Bringing doctors together

or

A free, private network for verified physicians

Leading voices

Dr. Shashank Sinha

Advanced Heart Failure · Inova Fairfax Hospital

View

Dr. Javed Butler

Advanced Heart Failure · Baylor Health Care System

View

Dr. Andrew Sauer

Advanced Heart Failure · Saint Luke's

View

Dr. Anu Lala

Advanced Heart Failure · Mount Sinai

View

Dr. Robert Mentz

Advanced Heart Failure · Duke University

View

Dr. Sunny Varshney

Advanced Heart Failure · Stanford

View

Dr. Stephanie Hsiao

Advanced Heart Failure · Stanford

View

Dr. Michelle Kittleson

Advanced Heart Failure · Cedars-Sinai

View

Dr. Adrian Hernandez

Advanced Heart Failure · Duke

View

Dr. Dorairaj Prabhakaran

General Cardiology

View
  • Discover your community

  • Stay current

  • A profile that represents you

  • Find your people

About RoonPrivacyTerms

    Overview

    Leveraging Single-Cell Technologies to Advance Understanding of Myocardial Disease.

    Jan 2, 2026DOI: 10.1161/circresaha.125.326002Journal ArticleAHA

    Written by: Robert S GardnerNathan R TuckerKaushik Amancherla

    Find these authors on Roon

    KA
    Dr. Kaushik Amancherla
    Advanced Heart Failure · VUMC

    Introduction

    Cardiovascular disease (CVD) remains the leading cause of death globally, with heart failure (HF) representing a common final pathway for many forms of CVD1,2. Despite major advances in modification of risk factors through dietary/lifestyle and pharmacotherapeutic approaches, the incidence of HF continues to rise and its prevalence is estimated to be >10 million in the United States of America by 20403. Studying CVD progression and HF within the context of human disease, as a complement to model system research, may accelerate drug and biomarker development. However, the adult human myocardium is complex, with a diverse range of cell types and molecular heterogeneity. Traditional “bulk” approaches to delineate gene expression patterns at the tissue-level have revealed broad signatures of disease but cannot resolve the cell-type-specific gene expression programs or identify rare cell types, i.e., those with relatively low abundance, that may underlie CVD trajectory.In contrast, single-cell transcriptomic approaches allow for profiling gene expression in individual cells or nuclei, capturing both cellular heterogeneity and rare cell populations (Figure 1). Initial studies demonstrating the technical feasibility of this approach prioritized accessible tissue specimens, such as peripheral blood mononuclear cells and tumor biopsies. In turn, results from such approaches have revolutionized clinical oncology and informed precision diagnostic and therapeutic strategies4–7. By the late 2010s and early 2020s, these approaches, complemented by high-resolution- spatial and multiomic platforms, were applied to human hearts and have allowed investigators to deconstruct the human myocardium with unprecedented granularity8–10.In this review, we first provide a historical overview of the evolution of single-cell and single-nucleus approaches and their limitations, followed by their applications to the human myocardium, highlighting unique challenges in working with this tissue. Next, we describe the cellular landscape of the healthy heart, followed by a discussion on the specific contributions of each major cell type – cardiomyocytes, fibroblasts, endothelial and mural cells, and immune cells – to human myocardial disease, emphasizing dynamic changes and intercellular interactions uncovered by single-cell approaches. We then clarify the limitations of single-cell studies to date in CVD and briefly comment on several emerging technologies within this space. Finally, we consider clinical translation and future directions—how these tools might advance diagnostics, risk prediction, and precision therapies, and how they can be incorporated into clinical CVD research.

    Single-cell technologies: a historical perspective

    Single-cell interrogation originated with histology and flow cytometry workflows to resolve differentiated cell populations from bulk tissues on the basis of morphology or a limited set of RNA/protein markers. Early single-cell preparations paired with qPCR quantified pre-selected gene expression in individual cells11, but remained low-throughput and susceptible to amplification bias. Microarray platforms12 expanded the gene panel per experiment but remained reliant upon pre-designed probes that continued to limit their utility. The advent of next generation sequencing (NGS13) enabled simultaneous sequencing of millions of cDNA fragments, ushering in the era of high-throughput transcriptomics. Leveraging NGS, RNA sequencing methods in “bulk” tissues (i.e., whole tissue)—typically coupled with poly-A capture or rRNA depletion followed by reverse transcription—eliminated the need to pre-select targets, allowing relatively unbiased transcript quantification14–16.
    Building upon these innovations, the first single-cell RNA-sequencing (scRNA-seq) method was introduced in 200917, demonstrating that whole transcriptome profiling in a single cell was feasible and could enable measurement of gene expression patterns from cell states/populations otherwise obscured in bulk tissue experiments. Following this demonstration, there has been rapid development of additional single-cell transcriptomic methodologies18,19. Ultimately, all technologies rely upon the ability to segregate cells for unique molecular barcoding of transcripts, which may then be read out post-sequencing. Plate-based workflows isolate cells into individual wells where barcoded cDNA is synthesized. Smart-seq20,21, for example, captures full-length transcripts, enabling study of splice variants and allele-specific expression, while MARS-seq22 combines fluorescent activated cell sorting with plate, cell, and molecular barcoding methods to increase throughput and identify PCR duplicates. Subsequent droplet-based approaches using microfluidic partitioning—inDrop, Drop-seq, and the 10X Genomics Chromium system23—dramatically increased the number of cells that could simultaneously undergo transcriptomic profiling. These methods partition individual cells or nuclei into nanoliter-scale droplets formed by water-in-oil emulsions, wherein cDNA is generated and barcoded by cell and unique transcript (unique molecular identifier [UMI]) by reverse transcription. Compared to select plate-based methods, gene detection sensitivity per cell is reduced in microfluidic droplet-based approaches, but due to sheer throughput, the detection of cell populations composed of relatively few cells is enhanced. Notably, gene expression is a dynamic, stochastic process that can occur in bursts24, such that the number of transcripts associated with a given gene fluctuate over time (e.g., lowly-expressed genes may not always be transcribed) even within the same cell type. Such phenomena lead to cell-to-cell transcriptional variability, and when considered alongside drop-out events stemming from failure to capture expressed transcripts in single-cell sequencing techniques, can lead to relatively sparse datasets25. Higher-throughput increases the likelihood of capturing lowly- and dynamically-expressed genes that may not be detected in all cells of a particular cell type, simultaneously permitting estimation of gene dispersion on a cell-by-cell basis.
    Additional high-throughput partitioning approaches have emerged that lower per cell costs, including use of combinatorial indexing with SPLiT-seq26, micro-well arrays with Seq-well27, and microfluidic-free droplet generation with PIP-seq28 (Figure 2A). Moreover, transcriptome-wide probe-based RNA capture techniques are now available, including those that permit use of formalin-fixed tissues (Flex from 10x Genomics29).
    Several state-of-the-art platforms integrate multiple molecular layers and provide spatially-resolved measures. Multiome approaches link within individual cells/nuclei transcriptomics to chromatin accessibility (10x Multiome) and cell-surface protein expression (CITE-seq30), whereas spatial transcriptomics are approaching single cell resolution, both with targeted panels and at the whole transcriptome level31. Collectively, this rapidly-expanding tool kit is paving the way for an explosion of cellular-resolution genomics and transcriptomics studies.
    While single cell applications in human tissue were initially capitalized in oncology, benefiting from immediate access to surgical and biopsy specimens obtained during routine clinical care, these approaches have recently been applied to cardiac biology to better delineate cardiac development, the healthy adult myocardium, and disease-associated cell-specific gene expression programs. Indeed, large-scale efforts to construct an integrated cellular atlas of the human heart are underway (Heart Network from Human Cell Atlas; https://data.humancellatlas.org/hca-bio-networks/heart). Moreover, interactive repositories such as the Broad Institute’s Single Cell Portal (https://singlecell.broadinstitute.org/single_cell), the GTEx Portal (https://gtexportal.org), and the Chan-Zuckerberg Initiative’s CellxGene (https://cellxgene.cziscience.com) house a growing catalog of cardiac single-cell datasets, enabling scientific data sharing, hypothesis generation, and accelerated biologic discovery.

