemental Figures 5–6, Supplemental Table 5).

After QC, 48,866 nuclei remained from 19 HFpEF and 24 controls. Marker genes for each nuclei cluster (Supplemental Table 6, Supplemental Figure 7) were used to annotate cell-types in the final UMAP (Figure 1a). Two distinct endothelial cell clusters were identified, labeled endothelial1 (EC1, the more abundant cluster) and endothelial2 (EC2). The most abundant cell types were cardiomyocytes, representing on average 26.6% of nuclei in each patient, followed by fibroblasts, EC1, pericytes, and macrophages (24.1%, 16.7%, 10.4%, and 5.5% of nuclei, respectively (Figure 1b). While proportions of cell types varied among individuals (Supplemental Figure 8), similar composition was found in HFpEF and controls for all but endocardial cells (Supplemental Table 7, Figure 1b).

Figure 1. Aggregation of 48,866 nuclei in a comprehensive single-nucleus map.

a) UMAP embedding of 48,866 nuclei from 8 pools across 43 patients, colored by cell type. b) Box plots demonstrating compositional differences between HFpEF and controls. Cell types are colored as in panel a and inset depicts the composition of rarer cell types for improved viewability. Significant compositional shifts (scCODA) are denoted with an *. Center line, median; Box limits, upper and lower quartiles; Whiskers, 1.5x interquartile range.

Differential gene expression (DE), functional annotation, and subclustering analysis in HFpEF vs NF-controls identifies cell-type-specific subclusters and enriched pathways

Nine of the 14 identified cell-types had sufficient nuclei in both HFpEF and controls to perform DE analysis (at least 20 nuclei per cell type in at least 3 individuals from both HFpEF and NF-controls). Patient level principal component analysis (PCA) of the top 500 most highly variable genes meeting expression thresholds (as described in the Supplemental Methods) demonstrated separation of HFpEF vs NF-controls for each cell type (Supplemental Figure 9). The Supplemental Materials detail the number of DE genes for each cell type (Supplemental Figure 10), full DE results (Supplemental Table 8), gene set enrichment analysis (GSEA) of Reactome pathways by cell type (Supplemental Table 9), and sub-cluster markers (Supplemental Table 10).

Cardiomyocytes

The top down-regulated protein-coding genes (based on lowest P-value, Figure 2a; values from CellBender log2 fold change [logFC], adjusted P-value, respectively) were related to calcium handling and arrhythmogenesis (FKBP5 (−2.322, 3.52e-12)), mitochondrial fission and function (MTFR1 (−0.872, 3.43e-10)), redox homeostasis (HMOX2 (−1.047, 9.16e-11), NMRAL1 (−0.839, 2.44e-10)), hypoxia signaling (EGLN1 (−1.043, 9.16e-11)) and inflammatory lipid metabolism (PLPP3 (−1.675, 3.71e-11)). The top up-regulated protein-coding genes were related to membrane repair and regulation of inflammation (MYOF (1.746, 1.10e-12)), progesterone and glucocorticoid signaling (PGR (1.122, 6.38e-10), HSD11B1 (1.122, 1.86e-09)), microtubule-based transport and organization (MID1 (3.439, 1.47e-09), EFCAB11 (1.113, 7.65e-10)), zinc-finger transcription factors (ZMAT1 (0.965, 8.93e-11), ZNF20 (0.873, 1.28e-09), ZFP2 (0.770, 2.18e-09)), and sodium/calcium handling (EFCAB11 (1.113, 7.65e-10), SCN2B (1.705, 2.60e-09)). GSEA identified broad categories of downregulated pathways related to cellular signaling, metabolism, protein quality control, and transcription/translation, and upregulated pathways related to calcium/sodium handling, microtubules (Cilium Assembly) and extracellular matrix remodeling (ECM, Figure 2b,c, Supplemental Table 9). Subcluster analysis did not identify unique cell states, instead likely reflecting overall changes in cardiomyocyte gene expression (Supplemental Figure 11).

Figure 2. Cardiomyocyte analysis.

a) Volcano plot for differential expression (DE) between HFpEF and NF-control in cardiomyocytes. The x-axis represents log2 fold-change (logFC) estimates of HFpEF versus NF-control; y-axis the -log10(raw P-value), based on CellBender. Red dots identify DE genes with low background probability heuristic (<0.40), FDR<0.05 and concordant effects size assayed by CellBender and CellRanger. The top 10 up- and down-regulated protein coding genes are labeled. b) Dot plots for Reactome pathways significantly enriched in HFpEF vs controls in cardiomyocytes. Pathways were curated using a Jaccard similarity index to identify redundant pathways and pathway names were shortened for visibility. Shading represents the normalized enrichment score (NES) and size represents the -log10(raw P-value). All pathways shown have FDR <0.05. c) Boxplots showing the log2 counts per million (log₂(CPM)) across control and HFpEF for the top significantly DE genes in selected pathways from panel b.

