Single-cell transcriptomics for the assessment of cardiac disease

Abstract

Cardiovascular disease is the leading cause of death globally. An advanced understanding of cardiovascular disease mechanisms is required to improve therapeutic strategies and patient risk stratification. State-of-the-art, large-scale, single-cell and single-nucleus transcriptomics facilitate the exploration of the cardiac cellular landscape at an unprecedented level, beyond its descriptive features, and can further our understanding of the mechanisms of disease and guide functional studies. In this Review, we provide an overview of the technical challenges in the experimental design of single-cell and single-nucleus transcriptomics studies, as well as a discussion of the type of inferences that can be made from the data derived from these studies. Furthermore, we describe novel findings derived from transcriptomics studies for each major cardiac cell type in both health and disease, and from development to adulthood. This Review also provides a guide to interpreting the exhaustive list of newly identified cardiac cell types and states, and highlights the consensus and discordances in annotation, indicating an urgent need for standardization. We describe advanced applications such as integration of single-cell data with spatial transcriptomics to map genes and cells on tissue and define cellular microenvironments that regulate homeostasis and disease progression. Finally, we discuss current and future translational and clinical implications of novel transcriptomics approaches, and provide an outlook of how these technologies will change the way we diagnose and treat heart disease.

Key points

• A good experimental design requires a matching of the protocol workflow to the cells of interest and scientific goals.

• The generation of reference heart cell atlases and standardized annotations is necessary for cross-study comparisons and accurate data interpretation.

• Emerging disease gene signatures reveal cell-state-specific changes, which will facilitate the generation of novel putative biomarkers and therapeutic targets.

• The definition of cellular microenvironments requires deconvolution with spatial and multi-omics approaches.

• The density of information from these novel omics approaches will contribute to the design of computational models to predict disease, stratify patients and facilitate drug discovery.

Introduction

The human heart is a dynamic organ composed of four morphologically and functionally distinct chambers, as well as highly specialized subdomains including the conduction system and valvular apparatus, all working in synchrony due to the perfectly coordinated actions of billions of cells. In each chamber and anatomical subdomain, diverse stimuli converge on the local cells, priming gene expression patterns that drive phenotypic and functional adaptations, cumulatively determining organ function. Indeed, this fine cellular orchestration facilitates not only continuous contraction and relaxation of cardiomyocytes, but also the effective responses to haemodynamic changes during prenatal and postnatal development, and adulthood.

During embryonic development, the cellular progenies of first and second heart fields are primed in utero, and exposure to haemodynamic changes contributes to their gene expression programmes and maturation after birth^1,2. In disease states, perturbations to the normal cellular repertoire and microenvironments caused by noxious stimuli, such as mechanical, electrical, chemical or ischaemic damage, can disrupt the transcriptional landscape^3,4,5,6. Gene expression is not merely an epiphenomenon, but rather an essential step in the implementation and amplification of pathogenetic circuits^3,7. Therefore, studying cellular transcriptional signatures is essential for gaining a robust understanding of organ and tissue function.

Cardiovascular disease remains the leading cause of death globally⁸. Ischaemic heart disease, more specifically, is the foremost cause of death in both men and women, despite robust efforts in the field of preventative cardiology. Furthermore, heart failure (HF), a cardiac functional impairment secondary to many aetiologies, is a rising global epidemic and, despite a growing number of treatments^9,10,11, transplantation remains the only definitive cure. Therefore, an urgent need exists for novel effective and targeted therapies with more precise risk stratification, which necessitates a deeper understanding of the underlying molecular mechanisms driving the progression of cardiac disease.

Single-cell omics technologies, and especially transcriptomics, have revolutionized the way we investigate organs and organisms, allowing an unprecedented level of resolution in the assessment of cell demographics during both health and disease¹². Single-cell transcriptomics provides information on gene expression prevalence and heterogeneity as well as co-expression of genes at the individual cell level to facilitate a cell-centric outlook. This approach involves the definition of novel cell markers and transcriptional signatures to delineate cell types (well-established cellular lineages) and cell states (encompassing subtypes or transient, functional cellular transcriptomics signatures) with high accuracy, either manually or using automated computational pipelines^13,14,15. Analysis of gene sets enriched in cell types or states allows functional inferences on intracellular regulatory networks and on intercellular pathways across specific cells by focusing on genes encoding ligand–receptor pairs^16,17. Such analyses highlight the processes underlying coordinated cellular communication that are otherwise masked in bulk analysis.

In this Review, we discuss the latest findings obtained using single-cell transcriptomics that advance our understanding of cardiac biology, development and disease. We explain how we can maximize the implementation of these technologies to study cardiac disease and provide an outlook on multimodal integration with spatial transcriptomics and epigenetics.

Successes and challenges in experimental design

The implementation of single-cell transcriptomics consists of several steps, from tissue processing for cell isolation to single-cell capture, reverse transcription, complementary DNA amplification and library construction, which are followed by sequencing and computational analysis^12,18,19. In this section, we describe the most successful experimental designs utilized to profile cardiac cells, discuss their advantages and disadvantages, and highlight the need for careful experimental planning and validation options.

Cell types of interest dictate protocol choice

Most single-cell capturing methods including flow cytometry, microfluidics and microdroplet-based systems have an upper size limit for cells of ~25–40 µm^20,21. Therefore, if the study focuses on cardiac stromal, vascular and immune cell compartments with no major size limitations, these capturing techniques can be used to select single cells for RNA sequencing^{22,23,24,25,26} (Fig. 1). However, in the mammalian adult heart, cardiomyocytes pose a major challenge because they are large, rod-shaped cells of approximatively 20 µm in width and 100 µm in length²⁷. Micromanipulation or laser capture of cardiomyocytes is possible, but these approaches are limited by low throughput and are operator-dependent⁷. Flow cytometry sorting of cardiomyocytes has been reported, although the length of sorted cardiomyocytes was ~50 µm, suggesting a bias for smaller cells or contracted cells during processing^28,29. The sorting of cardiomyocytes with preserved RNA quality and function has been achieved using a specialized fluorescence-activated cell sorting (FACS) instrument with a large 500 µm nozzle. However, this type of instrument is not routinely available. Furthermore, the complexity and length of the cardiomyocyte isolation process can potentially affect the resulting transcriptional signature³⁰. Single-cell RNA sequencing (scRNA-seq) technologies are limited not only by inefficiency in capturing adult cardiomyocytes, but also by the fact that adipocytes cannot be captured using single-cell isolation protocols. Moreover, an under-representation of cells such as pericytes and fibroblasts is also common in scRNA-seq^27,31. Therefore, if the aim of the study is to assess and capture all cardiac cell types, profiling single nuclei by using single-nucleus RNA sequencing (snRNA-seq) is the most effective method^{27,32,33,34,35,36,37}. Importantly, snRNA-seq generates largely overlapping molecular signatures with scRNA-seq in the adult human heart²⁷ and during cardiomyocyte differentiation in vitro³⁸. Furthermore, unlike snRNA-seq, single-cell isolation for scRNAseq requires fresh tissue, and enzymatic and mechanical methods used for single-cell dissociation can risk inducing stress-related transcriptional artefacts, especially when performed at 37 °C for >1 h³⁹. The testing of RNA quality from a fresh or frozen tissue sample is recommended. Of note, there are also some caveats in the use of snRNA-seq. snRNA-seq is not ideal for cells such as neurons, in which crucial mRNA trafficking along the axon to the synapse would be missed⁴⁰. Likewise, with snRNA-seq, data from binucleated or multinucleated cardiomyocyte nuclei and data from single-nucleated cardiomyocyte nuclei cannot be distinguished. Additionally, under-representation of endothelial cells in snRNA-seq studies has been reported, which varies according to the protocol used^27,31. Taken together, snRNA-seq is an effective tool for obtaining comprehensive data from cardiac tissue, and its integration with scRNA-seq maximizes the coverage of cell types and states²⁷.

Fig. 1: Workflow for single-cell and single-nucleus transcriptomics.

The design and execution of single-cell experiments involves a multistep process that requires careful planning. The assessment of cardiomyocytes is challenging given that they are too large to be selected as single cells using fluorescence-activated cell sorting (FACS) or most microfluidics-based methods. If all cardiac cell types need to be analysed, the isolation of single nuclei from frozen samples is currently the most commonly used approach, given that this technique allows capture of all known cardiac cell types at the highest throughput. Nucleus isolation requires enzymatic and mechanical dissociation, often with the use of a Dounce homogenizer to help release the nuclei. Before proceeding to library preparation, nuclei can be purified by filtering and then by FACS, or less-stringent protocols include differential centrifugation before filtering and alternatively density gradient centrifugation. If only non-cardiomyocyte cells are needed, mechanical and enzymatic tissue dissociation of fresh samples allows the recovery of stromal and interstitial cells. For rare cell types of interest, the potential enrichment for specific populations can be achieved via antibody labelling of cell surface markers, followed by magnetic bead-based enrichment or flow cytometry sorting. These enriching methods can also be used to select cells expressing fluorescent proteins, such as in lineage-tracing experiments. After the isolation and purification steps, single cells and nuclei are captured, labelled (by barcoding) and incorporated into a library preparation using a variety of single-cell platforms: droplet-based approaches, such as the Chromium Controller from 10× Genomics, which have a high throughput; nanowell-based methods, such as: the ICELL8 instrument from Takara Bio, which allows the selection of a wide range of cell sizes with low-to-medium throughput; microfluidics approaches, such as the Fluidigm C1 platform; multiwell plate-based protocols, such as Smart-seq3xpress; and methods based on multiple rounds of splitting and pooling of cells, such as SPLiT-seq, which allow the barcoding of single cells without the need for physical separation. scRNA-seq, single-cell RNA sequencing; snRNA-seq, single-nucleus RNA sequencing.