    Single-nucleus workflows: challenges in the human heart

    The overwhelming majority of studies applying single-cell techniques to human cardiac tissues has prioritized droplet-based microfluidic approaches, with some supplementation from spatial transcriptomic and multiomic methods8–10,32–40. Here, we highlight design considerations with a primary focus on the use of droplet-based microfluidics for cardiovascular research.
    Human cardiac tissue for research has been largely limited to specimens collected from patients with end-stage heart disease undergoing advanced cardiac therapies, such as heart transplant or left ventricular assist device (LVAD) implantation, and from healthy/non-diseased donors collected postmortem. Given logistical complications of procurement and processing, samples are typically flash-frozen, which necessitates use of nuclear suspensions rather than intact cells due to plasma membrane rupture during the freeze-thaw cycle. Conventional whole-cell approaches perform poorly in freshly collected heart tissues as the large size of adult cardiomyocytes (~60–140+ microns long41) precludes efficient passage through the microfluidics or cell sorting devices predominantly utilized. We note, however, select studies have implemented intact single cell approaches immediately following tissue collection, primarily focusing on non-cardiomyocytes8,42–44, while emerging large-particle cell sorting and microfluidic-free partitioning techniques show promise to target intact adult cardiomyocytes26,28,45,46.
    Figure 2B summarizes the advantages and disadvantages of using single cell vs single nuclear suspensions, highlighting important trade-offs between the two approaches. Advantages of nuclear suspensions include compatibility with cryopreserved, fibrous, or fatty tissues that are difficult to cleanly dissociate; mitigation of capture biases induced by differences in cell sizes or vulnerabilities; and avoidance of transcriptional impacts induced by harsh whole cell isolation protocols47,48. For example, the process of generating whole cell suspensions is associated with higher expression of activity-related gene expression as shown in matched single cell and nucleus preparations in brain48 and other tissues49. These dissociation-induced transcriptional artifacts can, in turn, confound downstream analyses. On the other hand, nuclear (but not whole cell) suspensions are also compatible with select multi-omic platforms, such as those pairing single-nucleus RNA-sequencing (snRNA-seq) and single-nucleus Assay for Transposase-Accessible Chromatin sequencing (snATAC-seq)50. In snATAC-seq, transposase is added to penetrate the nuclear membrane to cut and tag regions of accessible chromatin for later sequencing, which requires nuclear suspensions.
    However, challenges and limitations persist when using nuclear suspensions. Relatively transcriptionally quiescent cells have relatively low levels of nuclear RNA and nuclear preparations typically contain increased levels of ambient RNA in the extranuclear supernatant51 due to the release of cytoplasmic RNA from lysed cells. Compounding these challenges are low cell yields in select heart regions (e.g., atria), that limit wash steps, further increasing ambient RNA. Moreover, the presence of multinucleated cardiomyocytes, thought to be ~25% of all cardiomyocytes in human52, may bias estimates of relative cell type proportions.
    The transcripts present in nuclear suspensions also differ from those in whole cells. Nascent/pre-mRNA is enriched in nuclear suspensions, which lack cytoplasmic RNA, and yield fewer detectable genes and transcripts35,48,53. The lower diversity of mature RNA in nuclear suspensions may also limit classifications and analyses that rely on estimates of spliced and unspliced transcripts (e.g., to infer RNA velocity), although recent findings suggest high correlations of RNA velocity measures from matched samples of single cell and single nucleus preparations54. While detection of mitochondrial transcripts can help identify low quality nuclei contaminated with cytoplasmic RNA, their capture is inherently lacking in single nucleus preparations53. This constraint hampers downstream analyses to detect changes in mitochondrial transcript expression reported to be sensitive to disease states in other organ systems55. In contrast, long non-coding RNAs are enriched in nuclear preparations53. As expected, however, pseudobulk analysis of single-cell versus single-nucleus data align better with data from bulk RNA sequencing experiments53.
    Although cell type markers generally correlate between snRNA-seq and scRNA-seq experiments, the relative abundance of each cell type and detection of rare or minor populations, e.g., adipocytes or endocardial cells, or distinct cell states can differ8,35,53,56,57. Notably, in cardiac tissues, the use of nuclear suspensions (compared to cellular suspensions) captured more reads from populations of mast cells, adipocytes, endocardial cells, epicardial cells, and lymphatic endothelial cells, an effect suggested to stem from inefficient recovery in single cell approaches due to processing methods (e.g., enzymatic digestion)35. Other reports corroborate enhanced detection of adipocytes in single nucleus preparations8.
    Typically, using microfluidic droplet-based approaches, frozen sections of heart tissue are mechanically and/or enzymatically dissociated to generate single nuclear suspensions, with removal of debris through centrifugation and/or flow cytometry. Single nuclear suspensions are partitioned into nanoliter-scale droplets along with lysis buffer, reverse transcription reagents, and oligonucleotides containing unique nuclear and molecular barcodes and poly-dT sequences to capture polyadenylated RNA. Within each droplet, nuclei are lysed, polyadenylated transcripts bind to the oligonucleotide, and barcoded cDNA is synthesized by reverse transcription. Incorporation of molecular and nucleus-specific barcodes allows for removal of PCR duplicates and identification of the transcript’s nuclear origin, respectively. After emulsion breakage, cDNA is pooled, purified, amplified with PCR, fragmented, and adapters added for NGS. Adaptations to this method include various multiplexing strategies prior to partitioning, thus limiting costs without sacrificing throughput58–60. The resulting cDNA libraries undergo paired-end sequencing, with generated reads containing the unique molecular barcode, the nucleus-specific barcode, and a portion of the transcript sequence. Short-read sequencing is most commonly used and enables gene-level quantification. While 90 base-pair reads corresponding to the captured transcript, as generated by the 10X Genomics Chromium platform, permit gene-level count data, the reads may not span spliced junctions and pose a challenge for isoform detection, particularly when intron-rich regions are highly represented.
    Here, we present a general framework for analysis of single cell/nucleus RNA sequencing data (Figure 3). We stress, however, that single cell analysis remains an evolving field, particularly as applied to cardiac datasets, with highly dynamic best practice standards. Critically, the computational approach, and the user-defined parameters applied at each step, can substantially influence downstream results. Moreover, optimal choices can be specific to each dataset. Therefore, a thorough understanding of each computational tool is essential when designing an analysis pipeline. To this end, we recommend systematic exploration of parameter selection and use of multiple tools, prioritizing convergent findings across analyses, which can be validated using orthogonal experiments. This is particularly important for analyzing cardiovascular datasets, as computational tools may be developed without ground truth comparisons, relying on simulated datasets61,62 or are tested on datasets from other tissues (e.g., peripheral blood mononuclear cells).
    The sequencing base call files are demultiplexed and FASTQ files are generated for each sample. Reads are aligned to a reference transcriptome using a splice-aware aligner63 (e.g., STAR64) while retaining intronic regions to capture reads from pre-mRNA and enhance the number of detected genes48,53. STAR alignment is integrated into pipelines such as CellRanger (10x Genomics) or STARsolo, which additionally remove PCR duplicates and call cells, generating cell-by-gene count matrices. High levels of ambient RNA complicate cell calling, as droplets with high amounts of ambient RNA may be misclassified as cells while true low RNA-containing nuclei (e.g., some types of immune cells) may be mistakenly discarded. Importantly, ambient RNA can obscure/contaminate the transcriptomic profiles of correctly called cells, confounding downstream analyses. To address such issues, computational tools have been developed65 (e.g., SoupX66 and Cellbender67) to model the ambient RNA in empty droplets and more confidently identify cells and/or adjust gene-level counts to subtract the ambient RNA profile.