Fibroblasts

Top upregulated genes in fibroblasts (Figure 3a) were related to cellular signaling (NR3C1 (1.066, 5.26e-15), LDB2 (2.713, 6.48e-15)), structural/ECM remodeling (TLL2 (3.959, 1.73e-17), FREM1 (2.395, 1.35e-14), CYS1 (2.977, 3.64e-16)), and ribosomal biogenesis and nucleotide metabolism (NMD3 (1.557, 2.42e-15), PRTFDC1 (1.749, 1.74e-14)). Top down-regulated genes were involved in inflammation and cellular stress regulation (FKBP5 (−3.621, 2.80e-18)), epigenetic control (JADE1 (−1.875, 2.80e-18)), redox homeostasis (MGST1 (−2.184, 7.12e-16), AOX1 (−2.930, 7.41e-18)), ECM and mechanotransduction (ITGA5 (−2.055, 2.02e-16), NID1 (−1.715, 9.15e-17), TGFBR3 (−1.802, 2.69e-15)), and cytoskeletal organization/cellular motility (WASF3 (−1.710, 1.09e-17)). GSEA predicted upregulation of pathways related to NO and GPCR signaling, ion channels, and platelet homeostasis; downregulated pathways were related to cellular signaling (stress, inflammation, hypoxia, mTOR), transcription/translation, autophagy, metabolism, and ECM remodeling (Figure 3b). Subcluster analysis identified a canonical activated fibroblast population, but this was only identified in a subset of NF-controls, not HFpEF (Figure 3c-3e). The shift in FB-0 and FB-1 were likely related to global changes in gene expression. Though the canonical activated fibroblast population was not present, there were several other genes changes that support ECM remodeling in HFpEF (FAP (1.706, 1.59e-04), ACTA2 (1.104, 3.03e-06), LUM (2.141, 1.24e-05), TGFBI (1.163, 2.02e-02), CXCL12 (2.132, 9.96e-08), and several COL genes).

Figure 3. Fibroblast analysis.

a) Volcano plot of differential gene expression between HFpEF and NF-control fibroblasts; format as in Figure 2a. b) Dot plots for enriched Reactome pathways in fibroblasts; format as in Figure 2b. c) Re-estimated UMAP embedding and clustering of 12,562 fibroblasts based on scVI batch-corrected latent space generated by restricting to fibroblast nuclei, colored by Leiden sub-clusters (left) and disease status (right). d) Box plots showing the proportion of each fibroblast sub-cluster among all fibroblasts in patients, separated by HFpEF and control. A red bar indicates a significant change in composition as detected in scCODA at false discovery rate of 0.05. Center line, median; Box limits, upper and lower quartiles; Whiskers, 1.5x interquartile range. e) Dot plot for expression of the top marker genes in each fibroblast sub-cluster. Dot size reflects the percent of nuclei expressing the gene at non-zero levels (%Expr>0) and color the average log-normalized expression scaled to the maximum value across sub-clusters for each gene (Avg norm expr).

Pericytes and Vascular Smooth Muscle Cells (VSMCs)

Pericytes displayed findings similar to fibroblasts (based on GSEA), with additional DE genes related to vascular stability (ANTXR1 (0.793, 1.29e-10), ITIH5 (1.744, 3.11e-10), NREP (1.524, 1.23e-10), PLA2G4C (1.181, 5.68e-10), DGKB (1.233, 1.35e-09), SPARCL1 (−1.029, 3.22e-12), LPP (−0.925, 1.35e-11), CDC42EP4 (−1.746, 1.17e-14)) and neurovascular signaling (NTF3 (2.018, 6.34e-10), PITPNM2 (1.085, 2.22e-10)) (Supplemental Figure 12a,c).

Top down-regulated genes/pathways in VSMCs were related to contractile function and cytoskeletal organization (RBPMS (−1.563, 3.90e-03), TMOD1 (−1.885, 4.37e-02), LPP (−1.700, 6.21e-03)) and Leptin-JAK/STAT signaling, while the top up-regulated genes reflected stress-responsive cellular signaling (MEF2C (1.676, 3.58e-03), DENND2A (2.210, 3.30e-04), GUCY1A1 (1.576, 6.52e-03)), cytoskeletal remodeling (SSH2 (1.472, 6.52e-03), SEMA5A (1.657, 6.94e-03), SEMA5B (4.094, 9.89e-03)), and calcium handling (CASQ2 (1.502, 1.03e-02)) consistent with VSMC phenotypic modulation and ECM remodeling (Supplemental Figure 12c). Together these changes suggest VSMCs in HFpEF may display a more synthetic/proliferative phenotype.