Purification and enrichment

The assessment of large numbers of cells that are present at low frequency, such as immune cells, requires antibody-based approaches for enrichment by flow cytometry or immunomagnetic beads. In animal models, specific cell lineages or rare cells, including epicardial cells, can be enriched using methods involving cell tracers such as genetic fate mapping, which involves insertion of a heritable gene encoding a fluorescent protein^41,42,43. After isolation, the nuclei are purified to remove the cytoplasmic debris resulting from deliberate disruption of the cytoplasm to release the nucleus. This filtering step can be performed by FACS^{27,37,44,45,46}, differential centrifugation or straining^33,47, or by density gradient separation⁴⁸ (Fig. 1). The effectiveness of differential centrifugation depends on pipetting skills, and the use of filters with a pore size as small as 10 µm can introduce bias⁴⁷. Indeed, the nuclei of mammalian cells have an estimated size of 8–12 µm³⁷, or even larger under certain conditions such as in hypertrophic cardiomyocytes⁴⁹. The use of a strainer alone can let through small tissue fragments and debris that can increase background noise³³. Density gradient centrifugation is also highly operator-dependent and is unsuitable for small volumes with few nuclei⁵⁰. Although FACS purification requires specialized instruments and expertise, this approach ensures good purification of nuclei given that the nozzles are typically ≥70 µm^32,33,47. Visual inspection of nuclei to confirm nuclear membrane integrity and absence of blebbing and tissue debris is recommended after purification³⁷. Of note, a FACS protocol based on immunolabelling of the cardiomyocyte-specific protein pericentriolar material 1 and Hoechst staining for DNA content allow the isolation and profiling of diploid versus tetraploid nuclei, a key step in defining the transcriptional changes occurring in neonatal proliferating cardiomyocytes^48,51. However, certain dyes used for nuclei sorting can intercalate between DNA base pairs and disrupt chromatin structure⁵². Therefore, when downstream analysis includes single-nucleus assay for transposase-accessible chromatin sequencing (snATAC-seq), a recommended dye such as 7-aminoactinomycin D should be used. In addition, during FACS, nuclei shear stress, high hydrodynamic pressure and osmotic changes could induce chromatin rearrangement⁵³. By contrast, these effects are unlikely to influence gene expression when the nuclei are isolated from frozen tissue and experiments are performed at 4 °C, which prevents the activation of transcription^{27,37,44,45,46}.

Available single-cell or single-nucleus platforms

After choosing the best protocol for cell and nucleus isolation, the next crucial decision is the optimal method for single-cell and single-nucleus capture and library construction^20,54,55 (Fig. 1). Microfluidic systems (such as the Fluidigm C1 platform) were among the first systems used to study heart cells; however, a low-to-medium throughput and high costs have limited their use^20,23. In droplet-based platforms (such as 10× Genomics and Drop-seq technologies), thousands of isolated single cells or nuclei are moved through advanced microfluidic devices, where they are individually partitioned with uniquely barcoded beads into nanolitre-sized gel emulsions^56,57. These platforms allow the capture of typically 5,000–10,000 cells or nuclei per sample with low costs, enabling a wide representation of cell populations, including rare cell types. Technologies based on full-length transcriptome sequencing, such as Smart-seq3, allow the characterization of transcript isoforms facilitating the detection of a larger number of transcripts per cell than droplet-based approaches, but at higher costs per cell and with lower throughput⁵⁸. However, protocol updates in the past year (Smart-seq3xpress) have substantially increased the throughput while maintaining high sensitivity⁵⁹. Nanowell-based technologies include ICELL8 (Takara), a mid-throughput platform that combines imaging and dispensing of single cells into nanowells to capture hundreds of cells with a wide range of sizes, including cardiomyocytes^60,61,62. Finally, SPLiT-seq is an alternative affordable method based on combinatorial barcoding that does not require single-cell capture, is compatible with fixed cells and nuclei, and can be used for large cardiomyocytes⁶³.

Capturing anatomical diversity

Designing cardiac single-cell or single-nucleus transcriptomics experiments requires careful consideration of the multiple (sub)anatomical regions of the heart, which consists of highly specialized structures with a diverse cellular composition. The most comprehensive human heart reference cell atlas involved the collection and analysis of six anatomical regions: the right ventricle (RV) and left ventricle (LV) free wall, apex, interventricular septum, and right and left atria²⁷. According to snRNA-seq data, the atria and ventricles of donor hearts have different cellular compositions, with an inverse correlation between the proportion of fibroblasts and cardiomyocytes, in accordance with the role of the ventricles as the primary pumping chambers²⁷. Analysis of the aortic valve and aorta was challenging, owing to the paucicellular nature of the tissue and its richness in extracellular matrix (ECM)^64,65. Although some studies have explored anatomical differences during mouse development to define progenitor cells beyond the known first and second heart fields^23,41,42,66, an unmet need exists for a systematic analysis of chamber differences in the adult mouse heart. Indeed, analyses with higher precision and broader spectrum of anatomical regions are needed for consistent comparisons across studies, species and disease phenotypes.

Power calculations

Two additional aspects are crucial in the design of scRNA-seq and snRNA-seq experiments: the number of single cells or nuclei to be analysed per sample and the depth of sequencing, which influences the number of genes obtained per cell. Most experiments are designed to obtain data from thousands of single cells with relatively shallow sequencing, or from hundreds of cells with deeper sequencing. The choice of approach can be qualitatively assessed in exploratory studies or on the basis of emerging tools for power calculations. The web-based single-cell one-sided probability interactive tool (SCOPIT) estimates the necessary number of cells to be sequenced to resolve cell types present at different frequencies⁶⁷, whereas the statistical framework scPower models the relationship between the number of cells per individual, sequencing depth, sample size and power of differentially expressed genes within cell types to compare a multitude of experimental designs and to optimize the design within a limited budget⁶⁸. In general, the sequencing of a large number of cells at a lower depth leads to higher power compared with sequencing fewer cells at a higher depth⁶⁸. Nevertheless, deeper sequencing of hundreds of cells can be useful to characterize prospectively sorted rare populations. Regarding biological replicates, proof-of-feasibility experiments have largely been conducted with a low number of replicates^22,26, but now the field has reached a phase in which variability can be predicted for both animal and human studies. In human cardiomyopathies involving pathogenic variants affecting the same gene, a cohort of five individuals has been shown to be sufficient to highlight statistically significant differences in gene expression and other related parameters, including cellular composition⁶⁹.

The importance of metadata

For best practice, to reduce the risk of technical bias, studies should, if possible, incorporate cells or nuclei obtained with the same protocol, especially when performing comparisons across different treatment groups or diseases and controls. Indeed, the various protocols for the isolation and purification of cells and nuclei can have different effects on the proportions of the types and states of cells retrieved, as well as on their transcriptional and epigenetic signatures. Moreover, substantial structural changes in pathological tissue or analysis of different anatomical regions of the heart, such as highly fibrotic tissue, could affect the release of certain cell types, even when using the same protocol. Therefore, reporting technical metadata and histopathological evaluation of the tissue microstructure for each sample is important to facilitate accurate data interpretation and integration across studies. Consequently, cell annotation and compositional analysis need to be evaluated in the context of the experimental design, and require validation with multiple platforms, including high-resolution spatial transcriptomics.

Overview of the computational workflow

Computational analysis of scRNA-seq or snRNA-seq data is a complex multistep process that requires specialized expertise (Fig. 2). In this section, we provide an overview of the crucial steps⁷⁰.

Fig. 2: Computational analysis workflow.

After the sequencing of libraries, the overall bioinformatics pipeline is largely the same, regardless of which sampling procedure was chosen. After aligning the reads to a reference genome, comprehensive single-cell packages (such as Scanpy and Seurat) are used for quality control, batch correction and data integration. A key step in this process is the clustering and annotation of cell types and states, which is performed on the basis of the expression of known marker genes and the interpretation of the transcriptional signatures of each cluster. Of note, deep learning-based approaches are emerging for automated annotation. Comprehensive annotation facilitates increased resolution in complex downstream analyses, such as: compositional analysis (the proportion of various cell types present in a tissue across various conditions); differential expression analysis between different diseases or treatments; and inferences on intercellular communication on the basis of the expression of genes encoding ligand receptors in different cell types and states. FC, fold change; SMCs, smooth muscle cells.

Quality control and data integration

Analysis pipelines for processing raw data, such as Cell Ranger⁵⁷, SEQC⁷¹ and zUMIs⁷², perform initial quality checks⁷³ on the sequencing reads, demultiplex data by assigning reads to their cellular barcodes and mRNA molecules of origin, and facilitate genome alignment and quantification. In droplet-based scRNA-seq and snRNA-seq analyses, the degree of contaminating ambient transcripts released into the cell or nucleus suspension during tissue dissociation can vary according to the dominant cell types and might lead to misinterpretation of the results. In the human heart, cardiomyocyte nuclei are the major contributors to ambient contamination. Software tools, such as CellBender, SoupX and DecontX, can minimize technical artefacts in data^74,75,76. For example, CellBender can estimate ambient RNA from empty droplets and correct the expression metrics by removing counts related to ambient RNA molecules and even random barcode swapping^74,75,76.