    Following cell calling, several computational workflows are available for data analysis and visualization. Most commonly, Seurat68 or Scanpy69 are used with additional tools available that can be integrated into these workflows. Processing steps include removal of low-quality nuclei based on the total number of detected genes, counts, and/or proportions of counts corresponding to mitochondrial/ribosomal genes.
    It is expected that during cell or nucleus partitioning using microfluidic approaches, a proportion of droplets contain more than one cell or nucleus (called “multiplets”). These multiplets are identified computationally, largely based on tools that construct populations of simulated doublets by merging pairs of randomly selected cells/nuclei within the dataset70–73. A doublet score can then be assigned to each cell/nucleus based on the proportion of its nearest neighbors that are simulated doublets, which can then be used to filter out cells/nuclei or clusters of cells/nuclei identified as likely doublets. These tools primarily detect barcodes associated with two or more distinct cell types, rather than doublets containing multiples of the same cell type, which remains a limitation.
    Quantification of intronic and exonic reads (e.g., scR-Invex [https://github.com/getzlab/scrinvex]) can further assess cytoplasmic contamination in nuclear preparations. Raw counts are normalized to a size factor to account for differences in sequencing depth. While scaling is often applied to each nucleus, additional methods apply normalization to pools of nuclei which are deconvolved to estimate nucleus-specific size factors74 or use gene-specific normalization that models the relationship between gene expression and library size across nuclei, grouping genes with similar scaling behavior75. These latter approaches were developed to better account for noise in single-cell datasets or better represent gene expression that does not scale linearly with sequencing depth76. Subsequently, counts are log-transformed (log1p) or variance-stabilized due to the skewed/zero-inflated, heteroskedastic characteristics typical of snRNA-seq datasets77.
    Principal components analysis (PCA) is applied to highly variable genes to identify the primary axes of variation within each dataset. Nearest neighbor (k-NN) graphs are constructed, whereby pairwise distances between all nuclei are computed (e.g., Euclidean distance) in the PCA-reduced dimensions. The k-NN graph is then used to identify local neighborhoods of transcriptionally similar nuclei (e.g., as done with Milo78), or more commonly as input into unsupervised clustering algorithms (e.g., Leiden clustering79) which optimally partition individual nuclei into formal clusters. Neighborhoods or clusters are annotated (e.g., to identify the cell type or cell state contained therein) manually through usage of canonical marker genes, computational approaches, and/or automated reference-based tools80,81, including large language models82. This process can be iterative, adjusting the resolution used in the clustering algorithm, while sub-clustering can identify unique transcriptional states within a given cell type. Additional methods annotate at the individual cell level (e.g., CellTypist83) as opposed to a broader cluster level, potentially limiting misclassification of nuclei contained within a heterogeneous group81.
    A major challenge in single-cell approaches relates to preservation of biological variation of interest while minimizing systematic technical variation stemming from distinct preparation methods, sequencing runs, or sample-to-sample variability. Screening principal components for correlation with known sources of potential technical variation is a useful first step to identify batch effects. However, numerous tools have been developed for batch-effect correction76 with those that rely on cluster-based and deep neural network approaches (e.g., Harmony84 and scVI85) most commonly utilized. As an example, Harmony uses an iterative clustering-based approach to adjust coordinates in low-dimension space. The tool favors nuclei alignment with clusters showing distributed representation across a majority of the defined batches. The goal is to then identify and align nuclei with similar transcriptional profiles (e.g., those from the same cell type or activation state) even when from unique individuals, tissues, or experimental conditions. This approach can be advantageous when the transcriptional variability explained by the biological classification of interest (e.g., cell type) is well-correlated across batches, but less optimal to integrate (e.g., across rare or variably observed cell types) that may not be well-represented. In cardiac tissues, which may show distinct representations of cell types due to sampling biases, or distinct representations of cell states due to disease or individual-to-individual variability, batch integration across cell types can prove challenging. Benchmarking batch integration tools demonstrates that an optimal approach depends on several factors, including feature selection, data normalization, and the complexity of integration, suggesting careful exploration of available options for each dataset86. Moreover, the majority of tools require the user to define batches of interest, leaving the potential for unknown batch effects to persist.
    Following quality control, cell-type annotation, and batch integration, detailed investigations can be undertaken to quantify cell type composition, differential gene expression between conditions, and pathway enrichment. Cell-type composition can be inferred from the relative counts of nuclei across distinct cell types, enabling comparisons across cardiac chambers or conditions, with specific tools developed to account for dependencies between cell-type proportions, or to characterize differentially abundant cell states78,87. Similarly, testing for differentially expressed genes across conditions—typically performed within clusters (i.e., cell types) or neighborhoods—can assess cell-type specific transcriptomic signatures associated with cardiac dysfunction and identify unique, rare, or previously uncharacterized cell states88.
    Several statistical approaches for testing of differential expression are employed89,90. Traditional statistical tests (e.g., the non-parametric Wilcoxon rank sum test) treat the cell as an independent unit of analysis, but do not account for correlations among cells from the same sample, limiting their utility in multi-sample datasets and resulting in enhanced false positive type I errors89. Tests developed specifically for single cell approaches, e.g., MAST91, can model the subject as a random effect to account for intra-subject variability. Alternatively, pseudo-bulk approaches aggregate expression measures across cells of interest for each sample and identify differential expression using methods developed for bulk tissue RNA sequencing data92,93. Additional tools consider complementary measures, including differences across conditions in gene expression variability and distribution94,95.
    Further interrogation identifies biological functions and molecular pathways associated with the observed transcriptomic changes96, with additional tools used to construct gene co-expression networks and infer transcription factors (e.g., SCENIC97) that may drive disease-associated signatures. Orthogonal approaches can clarify cell-cell communication based on ligand-receptor interactions across cell types (e.g., CellChat98 and CellPhoneDB99), determine cell trajectories (e.g., Monocle100, Slingshot101, Scanpy69), and integrate RNA velocity to identify cell-state transitions (e.g., Cell2fate102 and scVelo103).
    We stress that aspects inherent to single cell approaches (e.g., the stochastic nature of gene expression and failure to capture or amplify select transcripts) can introduce noise that can lead to false identification of differentially-expressed genes, distinct cell states, and/or trajectories. These limitations can be exacerbated when the sequencing depth per cell/nucleus is low, or in the presence of homotypic multiplets (barcodes containing multiple cells of the same type) that resemble transitional cell states. Scrutinizing both the list of highly variable genes used to determine the axes of variation, and the list of marker genes used to define cell states can help to ensure lowly-expressed genes are not driving the classifications. Recently, tools are being developed to identify and prevent spurious clustering104,105. Despite these challenges, the resulting data stem from relatively unbiased sampling at large scale and can be used to inform further orthogonal and mechanistic experiments for validation. Indeed, single-cell findings prove useful both to corroborate well-established findings and in generating novel hypotheses for subsequent validation.