Immune cells

HFpEF lymphocytes (Supplemental Figure 13) displayed up-regulation of genes related to T-cell activation (NFATC2 (0.899, 1.24e-03)) and T cell receptor/NF-kB signaling (ZAP70 (1.079, 1.65e-02), CARD11 (0.910, 1.96e-03), PRKCB (0.500, 3.05e-02)), cellular adhesion/migration (ITGAL (1.164, 4.65e-04), CX3CR1 (2.471, 1.93e-04), MYO1F (0.894, 2.36e-04), STK10 (0.909, 4.87e-04)), and cytotoxic activity (NKG7 (1.676, 4.48e-02), GNLY (1.552, 5.51e-03)). In contrast, genes related to T cell quiescence (FOXO1 (−1.257, 3.22e-05), IL7R (−2.304, 6.70e-03)) and inhibition (PIK3IP1 (−1.842, 1.46e-06), SLA (−1.119, 7.62e-05), RASA1 (−1.029, 1.18e-04), CBLB (−1.019, 9.35e-04), SLAMF6 (−1.187, 7.62e-05)) were downregulated. There were two upregulated inhibitory receptors (CD300A (2.807, 1.43e-02), CD244 (2.229, 6.86e-03)) that might reflect compensation in the setting of chronic activation. Subclustering analysis revealed a decrease in one population of resting T-cells (LC-0, top marker genes include IL7R, RCAN3, CAMK4, LEF1, NELL2, ADAM19, INPP4B, ST8SIA1). Overall, the data indicate a shift toward effector T cells over quiescent T cells in HFpEF.

Top upregulated genes in HFpEF macrophages (Figure 4a) suggest a shift toward antigen presentation (ITPR2 (1.093, 1.29e-13), BLNK (1.354, 5.50e-10), TRAF3IP3 (1.836, 7.84e-10)), macrophage recruitment and survival (CX3CR1 (3.060, 7.46e-11), CSF1R (0.983, 7.46e-11)), and cellular remodeling pathways (P3H2 (1.856, 2.53e-11), PLXDC1 (1.899, 1.62e-10), RUBCNL (1.446, 2.52e-10), CPED1 (1.257, 3.09e-10), WWP1 (1.345, 3.27e-10), RTN1 (1.087, 1.62e-09)). Top downregulated genes involved macrophage clearance programs (CD163 (−2.392, 2.11e-20), MERTK (−1.852, 5.35e-15), VSIG4 (−1.551, 2.63e-12)) and regulation of TLR signaling (IRAK3 (−1.519, 4.67e-15), TLR2 (−1.322, 2.42e-14)). GSEA supported this shift away from innate immune system functions towards antigen presentation/immune checkpoint modulation (Figure 4b,c). PD-L1 (PDCD1LG2 (1.419, 5.28e-05)) and multiple HLA class II molecules were upregulated, while several phagocytic genes were downregulated (MARCO (−2.539, 8.71e-04), ANPEP (−1.979, 4.75e-05), CTSD (−0.675, 3.99e-02)). Subclustering analysis showed an increase in monocytes (MP-4, top marker genes CROCC2, TCF7L2, ADGRE2, RIPOR2, NEURL1 along with nonclassical monocyte marker FCGR3A), while the shift in resident macrophage subclusters MP-0 and MP-2 likely reflected overall differential gene expression in HFpEF vs NF-controls (Figure 4d-f). These data support a shift away from tissue-resident clearance toward antigen-presenting phenotypes. Of note, the adaptive immune system pathway was enriched in cardiomyocytes and macrophages, yet most core enrichment genes in cardiomyocytes were down-regulated and distinct from core enrichment genes in macrophages that were up-regulated (Supplemental Figure 14). This highlights the utility of single cell analysis to resolve discordant gene expression changes in distinct cell types.