Standard quality control includes establishing minimum and maximum numbers of reads and genes per cell or nucleus, and determining a threshold for the highest percentage of genes encoding ribosomal proteins and mitochondria per cell or nucleus, above which a cell is defined as poor quality or unhealthy⁷³. Single-cell analysis of cardiomyocytes needs to take into consideration the high proportion of mitochondria normally present in cardiomyocytes compared with other cells⁷⁷. Two or more cells or nuclei that are attached, captured within the same droplet or microwell, and represented by the same barcode can generate a hybrid transcriptome, an artefact that violates the fundamental principle of single-cell technology and results in incorrect inferences. Doublet or multiplet detection is best performed using unbiased methods such as Scrublet and SOLO, which involve simulation of doublets or multiplets to create a training set for a machine learning classifier^70,78,79. Correct and precise identification of doublets is necessary to avoid the risk of confusing artefactual chimeric hybrids derived from two (or more) cells adhered together with transitional cell states.

To overcome batch effects and unwanted technical variation while retaining biological differences, multiple integration methods can be applied^80,81. The assembly of one of the largest human donor heart cell atlases involved the successful integration of data from single nuclei and cells from 14 donor hearts, in which differences between ventricles and atria were retained, as well as left and right specificities²⁷. Successful data integration was also observed in two studies of human hearts from individuals with HF that included analyses of >800,000 nuclei^47,69.

Defining cell types and states

The annotation of cell types and states is a complex task that is necessary for data interpretation. The major cardiovascular cell types include cardiomyocytes, fibroblasts, endothelial cells, smooth muscle cells (SMCs), pericytes, immune cells, neuronal and glial cells, and adipocytes. Most of these cell types have multiple identifiable cell states with anatomical specificities^{22,23,24,27,33,82}. Cell annotation requires unbiased clustering and the analysis of the expression of known marker genes and novel gene signatures. Unbiased clustering is a key step, the results of which depend on algorithm resolution and the number of droplets analysed: a high resolution will determine a high number of cell states or types, whereas lower granularity will be obtained with lower resolution. Each study will apply a given resolution with a subjective final decision that influences the results, which together with the effectiveness of data integration, doublet exclusion and ambient contamination subtraction can affect the accuracy of comparisons across studies. Computational tools that can be used to estimate the best resolution include SCCAF⁸³ and MultiK⁸⁴. However, it is imperative to assess whether differentially expressed genes across the various clusters are clearly defined and are biologically meaningful. The annotation of the identified cell clusters (each of which represents a cell type or state) is mostly performed manually on the basis of transcriptional signatures. However, owing to the availability of open reference data, label transfer approaches are possible for some tissue systems such as immune cells. CellTypist and scNym can automatically annotate cell types or states in a query dataset by mapping them onto a reference^13,85. The creation of databases and collaborative initiatives will be needed to form a consensus on the optimal approach to the annotation of cardiac cell types and states for accurate comparisons across studies. Furthermore, the use of novel deep learning strategies such as scArches, which is based on transfer learning, enables efficient building and sharing of reference atlasing data and to retain disease variation when mapping to a health reference, as shown for coronavirus disease 2019 (COVID-19) datasets⁸⁶. Such approaches will be key to ensuring the efficient use of the human cardiac reference atlas to improve our understanding of disease and to create model organism atlases that will facilitate functional validations.

Integration of multimodal omics approaches

The integration of data across multiple modalities such as scRNA-seq and snRNA-seq, ATAC-seq or proteomics can generate standardized cell state labels^87,88,89. Computational tools such as cell2location and Giotto allow mapping of newly defined cell types and states onto the physical 2D space by combined analysis of scRNA-seq or snRNA-seq data with spatial transcriptomics (analysis of gene expression on tissue sections)^27,90,91,92. This integration is needed because of the intrinsic nature of current spatial transcriptomics technologies that are based on the analysis of microtiles of tissue typically encompassing 5–15 cells, although this number varies depending on the size of the tile and the size of the cells captured, resulting in microbulk gene expression data. Integration tools allow the deconvolution of this microbulk information and the mapping of specific cells in space, which guides the definition of cellular niches. One typical spatial transcriptomics method, such as the commercially available Visium by 10x Genomics, is based on positioning tissue samples on slides covered with unique barcoded mRNA-binding oligonucleotides, which facilitates the capture of RNA from the tissue with high spatial resolution (protocols are available for frozen sections and formalin-fixed paraffin-embedded (FFPE) sections)^93,94. Alternatively, individual RNA molecules can be directly profiled using the Nanostring GeoMx Digital Spatial Profiler, which assigns fluorescently labelled barcoded probes to genes of interest that are hybridized on the tissue and subsequently counted by a computerized optical lens without the need for amplification (compatible with FFPE sections)⁹⁵. Non-commercially available protocols such as Slide-seq have also been developed, which involve the use of barcodes to capture RNA with a resolution of 10 µm⁹⁴.

Together, the assembly of large atlases from publicly available datasets representing tens to hundreds of individuals will contribute to the definition of consensus cardiac cellular maps by characterizing common signatures across multiple studies⁹⁶. Larger multi-organ studies such as the Tabula Sapiens⁹⁷ overcame difficulties of integrating data from diverse organs, including the heart, paving the way to a new series of studies focusing on the definition of shared and organ-specific molecular signatures of cells present across the whole body, such as fibroblasts, vascular cells and immune cells⁹⁸. This step is crucial for the definition of putative organ-specific and tissue-specific cellular therapeutic targets. Likewise, the computational analysis of scRNA-seq and snRNA-seq data and its integration with other modalities is a fundamental and demanding step that requires interdisciplinary expertise for accurate hypothesis generation and functional inferences.

Cardiomyocyte profiling in health and disease

Isolating large numbers of rod-shaped cardiomyocytes from cardiac tissue is inherently difficult and requires manual micropipetting or dispenser approaches using a large nozzle size. Consequently, only a few studies have characterized gene expression by scRNA-seq in single, freshly isolated adult human cardiomyocytes^7,61.

Cardiomyocytes in the healthy heart

snRNA-seq has emerged as a successful high-throughput transcriptomics approach for profiling adult human cardiomyocytes^27,32,33 and has revealed previously unknown inter-compartmental and intra-compartmental cardiomyocyte heterogeneity between cardiac regions of donor hearts^27,33 (Table 1). Although the transcriptional diversity between atrial and ventricular cardiomyocytes probably reflects different developmental origins and electromechanical stimulations, distinct genomic signatures of cardiomyocyte subpopulations within anatomical regions suggest additional functional diversity that might correspond to specific tissue microenvironments. Such subpopulations include cardiomyocytes enriched for retinoic acid-responsive genes and stress-response-related genes, as well as cardiomyocytes enriched for nuclear-encoded mitochondrial genes indicative of a high energetic state, which suggests that these cardiomyocytes are equipped for a higher workload²⁷. Interestingly, these cardiomyocyte states have been found in both atrial and ventricular cardiomyocytes²⁷. However, whether their localization is enriched in areas of increased wall stress within the ventricle or atrium and how they vary under different pathological conditions, such as mitral or tricuspid valve disease, remains unknown. Likewise, the differences between the transcriptional signatures specific to the cardiomyocytes of the spirally oriented ventricular muscle fibres that constrict the chamber and those of the trabecular cardiomyocytes remain unclear⁹⁹. A novel cardiomyocyte state enriched for Myoz2, which encodes the calcineurin inhibitor myozenin 2, was localized just below the epicardial surface of the mouse heart²⁸, supporting the hypothesis that specific cardiomyocyte states localize to defined microenvironments.

Table 1 Cardiomyocyte gene signatures identified with omics technology

Prenatal cardiomyocyte development

Analysis of human fetal hearts at 5–25 weeks of gestation has led to the identification of cell clusters composed of trabecular, ventricular and atrial compact myocardium¹⁰⁰. Key transcription factors, such as HAND2 and NR2F1 (encoding heart and neural crest derivative-expressed protein 2 and COUP transcription factor 1, respectively), were identified in atrial cardiomyocytes, whereas HAND1 and HEY2 (encoding hairy/enhancer-of-split related with YRPW motif protein 2) were found in ventricular muscle cells. At 5 weeks, proliferating cardiomyocytes were observed, as well as muscle cells with key left-side specification genes (IRX3, encoding iroquois-class homeodomain protein IRX 3 in ventricular compartments, and PITX2 encoding pituitary homeobox 2 in atrial compartments)¹⁰⁰. At 19–22 weeks, a population of proliferating cardiomyocytes showed high expression of TOP2A (encoding DNA topoisomerase 2α) and MKI67 (encoding proliferation marker protein Ki-67), as well as a lack of expression of TCAP (encoding telethonin, a protein needed for the assembly of mature myofibrils)¹⁰¹. An understanding of the gene expression signatures present in proliferating cells is key to identifying cardiomyocytes with a predisposition to divide and to define potential therapeutic strategies for heart regeneration. In this regard, a comparison of gene expression signatures across different time points in prenatal and postnatal periods is needed.

Of note, a comparison of the cardiac cellular landscape in human versus that in mouse fetal hearts suggested differences in gene expression signatures between the two species^{23,100,102,103}. Although cardiomyocytes were the most similar among all cell types, ventricular cardiomyocytes in humans expressed ECM-encoding genes earlier and more abundantly than those in mice. RELN (encoding reelin) expression was specific to human atrial trabecular cardiomyocytes and was absent in mice^{23,100,102,103}. Furthermore, CITED2 (encoding Cbp/p300-interacting transactivator 2) and CITED4 were enriched in developing human cardiomyocytes, whereas Cited1 was enriched in the mouse counterpart^{23,100,102,103} (Fig. 3).

Fig. 3: Single-cell and single-nucleus analysis revealed previously unknown complexities within fibroblasts.

Fibroblasts are key in homeostasis as well as in disease progression, given that fibrosis is a common feature of the response to injury. Single-cell and single-nucleus RNA sequencing have provided insights into the role of fibroblasts in the ageing heart and in different disease settings, such as myocardial infarction (MI) and dilated cardiomyopathy (DCM). A list of fibroblast populations annotated using single-cell and single-nucleus studies is shown in Table 2. COVID-19, coronavirus disease 2019; TAC, transverse aortic constriction.