    The healthy human heart: Insights from single nucleus transcriptomics

    The heart consists of diverse spatially and transcriptionally specialized cell types that work together to maintain its function as an effective pump. While cardiomyocytes are central to muscle contraction, complex interactions between numerous cell types support structural integrity, vascularization, and signaling dynamics. Here, we highlight findings from single cell experiments that focus on the healthy adult myocardium8–10,106.
    In cardiac tissues, a diversity of cell types is detected using snRNA-seq. Two seminal large-scale surveys of the healthy human myocardium published in 20208,9 identified 9 major cell classes across cardiac chambers—cardiomyocytes (~40% of all nuclei), fibroblasts (~25%), pericytes (~13%), endothelial cells (~10%), macrophages (~5%), vascular smooth muscle cells (VSMCs; ~3%), adipocytes (~2%), lymphocytes (~1%), and neuronal cells (~1%). Additionally, sub-clustering or cell enrichment methods uncovered additional cell populations, including lymphatic endothelial cells, endocardial cells, mast cells, and subsets of T cells, neutrophils, dendritic cells and monocytes8–10. These findings demonstrate the ability to capture cell populations with relatively low abundance. A negative correlation between the relative proportions of ventricular cardiomyocytes and fibroblasts and positive correlation between VSMCs and pericytes8, present the possibility of dynamic cellular interplay and highlight inter-individual variability which has long confounded bulk approaches.
    The relatively large scale of these single nucleus cardiac datasets has uncovered previously uncharacterized or unappreciated putative cell states. Cardiomyocytes exhibit the highest degree of chamber-specific transcriptional heterogeneity, with atrial cardiomyocytes enriched for NPPA, MYL7, and MYH6, and ventricular cardiomyocytes enriched for HEY2 and MYH78–10,106, suggesting unique properties in their contractile machinery. The most prominent transcriptional differences across the atrial-ventricular axis were observed between left atrium and left ventricle, suggesting laterality may shape chamber-specific gene expression programs9. An example of laterality is observed in the atrium, with left atrial cardiomyocytes showing relatively high expression of contractile genes while right atrial cardiomyocytes exhibit relatively high expression of gene programs implicated in cardiac conduction106. Transcriptional profiles of unique cardiomyocyte populations largely associate with muscle cell development and cardiac contraction9,10, but also show distinct and varied signatures, including enhancements in immune and metabolic programs10. Further demonstrating functionally-distinct cardiomyocyte sub-populations, atrial-ventricular comparisons revealed transcriptional signatures indicative of enhanced metabolic processes in the left ventricle, consistent with their higher energy demand10.
    Cell-type composition also varies according to chamber, with ventricular tissue containing a higher proportion of cardiomyocytes and lower proportion of fibroblasts in comparison to atrial tissue. Transcriptional diversity across chambers is also evident in non-cardiomyocytes9,90,106. For example, atrial macrophages preferentially express M2-associated transcripts while atrial fibroblasts exhibit profibrotic profiles, particularly within the right atria9, suggesting chamber-specific niches associated with immune modulation and extracellular matrix regulation.
    Beyond transcriptomics measured in isolation, studies combining snRNA-seq with snATAC-seq have further uncovered distinct region- and cell-type-specific gene regulatory programs107, while spatial multi-omic approaches33 have characterized novel cardiac niches. Altogether, single nucleus approaches to the healthy human myocardium highlight potential links between cell-specific transcriptional maps and chamber-specific physiology.
    Large-scale atlases constructed from single-cell studies have also been intersected with CVD genetics connecting genetic risk factors to expression profiles of specific cell types. Approximately 25% of the genes implicated in monogenic cardiomyopathies are enriched in cardiomyocytes9. Moreover, genome-wide association study (GWAS)-loci for atrial fibrillation, PR and QT intervals localize to genes selectively expressed in distinct cardiomyocyte populations8,9,108,109, whereas GWAS signals for QRS duration, QT interval, coronary disease, and hypertension are linked to expression profiles in non-cardiomyocytes, including adipocytes, mural, and neuronal cells8,9,110. snATAC-seq experiments observed enhancer regions linked to HF were selective accessibility in cardiomyocytes and fibroblasts. Regions differentially accessible in cardiomyocytes were proximal to genes associated with cardiac contraction and cation transport, whereas those differentially accessible in fibroblasts were proximal to genes linked to extracellular matrix organization (e.g., LUM, DCN) and connective tissue development107. Non-coding atrial fibrillation risk variants also map to regions showing selective accessibility in cardiomyocytes107, consistent with findings relying strictly on snRNA-seq signatures. Efforts prioritizing statistical fine-mapping and/or expression quantitative trait loci (eQTLs) are beginning to identify variant-to-gene interactions that contribute to disease107,111,112.
    CVD risk is also robustly associated with advanced age, whereas sex differences are observed in distinct phenotypes that lead to HF113,114. Importantly, age- and sex-dependent differences are characterized in single cell experiments, showing changes in cell composition and cell-type specific transcriptional profiles115,116, further providing a foundation to explore cellular diversity that may predispose individuals to cardiac dysfunction. Integrating snRNA-seq with chromatin accessibility in human hearts, sex- and age-associated transcriptional programs are being further delineated116. In early studies, female cardiomyocytes showed enrichment of fatty acid metabolism genes, while aging hearts displayed increased TGFβ signaling in fibroblasts and inflammation across several cell types spanning immune and non-immune cells (e.g., vascular endothelium). These data suggest that personalized insights—such as sex-specific drug responses or age-stratified risk profiles—may eventually emerge from cardiac single-cell atlases.
    Altogether, findings in the healthy human heart offer insight on cell-type-specific contributions to CVD risk, informing efforts to innovate precision medicine. These findings in the non-diseased heart additionally provide a comparative foundation which can be used to identify cellular and transcriptional mechanisms that associate with and may underlie cardiac dysfunction and HF.
    While this review primarily focuses on the adult heart, we note that single-cell efforts have also been applied to understand fetal and postnatal cardiac development,108,117–121, revealing prominent shifts in cell type proportions and transcriptional programs. Insights include observations of right ventricular enrichment of fibroblasts and immune cell subsets coincident with declines in cardiomyocytes in the postnatal vs fetal heart. Moreover, across development, cardiomyocytes undergo significant changes in gene expression programs with downregulation of cell cycle programs and upregulation of metabolic programs119. Beyond the scope of this review, biological discovery and key findings from the application of single cell approaches in model systems have also been summarized122,123.