Figure 4. Macrophage analysis.

a) Volcano plot for differential expression (DE) between HFpEF and control in macrophages; format as in Figure 2a. b) Dot plots for enriched Reactome pathways in macrophages; format as in Figure 2b. c) Boxplots showing the log2 counts per million (log₂(CPM)) across control and HFpEF samples for the top DE genes driving the pathway enrichment for the Adaptive Immune System. d) Re-estimated UMAP embedding and clustering of 3,050 macrophages based on scVI batch-corrected latent space generated after restricting to macrophage nuclei; format as in Figure 3c. e) Box plots showing the proportion of each macrophage sub-cluster in NF-controls and HFpEF; format as in Figure 3d. Samples with <10 macrophage nuclei are excluded from the composition analysis. f) Dot plot of top marker genes in each macrophage sub-cluster; format as in Figure 3e.

Endothelial cells

There were three distinct endothelial cell types (Figure 5a); endothelial1 (EC1), the dominant cluster; endothelial2 (EC2), a smaller cluster; and endocardial cells (EndoC). Top HFpEF upregulated genes in EC1 included NRP2 (sprouting angiogenesis and lymphangiogenesis, co-factor for VEGFs, (1.607, 3.20e-13)), EEPD1 (DNA repair in the setting of replication stress, (1.142, 1.14e-12)), and genes related to scaffolding, junctional adhesion, and cytoskeletal remodeling (SNTB1 (1.690, 2.19e-12), AFAP1L1 (1.254, 3.36e-12), JAM2 (1.292, 1.23e-11), RCSD1 (1.212, 4.20e-11)) (Figure 5b). Top HFpEF down-regulated genes in EC1 included RPGR (actin dynamics, mechano-transduction, (−1.625, 8.62e-15)), ST6GALNAC3 (glycocalyx sialylation that can affect cellular adhesion and barrier function, (−1.120, 6.48e-14)), hemostasis-related genes (VWF (−1.061, 1.26e-13) and F8 (−1.300, 3.20e-13)), and SLCO4A1 (cellular transport/uptake, (−3.119, 1.26e-13)). GSEA predicted downregulation of several pathways related to platelet activation, innate immune signaling, oxidative stress, translation initiation, cellular transport, and hypoxia signaling (Figure 5e). The smaller EC2 cluster had HFpEF upregulated genes EEPD1 (1.800, 1.78e-02) and others related to adhesion and barrier function (PPP1R16B (1.846, 1.08e-02), ITGA6 (1.216, 1.78e-02), ST8SIA6 (1.171, 2.18e-02)), cellular signaling (CTDSPL (1.546, 2.70e-02), EBF3 (1.146, 3.86e-02)), leptin (LEPR (1.406, 3.58e-02)), and interferon (STAT1 (1.548, 4.75e-02)) (Figure 5c). Downregulated genes included BMP6 (−2.503, 1.78e-02) and TEAD4 (promotes angiogenesis, −1.886, 2.18e-02) and others related to cellular adhesion (PCDH1 (−2.385, 1.82e-02), CD9 (−1.893, 2.23e-02), and PODXL (−1.484, 1.78e-02), notably PODXL is anti-adhesive). EndoC cells in HFpEF had key downregulated genes related to cytokine/interleukin/hypoxia signaling (OSMR (−1.150, 8.52e-05)), along with other down-regulated genes IL6ST (−0.720, 2.49e-03), IL4R (−0.946, 1.32e-02), IL18R1 (−1.172, 3.20e-03), HIF1A (−1.311, 1.17e-03), EPAS1/HIF2A (−0.727, 2.21e-03), angiogenesis/junctional integrity (PECAM1/CD31 (−0.863, 3.74e-03), CLEC14A (−0.914, 4.93e-03), ANGPT2 (−0.803, 1.21e-02), RHOJ (−0.888, 7.60e-04), PODXL (−0.830, 5.83e-03)), and metabolism/cellular survival (NAMPT (−1.994, 4.13e-05), BCL2L1 (−1.056, 1.93e-04), SLC2A3 (−1.494, 1.21e-03), ABCC1/4 ((−0.723, 6.99e-03), (−1.526, 2.09e-03)) (Figure 5d). Top HFpEF upregulated genes OPCML (3.158, 2.64e-07), PGM5 (1.358, 1.90e-05), GCNT1 (1.439, 5.24e-04), LRIG1 (1.339, 2.88e-04), TPCN1 (1.026, 5.24e-04), and GNA14 (0.837, 6.07e-04) were related to cellular adhesion/glycocalyx remodeling and regulation of tyrosine kinase receptor signaling. GSEA predicted downregulation of pathways related to immune/inflammatory signaling (Figure 5e). Shifts in composition of endothelial subclusters revealed a higher proportion of endocardial cells in HFpEF (EC-2) consistent with the global analysis, while the others likely reflected overall changes in transcription rather than distinct cell states (EC-0, EC-3, EC-5, EC-7, all capillary, Figure 5f-i). Altogether the data support a remodeled endothelial compartment with alterations in inflammation, angiogenesis, response to stress/hypoxia, and cellular adhesion/junctional integrity.