Cardiomyocytes in disease

Myocardial infarction

In a mouse model of myocardial infarction (MI), scRNA-seq identified a cardiomyocyte subset with upregulated expression of β2 microglobulin in response to ischaemic damage²⁹. In vitro experiments suggested that this upregulation drives fibroblast activation. Furthermore, a multimodal approach combining snRNA-seq, snATAC-seq and spatial transcriptomics was used to study cardiac samples from hearts explanted 2–5 days after the onset of clinical symptoms of MI, before the patients received a total artificial heart¹⁰⁴. This integrated method allowed the investigators to map putative enhancers controlling gene expression within distinct cardiomyocyte niches of injury, repair and remodelling, revealing a regional influx of immune cells in response to localized induction of cytokines, and defined RUNX1 (encoding runt-related transcription factor 1) as a potential driver of myofibroblast differentiation¹⁰⁴. The inclusion of spatial transcriptomics helped to determine localized differences at the site of ischaemic injury. ANKRD1 (encoding the transcriptional repressor ankyrin repeat domain 1, the overexpression of which has been shown to impair cardiomyocyte function¹⁰⁵) and NPPB (encoding natriuretic peptide B, which is widely used as a marker for cardiac disease and is upregulated in the border zone after MI in mice^106,107) were defined as markers for two niches of stressed cardiomyocytes¹⁰⁴. Importantly, integration with snATAC-seq facilitated the identification of T-box protein 3 (TBX3) and myocyte-specific enhancer factor 2D as regulators of an ANKRD1⁺NPPB⁻ pre-stressed cell state, and cyclic AMP-dependent transcription factor ATF3 as a driver of the ANKRD1⁺NPPB⁺ state¹⁰⁴.

Cardiac hypertrophy

A multimodal approach including epigenetic, morphological and functional assessment and co-expression network analysis of mouse and human single cardiomyocyte transcriptomes has facilitated the delineation of conserved mechanisms of pressure overload responses⁷. In early hypertrophy, cardiomyocytes were found to activate mitochondrial translation and oxidative phosphorylation, which correlated with morphological hypertrophy⁷. Subsequently, sustained overload induced p53 activation, leading to the disruption of adaptive hypertrophy transcriptional programmes and the promotion of HF⁷. This observation confirms that cardiomyocyte identity and morphological and functional phenotypes are encoded in transcriptional programmes.

A snRNA-seq study on pathological cardiac hypertrophy caused by aortic stenosis highlighted the downregulation of ephrins, the largest family of receptor tyrosine kinases, in cardiomyocyte hypertrophy⁴⁶. In particular, a downregulation of EPHB1 (encoding ephrin type B receptor 1) observed in the human hypertrophic heart was confirmed in a transverse aortic constriction (TAC) mouse model of pressure overload⁴⁶. In vitro treatment of cells with EFNB2, the ligand for EPHB1, rescued the hypertrophic phenotype⁴⁶. A TAC mouse model was used to mimic the progression towards pathological hypertrophy and led to the identification of early metabolic cardiomyocyte adaptation, which was evidenced by an upregulation in the expression of glycolysis-related genes, a continuous increase in the expression of the hypertrophy-related genes Nppa and Nppb, and a decrease in calcium handling-related genes¹⁰⁸. In addition, the number of cardiomyocytes enriched for ERBB4 (encoding receptor tyrosine protein kinase erbB4) and FGF12 (encoding fibroblast growth factor 12) was reduced in hypertrophied human hearts compared with healthy hearts⁴⁶. To advance our understanding of pathological cardiac hypertrophy, the cardiomyocyte profile from patients with cardiac hypertrophy of different aetiologies, such as hypertension and hypertrophic cardiomyopathy, should be assessed. Similarly, the specific cardiomyocyte states present in these diseases and their distribution within the tissue remains largely unexplored.

Cardiomyopathy

A convergence of cardiomyocyte transcriptomic changes is thought to be present in the setting of HF associated with dilated cardiomyopathy (DCM)⁴⁵. An analysis of 61 patients with DCM or arrhythmogenic cardiomyopathy (ACM) with defined genetic mutations showed a substantial proportion (20–40%) of differentially expressed genes, suggesting a degree of divergence even in late-stage failing hearts⁶⁹. Specific expression profiles included a general downregulation of MYH6 (encoding myosin 6) and cell-state-specific upregulation of SH3RF2 (encoding the anti-apoptotic protein E3 ubiquitin protein ligase SH3RF2) in patients with no known pathological gene variant, and an upregulation of FNIP2 (encoding folliculin-interacting protein 2, which is involved in the inhibition of oxidative metabolism) in patients with variants in LMNA (encoding prelamin A/C) or PKP2 (encoding plakophilin 2), but not in patients with variants in TTN (encoding titin) or RBM20 (encoding RNA-binding protein 20)⁶⁹. Different HF aetiologies also lead to different cardiomyocyte transcriptomic signatures, suggesting a divergence of pathogenic mechanisms. Cardiomyocytes from failing hearts due to ischaemic cardiomyopathy (ICM) show dysregulation of different gene ontologies from cardiomyocytes from hearts with DCM⁶². For example, ICM causes cardiomyocyte changes in energy metabolism and protein targeting to the endoplasmic reticulum, whereas DCM causes cardiomyocyte changes in muscle contraction⁶².

Congenital heart disease

Single-cell transcriptomics has also provided new insights into congenital heart disease. A study of hearts from nine paediatric patients with different aetiologies of congenital heart disease and four control individuals found that age had a minimal contribution to the cardiomyocyte transcriptome compared with disease status¹⁰⁹. A disease-specific cell state was identified in patients with congenital heart disease, which was characterized by increased EGF receptor (EGFR) and forkhead box protein O signalling, as well as insulin resistance. These transcriptomic changes were validated by scATAC-seq data that showed an increase in chromatin accessibility in 84–90% of differentially expressed genes. The expression of CORIN (encoding atrial natriuretic peptide-converting enzyme) was also strongly associated with healthy cardiomyocytes, whereas CRIM1 (encoding cysteine-rich motor neuron 1 protein) was enriched in diseased cardiomyocytes. This observation was validated by RNA in situ hybridization, which showed that the CRIM1 to CORIN ratio was higher in cardiomyocytes from patients with congenital heart disease than in those from controls¹⁰⁹.

Cardiac regeneration

The regeneration potential of adult mammalian hearts is very limited and the response after injury is inadequate to reconstitute ventricular cardiomyocyte numbers¹¹⁰. However, an analysis of mouse and human failing and non-failing adult hearts revealed subpopulations of cardiomyocytes with cardiac regeneration potential, underpinned by a capacity to dedifferentiate and upregulate cell cycle regulators after stress³⁴. In a zebrafish model of heart regeneration after cryoinjury, scRNA-seq defined a subset of cardiomyocytes from the border zone that could partially dedifferentiate, shift their metabolism towards glycolysis and proliferate¹¹¹. Neuregulin 1–ErbB2 signalling contributed to the switch to glycolysis and promotion of cell division in mice and zebrafish¹¹¹. Subsequent snRNA-seq studies of mouse hearts during the early postnatal regenerative window identified immature cardiomyocytes that enter the cell cycle after injury but disappear as the heart loses its regenerative capacity⁴⁸. According to findings from gain-of-function experiments, the unique transcriptional signature of these proliferative cardiomyocytes is related to the activity of nuclear transcription factor Y subunit α and serum response factor⁴⁸. In addition, the endoplasmic reticulum membrane sensors NFE2L1 and NFEL2 drive cardiomyocyte protection⁴⁸. The identification of proliferative cardiomyocytes provides novel inroads for future regenerative therapies.

Fibroblasts in health and disease

Cardiac fibroblasts are dynamic components of the heart’s cellular ecosystem that act as lineage progenitors, master conductors of ECM synthesis and remodelling, intercellular signalling hubs and electromechanical transducers¹¹². Fibroblasts are also the central component in cardiac fibrosis, which is observed in most forms of cardiac pathology (Fig. 3a). Although initially protective, unresolved fibrosis leads to chamber stiffening and HF, as well as sudden death due to arrhythmias¹¹². Many anti-fibrotic drugs have not been successful in improving end points in clinical trials, probably owing to factors such as biological differences between rodent and human hearts, and therapeutic target pleiotropy leading to adverse effects. Therefore, the identification of novel therapeutic targets in cardiac fibrosis is a priority^{113,114,115,116,117,118}. Single-cell genomics have begun to unravel diverse quiescent and activated fibroblast populations in adult hearts^{22,24,27,28,119,120,121} (Table 2).