    Cell-specific contributions to human myocardial disease

    Direct study of human myocardium has broadened our understanding and generated novel hypotheses of how diverse cell types and associated molecular programs influence normal cardiovascular function and contribute to CVD. Single-cell and single-nuclear techniques have enabled novel discovery of a coordinated set of derangements across multiple cardiac cell populations, otherwise undetectable with prior approaches, that coincides with the development and recovery from CVD, primarily heart failure – a common final pathway. Through the dissection of cell-specific gene expression patterns, these methods have characterized cellular and transcriptional signatures of clinically-relevant CVD phenotypes, including ventricular hypertrophy, various etiologies of end-stage heart failure, myocardial ischemia, and atrial fibrillation; Table 1).
    Below, we highlight cell-specific contributions to human CVD uncovered using single-cell and single-nucleus approaches. This review primarily highlights the transcriptional heterogeneity of the most abundant and largely explored cardiac cell types, which also show relatively prominent transcriptional shifts in CVD: cardiomyocytes, fibroblasts, mural cells, endothelial cells, and myeloid cells. However, it is important to note reports suggesting modest shifts in CVD in the relative cell numbers and transcriptional profiles of relatively low abundant or rare cell types. For example, HF has been associated with proportional changes and/or differential expression in neural cells, adipocytes, mast cells, lymphatic endothelial cells, and lymphocyte subsets35,39,124–126. Notable transcriptional changes in HF are associated with altered cytokine signaling in adipocytes and lymphatic endothelial cells39. Moreover, sub-clustering of cardiac neural cells has revealed transcriptional heterogeneity with a reported immune-related sub-population, and overall transcriptional shifts in DCM that are associated with axon guidance, neural migration, and response to calcium ions127. While not the focus of this review, these findings, taken together with those showing disease-associated GWAS signals localized to genes showing selective expression in several minor cell types128, warrant further exploration of the roles of these relatively low abundant cells in CVD.
    It is also important to acknowledge that research tissue is obtained mainly from patients undergoing advanced cardiac therapies for end-stage CVD or valve dysfunction (e.g., patients undergoing surgical replacement of mitral or aortic valve) and generally collected at a single time point. As a result, transcriptional profiles often reflect a single snapshot of advanced disease, and caution is warranted when inferring early pathogenic mechanisms that may be obscured by later compensatory and/or end-stage gene programs, as well as causal rather than correlative relationships.
    The cardiomyocyte is the primary functional unit of the heart, mediating the tightly regulated cycle of cardiac contraction and relaxation to ensure adequate pulmonary and systemic circulation. While cardiomyocytes account for ~40% of the cells in the heart, they account for ~70–80% of tissue volume and dominate RNA abundance relative to other cell types. Cardiomyocytes show transcriptional heterogeneity that corresponds to functional diversity, e.g., as noted across the chambers of the heart in their contractile properties and energetic demand10. During injury or disease states, cardiomyocytes can undergo maladaptive remodeling, displaying shifts in contractile, conductance, metabolic, apoptotic, and pro-inflammatory programs39,125. Transcriptomic profiling has verified prior findings and further uncovered unique disease-associated cardiomyocyte states that may offer insights into HF pathogenesis, identifying novel candidate therapeutic targets.
    Ventricular cardiomyocytes decline in proportion in the failing heart across diverse HF phenotypes32,35,36,39,126,129, consistent with findings of cardiomyocyte cell death32 and expansion of non-cardiomyocytes35,39,130. In contrast, in non-cardiomyopathic conditions (e.g., atrial fibrillation) overall cardiomyocyte proportions remained stable in left atrial tissue109, underscoring disease- and chamber-specific compositional alterations.
    Irrespective of compositional changes, cardiomyocytes broadly undergo dramatic transcriptional changes in disease that converge on a canonical HF signature126. Stress response and hypertrophic markers common in the failing heart include upregulation of natriuretic peptides (NPPA, NPPB), as well as shifts in structural and contraction-related genes (e.g., upregulation of MYH7, ANKRD1; and downregulation of MYH6)10,32,35,36,108,124,131–133. These effects are observed in both adult and pediatric HF populations108 and are consistent with a reversion to a fetal cardiomyocyte program106,132,134. Pathway enrichment analyses implicate changes in metabolic and sarcomere/contractile functions in HF10,126, aligning with studies linking impaired mitochondrial function to contractile impairment135,136. Additionally, inter-related dysregulated pathways include those related to apoptosis, ion channel conductance, muscle proliferation, extracellular matrix synthesis, cell adhesion, and immunomodulation/inflammation. These transcriptional changes may be linked to development of cardiac fibrosis, in conjunction with changes in fibroblast-specific gene programs, and conduction system abnormalities that accompany left ventricular dysfunction.
    Cardiomyocyte sub-populations in end stage heart failure are diverse. For example, disease-associated decreases are observed in separate clusters of MYH6+ and GRIK2+ cardiomyocytes that show enrichment in muscle contraction and metabolic transcriptional profiles. These decreases are concurrent with increases in ACTA1/ANKRD1+, NPPA/NPPB+, ADGRL3+, and BMPR1B+ cardiomyocytes distinguished by MAPK signaling, NOTCH signaling, and potassium channel regulation35. Interestingly, a NPPA+ cardiomyocyte subset was shown to express relatively low levels of contractile genes, e.g., TTN and RYR2, identifying changes in contractile properties that may underly impaired cardiac functions131. Taken together with additional findings of enriched of apoptotic and inflammatory transcriptional signatures in HF cardiomyocyte subsets129, these findings characterize unique cardiomyocyte cell states consistent with structural and functional remodeling in the dysfunctional myocardium. However, it is important to note that while discrete sub-populations of cardiomyocytes and other cell types have been described, they may represent differential cell states during a dynamic process and experimental validation is needed to confirm the presence of novel cell subtypes.
    Further linking a NPPA/ANKRD1+ cell signature to functional dysregulation of the heart, a decrease in NPPA/ANKRD1+ cardiomyocytes is associated with myocardial recovery following LVAD implant10,36, coincident with upregulation of a GATA4 transcriptional network and downregulation of an inflammatory network (STAT3, JUNB, JUND, NFKB1)36 in these same cells. These findings are consistent with prior data implicating GATA4 downregulation with LV dysfunction133. Furthermore, reversion to a non-diseased metabolic and contractile signature in cardiomyocytes was selectively observed in patients showing myocardial recovery10,36,137 while persistence of inflammatory signatures was indicative of a reduced likelihood of recovery. These experiments additionally identified novel genes (e.g., ABRA, encoding for an actin-binding protein) as potential biomarkers of recovery.
    The discovery of novel targets via single-cell approaches has subsequently led to new biological insights, elucidated by model system experiments. For example, ATRNL1, encoding a glycoprotein involved in cell adhesion and signaling primarily localized to the intercalated disc, is upregulated in atrial cardiomyocytes in atrial fibrillation109. Experiments modulating ATRNL1 expression in human embryonic stem cell-derived cardiomyocytes revealed an inverse correlation between expression levels and action potential duration as well as gene sets implicated in cardiac conduction, linking ATNL1 to an arrythmia phenotype109. Notably, ATRNL1 expression also appeared to have a cardioprotective profile by modulating metabolic and apoptotic pathways, suggesting a multifaceted role in cardiac disease109.
    Ligand-receptor analyses suggest an overall reduction of cell-cell interactions between cardiomyocytes and other cell types in the failing heart133,138,139. Notable impairments include predicted dysregulation of Eph-ephrin signaling between endothelial cells and cardiomyocytes, an effect subsequently validated with model system approaches. EPHB1, encoding the receptor for the ligand EFNB2, was shown to be selectively expressed in cardiomyocytes, dramatically downregulated in hypertrophic human heart, and inversely correlated with wall thickness in a pressure overload mouse model. Silencing of endothelial EFNB2 in vitro promoted cardiomyocyte hypertrophy, reduced contractile rates, and increased expression of stress-response genes, while exogenous EFNB2 or overexpression of cardiomyocyte EPHB1 reversed these effects133. These findings highlight the therapeutic potential for modulation of endothelial-cardiomyocyte interactions. Disease-associated changes in cell-cell signaling are also reported among cardiomyocyte subclusters132,138 and with other cell types35,108,133,138, with notable alterations in cardiomyocyte integrin signaling132,138,139, a pathway implicated in functional and structural cardiac remodeling140.
    In recent years, fibroblasts have emerged as crucial mediators of CVD, representing a novel axis for therapeutic intervention to mitigate cardiomyopathy progression34,141,142. Cardiac fibroblasts are highly plastic regulators of extracellular matrix homeostasis to maintain cardiac structural integrity. Beyond their role in cardiac scaffolding, fibroblast crosstalk with other cell types of the myocardium regulates conductance and contractile properties, and in coordination with immune cells helps modulate inflammatory tone and contribute to tissue homeostasis.
    Interstitial fibrosis is a hallmark of the failing heart, commonly observed with disease progression in cardiomyopathies, and is significantly associated with mortality143,144. Fibrosis is driven by activation and expansion of fibroblasts, in some cases to an inflammatory myofibroblast state, that results in pathological remodeling of the myocardium. In focal or replacement fibrosis, typically observed in the infarct zone following myocardial infarction, large areas of scar tissue form in response to cardiomyocyte cell death. This contrasts with interstitial fibrosis which occurs on a more microscopic scale, for example, in response to disease conditions such as pressure overload, or in areas remote to an infarct zone145. Single-cell approaches have helped corroborate and characterize transcriptomic signatures of maladaptive fibroblasts associated with heart disease and end stage HF34,141,142.
    Across multiple cohorts of cardiomyopathy (e.g., dilated and ischemic cardiomyopathies), fibroblasts demonstrate disease-associated increases in their overall proportion35,39,130. However, other studies show stability of fibroblast composition in atrial tissue in atrial fibrillation43,109, in septal tissue in HCM39,138, and in genetic cardiomyopathy (including arrhythmogenic cardiomyopathy)125. This variability may reflect sampling bias126, or inter-individual fibrosis burden (as not all individuals with cardiomyopathy exhibit myocardial fibrosis)109,146 147.