Figure 5. Endothelial analysis.

a) UMAP of all HFpEF and NF-control nuclei, with endothelial cell types labeled. b-d) Volcano plots for differential expression between HFpEF and NF-control; format as in Figure 2a. e) Dot plots for enriched Reactome pathways in HFpEF vs NF-controls in Endothelial1 (EC1) and Endocardial (EndoC); format as in Figure 2b. f-g) Re-estimated UMAP embedding and clustering of 11,315 EC, EndoC, and lymphatic endothelial (LEC) nuclei based on scVI batch-corrected latent space generated after restricting to these cell types; format as in Figure 3c. h) Box plots showing the proportion of each endothelial sub-cluster separated by NF-controls and HFpEF; format as in Figure 3d. Samples with <10 endothelial nuclei were excluded. i), Dot plot of top marker gene expression for each sub-cluster; format as in Figure 3e.

Comparison to bulk RNA-seq demonstrates the additive utility of snRNA-seq

Estimated logFC between HFpEF and NF-controls based on the snRNA-seq pseudo-bulk data across all cell types significantly correlated with the prior bulk RNA-seq data ³ (r = 0.68, 95% CI 0.67–0.69, Supplemental Figure 15). This was important because the prior bulk RNA-seq analysis was derived from 41 HFpEF, versus 19 in the present analysis, and encompassed a broader BMI range and higher proportion of self-identified White/Caucasian patients. Between 28–58% of cell-type-specific DE genes were not identified as DE in the pseudo-bulk analysis of snRNA-seq (and by extension, were not identified as DE in the prior bulk RNAseq analysis), highlighting the utility of snRNA-seq (Supplemental Figure 16).

Sensitivity analyses with covariate adjustment shows consistent effect estimates

The primary analysis included sex as a covariate. Testing for a sex*disease interaction found only 1 gene with a significant interaction across all cell types; a noncoding gene in fibroblasts (AC109466.1, Supplemental Table 11). As this might be due to limited sample size (15F, 4M) in HFpEF, we performed the same analysis in the bulk RNA-seq (24F, 17M) finding only 2 genes with significant sex*disease interaction (1 protein coding gene, FMO5, Supplemental Table 12). Self-identified race could not be included as a covariate since no controls were Black/African-American. Instead, the prior bulk RNA-seq was leveraged. Our prior bulk RNA-seq analysis had strong correlations in LogFC vs NF-controls in Black/African-American HFpEF (n=28) vs non-Black/African-American HFpEF (n=13, r=0.939, 95% CI 0.937–0.941), Supplemental Figure 17) with similar clinical features (Supplemental Table 13).

Sensitivity analysis additionally adjusted for age, BMI, diabetes, and renal function (eGFR). LogFC in HFpEF vs NF-controls in the minimally adjusted (sex only) versus the sensitivity analysis were strongly correlated for all major cell types (Supplemental Figure 18). There was an expected reduction in power due to differences in covariates between the two groups. These data suggest differential expression was not predominantly due to co-morbidities or demographics.

Validation of animal models identifies lack of correlation between animal models and human HFpEF

Given the growing use of pre-clinical models of HFpEF, we examined the overlap in DE genes between animal models versus human. There were two single cell RNA-seq^28,29 (one including only interstitial cells²⁸) and two bulk RNA-seq studies^30,31 with myocardial data provided for comparison, both using the high fat diet + L-NAME (“two-hit”) animal model of HFpEF in mice. While results from the two animal studies weakly correlated to each other, there were no significant correlations between either animal study and our human snRNA-seq study with r > 0.10 (Supplemental Figure 19). In the bulk RNA-seq, the two animal models were moderately correlated, while neither animal study positively correlated to bulk RNA-seq in human HFpEF (Supplemental Figure 20).