Table 2 Fibroblast populations annotated using single-cell and single-nucleus studies

Human fibroblasts

In the human donor heart cell atlas, in a study with one of the highest number of cells and nuclei assembled to date, five fibroblast populations were identified, including chamber-specific ECM-producing fibroblasts and fibroblasts enriched for cytokine receptor genes such as OSMR (encoding oncostatin-M-specific receptor subunit β) and Il6ST (encoding IL-6 receptor subunit β), which might modulate immune responses²⁷. This cell atlas identified activated fibroblasts that were shown to express pro-fibrotic genes²⁷; however, other studies did not show activated fibroblasts in healthy hearts^33,47. Shifts in the proportion of specific fibroblast populations and their gene signatures in disease states have been observed, including an increase in myofibroblasts enriched for ELN (encoding elastin) in patients with DCM^45,122 or ICM¹²², as well as an increase in lipogenic fibroblasts enriched for DLK1 (encoding protein delta homologue 1) in the RV of patients with DCM and in non-ischaemic areas of the LV in patients with ICM¹²². In addition, a decrease in ‘resting’ fibroblasts (enriched for PLA2G2A, encoding membrane-associated phospholipase A2) was observed in patients with DCM^45,122 or ICM¹²². Activated fibroblasts, which are characterized by the expression of POSTN (encoding periostin) and FAP (encoding fibroblast activation protein), are also consistently increased in HF^45,47,69,122. However, the extent of the increase in activated fibroblasts might depend on patient genotype, given that the increase was less evident in patients with variants in LMNA and RBM20 compared with patients with TTN variants or no known pathogenic variants⁶⁹. In a study of patients with DCM or HCM⁴⁷, the best marker of activated fibroblasts was expression of COL22A1 (encoding collagen α1 (XXII) chain), which was variably detected in patients with DCM or HCM⁴⁷. This finding should be interpreted cautiously, considering that a separate study of patients with DCM or ACM showed genotype-specific upregulation of different collagens, such as COL4A1 and COL4A2, in patients with pathogenic variants in LMNA, TTN or PKP2, and COL4A5 and COL24A1 in patients with no known pathogenic variants⁶⁹. Overall, these findings highlight unexpected nuances in the gene signatures of activated fibroblasts, and a need to further understand the relationship between different fibroblast states and their distribution in cardiovascular diseases. A CRISPR-based knockout screen of several genes expressed in cardiomyopathy-associated fibroblast populations uncovered several new regulators of fibrosis, such as PRELP and JAZF1 (encoding prolargin and juxtaposed with another zinc finger protein 1, respectively)⁴⁷.

Single-cell and single-nucleus transcriptomics studies of developing, healthy, diseased and partially recovered (unloaded) adult human hearts have confirmed the presence of fibroblast cell state heterogeneity and have highlighted additional regulators of fibroblast differentiation^{27,61,100,104,112,123}. Co-localization of myofibroblasts with phagocytic macrophages after MI has also been demonstrated using spatial transcriptomics¹⁰⁴. Finally, increased expression of ACE2 (encoding angiotensin-converting enzyme 2) was observed in the fibroblasts of hearts from patients with COVID-19 compared with healthy controls¹²⁴, in addition to an enrichment in fibroblasts with pro-thrombotic, ECM-producing, ECM-organizing and immune cell-related signatures^{27,124,125,126,127,128,129}.

Mouse fibroblasts

In mouse ventricles, a predominant quiescent fibroblast subtype (PDGFRA⁺SCA1^high fibroblasts, termed F-SH²⁴; also described as progenitor-like state fibroblasts²⁵) exists across all organs^{130,131,132,133,134}, and is enriched in cells with progenitor characteristics such as hypoxic niche and potential for multilineage differentiation, suggesting that these fibroblasts might be a lineage reserve for mobilization after injury^{131,132,133,135,136}. F-Act fibroblasts^24,132, an activated fibroblast population expressing Meox1 (encoding homeobox protein MOX1 (MEOX1), a transcriptional regulator involved in fibroblast activation during cardiac dysfunction¹³⁷) and Cilp (encoding cartilage intermediate layer protein 1), are present at low levels in uninjured hearts, but expand to account for 20–50% of all cardiac fibroblasts by day 3 after MI^24,132, seemingly deriving from more quiescent fibroblasts, or their pre-proliferative and proliferating counterparts²⁴. F-Act fibroblasts lack expression of the α-smooth muscle actin gene Acta2, and thus might correspond more closely to previously characterized ‘proto-myofibroblasts’¹¹². Interestingly, Cilp1-knockout mice have better cardiac function after MI and a reduced number of myofibroblasts than control mice¹³⁸, supporting a key role of Cilp1 in fibroblast activation and differentiation. F-Act fibroblasts resemble the human POSTN⁺FAP⁺ fibroblasts mentioned above, but their functional similarity needs validation¹¹². Another minor activated fibroblast population in mice, F-WntX²⁴, expresses genes encoding secreted antagonists of WNT signalling and other pro-fibrotic regulators, one of which, WIF1, is essential for restricting the response of inflammatory monocytes after MI^24,139.

scRNA-seq has enabled investigators to define fibroblast-dependent cytokine pathways that contribute to the inflammatory phase of MI repair¹⁴⁰. The levels of a fibroblast subpopulation known as injury response cells (IR) peak on day 1 after MI, before decreasing rapidly, and this cell subtype is likely to be involved in the early inflammatory response²⁵. IR cells might be similar to pro-inflammatory fibroblasts, previously described in a cardiac pressure overload model, which recruit Ly6C^high monocytes via NF-κB signalling and expression of genes encoding monocyte chemoattractants such as CC-motif chemokine 2 (CCL2) and CCL5 (ref.¹⁴¹).

The number of myofibroblasts, the most distinctive injury-related fibroblast subtype, peaks at about day 7 after MI in mice²⁵. Myofibroblasts can derive from the F-Act pool and/or other activated populations and as they differentiate show downregulation of stem cell markers and massive upregulation of ECM-related and contraction-related genes^25,142. However, myofibroblast subgroups have different pro-fibrotic or anti-fibrotic regulatory gene signatures, suggesting that fibrosis is self-limiting²⁴. Accordingly, myofibroblasts transition to a deactivated, post-proliferative cell type known as matrifibrocytes¹⁴³, which are associated with osteogenic and chondrogenic ECM signatures and persist within the scar, probably to direct its maintenance and remodelling^25,143. Activated fibroblasts that accumulate late in angiotensin II-induced cardiac hypertrophy also resemble matrifibrocytes¹⁴⁴, suggesting a role for matrifibrocytes in non-ischaemic heart disease, in which perivascular and interstitial fibrosis predominates.

scRNA-seq analyses of other mouse models of heart disease have highlighted the presence of inflammatory, angiogenic and osteogenic fibroblast signatures^{82,121,134,137,145,146}, which are more abundant with advanced age and might compromise fibroblast–endothelial intercellular signalling⁸². In mice, deletion of Hif1a (encoding hypoxia-inducible factor 1α) in fibroblasts leads to excessive fibrosis after MI¹³², whereas conditional deletion of Lats1 and Lats2 (encoding mechanosensitive Hippo pathway-related serine/threonine protein kinases) in fibroblasts leads to their spontaneous transition to myofibroblasts in uninjured hearts, and formation of a pervasive, non-compacted scar after MI via a mechanism involving the transcription factors YAP and TEAD¹²¹. Underlining the context-specific effect of the Hippo pathway, epicardial-specific deletion of Lats1 and Lats2 in mice was embryonically lethal, leading to defective coronary vasculature remodelling and an impairment in the differentiation of epicardial progenitors into cardiac fibroblasts¹²¹. scRNA-seq and scATAC-seq analyses in a mouse model of pressure overload, with and without pharmacological inhibition of epigenetic factors, have demonstrated that activated fibroblasts are capable of reverting to quiescent fibroblasts, and MEOX1 was revealed as a key regulator of fibroblast activation¹³⁷. Taken together, these findings show that the annotation of cardiac fibroblast states and inferences of their function is at present more granular in mouse hearts than in human hearts, and that for several mouse cell states, a human equivalent has not yet been identified.

Vascular cell signatures in health and disease

Cardiac vascular cell compartments include those within coronary arteries, capillaries and the endocardium, and have different developmental origins, structures and functions. Importantly, the myocardial microvasculature is the most abundant and capillary endothelial cells are in close contact with cardiomyocytes¹⁴⁷. Although this microarchitecture infers a close regulatory relationship between endothelial cells and cardiomyocytes, much remains to be understood about the drivers of vascular remodelling in cardiovascular disease.

Endothelial cells

In adult organ donor hearts, the transcriptional signatures of all major cardiac endothelial cell populations have been reported, including capillary, arterial, venous, endocardial and lymphatic endothelial cells²⁷. The highest heterogeneity was observed in capillary endothelial cells, which express RGCC (encoding a regulator of the cell cycle) and AQP1 (encoding aquaporin 1) (Fig. 4a), with several new cell states identified. One of these cell states was characterized by high expression of genes encoding transcription factors (ATF3, FOS and JUN)¹⁴⁸, which could be induced by endoplasmic reticulum stress¹⁴⁹ (as seen in atherosclerosis¹⁵⁰) and other stimuli such as DNA damage¹⁵¹ and oxidative stress¹⁵². A second cell state showed enrichment for cytokine-related genes, such as CX3CL1, CCL2 and Il16, and interferon-related genes, suggesting that these cells regulate the local immune response²⁷. This analysis remains the most in-depth study of endothelial cell populations, particularly for capillary endothelial cells, with other studies being limited by the numbers of cells and nuclei available, and conservative choices of clustering resolution^33,47. Indeed, although lymphatic and endocardial endothelial cells have been identified^33,47, the whole range of vascular endothelial cells have not yet been recapitulated⁴⁷. Some of the inconsistencies in endothelial cell states between studies are related to differences in annotations or markers used. For example, the annotation of a DKK2⁺ population⁴⁷ probably corresponds to arterial endothelial cells⁴⁵. This example again highlights the importance of standardizing annotations to facilitate more accurate comparison across studies.

Fig. 4: Species-specific expression of gene markers in endothelial cells of healthy and diseased hearts.

a, Illustration of the major vascular structures in the human heart, including the endocardium, and summary of key endothelial cell subsets and their gene markers described in the healthy human heart. b, Genes with different expression patterns in endothelial cells of healthy human and mouse hearts. ‘No cell state specificity’ indicates that the gene is enriched in several cell states without any clear specificity. c, Genes with differential expression in endothelial cells in diseased and healthy hearts, including human hearts with dilated cardiomyopathy (DCM) and ischaemic cardiomyopathy (ICM), and mouse hearts after myocardial infarction (MI). TGFβ, transforming growth factor-β.