    In contrast to equivocal changes in cell composition, the vast majority of studies demonstrates that fibroblasts are a major driver of disease-associated transcriptional changes39,44,148. Major populations associated with disease include activated fibroblasts (e.g., POSTN+/FAP+39,44,130), contractile myofibroblasts, and those with inflammatory or immunomodulatory profiles. Single cell approaches have also characterized a subset of fibroblasts that show high expression of POSTN, THBS4, and CLIP without concurrent expression of myofibroblast markers (e.g., ACTA2), defining a putative disease-associated matrifibrocyte population39,148, previously identified in model systems149, and thought to maintain scar tissue in late-stage disease. Populations of activated fibroblasts upregulate genes involved in extracellular matrix (e.g., TIMPs139,148) and profibrotic remodeling (e.g., COL1A1, COL1A2, COL3A1, COL5A1, among others39,139,148), leading to an aggregate increase in extracellular matrix deposition and consequent fibrosis. TGFβ signaling is prominently implicated in disease-associated fibroblasts, the transition to myofibroblasts, and pathological cardiac remodeling39,126,150. Top upregulated genes across pathologies are linked to the TGFβ signaling pathway, suggesting potential for therapeutic interruption of this axis. For example, functional knockdown of AEBP1, a TGFβ-sensitive gene dramatically upregulated in heart failure, attenuates markers of myofibroblasts and collagen deposition, and reduces fibroblast proliferation and contractile capacity following TGFβ stimulation, suggesting an anti-fibrotic profile42,148,150.
    Disease-associated fibroblasts also exhibit enrichment in potassium channel and neuronal system pathways39, consistent with alterations in both cardiomyocyte-fibroblast133,139 and neuronal-fibroblast signaling44. Spatial and ligand-receptor analyses also identify tight coupling between POSTN+ fibroblasts and inflammatory IL1β+ or SPP1+ macrophages in ischemic cardiomyopathy32. Model system studies show that inhibiting the NLRP3 inflammasome in a mouse model of arrhythmogenic right ventricular cardiomyopathy, for example, improves cardiac function44. Further findings of antigen-presenting fibroblast subsets, and those showing an expansion of inflammatory fibroblasts expressing CXCR3 ligands that spatially colocalize with conventional dendritic cells in ICI-associated myocarditis open the possibility of a multifaceted role for immune-fibroblast interactions in myocardial and electrical injury38 124,151.
    Transcriptional changes in cardiac fibroblasts are also associated with myocardial recovery. Within cohorts of individuals undergoing LVAD implant as a potential bridge-to-myocardial-recovery strategy, fibroblast signatures were prominently associated with recovery and distinguished responders from non-responders10,36. Successful myocardial recovery associates with downregulation of the transcription factor RUNX136, involved in TGFβ signaling, and downregulation of inflammatory gene sets implicated in TNFα/NFkB signaling. These single-cell findings were validated with in situ hybridization. RUNX1 expression in fibroblasts was negatively correlated with left ventricular ejection fraction and positively associated with inflammatory signatures in macrophages. In contrast, lack of success in myocardial recovery was associated with high levels of fibroblasts with inflammatory transcriptional profiles at the time of LVAD implantation, suggesting that inflammatory-cell-fibroblast cross talk is a key mediator of functional recovery. Consistent with this interpretation, altered gene sets in disease-associated fibroblasts reflect changes in oxidative stress, MAPK signaling, cytokine signaling, the complement cascade, and neutrophil degranulation39,148,150, further implicating an inflammatory or immunomodulatory profile of fibroblast subsets in the failing heart.
    The microvascular compartment—spanning arterial, capillary, venous, endocardial, and lymphatic endothelial cells—and mural cells (e.g., pericytes and VSMCs) show substantial heterogeneity in the healthy human heart8,9. Historically, the epicardial coronaries and other large arterial systems have been the focus of study. However, the microvasculature is getting increased attention in recent years as microvascular dysfunction in a variety of CVDs is being increasingly recognized152.
    Subsets of endothelial cells have been shown to be increased in chronically failing hearts of varying etiologies35,39,125. These subpopulations show increased expression of markers of angiogenesis and altered expression of genes involved in blood vessel development39,125. Furthermore, in DCM, arterial-associated endothelial cells were enriched for interferon and MAPK signaling, capillary-associated endothelial cells were enriched for NGF signaling, and venous endothelial cells were enriched for TGFβ and MAPK signaling, among others35. On the other hand, in ischemic hearts, capillary, venous, and lymphatic endothelial cell populations were increased32. Endothelial cells were enriched for hypoxia signaling within ischemic zones and TGFβ signaling in fibrotic zones. These data are consistent with a pro-angiogenic state immediately following ischemic injury and model system data suggesting that the lymphatic system influences the myocardial inflammatory response after myocardial infarction153. Within hypertrophied hearts, interruption of the interaction between EFNB2 in endothelial cells and EPHB1, a receptor tyrosine kinase, in cardiomyocytes was implicated in cardiomyocyte hypertrophy which may drive maladaptive remodeling133. In the transplanted heart, chronic rejection often involves the microvasculature, allowing for additional information to be gleaned regarding the role these endothelial cells may play in pathologic states154. In biopsies from patients with chronic rejection, endothelial cells exhibited a number of differentially expressed genes in comparison to non-rejecting hearts with these genes enriched for pathways of angiogenesis, activation, extracellular matrix remodeling, and others37. Interestingly, due to the presence of two distinct genomes—donor and recipient—recipient-derived endothelial cells (comprising ~22% of the total endothelial cell population in that study) were enriched for markers of endothelial-to-mesenchymal transition, whereas donor-derived endothelial cells were not. Taken together, these data underscore how endothelial populations within the myocardium not only shift in composition but also adopt disease-specific transcriptomic signatures that may modulate myocardial inflammation and remodeling.
    Crosstalk between pericytes, VSMCs, and endothelial cells is directly implicated in maintaining homeostasis and coordinated responses to myocardial injury155,156. Consistent with this, following myocardial infarction, pericytes were enriched in myogenic niches involved in post-infarct remodeling32. In cardiomyopathy, pericytes increased expression of genes involved in extracellular matrix remodeling and vascular adhesion, but it is unclear whether these may be pathogenic or in response to maladaptive remodeling39,125. In patients who do develop reverse remodeling and improvement in their cardiac function with mechanical unloading, the role of pericytes remains unclear though they showed a number of genes significantly associated with a recovery signature36. Similarly, VSMCs play key roles in vascular organization following myocardial infarction32. In cardiomyopathy, there was a reduction in VSMCs and a shift in transcriptomic profiles toward phenotypes that included contractile and vascular stiffness, suggestive of impaired diastolic function in cardiomyopathy, and extracellular remodeling39,125. These changes in pericyte and VSMC gene expression profiles in a context-specific fashion during ischemia and in failing hearts may reflect their contribution to pathological remodeling.
    The immune system plays a crucial role in cardiac development157, maintaining cardiac homeostasis157, and in response to infection and injury. CCR2− macrophages, for example, have been suggested to drive fetal coronary angiogenesis158, while cardiac resident macrophages contribute to not only housekeeping functions, but also actively participate in electrical conduction through connexin 43-containing gap junction formation with cardiomyocytes159. The role of lymphoid cells remains less clear in the healthy heart and may be context-specific in response to various types of injury160.
    Myeloid cells, particularly macrophages, help maintain steady-state physiology in the healthy heart. However, in dilated cardiomyopathy (DCM), myeloid cells—monocytes, macrophages, and dendritic cells—shift toward an inflammatory phenotype35. An expansion of monocytes and dendritic cells has been observed in DCM, with enrichment for pathways involved in cytokine signaling, T cell co-stimulatory signaling, and antigen presentation, among others35. Several studies found a reduction in the composition of proliferating myeloid cells and CCR2− macrophages in cardiomyopathy35,39,125, which are embryonically-derived from yolk-sac macrophages or fetal monocytes and enriched for pathways typically involved in repair, such as those for epithelial-to-mesenchymal transition and myogenesis161,162. On the other hand, and consistent with animal data, myeloid cell populations (e.g., SPP1+ macrophages, CCL18+ macrophages) significantly increase in the acutely ischemic heart and in its post-injury response32. Inflammatory myeloid niches were surrounded by fibrotic niches opening the possibility for a fibro-myeloid interaction that promotes post-infarct myocardial fibrosis. The presence of inflammatory macrophages was negatively associated with myocardial recovery36. Considering the intimate link between inflammation and fibrosis163, interactions between CCR2+ monocytes/macrophages with cardiac fibroblasts through IL-1β signaling has been shown to mediate cardiac fibrosis by facilitating the development of FAP/POSTN+ fibroblasts34. Importantly, these human myocardial-derived single-cell multi-omic findings were validated in mouse models lacking IL-1β specifically in CCR2+ monocytes and macrophages, supporting the notion that IL-1β derived from CCR2+ monocytes/macrophages promotes fibroblast activation during the development of cardiac fibrosis. Interestingly, downregulation of RUNX1 activity in CD163+ cardiac resident macrophages, which regulates FAP/POSTN+ fibroblast development, was previously shown to be associated with increased likelihood of myocardial recovery as the RUNX1 gene regulatory network promotes cardiac fibrosis34,36,141,142. Beyond the roles macrophages play, dendritic cell-fibroblast crosstalk has also been implicated in myocardial injury in ICI-associated myocarditis38. These data highlight the importance of the monocyte/macrophage population, in particular, in both the development of myocardial disease and potential for therapeutic modulation to promote functional recovery.
    In contrast to the cells of the innate immune system, significantly less is known about the role that cells of the adaptive immune system play in the healthy human heart. However, activated T cells have been shown to drive pathological remodeling in animal models of heart failure164. Consistent with model system data, single-cell approaches have shown compositional increases in CD4+ activated T cells and NK cells in human DCM39,125. Both CD4+ and CD8+ T cells exhibited markers of activation (e.g., interferon-gamma, proinflammatory cytokines, etc.) in DCM125. As strategies to enrich for lymphocytes evolve, it is anticipated that roles of T and B lymphocytes will be further linked with mechanistic studies to delineate their roles in the pathogenesis of human CVD.