Comparison of HFpEF to DCM reveals most divergent transcriptional changes occur in cardiomyocytes

Both HFpEF and DCM share pathobiology, including energetic defects, hypertrophic and fibrotic stimulation, immune activation, and vascular dysregulation. To identify the shared and distinct cell-type-specific transcriptional features, we combined current data with prior snRNA-seq analysis of DCM versus NF-controls from LV tissue. Clinical characteristics of the latter are in Supplemental Tables 14-15. Though the experiments differed in tissue sources (RVS vs LVS), a prior study showed gene expression in RVS vs LVS is highly correlated in DCM and NF-controls.³ The combined UMAP identified typical cell types in myocardial tissue with mixing of the disease groups within each cell type (Supplemental Figure 21). Patient level PCA revealed significant overlap between NF-controls from both experiments supporting comparable datasets (Figure 6a). Cell-type-specific expression disparities between HFpEF and DCM were found in various cell types, but most prominently in cardiomyocytes (n=1482 genes, Figure 6b, Supplemental Table 16). Sensitivity analysis with additional covariate adjustment revealed a strong correlation between the logFC in all cell types (Supplemental Figure 22), suggesting the difference was not driven by covariates. A subset of genes only DE in HFpEF or DE in opposite directions in DCM (vs NF-controls) are highlighted in Figure 6c (cardiomyocytes) and Supplemental Figure 23 (non-cardiomyocytes). These genes had an absolute logFC of at least 1 and a minimal threshold of expression for the reference group. Of cardiomyocyte genes, DE was generally consistent with prior bulk RNA-seq comparing HFpEF to DCM and NF-controls (all RVS, Supplemental Figure 24). Three genes (PLPP3, PTPRD, MAP2K6) were selected for validation of protein expression based on 1) their magnitude of gene expression change and 2) their potential role in HFpEF pathobiology. Expression of PLPP3 was significantly lower in both HFpEF and DCM (p=0.04, p=0.01 vs NF-controls, respectively), while expression of MAP2K6 was significantly higher in HFpEF (p=1.4e-7) and lower in DCM (p=0.02) vs controls. Protein expression of PTPRD was unchanged in HFpEF and DCM vs NF-controls (Supplemental Figure 25).

Figure 6. Transcriptional differences between HFpEF and DCM.

a) Principal component (PC) analysis by cell type comparing transcriptome of NF-controls, heart failure with preserved ejection fraction (HFpEF), and dilated cardiomyopathy (DCM). Percent of variation explained by each PC is shown. Normal-probability contours based on a 95% data ellipse are shown for each group. b) Bar plots for number of genes significantly differentially expressed, by cell type, for each group comparison. c) Forest plots display logFC in cardiomyocytes for selected genes uniquely down- or up-regulated in HFpEFvsControl, and either unchanged or differentially expressed in the opposite direction in DCMvsControl. Error bars represent 95% confidence intervals. Genes shown had 1) absolute log₂ fold change (logFC) ≥ 1; 2) adjusted p-value<0.05, 3) proportion of nuclei with detectable at least 0.1 in HFpEF (upregulated genes) or controls (downregulated genes), and 4) were protein coding. LogFC was calculated using CellBender. VSMC, vascular smooth muscle cell.

CONCLUSIONS

This study provides myocardial snRNA-seq analysis comparing HFpEF to NF-controls and integrates the findings with a comparison of DCM vs NF-controls. Our work demonstrates the feasibility of performing snRNA-seq on small human heart biopsies, pooling samples to increase nuclei yield, followed by genotype-based demultiplexing. Pathways controlling transcription/translation, immune activation, metabolism, and protein quality control are dysregulated across multiple cell types, most notably in cardiomyocytes and fibroblasts, the cell types that provided the most nuclei and likely dominate bulk RNA-seq. Cell-type-specific profiles revealed VSMCs have a more synthetic, proliferative phenotype; T-cells have a profile consistent with a shift toward effector T-cell activity and away from quiescent T-cells; macrophages display changes consistent with an antigen-presenting phenotype with downregulated clearance programs; and endothelial cells have transcriptome changes suggestive of alterations in inflammation, angiogenesis, and stress/hypoxia responses. Unlike DCM, we did not find a canonical activated fibroblast sub-population in HFpEF despite many DE genes in fibroblasts being involved with ECM remodeling. Lastly, we find cardiomyocytes have the most transcriptional divergence between HFpEF and DCM. These results reveal potential cell-type-specific mechanisms pertinent to the HFpEF heart.