SnRNA-seq analysis of patients with DCM have highlighted differences in the transcriptional signatures of venous, capillary and especially endocardial endothelial cells compared with healthy controls⁴⁵. The transforming growth factor-β (TGFβ) signalling pathway is enriched in both venous and capillary endothelial cells⁴⁵. Differential gene expression analysis identified major changes in the endocardium with upregulation of BMP4 and BMP6 (encoding bone morphogenetic protein 4 (BMP4) and BMP6, respectively) between healthy hearts and DCM hearts and a shift in expression from NRG3 (encoding membrane-bound pro-neuregulin 3 (NRG3)) in healthy hearts to NRG1 in DCM hearts^45,69, which mediates the compensatory response to stress^153,154. In addition, a genotype stratification study of samples of hearts with DCM and ACM demonstrated that BMP and NRG signalling does not change uniformly in all cardiomyopathies, but that genotype-specific and chamber-specific disruptions of these intercellular signalling pathways are driven by the endocardium⁶⁹. Indeed, in the LV, NRG1 and BMP6 are upregulated in the genotypes associated with DCM (variants in LMNA, TTN or RBM20, or no pathogenic variants), but not in ACM (variants in PKP2)⁶⁹. However, patients with variants in PKP2 showed upregulation of NRG1 and BMP6 in the RV, whereas patients with no known pathogenic variants did not⁶⁹. Of note, Nrg1 was described as an endocardial marker in healthy mouse hearts¹⁵⁵, suggesting potential species-specific differences in the role and transcriptional profile of the endocardium (Fig. 4b). By contrast, NRP3 expression is conserved in both the human and mouse endocardium^28,118,122. Finally, a potentially angiogenic capillary endothelial cell population that expressed TMEM163 (encoding transmembrane protein 163) and KIT (encoding mast/stem cell growth factor receptor KIT) was shown to increase in both DCM and HCM, suggesting an increase in the formation of small vessels to compensate for impaired cardiac function⁴⁷.

In mice, endothelial cells showed increased expression of genes encoding cytokines such as CXCL2 and CCL9 on day 3 after MI, and a proliferative signature on day 7 after MI¹⁵⁶ (Fig. 4c), which was accompanied by an increase in cells enriched for interferon signalling¹⁵⁷. Upregulation of Plvap (encoding plasmalemma vesicle-associated protein (PLVAP), a membrane protein that contributes to permeability) and an increase in endothelial fenestration, stomata of caveolae and transendothelial channel formation was also detected¹⁵⁷. In human hearts, increased PLVAP expression was found in venous endothelial cells and the endocardium, and widespread localization of PLVAP was observed in ischaemic hearts, especially in fibrotic areas^27,157. Together, these data suggest a potential phenotypic shift in endothelial cells after injury away from the previously described continuous capillary cardiac endothelium¹⁴⁷. In addition, in a TAC mouse model, scRNA-seq of cadherin 5 lineage-traced cells showed an increase in VEGF, WNT, EGFR and MAPK pathways in a cell state that was enriched for genes related to angiogenesis¹⁵⁸. Of note, in the adult human heart, expression of VWF (encoding von Willebrand factor (vWF)) is detected in most endothelial cell states except lymphatic endothelial cells²⁷. By contrast, in the mouse heart, Vwf is enriched only in endocardial and venous endothelial cells^155,159 (Fig. 4b). Given the role of vWF in haemostasis, inflammation, vascular permeability and angiogenesis, this finding could have important implications in the interpretation of mouse models of disease for clinical translational purposes^160,161. Likewise, single-cell and single-nucleus data identified ACKR1 (encoding a chemokine-scavenging receptor) as a venous endothelial cell marker in the human heart. However, this gene has not been reported in the mouse heart, suggesting a different modulation of chemokine bioavailability and, consequently, leukocyte recruitment across species.

Finally, a single-cell analysis of 4,000 cardiac cells from human fetuses (aged 5–25 weeks) identified four main states of endothelial cells: endocardial, valvular, coronary and vascular¹⁰⁰. Of note, CDH11 (encoding cadherin 11) is broadly expressed in human endocardial cells, whereas in equivalent stages of mouse development, its expression is restricted to the endocardium of the valves^64,100. Expression of RNASE1 (encoding ribonuclease pancreatic, shown to protect endothelial cells during inflammation¹⁶²) was found in human cardiac fetal endothelial cells only, whereas expression of Icam2 (encoding intercellular adhesion molecule 2, which is involved in regulating vascular permeability¹⁶³) was specific to the mouse embryo¹⁰⁰ (Fig. 4b). A separate study that analysed 17,000 cardiac from cells human fetuses (aged 19–22 weeks) defined six different endothelial states, including two endocardial populations enriched in NPR3 (encoding natriuretic peptide receptor 3)¹⁰¹. One of these populations co-expressed INHBA (encoding inhibin-βa chain, which is selectively expressed in adult endocardium^27,64,100) and MEIS2 (encoding meis homeobox 2, a transcriptional regulator essential for cardiac neural crest development¹⁶⁴). The expression of MEIS2 might reflect a more immature subset of endocardial cells or a different subset of cells that disappears in adulthood.

Smooth muscle cells and pericytes

SMCs from human adult donor hearts include cells expressing classic gene markers such as MYH11 and ACTA2 and a second population with high expression of CNN1 (encoding calponin 1), which probably represents arterial SMCs^27,165. Pericyte subtypes include typical ABCC9⁺ and KCNJ8⁺ cells, and a population enriched for AGT (encoding angiotensinogen), the expression of which is downregulated in DCM and indicative of dysregulated vasoconstriction^27,45. scRNA-seq analysis of vascular SMCs (VSMCs) defined a synthetic state enriched in coronary artery SMCs compared with aorta and pulmonary artery SMCs¹⁶⁶, which might indicate a specific adaptation of coronary arteries that are constantly exposed to local pressure changes due to cardiac contraction and relaxation. A proliferative VSMC state expressing FABP4 (encoding a fatty acid-binding protein) was identified and the numbers were found to be expanded in atherosclerosis¹⁶⁶. scRNA-seq of atherosclerotic lesions showed that VSMCs undergo phenotypic modulation to transform into unique fibroblast-like cells (fibromyocytes) that contribute to the lesion and fibrous cap in both humans and mice¹⁶⁷. Loss-of-function and gain-of-function validation studies in mouse models showed that TCF21 (encoding transcription factor 21) is essential for this phenotypic modulation and that higher levels of TCF21 are associated with decreased risk of coronary artery disease in humans¹⁶⁷. Interestingly, in a study of human hearts with DCM or ACM, SMCs and pericytes showed an upregulation in the long non-coding RNA ADAMTS9-AS2 and simultaneous downregulation in ADAMTS9 (encoding disintegrin and metalloproteinase with thrombospondin motifs 9, which is involved in ECM remodelling)⁶⁹. In addition, genotype-specific changes were also observed. For example, NOTCH3 (encoding neurogenic locus notch homologue protein 3) was downregulated in the pericytes of patients with pathogenic variants in TTN or PKP2, whereas SLIT3 (encoding slit homologue 3 protein, a ligand that regulates fibroblast activity) was upregulated in ELN⁺ SMCs from patients with variants in PKP2 or LMNA⁶⁹.

The integration of fate mapping with Wnt1–Cre and scRNA-seq helped establish the heterogeneity of cardiac neural crest cell (CNCC)-derived cardiac cell populations from embryonic day (E) 10.5 to postnatal day 7 in mice¹⁶⁸. As expected, on postnatal day 7, most of the CNCCs localized to the aorta, pulmonary arteries and the coronary vasculature. Nine VSMC populations were identified, in addition to microvascular SMC (mVSMC) and pericyte states. An analysis of lineage trajectories (computational inference of differentiation paths) using RNA velocity revealed a transition from pericytes to mVSMCs with Notch3, Tbx2, Fosb (encoding protein FoxB) and Klf2 (encoding Krueppel-like factor 2 (KLF2)) as potential key regulators^168,169,170. Importantly, when analysing potential differentiation paths, the inferred latent time based on RNA velocity was similar to the developmental time. The identification of several differentiation paths included one rooted on a Crabp1⁺ (encoding cellular retinoic acid-binding protein 1) cell population, branching into both mesenchymal cells and VSMCs¹⁶⁸, and highlighting previously unknown lineage relationships.

Dynamic adaptations of immune cells

Previous studies have shown that healthy adult human and mouse hearts contain major immune cell populations of both lymphocytes and myeloid cells^27,171. Resident immune cells, such as cardiac macrophages, are interspersed across the heart between cardiomyocytes, the epicardium, the endocardium, valves and the nodes, where they contribute to organ homeostasis.

Myeloid cells

Myeloid cells in the healthy adult human heart comprise monocytes, dendritic cells and several subtypes of macrophages, such as LYVE1⁺ macrophages (which include MHC-II^high and MHC-II^low populations), MHC-II⁺TREM2⁺ macrophages, fibroblast-interacting macrophages and monocyte-derived macrophages²⁷. In addition, comparisons with skeletal muscle and kidneys revealed that cardiac myeloid cells, including LYVE1⁺ macrophages, have cardiac-specific features, which might be attributable to the tissue-specific adaptability of myeloid cells^27,31,172. scRNA-seq findings in mice showed that cardiac macrophages are enriched for ion channels and facilitate electrical conduction through the distal atrioventricular node¹⁷³.