    Limitations to single-cell transcriptomics

    Despite their transformative potential, single-cell and single-nucleus RNA sequencing approaches carry several limitations that warrant consideration. Most notably in CVD, at this early stage, studies have largely reaffirmed known biology (e.g., canonical cardiac cell types or stress-responsive gene programs, etc.) rather than identifying entirely novel populations or mechanisms. This derives from the primary focus being the construction of single-cell “atlases” in health and disease. Additionally, current analyses are largely cross-sectional in design and limited in temporal resolution, making it difficult to infer dynamic processes or causality without complementary time-course or perturbation data. Thus, while single-cell atlases serve as powerful platforms for hypothesis generation, their biological interpretations must be validated through orthogonal methods. Emerging studies have begun to address these limitations by incorporating CRISPR perturbation screens165, induced pluripotent stem cell (iPSC)-derived cardiac organoids166, and in vivo lineage tracing34,167 to directly test predictions from single-cell data. While such efforts remain limited in number, they represent a crucial step toward transitioning from descriptive atlases to mechanistic insight and are discussed in greater detail below.

    Emerging complementary and supportive technologies

    Recent advances in cardiovascular single-cell transcriptomics have been accelerated via the integration of emerging modalities that extend beyond gene expression alone. These complementary technologies—including spatial transcriptomics, multi-omics, perturbational genomics, and machine learning, among others—have the potential to transform the field by enabling mechanistic discovery, functional validation, and clinical translation.
    In contrast to scRNA-seq and snRNA-seq, which require enzymatic or mechanical tissue dissociation, spatial transcriptomics technologies (e.g., 10X Genomics’ Xenium168 and Visium platforms, MERFISH169, etc.) enable the direct analysis of gene expression in tissues at near-cellular or sub-cellular resolution. These approaches offer several advantages: 1) they are compatible with formalin-fixed, paraffin-embedded (FFPE) samples, facilitating study of archived clinical specimens; 2) they avoid tissue dissociation-induced transcriptional artifacts; and 3) they allow for the contextual analysis of cellular neighborhoods, supporting direct inference of cell–cell interactions and spatially-regulated transcriptional programs. However, despite these advantages, current spatial methods face significant trade-offs. Image-based platforms (e.g., Xenium, MERFISH, etc.) offer sub-cellular resolution but are restricted to targeted gene panels that typically detect hundreds to thousands of genes, with recent advances targeting up to ~10,000 genes, while sequencing-based approaches (e.g., Visium) provide whole transcriptomic coverage but lack true single-cell resolution.
    Leveraging a combination of single-cell multi-omics and spatial transcriptomics, study of eight distinct anatomical regions of the human myocardium—including the conducting system—defined distinct cellular niches within the sinoatrial node, comprising a central region enriched for P cells, activated fibroblasts, and NGF+ glial cells and a peripheral region enriched for several immune cell types33. Importantly, spatial data mapped the physical relationships of these cell types, reinforcing inferred trans-synaptic neuro–cardiac interactions via glutamatergic signaling pathways between P cells and glia. Furthermore, spatially organized immune niches were identified within the epicardium along with previously unrecognized IgA+ and IgG+ plasma cells, suggesting possible protective roles against direct pathogens (e.g., from the lungs)33. A similar strategy delineated nine unique cellular niches within the infarcted human myocardium, corresponding to the infarct zone, border zone between infarcted and healthy tissue, and zones remote from myocardial injury32. For example, SPP1+ macrophages co-localized with POSTN+ myofibroblasts within fibrotic niches in ischemic tissue, reflecting coordinated spatial remodeling dynamics in response to ischemic injury. Building upon such cross-sectional insights, serial endomyocardial sampling of transplanted hearts during acute rejection has uncovered substantial cell-specific transcriptomic heterogeneity even within the same rejection grades masked by standard H&E staining alone170. Distinct differential gene expression patterns were observed between individuals whose cellular rejection resolved with treatment versus those with persistent rejection. Importantly, cell-type–specific expression profiles during and following acute rejection were associated with chronic rejection (cardiac allograft vasculopathy [CAV]), underscoring the potential of spatial transcriptomics to identify prognostic tissue signatures not captured by conventional pathology.
    Together, these spatially resolved studies extend single-cell atlases by anchoring transcriptional signatures within their native tissue context and enabling direct inference of cell–cell communication, lineage localization, and regional regulatory programs. While technical limitations currently constrain fully unbiased spatial analyses, they are expected to diminish with ongoing advances in resolution, sensitivity, and gene panel scalability. This tissue-centric mapping will be essential in future studies for understanding how spatial architecture guides cellular behavior in both health and disease.
    Single-cell multi-omics technologies are further revolutionizing our understanding of gene regulation and cellular identity in the human heart. Several integrative assays (e.g., 10X Genomics’ Multiome171, etc.) enable the joint profiling of RNA and chromatin accessibility (ATAC-seq) in individual nuclei. While not widely utilized yet, these platforms have been applied in landmark studies to dissect the regulatory architecture underlying cardiac development, homeostasis, and disease. Pairing snRNA-seq with snATAC-seq, the FOXP2 transcription factor was implicated in P cells within the sinoatrial node33 and linked chromatin accessibility to gene expression changes in fibroblasts and macrophages32. Combining snRNA-seq, snATAC-seq, and surface protein marker analysis (CITE-seq) across healthy, infarcted, and failing human hearts enabled the identification of distinct fibroblast activation trajectories—encompassing POSTN+ matrifibrocytes and ACTA2+ myofibroblasts—and suggested immunologic cues, including IL-1β signaling from CCR2+ macrophages, as potential upstream modulators of fibroblast state transitions. Beyond characterizing regulatory networks in health and disease, joint analysis of gene expression and chromatin accessibility has also shed light on the impact of age and sex in the healthy human heart116. Sex-biased gene regulatory programs (e.g., cholesterol metabolism, oxidative phosphorylation, etc.) and age-related increases in fibroblast senescence and lymphocyte-associated accessibility were identified, underscoring the utility of multi-omic approaches for capturing inter-individual variation in cardiac biology.
    These combination platforms allowing simultaneous multi-modality measurements within the same cell or nucleus offer a more comprehensive and nuanced framework for characterizing cardiac cell types, states, and responses to physiological or pathological stimuli. While the insights gained from multi-omic single-cell analyses remain largely hypothesis-generating, they provide a foundational basis for identifying candidate regulatory mechanisms and transcriptional networks that merit functional validation. As throughput, resolution, and modality integration continue to improve, multi-omic technologies are poised to play a central role in uncovering the complex molecular architecture of cardiovascular health and disease.
    Excitingly, single cell partitioning and spatial transcriptomics approaches are being adapted172–174 for compatibility with long-read sequencing (e.g., PacBio, Oxford Nanopore Technologies) enabling capture of full-length transcripts and detection of transcript isoforms. Indeed, isoform detection is an emerging priority in multiple fields including CVD, with findings in bulk tissues indicating that alterations to RNA splicing are associated with HF175,176. Application of long-read sequencing in the mouse brain demonstrates spatial specificity of isoform expression across distinct regions177, while in bulk cardiac tissues long-read sequencing has revealed novel alternative transcripts validated by protein assay178 and shifts in isoform expression of sarcomere genes in HF179. While findings in bulk tissues demonstrate the ability of long-read sequencing to uncover disease-associated changes in isoform expression, future experiments using these platforms paired with single-cell approaches show promise to identify the diversity of spliced transcripts in the healthy and diseased heart at cellular resolution, Challenges using long-read sequencing for single cell applications persist, with relatively high intrinsic error rates particularly in early long-read sequencing platforms hindering cell and molecular barcode assignments, though hybrid approaches using simultaneous application of short read sequencing facilitate barcode assignments whereas recent advances (e.g., duplex sequencing by ONT or high-fidelity reads by PacBio) offer significantly reduced error rates. The number of total bases sequenced in long-read approaches requires higher sequencing throughput, leading to relatively higher costs and limiting the number of cells/nuclei typically sequenced.
    Single-cell transcriptomic studies are powerful tools for identifying putative cell states, signaling pathways, and disease-associated cell compositional changes or transcriptional programs. However, these observations remain largely hypothesis-generating and require functional validation to determine causality. Recent studies have increasingly coupled single-cell analyses with perturbational and lineage-tracing approaches, such as CRISPR-based genetic screens, to test whether transcriptionally defined states drive phenotypes.
    In human heart failure, following identification of genes via snRNA-seq whose expression patterns changed during fibroblast activation, targeted CRISPR knockout in primary human fibroblasts demonstrated that several of these genes (e.g., PRELP, JAZF1, COL22A1, etc.) were required for TGFβ1-induced myofibroblast activation, functionally linking transcriptional state to fibrotic remodeling39. A similar strategy was used to test spatially predicted developmental signals in the fetal heart. Following the identification of PLXN-SEMA signaling in the myocardium using MERFISH and scRNA-seq, human pluripotent stem cell-derived cardiac organoids and conditional mouse knockout models identified the role of SEMA3C in guiding ventricular organization121. Extending beyond in vitro validation, multi-omic single-cell findings have also been validated in mouse models using lineage tracing, confirming that FAP+ fibroblast precursors preferentially give rise to matrifibrocytes and providing in vivo support for transcriptionally inferred trajectories34. While most cardiac single-cell studies to date remain descriptive atlases, the growing emphasis on functional validation is critical for transitioning from hypothesis generation to providing mechanistic insight.
    Artificial intelligence (AI) is rapidly reshaping clinical cardiology. Deep learning models trained on large-scale electrocardiogram and echocardiography datasets are now capable of detecting structural heart disease, valvular abnormalities, and even rare conditions like cardiac amyloidosis with high diagnostic accuracy180–183.
    Similar machine learning (ML) approaches are beginning to be applied to single-cell and multi-omic datasets, where the complexity and dimensionality of the data often exceed what traditional methods can readily interpret. ML-based tools, such as autoencoder-based pipelines, have demonstrated improved resolution of cardiomyocyte subtypes and demonstrated the role of HIPPO/YAP and MAPK signaling pathways in cardiomyocyte proliferation. Other tools such as scVI85 (variational inference), which leverages deep neural networks, are increasingly used to correct for batch effects. More recently, foundation models such as Geneformer184 and scGPT185, which are pre-trained on tens of millions of single-cell multi-omic profiles, have shown promise in tasks such as cell-type annotation, transcriptional trajectory modeling, and gene perturbation inference. In cardiovascular research, these models have been used to suggest candidate regulators of cardiomyopathy—recapitulating known biologically-relevant genes that associate with hypertrophic and dilated cardiomyopathies—and nominate candidate therapeutic targets through in silico perturbation experiments184. Importantly, CRISPR knockout of GSN and PLN—targets predicted to improve contractility with therapeutic modulation in dilated cardiomyopathy—using iPSC-derived cardiac organoids harboring titin truncating variants (TTN+/−; a common genetic cause of dilated cardiomyopathy186,187) significantly improved contraction in these organoids, validating predicting genes184. While findings from current applications continue to require orthogonal validation, they demonstrate the potential of AI-based approaches to support hypothesis generation in settings where direct experimental data may be limited. As pre-trained foundation models, such as those described, become more accessible, these approaches may enable more generalizable and scalable analyses particularly when fine-tuned on focused cardiovascular datasets. However, their utility depends on careful benchmarking, task-specific training, and experimental validation to ensure biological relevance.