The snRNA-seq analysis identified 14 clusters with similar cell types and proportions in prior snRNA-seq analyses of human myocardium.^{9,12,13,27,32} Similar to DCM and HCM, the largest number of DE genes were in cardiomyocytes and fibroblasts.⁹ Downregulated pathways in cardiomyocytes were similar to those identified in the proteome from similar HFpEF biopsies.⁵ Metabolic pathways were broadly downregulated in multiple cell types, suggesting general factors such as obesity, diabetes, and hemodynamic and neurohormonal stressors are likely drivers for such changes. Prior HFpEF studies found evidence of impaired metabolism of fats, glucose, and branched-chain amino acids in the heart ^3–6, including a study of trans-cardiac metabolomics. ³³ Similar patterns have been reported in DCM, which like HFpEF, has depressed fat metabolism genes in multiple cell types ^9,10 despite less obesity/diabetes. Mitochondrial ultrastructure is abnormal in HFpEF cardiomyocytes⁷, consistent with downregulated genes related to mitochondrial structure/function, including MTFR1, a regulator of mitochondrial fission. While genes related to oxidative phosphorylation were previously found upregulated in HFpEF,³ in the current analysis they were downregulated in cardiomyocytes. This disparity may relate to differences between nuclear vs cytosolic transcriptomics affecting oxidative phosphorylation genes. ³⁴ Upregulated genes/pathways were related to sodium/calcium handling and cytoskeletal/ECM remodeling. Remodeling of the cytoskeletal microtubule network in cardiomyocytes can impact mechanotransduction, excitation-contraction coupling, and myocardial stiffness in HF. ³⁵ These data support alterations in metabolism, translation, cytoskeletal/ECM remodeling, and protein quality control in HFpEF.

Cell-type-specific transcriptomic profiles in the vascular compartment are consistent with a high prevalence of coronary microvascular dysfunction in HFpEF. ³⁶ The transcriptome of VSMCs in HFpEF reflected a more synthetic/proliferative phenotype with downregulation of RBPMS. ³⁷ ACTA2 was also downregulated, a consistent marker of VSMC phenotypic remodeling. ³⁸ The endothelial compartment had expression changes suggesting downregulation of hypoxia/oxidative stress responses, structural/cytoskeletal remodeling, altered angiogenesis, and responses to inflammatory/immune signaling. These results align with endothelial dysfunction playing a role in HFpEF. ³⁹

While proteomic analyses of peripheral blood in HFpEF have identified higher levels of inflammatory proteins,⁴⁰ prior bulk RNA-seq of HFpEF vs NF-controls from heart tissue ³ did not find enriched inflammatory pathways. In contrast, snRNA-seq detected enrichment of inflammation/immune activation in multiple cell types, yet they were often upregulated in immune cells but downregulated in non-immune cells as reported in DCM⁹. Several of these pathways had many genes ubiquitously expressed in other cell types, and these may serve non-immune/inflammatory functions in such cells. The transcriptional signature of HFpEF lymphocytes implies recruitment of activated, effector T cells, ^41,42 which has been reported in preclinical models of HFpEF. Whether this results from more systemic inflammation⁴⁰ or a myocardial inflammasome ⁴³ remains uncertain. One of the lymphocyte clusters identified as quiescent/resting T cells (high IL7R and LEF1 expression⁴⁴) were reduced in HFpEF. Macrophages were similar in abundance between HFpEF and NF-controls but one sub-cluster identified as monocytes^10,27,32 were more abundant in HFpEF. Macrophages in HFpEF appear to shift toward immune modulation/antigen presentation, potentially activating effector T cells. Concurrent downregulation of CD163 and MERTK suggests impaired clearance programs, ⁴⁵ which might contribute to myocardial inflammation and microvascular dysfunction. Histologic studies show increased macrophage infiltration in HFpEF myocardium,^14,46 supporting these findings.

Unlike DCM, ⁹ sub-clustering did not reveal a distinct, canonical activated fibroblast population in HFpEF and marker genes were not DE in HFpEF vs NF-control fibroblasts (FN1, TNC, POSTN, COL1A1/COL1A2). However, the global gene expression profile of HFpEF fibroblasts suggested fibroblast activation (FAP, ACTA2) and ECM remodeling (several COL genes, LUM, TGFBI, CXCL12), along with many DE genes consistent with altered cellular signaling and metabolism.

The purpose of comparing HFpEF to DCM was not to show these diseases differ, but to identify cell-type-specific transcriptional similarities. Further, given the long history of therapeutic failure in HFpEF, we thought it useful to probe for unique gene signatures between HFpEF and DCM that might lead to future effective treatments. Despite their distinct clinical phenotypes, we observed strikingly similar transcriptional shifts in most cell types, with the notable exception of cardiomyocytes. Among the cardiomyocyte genes most disparate between the HF subtypes was MAP2K6 (also MKK6), the upstream kinase that principally activates p38α and p38β MAP kinase. Intriguingly mice with activated-MKK6 in cardiomyocytes develop marked diastolic dysfunction with a preserved EF⁴⁷. Other genes downregulated only in HFpEF cardiomyocytes were phospholipid phosphatase PLPP3 which has effects on lipid signaling, cardiac hypertrophy, and decreased mitochondrial bioenergetics⁴⁸, and tyrosine phosphatase receptor type D (PTPRD), which activates PPAR gamma 2 and regulates insulin signaling⁴⁹.