Several cardiac monocyte or macrophage populations have been described, with different origins, mechanisms of self-replenishment and distinct functions in the context of inflammation, fibrosis and tissue repair^174,175,176. scRNA-seq studies based on mouse models of MI defined a gene expression signature to discriminate a self-renewing macrophage population that originated partly from monocytes (Ccr2⁻MHCII^high), in addition to the previously described monocyte-derived macrophages (Ccr2⁺MHC-II^high) and self-renewing tissue-resident macrophages (Ccr2⁻MHC-II^lowLyve1⁺Timd4⁺)^177,178. Further analysis revealed a homeostatic and reparative function of tissue-resident macrophages, whereas Ccr2⁺ macrophages are enriched for classic inflammatory pathways¹⁷⁸. In mouse models, monocytes and monocyte-derived macrophages present at the early stages after MI were found to be recruited via pro-inflammatory circuits^24,177. Loss-of-function and scRNA-seq studies demonstrated a crucial homeostatic and anti-inflammatory role of tissue-resident macrophages, whereby depletion of these macrophages increased monocyte recruitment and worsened ventricular dysfunction after cardiac injury and hypertension^177,178,179. In addition, a putative self-renewing population of tissue-resident macrophages (MHC-II^lowLyve1⁺) was shown to contribute to the formation of the lymphatic network in a pressure-overload HF model and during heart development in mice^180,181.

scRNA-seq also revealed immune biphasic cell recruitment in mice after MI²⁴. Diffusion map analysis showed a trajectory from early infiltrating monocyte-derived macrophages to inflammatory macrophages, which led to a later peak of tissue-resident reparative macrophages with a pro-regenerative upregulation of Igf1 (encoding insulin-like growth factor 1)²⁴. Combining snRNA-seq and spatial transcriptomics to study samples from patients with MI-induced cardiogenic shock helped to uncover an accumulation of CCR2⁺ macrophages in the injury zone, which predicted the localization of specific fibroblasts and TGFβ2 pathway activity¹⁰⁴. Specifically, SPP1⁺ (encoding osteopontin) macrophages were enriched in ischaemic tissue samples and colocalized with myofibroblasts, whereas CCL18⁺ macrophages were enriched in fibrotic tissue samples. Finally, in a TAC mouse model, activation of pro-inflammatory macrophages was identified as a key event during the transition from normal to reduced cardiac function¹⁰⁸. At 5 weeks, macrophages showed upregulation of several chemokines of the CCL subclass, and treating cardiomyocytes in vitro with the conditioned medium of macrophages from the 5-week post-MI samples led to an increase in Nppa and Nppb expression, typical stress-related genes that are upregulated in HF.

Lymphoid cells

Lymphocytes comprise various subtypes in the healthy adult human heart, including T cell subsets, natural killer (NK) cells and B cells²⁷. CD4⁺ T cells can be classified into naive, effector and regulatory CD4⁺ T cells, whereas CD8⁺ T cells include a population of cytotoxic cells characterized by high expression of granzymes and perforin. Cell-mediated cytotoxicity and CD4⁺ effector T cells have been implicated in the pathogenesis of cardiomyopathy and myocarditis^182,183,184. In a snRNA-seq study of HCM and DCM hearts, an increase in lymphocytes expressing LINGO2 (encoding leucine-rich repeat and immunoglobulin-like domain-containing nogo receptor-interacting protein 2) and several known NK cell markers was identified, but their function remains unclear⁴⁷. NK cells have been shown to have a protective effect against cardiac fibrosis¹⁸⁵, suggesting a potential immune-mediated protective mechanisms during HF. To summarize, enrichment of immune cells before scRNA-seq analysis will help to capture these highly diverse and dynamic cells and improve our understanding of their role in homeostasis and disease.

Functionally important rare cell types

Conduction system and neuronal cells

Neuronal cells represent approximately 1% of human cardiac nuclei and are defined by expression of NRXN1 (encoding neurexin 1) which is involved in the formation of synaptic contacts²⁷. Given that these cells also express PLP1 (encoding myelin proteolipid protein, a well-known Schwann cell marker), they might plausibly include glial cells or represent doublets of neurons and glial cells^27,186. In mice, scRNA-seq identified neuronal-like cells based on the expression of Kcna1 (encoding potassium voltage-gated channel subfamily A member 1) which is involved in regulating nerve signalling²⁵. An analysis of 3,961 human cardiac neuronal cells revealed that 80% of these cells are prototypic neurons²⁷. Furthermore, a cell state enriched for the WNT signalling receptor gene LGR5 (encoding leucine-rich repeat-containing G protein coupled receptor 5) and genes encoding myelin constituents was identified as potential Schwann cells²⁷.

To define the transcriptional signature of the conduction system, specific anatomical regions should be profiled. A multi-species study involving mice, rabbits and cynomolgus monkeys characterized the cells of the mammalian sinoatrial node¹⁸⁷. Vsnl1 (encoding the calcium-sensing protein visinin-like protein 1) was defined as a core marker of pacemaker cells in the three species studied. Indeed, disruption of this gene led to reduced beating rates in human induced pluripotent stem cell-derived cardiomyocytes (hiPSC-CMs) and decreased heart rate in mice¹⁸⁷. Of note, Kcnj8 was highly expressed in pacemaker cells of cynomolgus monkeys, but not in mice or rabbits. Moreover, analysis of the sinoatrial node from mouse hearts at E16.5 showed that 25% of genes previously reported to be sinoatrial node-specific by bulk RNA sequencing were in fact expressed in cells other than pacemaker cells¹⁸⁸, confirming the value of single-cell approaches¹⁸⁸. Studying the human conduction system at the single-cell level is a crucial next step that is currently hampered by limited availability of tissue.

Adipose cells

Adipocytes identified by snRNA-seq analysis using classic markers such as ADIPOQ (encoding an adipokine involved in the control of fat metabolism and insulin sensitivity) and GPAM (encoding mitochondrial glycerol-3-phosphate acyltransferase 1) represent 0.2–0.5% of the droplets obtained from human hearts²⁷. A potential fibrogenic adipocyte precursor expressing ECM-related genes, as well as a cytokine-enriched population were reported; however, the distinction between white and brown cardiac adipocytes cannot yet be made²⁷. A disease-specific population of adipocytes identified in DCM and ACM samples is characterized by changes in the expression of genes related to fatty acid metabolism and is enriched in patients with pathogenic variants in PKP2, LMNA or RBM20 (ref.⁶⁹). Given the potential role of adipocytes in diseases such as metabolic syndrome and arrhythmia, efforts should be intensified to study this cell population.

Epicardial cells

Epicardial cells are identifiable by their expression of known markers such as MSLN and WT1 (encoding mesothelin and Wilms tumour protein, respectively²⁷), and are one of the rarest cell populations in the human heart, thus necessitating enrichment for a full exploration of its complexity. Using a Wt1 lineage tracing system in a mouse model of MI, two Wt1⁺ epicardial cell populations were identified, one of which was enriched for Msln and keratin-encoding genes, whereas the other was characterized by a proliferation-associated gene signature⁴³. Interestingly a Cd44⁺Wt1⁻ epicardial population was also described, suggesting a heterogeneity that goes beyond Wt1⁺ cells⁴³. Furthermore, in a mouse model of neonatal heart regeneration, a combination of scRNA-seq and scATAC-seq analyses revealed an increase in epicardial cells after MI that was accompanied by increased transcriptional activation¹⁵⁵. Transcription factor motifs were preferentially enriched in cis-regulatory elements of epicardial MI-induced genes, including the KLF14 motif specific to the regenerative phase and FOS-related or JUN-related motifs enriched in the non-regenerative phase¹⁵⁵.

Using scRNA-seq, it was possible to assess the cellular heterogeneity of the cardiac crescent in E7.5–E8.0 mouse embryos⁶⁶. Findings from this study led to the identification of a new cardiac field, denoted as the juxtacardiac field, which is characterized by the expression of Mab21l2. Mab21l2 fate mapping showed that the juxtacardiac field contains a pool of progenitor cells for cardiomyocytes and epicardial cells⁶⁶. Although enrichment of rare cell types can be easily achieved in mouse models using lineage tracing tools, in human hearts, it is necessary to select specific anatomical microdomains or identify novel surface markers for prospective sorting.

Cellular networks

Intercellular communication in the heart contributes to development, maintenance of homeostasis and disease progression. Novel computational tools facilitate the mapping of cell states identified using single-cell and single-nucleus transcriptomics integrated with spatial transcriptomics, allowing the unbiased discovery of cellular microenvironments and the prioritization of cell–cell interactions based on niches containing specific cell types or states^90,189 (Fig. 5).

Fig. 5: Integrating scRNA-seq and snRNA-seq data with spatial transcriptomics analysis to define cellular microenvironments and their dynamic intercellular signalling networks.

Single-cell RNA sequencing (scRNA-seq) and single-nucleus RNA sequencing (snRNA-seq) allow the definition of cell states and the prediction of cellular interactions based on the expression of genes encoding ligands and receptors. However, these data do not provide information about cellular proximity. Spatial transcriptomics allows the mapping of gene expression within tissue sections across microtiles that represent microbulk transcriptomics data. By mapping scRNA-seq and snRNA-seq data onto spatial data using computational tools such as cell2location and Giotto, it is possible to predict the localization of cell types and states within microtiles and microanatomical niches. The cellular proximity within the defined cellular niches facilitates inferences on the most probable cellular interactions and the identification of putative functional microenvironments.