    Clinical translation and future directions

    Single-cell technologies have the potential to significantly advance precision approaches to CVD through acceleration of biomarker discovery and improving our understanding of cardiac biology (Figure 4). Although they have already catalyzed major discoveries in oncology, these approaches remain relatively underexplored withing the CVD space. Until recently, a major limitation to myocardium-specific single-cell approaches in clinical cardiology has been that endomyocardial biopsy (EMB) is performed infrequently due to risk of adverse events (e.g., ventricular perforation, pericardial effusion, death, etc.)188. Consequently, it is only recommended in highly context-specific scenarios (e.g., fulminant myocarditis, inflammatory heart disease, rejection surveillance following heart transplant, etc.)189. However, with the ability to isolate nuclei from snap-frozen cardiac tissue, as well as newly-developed methods to isolate nuclei from archived FFPE samples collected during routine clinical care opens novel avenues to understand CVD-relevant biology directly at the level of human myocardium190–192.
    To date, most studies of the myocardium have prioritized end-stage CVD34–36,39,108,125. While these studies have been highly informative about the cellular landscape of healthy and diseased tissue and have been critical to technological development for single cell methods, data suggest a common transcriptomic signature—regardless of etiology of heart failure—at this late stage. The limited therapeutic options at this late stage–heart transplantation or left ventricular assist device implantation—underscore the urgent need to investigate CVD at earlier stages of disease, which could be more informative in prioritizing targets for novel therapeutic strategies to modify disease trajectory or for identifying disease-specific biomarkers.
    Integrating collection of biopsies and peripheral blood into prospective clinical studies may shed light on CVD progression and heterogenous therapeutic responses. Analogous strategies undertaken in clinical oncology have already generated clinically-actionable strategies to inform therapy and monitor response5,7,193,194. A similar model could be adapted for CVD trials, where a subset of trial participants undergo simultaneous EMB and peripheral blood phenotyping. With appropriate informed consent, some studies have already enrolled individuals with early-stage CVD38,195 toward collection of research EMBs during clinically-indicated procedures. While oncology has found the most success in employing single-cell approaches to samples obtained during clinical trials to date193,194,196, their prior experience helps inform trial design in CVD. Such designs could reveal how molecular phenotypes evolve in response to standard-of-care or experimental therapies, potentially accelerating the identification of drug targets and predictive biomarkers.
    Interestingly, one area that may benefit from longitudinal EMBs obtained from the same patient is in heart transplantation, where EMB is standard of care for the diagnosis of acute rejection197,198. However, tissue size of EMBs (~3–5 mg, typically) still poses a technical challenge in isolating nuclei or cells and few groups have employed single-cell genomics approaches in this population37,38. As immune-cardiac interactions and the role of donor- versus recipient-derived cells’ contribution to acute and chronic rejection in heart transplantation are incompletely understood, the routine access to EMBs offers a powerful opportunity to accelerate knowledge in transplant biology. Furthermore, as immunohistopathology is used to diagnose acute rejection but does not fully explain the breadth of clinical presentation, interrogating the molecular underpinnings of acute rejection may inform novel molecular phenotyping of rejection and the development of predictive models of therapeutic response. Importantly, serial sampling of tissue from the same individual represents a unique opportunity to move beyond cross-sectional study designs (which imperfectly infer trajectories) to directly assess cell-specific transcriptomic dynamics over the course of disease development and progression. Such approaches may not only have the potential to substantially improve post-transplant outcomes but also yield insights relevant to the broader areas of cardio-immunology and cardio-oncology, xenotransplantation, and inflammatory heart disease.
    Direct study of biobanked snap-frozen cardiac tissue or clinically-archived FFPE samples may improve our understanding of earlier disease pathogenesis. Specifically, the latter takes advantage of previously-collected cardiac specimens obtained during routine clinical care (e.g., during suspected inflammatory heart disease, etc.) to improve understanding of early mechanisms of CVD. Further, the advent of these technologies may inform institutional protocols to be revised in order to prospectively biobank typically difficult-to-obtain cardiac specimens from individuals undergoing EMB (e.g., early during clinical course) or specimens not suitable for donation to accelerate CVD research. Beyond this, studies have suggested that leveraging single-cell genomics can accelerate drug discovery and repurposing to a similar scale as that achieved by human genetics199–203. Furthermore, linking tissue-specific gene expression to a more accessible biospecimen (e.g., peripheral blood) in the same patient may identify biomarkers of disease progression while obviating the need for invasive biopsies204.
    However, the potential for translating single-cell genomics into clinical practice remains challenging due to several barriers. Specific to CVD, indications for biopsy and the technical challenges of processing and analyzing cardiac tissue make standardization across centers difficult for CLIA style clinical laboratory grade testing. Ethical considerations are central for tissue procurement and informed consent must explicitly distinguish research conducted on residual tissue versus procurement of additional cores. Despite technological advances and emerging multiplexing strategies that allow sample pooling to reduce cost per cell and batch variability58–60, the high per-sample cost of current single-cell assays remains a major barrier for widespread clinical implementation. Additionally, in comparison to bulk omics methods, most single-cell studies remain underpowered, with relatively small sample sizes that reduce generalizability and the discovery of robust disease-associated biomarkers. Moreover, differences in tissue handling (e.g., fresh vs. frozen, single-cell vs. single-nucleus, etc.), sequencing platforms, and processing pipelines introduce technical noise that can obscure biological signals and hinder cross-institutional (or even, cross-laboratory) comparability. Systematic sources of variability, such as dissociation-induced gene expression artifacts, differential cell-type recovery, and others, must be controlled through harmonized wet lab and computational standard operating procedures. Moving toward clinical validity will require 1) large, deeply phenotyped cohorts; 2) automated, reproducible computational pipelines that incorporate appropriate statistical frameworks (e.g., pseudobulk modeling, compositional analysis, etc.) and are widely accessible; and 3) rigorous confirmation of findings or hypotheses generated from single-cell analyses through orthogonal validation strategies (e.g., in vivo lineage tracing, spatial transcriptomics, multimodal profiling, and functional perturbations). Furthermore, large-animal (e.g., porcine205,206) platforms that enable catheter-based delivery, serial imaging/hemodynamics, safety/tolerability assessment, and repeated myocardial sampling for cross-species state mapping may complement human iPSC-derived cardiomyocytes, engineered heart tissues, and heart-on-a-chip systems to provide controllable settings to prioritize causal targets and dosing regimens prior to clinical qualification. Finally, safeguards must be prioritized for data governance as studies suggest that even raw counts matrices can compromise privacy207. While recent work suggests that federated learning and other approaches minimize risk of data leakage, safeguards must continually evolve to maximize sustainable protection208,209. Despite these barriers, ongoing efforts to develop harmonized single-cell reference atlases, population-scale initiatives, and validated ML/AI-based frameworks offer a potential path toward clinical application.
    The integration of single-cell genomics within clinical cardiology has moved from a theoretical possibility to a rapidly emerging reality. Enhanced methods for isolating nuclei from small or archival tissue samples, combined with the precedent of EMB in select patient groups, open a path to systematically map the molecular underpinnings of CVD in humans long before irreversible cardiac remodeling has occurred. Moving forward, embedding single-cell analyses into well-designed clinical trials can accelerate biomarker discovery, refine drug targeting, and ultimately shift the therapeutic paradigm to earlier, more effective interventions.
    KADr. Kaushik Amancherla

    Dr. Kaushik Amancherla

    Cardiology · Advanced Heart Failure & Transplantation
    Vanderbilt University Medical Center
    kaushik