One of the major challenges in the field is developing pre-clinical models of HFpEF. While some aspects of human HFpEF have been recapitulated in animal models, particularly individual pathways of interest, we found limited correlation between broader transcriptomic profiles in bulk or single cell RNA-seq in the “two-hit” animal model of HFpEF in mice. The single cell data ^28,29 were limited by sample size (n=2–4); however, the bulk RNA-seq ^30,31 also did not correlate to human HFpEF. These data highlight the need for better animal models to recapitulate the transcriptomic changes in human HFpEF hearts.

This study has several limitations. The sample size is modest, so may not capture the full extent of patient variability and features found in the broader human HFpEF population. As noted, HFpEF and NF-controls differed in self-identified race, limiting our ability to address race disparities. While cell-type-specific race disparities may pertain, our prior bulk RNA-seq analysis found very few genes impacted by race in a broader HFpEF cohort. NF-controls were also younger and less obese. Tissue harvesting methods also differed. While NF-control tissue was obtained from the same endocardial region as HFpEF, the NF-control and DCM samples were from whole excised hearts after anterograde perfusion with cold cardioplegia. This procedure was not possible in HFpEF patients. Pooling samples may have reduced the recovery of less transcriptionally complex cell types, as genetic demultiplexing relies on reads from the snRNA-seq itself. The combination of this computational demultiplexing, potential differences in nuclei dissociation efficiency by cell type, and differences in intramyocardial (NF-controls) vs endomyocardial (HFpEF) sampling could impact cell type proportion estimates, so cell type proportions from snRNA-seq should be interpreted with caution. In rare instances, discrepancies arose between our snRNA-seq and bulk RNA-seq data, potentially due to differences between nuclear and cytoplasmic RNA isolation. Differences between DCM and HFpEF in cell types with low transcriptional activity or abundance, such as lymphocytes, made it harder to detect differences in these cells. Lastly, while our Western blots showed clear bands at the appropriate molecular weight for our target proteins, some of the blots had non-specific bands at different molecular weights, a limitation of commercially available polyclonal antibodies.

Recognizing these limitations, our results provide a cell-type-specific analysis of HFpEF that should prove valuable for future mechanistic preclinical studies and improved targeting of molecular defects in HFpEF.

NOVELTY AND SIGNIFICANCE.

What is known?

• Heart failure with preserved ejection fraction (HFpEF) is a multi-system disease with high morbidity and mortality, but its myocardial pathobiology remains poorly defined.

• Prior studies of human heart tissue from HFpEF identified distinctive alterations in metabolism, protein translation/quality control, and immune activation.

• Single-nucleus RNA sequencing revealed important cell-type-specific transcriptomics in other cardiac diseases, while cell-type-specific gene expression in HFpEF is unknown.

What new information does this article contribute?

• Cell-type-specific differential gene expression in HFpEF identified dysregulation of immune activation, transcription/translation, and metabolism pathways in multiple cell types.

• Distinct and directionally-opposite changes in multiple genes were detected in cardiomyocytes from HFpEF versus dilated cardiomyopathy.

• In HFpEF, vascular smooth muscle cells had a more synthetic phenotype, while immune analysis suggested enhanced T cell activation and reduced macrophage clearance programs.

While bulk analyses of heart tissue from HFpEF yielded hypotheses regarding its pathobiology, this study advances our knowledge of transcriptomics at the single cell resolution. Analyses of lower abundant cell types suggest remodeling of the vascular/immune compartments in HFpEF. These findings were largely shared with dilated cardiomyopathy, while major discrepancies were identified in cardiomyocytes between the two heart failure phenotypes. These findings may aid development of novel therapeutics for HFpEF.

NON-STANDARD ABBREVIATIONS AND ACRONYMS

CM

cardiomyocytes

DCM

dilated cardiomyopathy

EC

endothelial cells

EndoC

endocardial cells

GSEA

gene set enrichment analysis

HFpEF

heart failure with preserved ejection fraction

logFC

log-2 fold change

LVS

left ventricular septum

OCT

optimal cutting temperature

PCA

principal component analysis

qPCR

quantitative polymerase chain reaction

RVS

right ventricular septum

snRNA-seq

single nucleus RNA sequencing

TCA

tricarboxylic acid

UMI

unique molecular identifier

VSMC

vascular smooth muscle cells