In the adult human heart, CellPhoneDB, a public repository of ligand receptors and their interactions, was used to predict intercellular communications within NOTCH pathways between vascular endothelial cells and VSMCs, uncovering venous-specific and arterial-specific interactions²⁷. Furthermore, interactions between fibroblasts and macrophages were predicted, including one that depended on the macrophage migration inhibitory factor (MIF)–CD74 receptor–ligand interaction, which was confirmed by co-expression in spatial transcriptomics²⁷. A 2022 study of pressure overload-induced hypertrophy used CellPhoneDB and CellChat to investigate changes in signalling between cardiomyocytes and other cell types⁴⁶. A downregulation in the expression of EPHB1 and EFNB2 was observed in cardiomyocytes and endothelial cells, respectively. A validation study using rat cardiomyocytes in 2D cultures and hiPSC-CM organoids with gain-of-function or loss-of-function approaches confirmed that EFNB2–EPHB1 signalling in cardiomyocytes regulates hypertrophy, contraction and stress responses, implicating EPHB1 as a putative therapeutic target in HCM⁴⁶. In a large study of patients with DCM and ACM with different pathogenic variants, an analysis of cell–cell interactions was crucial in identifying genotype and chamber-dependent signalling pathways involved in disease pathogenesis⁶⁹. For example, endothelin signalling originating from the endocardium was increased in the LV of patients with pathogenic variants in LMNA and in the RV of patients with pathogenic variants in PKP2, but not in patients with other genotypes. Of note, the dynamics of some pathways are cell type-dependent. For example, in the LV of patients with pathogenic variants in LMNA, the predicted BMP signalling from SMCs and pericytes to cardiomyocytes is decreased, whereas signalling from endothelial cells to cardiomyocytes is increased⁶⁹.

In the adult mouse heart, analysis of non-cardiomyocyte cell populations identified fibroblasts as a crucial intercellular communication hub²². After MI, among the non-cardiomyocyte populations, myofibroblasts have the highest number of differentially expressed ligands (consisting mostly of ECM-related genes^24,156), followed by macrophages²⁴. Unexpectedly, endothelial cells express the highest number of differentially expressed receptors, suggesting a role as downstream effectors of paracrine and juxtacrine mechanisms after injury^24,156. In neonatal mouse hearts, the epicardium was defined as a source of paracrine signalling during both regenerating and non-regenerating phases¹⁵⁵. As confirmed by in vitro experiments, epicardial induction of Rspo1 (encoding R-spondin 1, a potent activator of the WNT–β-catenin signalling pathway) is a likely driver of pro-angiogenic signalling. Moreover, given the epicardial upregulation of Ltbp3 (encoding latent-transforming growth factor-β-binding protein 3 (LTBP3)) and the increased proportion of fibroblasts in the S-phase after treatment with recombinant LTBP3 in vitro, epicardium-derived signalling was suggested to contribute to fibroblast proliferation¹⁵⁵. Finally, in a TAC mouse model, loss of protective signalling from fibroblasts to cardiomyocytes and increased macrophage–cardiomyocyte interactions were suggested to contribute to the initiation and progression of hypertrophy^24,108,156.

During cardiac development in the human fetus (age 5–25 weeks), a cardiomyocyte-to-endothelial cell BMP paracrine signalling pattern was predicted on the basis of increased cardiomyocyte expression of BMP5 and BMP7 and enrichment for BMP receptor genes in endothelial cells, suggesting a potential role for BMP in driving endothelial-to-mesenchymal transition during endocardial cushion formation¹⁰⁰. In mouse embryonic development, a MIF–CXCR2 interaction was predicted in the second heart field of embryos at E8.5 (ref.¹⁹⁰). Inhibition of MIF and CXCR2 using small-molecule compounds impaired the elongation of the outflow tract and RV, probably due to defective cell migration. Taken together, to prioritize validation among hundreds of predicted receptor–ligand pairs, filtering criteria require computational tools, which take into consideration the activation of downstream cascades in the cells receiving the signalling cues^90,189.

Future directions for omics approaches

Transforming diagnosis and therapy

Single-cell, single-nucleus and spatial transcriptomics accompanied by machine learning-based analysis is likely to contribute to the identification of markers for diagnostic and prognostic evaluation of cardiovascular diseases. In particular, these technologies can provide new insights into cell states (including those of rare cells) and cell-type composition of diseased hearts, and define cell-specific theraputic targets^45,46,191. An early scRNA-seq study identified the source of IL-11, a novel anti-fibrotic target for the cardiovascular system¹⁴⁵, demonstrating the potential of single-cell omics approaches in developing innovative therapies. 

Capturing the earliest and most specific therapeutic cellular targets among the numerous maladaptive cardiomyocyte responses driving HF is crucial in hereditary DCM compared with HCM, in which variants in the same culprit gene can lead to opposite phenotypes¹⁹². Although a study of patients with end-stage HCM or DCM showed changes in cellular composition across the two groups, only minimal differences were detected at the transcriptional level⁴⁷, highlighting a need to stratify patients on the basis of aetiology and to study earlier stages of disease progression. Indeed, analysis of genotype-stratified DCM samples identified specific transcriptional changes in different pathogenic variants groups, in addition to  shared gene signatures. The number of differentially expressed genes was sufficient to develop a deep learning method that could predict the genotype on the basis of gene expression signatures of cardiomyocytes, endothelial cells, fibroblasts and myeloid cells⁶⁹. This finding indicates that even at later stages of HF, the underlying cellular landscape and implicitly maladaptive mechanisms differ depending on the underlying cause. To understand the mechanisms involved in progression of DCM to HF, multimodal omics analysis should be performed in both early and late phases of disease, and patients should be stratified according to disease aetiology. Likewise, in ICM, understanding the cellular changes occurring before the onset of HF will be crucial to the stratification of patients in minimizing the current heterogeneity in response to therapy and towards personalized medicine approaches. Taken together, such information can improve our understanding of the pathobiological mechanisms underlying HF and guide the identification of prognostic biomarkers and novel druggable targets for cell-targeted interceptive medicine¹⁹³.

hiPSC-derived cardiovascular cells

HiPSC-derived cardiovascular cells and cardiovascular cells derived from direct reprogramming hold great promise, not only for cardiovascular regeneration but also for drug screening and disease modelling, which is especially relevant for cardiomyocytes, given the lack of reliable cell lines and the challenges in working with large numbers of primary cells for an extended amount of time. To ensure the replicability of hiPSC-CMs as in vitro models, the maturation process needs to be improved and the phenotypic heterogeneity of the cells needs to be functionally evaluated. This latter step is necessary to establish precise models of the cellular subtypes identified in vivo by single-cell and single-nucleus approches. Indeed some of these cell states might be better therapeutic targets than others or even therapeutic products for cell therapy. The integration of single-cell and single-nucleus transcriptomics data from human hearts from all stages of development with those from hiPSC-derived cardiovascular cells contributes to our understanding of the variability in differentiation and maturity occurring in vitro compared with the processes occurring within the cardiac cellular landscape in vivo^{124,194,195,196,197,198}. One such study described the role of the progesterone receptor in driving sex-specific metabolic programmes and maturation of cardiomyocytes¹⁹⁹. Importantly, transcriptomics and epigenomic single-cell analyses have guided improvements in protocols for differentiation and maturation of cardiovascular lineages such as cardiomyocytes and endothelial cells in hiPSC-based 2D and 3D platforms, as well as in direct reprogramming approaches^{38,200,201,202,203}. Furthermore, single-cell studies of hiPSC-derived cardiovascular cells were not only highly informative in reproducing the differentiation steps occurring in cardiac development, but also contributed to the modelling of congenital and other genetic disorders^{194,204,205,206}. Such studies provide a comprehensive overview of the transcriptional changes that occur in specific cells at various stages of differentiation  to facilitate precise mechanistic insights, as seen in studies showing a dosage-dependent effect of TBX5 in different cell states in the setting of congenital heart disease, as well as the convergence of cell cycle dysregualtion and autophagy as pathogenic mechanisms of hypoplastic heart disease^204,207.

Age, sex and comorbidities

As presented throughout this Review, great effort has already been made to understand the complexity of the cardiac cellular landscape. However, some knowledge gaps remain. An urgent need exists for the development of a longitudinal heart atlas from the tissues of infants, children, adolescents and young adults (aged 20–40 years) that recapitulates the substantial morphological and haemodynamic changes occurring from birth until adulthood. Understanding sex-related differences in health and disease is a developing area in cardiology. A single-cell study showed differences in the prevalence of fibroblast subtypes and gene expression patterns between male and female mice^22,144 and between inbred strains²⁵. Furthermore, a study based on limited numbers of donor hearts identified the progesterone receptor as a key mediator of sex-dependent transcriptional programmes during cardiomyocyte maturation¹⁹⁹. Therefore, given the effect of biological variables such as sex and ancestry on cardiovascular risk, it is imperative to integrate datasets from increasing numbers of donors to evaluate the influence of these variables on the human cardiac cellular landscape. Moreover, patients with cardiovascular disease, especially those of advanced age, often have comorbidities such diabetes mellitus, obesity and high blood pressure, which will need to be considered in the analysis and interpretation. The stratification of patients and organ donors will be possible through the integration of data from hundreds of individuals and the collection of appropriate clinical metadata.

Conclusions

Single-cell and single-nucleus omics technologies have provided invaluable novel insights into the cardiac cellular landscape to facilitate a better understanding of disease mechanisms and more tools for accurate risk stratification and precision medicine. However, newly defined cell states not only require thorough validation, but also cooperative and interdisciplinary efforts to standardize definitions and nomenclature within and across species, which will facilitate more accurate comparison between studies and the optimization of experimental designs for disease models. Finally, an urgent need exists to ensure the availability of computational resources and training for the new generation of scientists and clinicians, to allow improved interrogation of the biology and to accelerate the application of these transformative technologies to the study and treatment of cardiac disease.